Revista Facultad de Ingeniería, Universidad de Antioquia, No.110, pp. 31-47, Jan-Mar 2024
Evaluating the IMERG precipitation satellite
product to derive intensity-duration-frequency
curves in Colombia
Evaluación del producto satelital de precipitación IMERG para derivar curvas
intensidad-duración-frecuencia en Colombia
Erasmo Rodríguez 1, Camila García-Echeverri 1, Ana González 1, John Sandoval 1, Manuel
Patarroyo-González 1, Daniela Estefanía Agudelo-Duque 1
1Departamento de Ingeniería Civil y Agrícola, Facultad de Ingeniería, Universidad Nacional de Colombia. Carrera 45 #
26-85. C. P 111321. Bogotá, Colombia.
CITE THIS ARTICLE AS:
E. Rodríguez, C.
García-Echeverri, A. González,
J. Sandoval, M.
Patarroyo-González, D. E.
Agudelo-Duque, ”Evaluating
the IMERG precipitation
satellite product to derive
intensity-duration-frequency
curves in Colombia”, Revista
Facultad de Ingeniería
Universidad de Antioquia, no.
110, pp. 31-47, Jan-Mar, 2024
[Online]. Available: https:
//www.doi.org/10.17533/
udea.redin.20230212
ARTICLE INFO:
Received: April 12, 2021
Accepted: February 06, 2023
Available online: February 07,
2023
KEYWORDS:
Precipitation; Remote Sensing;
Colombia
Precipitación; Sensoramiento
Remoto; Colombia
ABSTRACT: This article explores the potentialities of the IMERG V06B FINAL product
for estimating Intensity-Duration-Frequency curves in Colombia, using the in-situ data
available for 110 rain gauges. From observed data for 76 of these stations, we validated
the satellite IMERG precipitation data for the period 2001-2019, at daily, monthly, and
annual resolutions. For 60 stations, better results were obtained for the monthly time
aggregation, followed by the yearly and daily scales, suggesting that seasonality is the
main rainfall characteristic captured by the product. Concerning the occurrence of daily
precipitation, results indicate that both the probability of detection and the probability of
false detection are high. In general terms, the comparison between intensities from
existing IDF curves and those derived from IMERG showed underestimations of the
rainfall intensities for the short durations studied (0.5 and 1 h), with mean relative errors
in the range [-69%,+56%], and overestimations for the large durations of 2 and 6 h, with
mean relative errors in the range [-61%,+171%]. Results also suggest that the IMERG
product at this moment is not able to capture the daily rainfall distribution in most of the
stations. Nevertheless, for almost 20% of the rain gauges, located mainly in the Amazon,
Orinoco, and Pacific Regions, the analysis showed that the maximum intensities derived
from IMERG are within +/-25% relative error, compared with the ones calculated using
the traditional approach.
RESUMEN: El artículo investiga las potencialidades del producto de precipitación satelital
IMERG V06B FINAL para estimar intensidades de curvas intensidad-duración-frecuencia
en Colombia, a partir de información de 110 estaciones. Utilizando datos para 76 de
estas estaciones en el período 2001-2019, se validó el producto IMERG a nivel diario,
mensual y anual. En 60 de estas estaciones se ha encontrado que el producto obtiene
mejores resultados en la escala mensual, seguida de las escalas anual y diaria, lo cual
indica que la estacionalidad es la principal característica de la lluvia adecuadamente
capturada por el producto. En general, la comparación entre curvas IDF existentes
y las derivadas a partir de IMERG mostró subestimaciones para las intensidades de
precipitación correspondientes a las duraciones más cortas (0.5 y 1 h), con errores
relativos medios en el rango [-69%, +56%], y sobreestimaciones para las duraciones
más grandes de 2 y 6 h, con errores medios relativos en el rango [-61%, +171%]. Los
resultados obtenidos sugieren que el producto IMERG en la actualidad no es capaz de
capturar la distribución de la lluvia diaria en la mayoría de las estaciones. A pesar de
esto, para cerca del 20% de las estaciones, ubicadas principalmente en las regiones
Orinoquía, Amazonía y Pacífica, los análisis muestran que las intensidades
31
* Corresponding author: Erasmo Rodríguez
E-mail: earodriguezs@unal.edu.co
ISSN 0120-6230
e-ISSN 2422-2844
DOI: 10.17533/udea.redin.20230212 31
Evaluating the IMERG precipitation satellite
product to derive intensity-duration-frequency
curves in Colombia
Evaluación del producto satelital de precipitación IMERG para derivar curvas
intensidad-duración-frecuencia en Colombia
Erasmo Rodríguez 1, Camila García-Echeverri 1, Ana González 1, John Sandoval 1, Manuel
Patarroyo-González 1, Daniela Estefanía Agudelo-Duque 1
1Departamento de Ingeniería Civil y Agrícola, Facultad de Ingeniería, Universidad Nacional de Colombia. Carrera 45 #
26-85. C. P 111321. Bogotá, Colombia.
CITE THIS ARTICLE AS:
E. Rodríguez, C.
García-Echeverri, A. González,
J. Sandoval, M.
Patarroyo-González, D. E.
Agudelo-Duque, ”Evaluating
the IMERG precipitation
satellite product to derive
intensity-duration-frequency
curves in Colombia”, Revista
Facultad de Ingeniería
Universidad de Antioquia, no.
110, pp. 31-47, Jan-Mar, 2024
[Online]. Available: https:
//www.doi.org/10.17533/
udea.redin.20230212
ARTICLE INFO:
Received: April 12, 2021
Accepted: February 06, 2023
Available online: February 07,
2023
KEYWORDS:
Precipitation; Remote Sensing;
Colombia
Precipitación; Sensoramiento
Remoto; Colombia
ABSTRACT: This article explores the potentialities of the IMERG V06B FINAL product
for estimating Intensity-Duration-Frequency curves in Colombia, using the in-situ data
available for 110 rain gauges. From observed data for 76 of these stations, we validated
the satellite IMERG precipitation data for the period 2001-2019, at daily, monthly, and
annual resolutions. For 60 stations, better results were obtained for the monthly time
aggregation, followed by the yearly and daily scales, suggesting that seasonality is the
main rainfall characteristic captured by the product. Concerning the occurrence of daily
precipitation, results indicate that both the probability of detection and the probability of
false detection are high. In general terms, the comparison between intensities from
existing IDF curves and those derived from IMERG showed underestimations of the
rainfall intensities for the short durations studied (0.5 and 1 h), with mean relative errors
in the range [-69%,+56%], and overestimations for the large durations of 2 and 6 h, with
mean relative errors in the range [-61%,+171%]. Results also suggest that the IMERG
product at this moment is not able to capture the daily rainfall distribution in most of the
stations. Nevertheless, for almost 20% of the rain gauges, located mainly in the Amazon,
Orinoco, and Pacific Regions, the analysis showed that the maximum intensities derived
from IMERG are within +/-25% relative error, compared with the ones calculated using
the traditional approach.
RESUMEN: El artículo investiga las potencialidades del producto de precipitación satelital
IMERG V06B FINAL para estimar intensidades de curvas intensidad-duración-frecuencia
en Colombia, a partir de información de 110 estaciones. Utilizando datos para 76 de
estas estaciones en el período 2001-2019, se validó el producto IMERG a nivel diario,
mensual y anual. En 60 de estas estaciones se ha encontrado que el producto obtiene
mejores resultados en la escala mensual, seguida de las escalas anual y diaria, lo cual
indica que la estacionalidad es la principal característica de la lluvia adecuadamente
capturada por el producto. En general, la comparación entre curvas IDF existentes
y las derivadas a partir de IMERG mostró subestimaciones para las intensidades de
precipitación correspondientes a las duraciones más cortas (0.5 y 1 h), con errores
relativos medios en el rango [-69%, +56%], y sobreestimaciones para las duraciones
más grandes de 2 y 6 h, con errores medios relativos en el rango [-61%, +171%]. Los
resultados obtenidos sugieren que el producto IMERG en la actualidad no es capaz de
capturar la distribución de la lluvia diaria en la mayoría de las estaciones. A pesar de
esto, para cerca del 20% de las estaciones, ubicadas principalmente en las regiones
Orinoquía, Amazonía y Pacífica, los análisis muestran que las intensidades
31
* Corresponding author: Erasmo Rodríguez
E-mail: earodriguezs@unal.edu.co
ISSN 0120-6230
e-ISSN 2422-2844
DOI: 10.17533/udea.redin.20230212 31
E. Rodríguez et al., Revista Facultad de Ingeniería, Universidad de Antioquia, No. 110, pp. 31-47, 2024
máximas derivadas de IMERG se encuentran dentro
de +/-25% de error relativo, en comparación con las
deducidas utilizando el método tradicional.
1. Introduction
Intensity-Duration-Frequency (IDF) curves summarize the
extreme rainfall characteristics for short time durations
at a point in space [1]. They are a fundamental input for
hydrological design as they provide data for estimating
flows in small ungauged basins and for constructing
design storms in medium and large-size watersheds
[2]; they are also a basis for flood risk assessments in
mountainous areas [3] and for studying rainfall triggering
landslides [4, 5] among others. They are calculated
from rainfall data using either manual, semi-automatic
or automatic procedures that extract information from
pluviograph strips or digital records [6]. The traditional
parametric method for constructing IDF curves is based
on a statistical analysis of annual maximum series [7, 8].
The method consists of calculating annual maximum
intensities for different durations (usually 15, 30, 60,
120, and 360 minutes), followed by the fitting of a
probability distribution that usually corresponds to one
from the GEV (Generalized Extreme Value) family [9, 10].
This distribution is used to calculate intensities for
selected return periods (usually 2, 5, 10, 25, 50, and
100 years) [11–13]. When pluviograph information is
not available, as usual in several parts of the world,
synthetic or generalized IDF curves can be estimated
from rainfall information using regionalization, scaling,
or other methods based on rainfall characteristics [14–17].
Recent products that derive rainfall intensity information
from meteorological satellite observations complement
the in-situ data and are beginning to be used for several
purposes, including, for example, the validation and
intercomparison of satellite rainfall estimates [18], the
analysis of flash floods [19, 20], hydrological modeling
[21–25], downscaling of rainfall extremes [26], and
the estimation of IDF curves [27–30]. Among these
products is the Integrated MultisatellitE Retrievals from
GPM-IMERG (or in short IMERG) produced by NASA,
a quasi-global dataset (60°N - 60°S), calibrated using
monthly rain gauge precipitation values, which provides
rainfall intensities at 0.5 h intervals, with a spatial
resolution of 0.1° (approximately 11 km at the equator).
As part of the GPM-era satellites [31], IMERG is the
successor of the former GPM-based rainfall estimates v3,
v4, and v5, which use a multi-satellite fusion algorithm
to generate precipitation data. The latest IMERG V06B
product, additionally incorporates data coming from
the TRMM mission, extending the period with available
data back to 2001. As the GPM mission, TRMM was a
research satellite mission launched in 1997, developed
by NASA and JAXA, that came to an end in 2015. Using
the TRMM multi-satellite precipitation analysis (TMPA),
gridded products at different spatiotemporal scales were
developed by NASA and JAXA, including versions 3B42
(monthly) and 3B43 (daily), both with a resolution of 1°,
which are the TMPA satellite products usually assessed
for hydrometeorological purposes.
We have selected the IMERG V06B product for conducting
this research based on four considerations: i) IMERG
has been applied and validated in a vast number of
regions around the world [32], showing that in several
areas, the product overestimates rainfall in lowlands and
underestimates rainfall at high altitudes. Evaluations of
the IMERG product in South American countries like Brazil
[33–35], Chile [27, 36] Colombia [37], and Ecuador [38],
have shown that the IMERG product correctly represents
the spatial pattern of orographic precipitation in the
tropical Andes; that IMERG captures nicely the spatial and
temporal rainfall characteristics in the rainiest region in
Colombia, and that better performances of the product
are obtained at regional rather than at grid scales ii)
Comparisons of IMERG with other precipitation products
such as TRMM [37], TMPA [38] or GSMaP [39] in regions
of South America suggest that both IMERG and TRMM
perform well in the Choco area in Colombia. Regarding
TMPA, results indicate that IMERG is a better product
than TMPA in detecting and quantifying rainfall in areas
in Ecuador and Peru, especially in the high Andes, and
that IMERG shows a better estimation of in-situ observed
rainfall than TMPA. Concerning GSMaP, evaluations in a
flat area in Brazil indicate that IMERG outperforms GSMaP
at annual and monthly scales, but it is slightly worse at
daily resolution iii (Compared with other satellite rainfall
products used to derive IDF curves, IMERG outperforms
TRMM and GSMaP [40] iv) IMERG represents the state of
the art of freely available homogeneous and consistent
rainfall products with temporal resolutions compatible
with those required when constructing IDF curves.
Various studies, conducted with different versions of
the IMERG products and in different parts of the world,
have evaluated and reported its use. A study in southern
Austria [41] describes the evaluation of the EARLY, LATE
and FINAL IMERG V03 products for 2014. They found
that the rainfall IMERG FINAL estimations outperformed
the EARLY and LATE products and that the IMERG FINAL
half-hourly estimates correspond approximately to 25
minutes of observed accumulations, with an offset of about
+40 minutes. Another study in Iran [42] describes the
evaluation of the IMERG dataset for the period 2014-2017
using data from 370 stations, reporting that the best
performance is obtained with the FINAL product and the
worst one with the EARLY product. On the island of Bali in
Indonesia, a study [43] tested, at multiple time scales, the
32
máximas derivadas de IMERG se encuentran dentro
de +/-25% de error relativo, en comparación con las
deducidas utilizando el método tradicional.
1. Introduction
Intensity-Duration-Frequency (IDF) curves summarize the
extreme rainfall characteristics for short time durations
at a point in space [1]. They are a fundamental input for
hydrological design as they provide data for estimating
flows in small ungauged basins and for constructing
design storms in medium and large-size watersheds
[2]; they are also a basis for flood risk assessments in
mountainous areas [3] and for studying rainfall triggering
landslides [4, 5] among others. They are calculated
from rainfall data using either manual, semi-automatic
or automatic procedures that extract information from
pluviograph strips or digital records [6]. The traditional
parametric method for constructing IDF curves is based
on a statistical analysis of annual maximum series [7, 8].
The method consists of calculating annual maximum
intensities for different durations (usually 15, 30, 60,
120, and 360 minutes), followed by the fitting of a
probability distribution that usually corresponds to one
from the GEV (Generalized Extreme Value) family [9, 10].
This distribution is used to calculate intensities for
selected return periods (usually 2, 5, 10, 25, 50, and
100 years) [11–13]. When pluviograph information is
not available, as usual in several parts of the world,
synthetic or generalized IDF curves can be estimated
from rainfall information using regionalization, scaling,
or other methods based on rainfall characteristics [14–17].
Recent products that derive rainfall intensity information
from meteorological satellite observations complement
the in-situ data and are beginning to be used for several
purposes, including, for example, the validation and
intercomparison of satellite rainfall estimates [18], the
analysis of flash floods [19, 20], hydrological modeling
[21–25], downscaling of rainfall extremes [26], and
the estimation of IDF curves [27–30]. Among these
products is the Integrated MultisatellitE Retrievals from
GPM-IMERG (or in short IMERG) produced by NASA,
a quasi-global dataset (60°N - 60°S), calibrated using
monthly rain gauge precipitation values, which provides
rainfall intensities at 0.5 h intervals, with a spatial
resolution of 0.1° (approximately 11 km at the equator).
As part of the GPM-era satellites [31], IMERG is the
successor of the former GPM-based rainfall estimates v3,
v4, and v5, which use a multi-satellite fusion algorithm
to generate precipitation data. The latest IMERG V06B
product, additionally incorporates data coming from
the TRMM mission, extending the period with available
data back to 2001. As the GPM mission, TRMM was a
research satellite mission launched in 1997, developed
by NASA and JAXA, that came to an end in 2015. Using
the TRMM multi-satellite precipitation analysis (TMPA),
gridded products at different spatiotemporal scales were
developed by NASA and JAXA, including versions 3B42
(monthly) and 3B43 (daily), both with a resolution of 1°,
which are the TMPA satellite products usually assessed
for hydrometeorological purposes.
We have selected the IMERG V06B product for conducting
this research based on four considerations: i) IMERG
has been applied and validated in a vast number of
regions around the world [32], showing that in several
areas, the product overestimates rainfall in lowlands and
underestimates rainfall at high altitudes. Evaluations of
the IMERG product in South American countries like Brazil
[33–35], Chile [27, 36] Colombia [37], and Ecuador [38],
have shown that the IMERG product correctly represents
the spatial pattern of orographic precipitation in the
tropical Andes; that IMERG captures nicely the spatial and
temporal rainfall characteristics in the rainiest region in
Colombia, and that better performances of the product
are obtained at regional rather than at grid scales ii)
Comparisons of IMERG with other precipitation products
such as TRMM [37], TMPA [38] or GSMaP [39] in regions
of South America suggest that both IMERG and TRMM
perform well in the Choco area in Colombia. Regarding
TMPA, results indicate that IMERG is a better product
than TMPA in detecting and quantifying rainfall in areas
in Ecuador and Peru, especially in the high Andes, and
that IMERG shows a better estimation of in-situ observed
rainfall than TMPA. Concerning GSMaP, evaluations in a
flat area in Brazil indicate that IMERG outperforms GSMaP
at annual and monthly scales, but it is slightly worse at
daily resolution iii (Compared with other satellite rainfall
products used to derive IDF curves, IMERG outperforms
TRMM and GSMaP [40] iv) IMERG represents the state of
the art of freely available homogeneous and consistent
rainfall products with temporal resolutions compatible
with those required when constructing IDF curves.
Various studies, conducted with different versions of
the IMERG products and in different parts of the world,
have evaluated and reported its use. A study in southern
Austria [41] describes the evaluation of the EARLY, LATE
and FINAL IMERG V03 products for 2014. They found
that the rainfall IMERG FINAL estimations outperformed
the EARLY and LATE products and that the IMERG FINAL
half-hourly estimates correspond approximately to 25
minutes of observed accumulations, with an offset of about
+40 minutes. Another study in Iran [42] describes the
evaluation of the IMERG dataset for the period 2014-2017
using data from 370 stations, reporting that the best
performance is obtained with the FINAL product and the
worst one with the EARLY product. On the island of Bali in
Indonesia, a study [43] tested, at multiple time scales, the
32
E. Rodríguez et al., Revista Facultad de Ingeniería, Universidad de Antioquia, No. 110, pp. 31-47, 2024
use of the IMERG FINAL product for the period 2014-2019.
They found that, in general, rainfall estimations with
IMERG were lower than observations. Besides this, IMERG
estimations were highly accurate at monthly and seasonal
scales, but less exact at daily time resolution, although
the probability of detection of rainfall events was correct.
In Spain [44] were investigated the effects of orographic
biases of IMERG V06B-FINAL estimates in the Ebro River
basin for the period 2014-2018. Results showed that the
performance of IMERG depends highly on altitude (with
discrepancies increasing with height), on the precipitation
regime, and on the rainfall gauge density considered to
create the observed gridded product, used as ground
validation data.
A comprehensive study in Canada [45] made an evaluation
of the IMERG V06B FINAL product for the period 2014-2018.
They found that over the coastal eastern and western parts
of the country, IMERG tends to overestimate the hourly
precipitation intensities by approximately 25% and that
the discrepancies with ground truth data are larger
in areas with larger rainfall. Overall, the results of
this study showed that IMERG is a good product for
investigating rainfall at high spatiotemporal resolutions.
Best performances occurred in the plain areas, with
lower uncertainties during warm months. Researchers in
Pakistan [46] investigated the performance of the IMERG
products using 62 rain gauge stations and found that on
a daily scale, the FINAL product performs well in terms
of frequency and intensity and that this product captures
relatively well extreme precipitation events. At monthly
time resolution, they found that the IMERG FINAL product
produces good results in plain and medium elevation
regions but has limitations at higher altitudes.
Additional studies have investigated the performance
of IMERG and/or TMPA products in Colombia. In the
western part of Colombia, researchers evaluated the
performance of the IMERG V06B product for the period
2014-2017 over complex terrain in the Chocó Region by
using daily time series of rainfall from 185 stations [37].
The results showed that IMERG represents well the spatial
and temporal variations in the mean daily precipitation in
the study area. Overestimations appeared for rainfall in the
relatively low precipitation and medium-to-high altitude
areas, and underestimations for mean daily precipitation
in areas with very high precipitation and medium-to-low
altitude. A study for the whole country, analyzed the
performance of the TRMM 3B43 V7 precipitation product
for the 1998-2015 period, at a monthly scale, using 1.180
rain gauge stations [47]. Their findings suggest that the
product performs well in the plain areas of the Amazon,
Orinoco, and Caribbean regions. Over the complex relief
of the Andes region, the product tends to overestimate
precipitation, while in the wet Pacific region, precipitation
is largely underestimated. Also, the performance of TRMM
3B43 V7 decreases during wet seasons. Another finding
of this research indicates that the product frequently
misses light rainfall events and less frequent but very
heavy storms, which causes rainfall overestimations in the
Andes region and underestimations in the Pacific region.
Several studies described above have shown that sub-daily
rainfall information is the basis for calculating IDF curves
using a traditional approach [1, 10, 17]. However, the
development of IDF curves continues to be a challenge
in the whole world, especially in countries from the
Global South that have limitations, if not lack of this
type of information [48]. Additionally, in Colombia and
several parts of the world, the number of rainfall stations
providing this information has decreased since the 1980s
[24]. Considering that IDF curves are the main input for the
hydrological design of hydraulic structures and that they
represent an important investment for the governments,
they need to be rigorously constructed and updated
continuously, as they are derived from extreme rainfall
that are non-stationary [10]. Constructing and updating
IDF curves from analog information, like in Colombia,
is a time-consuming and costly process [17]. Although
satellite rainfall data may be used to develop IDF curves,
few studies have reported this. Given all these reasons,
the aim of this study is to contribute to the knowledge of
the use of satellite-based information to construct IDF
curves, bridging some of these gaps [26–30].
Accordingly, in this study, we first validated at annual,
monthly, and daily resolutions the IMERG V06B FINAL
product using ground validation data for 76 rain gauge
stations in Colombia for the period 2001-2019, using a
pixel-point methodology and calculating thirteen different
statistical and contingency metrics. Then, we estimated
and validated the maximum intensities of IDF curves at 110
stations obtained from the IMERG V06B FINAL product,
using the corresponding IDF curves at the same locations,
constructed with the traditional method and in-situ data
[17].
2. Study area
This study uses IDF curves, calculated up to the year
2010 using the traditional approach, from 110 stations in
Colombia. Due to daily rainfall information availability, only
76 out of the 110 rain gauges have been used for evaluating
the IMERG V06B product. Figure 1a shows the spatial
distribution of the stations, which is heterogeneous within
the country, with most of the gauges (70%) located in the
central (Andes), 14% in the northern region (Caribbean),
9% in the eastern Orinoco region, 4% in the Amazon
southeastern region, and 3% in the Pacific western region.
33
use of the IMERG FINAL product for the period 2014-2019.
They found that, in general, rainfall estimations with
IMERG were lower than observations. Besides this, IMERG
estimations were highly accurate at monthly and seasonal
scales, but less exact at daily time resolution, although
the probability of detection of rainfall events was correct.
In Spain [44] were investigated the effects of orographic
biases of IMERG V06B-FINAL estimates in the Ebro River
basin for the period 2014-2018. Results showed that the
performance of IMERG depends highly on altitude (with
discrepancies increasing with height), on the precipitation
regime, and on the rainfall gauge density considered to
create the observed gridded product, used as ground
validation data.
A comprehensive study in Canada [45] made an evaluation
of the IMERG V06B FINAL product for the period 2014-2018.
They found that over the coastal eastern and western parts
of the country, IMERG tends to overestimate the hourly
precipitation intensities by approximately 25% and that
the discrepancies with ground truth data are larger
in areas with larger rainfall. Overall, the results of
this study showed that IMERG is a good product for
investigating rainfall at high spatiotemporal resolutions.
Best performances occurred in the plain areas, with
lower uncertainties during warm months. Researchers in
Pakistan [46] investigated the performance of the IMERG
products using 62 rain gauge stations and found that on
a daily scale, the FINAL product performs well in terms
of frequency and intensity and that this product captures
relatively well extreme precipitation events. At monthly
time resolution, they found that the IMERG FINAL product
produces good results in plain and medium elevation
regions but has limitations at higher altitudes.
Additional studies have investigated the performance
of IMERG and/or TMPA products in Colombia. In the
western part of Colombia, researchers evaluated the
performance of the IMERG V06B product for the period
2014-2017 over complex terrain in the Chocó Region by
using daily time series of rainfall from 185 stations [37].
The results showed that IMERG represents well the spatial
and temporal variations in the mean daily precipitation in
the study area. Overestimations appeared for rainfall in the
relatively low precipitation and medium-to-high altitude
areas, and underestimations for mean daily precipitation
in areas with very high precipitation and medium-to-low
altitude. A study for the whole country, analyzed the
performance of the TRMM 3B43 V7 precipitation product
for the 1998-2015 period, at a monthly scale, using 1.180
rain gauge stations [47]. Their findings suggest that the
product performs well in the plain areas of the Amazon,
Orinoco, and Caribbean regions. Over the complex relief
of the Andes region, the product tends to overestimate
precipitation, while in the wet Pacific region, precipitation
is largely underestimated. Also, the performance of TRMM
3B43 V7 decreases during wet seasons. Another finding
of this research indicates that the product frequently
misses light rainfall events and less frequent but very
heavy storms, which causes rainfall overestimations in the
Andes region and underestimations in the Pacific region.
Several studies described above have shown that sub-daily
rainfall information is the basis for calculating IDF curves
using a traditional approach [1, 10, 17]. However, the
development of IDF curves continues to be a challenge
in the whole world, especially in countries from the
Global South that have limitations, if not lack of this
type of information [48]. Additionally, in Colombia and
several parts of the world, the number of rainfall stations
providing this information has decreased since the 1980s
[24]. Considering that IDF curves are the main input for the
hydrological design of hydraulic structures and that they
represent an important investment for the governments,
they need to be rigorously constructed and updated
continuously, as they are derived from extreme rainfall
that are non-stationary [10]. Constructing and updating
IDF curves from analog information, like in Colombia,
is a time-consuming and costly process [17]. Although
satellite rainfall data may be used to develop IDF curves,
few studies have reported this. Given all these reasons,
the aim of this study is to contribute to the knowledge of
the use of satellite-based information to construct IDF
curves, bridging some of these gaps [26–30].
Accordingly, in this study, we first validated at annual,
monthly, and daily resolutions the IMERG V06B FINAL
product using ground validation data for 76 rain gauge
stations in Colombia for the period 2001-2019, using a
pixel-point methodology and calculating thirteen different
statistical and contingency metrics. Then, we estimated
and validated the maximum intensities of IDF curves at 110
stations obtained from the IMERG V06B FINAL product,
using the corresponding IDF curves at the same locations,
constructed with the traditional method and in-situ data
[17].
2. Study area
This study uses IDF curves, calculated up to the year
2010 using the traditional approach, from 110 stations in
Colombia. Due to daily rainfall information availability, only
76 out of the 110 rain gauges have been used for evaluating
the IMERG V06B product. Figure 1a shows the spatial
distribution of the stations, which is heterogeneous within
the country, with most of the gauges (70%) located in the
central (Andes), 14% in the northern region (Caribbean),
9% in the eastern Orinoco region, 4% in the Amazon
southeastern region, and 3% in the Pacific western region.
33
E. Rodríguez et al., Revista Facultad de Ingeniería, Universidad de Antioquia, No. 110, pp. 31-47, 2024
(a) (b)
Figure 1 (a) Spatial distribution of the rain gauges investigated within the five different regions in Colombia. (b) Mean annual
precipitation for each station
Interactions between several macroscale and mesoscale
meteorological phenomena, the Pacific, Atlantic, and
Caribbean Oceans, the Andes ranges and the Amazon
rainforest result in a complex spatiotemporal precipitation
pattern in Colombia [49]. On the interannual scale,
Colombia’s dominant driver of climate and weather
variability is the El Niño Southern Oscillation (ENSO),
which brings rainfall above/below normal during the La
Niña/El Niño phases [50, 51]. On the annual scale, the
movement of the Intertropical Convergence Zone (ITCZ)
exerts a strong control on the rainfall seasonality in
Colombia, with a bimodal rainfall regime in the Andes
region, and unimodal rainfall regimes in the Caribbean
and Amazon regions [49, 52]. Precipitation in the Pacific
region is mainly associated with a low-level westerly jet
(Chocó jet) that brings rainfall most of the year, following
a unimodal precipitation pattern [51].
In the Orinoco region, rainfall is produced mainly by
the South American low-level jet that brings Mesoscale
Convective Systems (MCS) from the Atlantic Ocean
and the Amazon rainforest, with a rainfall pattern of
high precipitation during the middle part of the year
[49]. Table 1 summarizes the rainfall characteristics
of Colombia’s five major natural regions. The complex
terrain and diverse rainfall climatology make the analysis
of precipitation in the study area challenging with either
point or gridded observed data or satellite products such
as IMERG.
3. Data
3.1 Ground-based observations
Daily time series of precipitation for 76 rain gauge stations
in Colombia, during the period 2001-2019, recorded by the
Instituto de Hidrología, Meteorología y Asuntos Ambientales
(IDEAM) were obtained from the hydrometeorological data
portal DHIME [53]. The list of stations used in this study,
their characteristics, and a summary of the statistical
and contingency criteria calculated are included in the
supplementary material (see supplementary material).
Rainfall intensity Tables and Equations for the IDF curves
for 110 rain gauges were also obtained from IDEAM [17].
3.2 IMERG data
EARLY and LATE (near real-time data) and FINAL
(post-processed data) IMERG products are freely available
online. IMERG EARLY, LATE, and FINAL datasets are
computed every 4 hours, 14 hours, and 3.5 months after
observation time, respectively. EARLY and LATE datasets
are computed an hour at a time, while FINAL is computed
a month at a time. According to the IMERG developers
34
(a) (b)
Figure 1 (a) Spatial distribution of the rain gauges investigated within the five different regions in Colombia. (b) Mean annual
precipitation for each station
Interactions between several macroscale and mesoscale
meteorological phenomena, the Pacific, Atlantic, and
Caribbean Oceans, the Andes ranges and the Amazon
rainforest result in a complex spatiotemporal precipitation
pattern in Colombia [49]. On the interannual scale,
Colombia’s dominant driver of climate and weather
variability is the El Niño Southern Oscillation (ENSO),
which brings rainfall above/below normal during the La
Niña/El Niño phases [50, 51]. On the annual scale, the
movement of the Intertropical Convergence Zone (ITCZ)
exerts a strong control on the rainfall seasonality in
Colombia, with a bimodal rainfall regime in the Andes
region, and unimodal rainfall regimes in the Caribbean
and Amazon regions [49, 52]. Precipitation in the Pacific
region is mainly associated with a low-level westerly jet
(Chocó jet) that brings rainfall most of the year, following
a unimodal precipitation pattern [51].
In the Orinoco region, rainfall is produced mainly by
the South American low-level jet that brings Mesoscale
Convective Systems (MCS) from the Atlantic Ocean
and the Amazon rainforest, with a rainfall pattern of
high precipitation during the middle part of the year
[49]. Table 1 summarizes the rainfall characteristics
of Colombia’s five major natural regions. The complex
terrain and diverse rainfall climatology make the analysis
of precipitation in the study area challenging with either
point or gridded observed data or satellite products such
as IMERG.
3. Data
3.1 Ground-based observations
Daily time series of precipitation for 76 rain gauge stations
in Colombia, during the period 2001-2019, recorded by the
Instituto de Hidrología, Meteorología y Asuntos Ambientales
(IDEAM) were obtained from the hydrometeorological data
portal DHIME [53]. The list of stations used in this study,
their characteristics, and a summary of the statistical
and contingency criteria calculated are included in the
supplementary material (see supplementary material).
Rainfall intensity Tables and Equations for the IDF curves
for 110 rain gauges were also obtained from IDEAM [17].
3.2 IMERG data
EARLY and LATE (near real-time data) and FINAL
(post-processed data) IMERG products are freely available
online. IMERG EARLY, LATE, and FINAL datasets are
computed every 4 hours, 14 hours, and 3.5 months after
observation time, respectively. EARLY and LATE datasets
are computed an hour at a time, while FINAL is computed
a month at a time. According to the IMERG developers
34
E. Rodríguez et al., Revista Facultad de Ingeniería, Universidad de Antioquia, No. 110, pp. 31-47, 2024
Table 1 Rainfall characteristics of the five major natural regions in Colombia
Elevation Main
Rainfall
Annual # Gauges with
Region Code Ranges Climatic
Regime
Rainfall IDF /daily
(MASL) Drivers (mm) rainfall data
ITCZ / Tropical
Caribbean 1 0-1400 Storms/ ENSO/ Unimodal 1500 16/11
MCS MJJ SON
ITCZ/Chocó Jet Bimodal
Andes 2 50-5700 ENSO/ MAM 1500 72/53
MCS OND
Orinoco 3 100-250 ITCZ Unimodal 2800 12/7
MCS MJJA
Amazon 4 100-300 ITCZ Unimodal 3500 4/3
MCS DJF MAM
Chocó Jet
Pacific 5 0-700 ITCZ Unimodal 5500 5/2
ENSO/MCS AMJJASOND
and several authors, the FINAL product provides the most
suitable estimations for research purposes [31].
IMERG V06B FINAL time series of rainfall intensity at
0.5 h intervals in (mm/h) and at a resolution of 0.10° ( ≈ 11
km) were downloaded for the period 2001-2019 from the
Giovanni NASA repository [54]. For 10 stations, the EARLY,
LATE, and FINAL IMERG products were also downloaded
for the same period, but initial validations of the three
products (results are included in the supplementary
material) showed that the FINAL product systematically
outperformed the other two, as reported by several
authors. For this reason, in the analysis reported here,
only the FINAL product was considered.
3.3 Quality control and data preparation
For the IDEAM rain gauge daily data, missing information
(blank data) was replaced by N.A. Stations with more
than 25% of missing data, or identified outliers, were
discarded from the analysis. In total, 76 stations were
considered in the validation of the IMERG product. IMERG
data in GMT, were converted to GMT-5 in order to obtain
IMERG estimations in local time. Then IMERG data
were converted from intensity to rainfall depth and then
aggregated for validation, at daily, monthly, and yearly
temporal resolutions, using an algorithm developed in R.
3.4 IDEAM IDF Curves
For each of the 110 rain gauges studied, there are IDF
curves available, updated to the year 2010, with different
starting years, as early as 1972 [17]. The development
of these updated IDF curves was based on data of
mass curves of precipitation for selected events at 1-min
resolution, covering either 24 h or one week, depending on
the type of pluviograph strip chart available. We compared
maximum intensities for selected durations and return
periods coming from IDEAM with those estimated from
IMERG data. Perhaps, different periods for comparison
can induce some differences in the curves that were not
considered in the analysis. However, UNAL [17] reports
that when updating the IDF curves up to the year 2010,
the differences in rainfall intensities for the new period
(updated) and the ones derived using the old period are
almost negligible, except for the short 15-minute duration.
Due to the IMERG temporal resolution (30 minutes), the 15
minutes duration was not investigated; then, we assumed
here that there are no significant differences in IDF
curves, associated with the different periods used in their
construction.
4. Methodology
4.1 Ground Validation
Similar studies where satellite products such as IMERG,
TMPA, APHRODITE, and PERSIANN-CDR are verified, have
adopted validation metrics and detection rates such as
Pearson’s correlation coefficient (r), mean absolute error
(MAE), root-mean-square error (RMSE), Nash–Sutcliffe
efficiency (NSE), relative bias (bias), relative error (RE),
and probability of detection (POD), false alarm ratio (FAR),
equitable threat score (ETS), frequency bias index (FBI),
critical success index (CSI), respectively [26, 55–59]. These
metrics have been key for the verification of the potential
of satellite precipitation products. Validation of the IMERG
V06B FINAL product was performed on a pixel-point
basis by comparing the IMERG gridded data with in-situ
information (IDEAM) at three different time resolutions:
daily, monthly, and yearly. Although several studies
35
Table 1 Rainfall characteristics of the five major natural regions in Colombia
Elevation Main
Rainfall
Annual # Gauges with
Region Code Ranges Climatic
Regime
Rainfall IDF /daily
(MASL) Drivers (mm) rainfall data
ITCZ / Tropical
Caribbean 1 0-1400 Storms/ ENSO/ Unimodal 1500 16/11
MCS MJJ SON
ITCZ/Chocó Jet Bimodal
Andes 2 50-5700 ENSO/ MAM 1500 72/53
MCS OND
Orinoco 3 100-250 ITCZ Unimodal 2800 12/7
MCS MJJA
Amazon 4 100-300 ITCZ Unimodal 3500 4/3
MCS DJF MAM
Chocó Jet
Pacific 5 0-700 ITCZ Unimodal 5500 5/2
ENSO/MCS AMJJASOND
and several authors, the FINAL product provides the most
suitable estimations for research purposes [31].
IMERG V06B FINAL time series of rainfall intensity at
0.5 h intervals in (mm/h) and at a resolution of 0.10° ( ≈ 11
km) were downloaded for the period 2001-2019 from the
Giovanni NASA repository [54]. For 10 stations, the EARLY,
LATE, and FINAL IMERG products were also downloaded
for the same period, but initial validations of the three
products (results are included in the supplementary
material) showed that the FINAL product systematically
outperformed the other two, as reported by several
authors. For this reason, in the analysis reported here,
only the FINAL product was considered.
3.3 Quality control and data preparation
For the IDEAM rain gauge daily data, missing information
(blank data) was replaced by N.A. Stations with more
than 25% of missing data, or identified outliers, were
discarded from the analysis. In total, 76 stations were
considered in the validation of the IMERG product. IMERG
data in GMT, were converted to GMT-5 in order to obtain
IMERG estimations in local time. Then IMERG data
were converted from intensity to rainfall depth and then
aggregated for validation, at daily, monthly, and yearly
temporal resolutions, using an algorithm developed in R.
3.4 IDEAM IDF Curves
For each of the 110 rain gauges studied, there are IDF
curves available, updated to the year 2010, with different
starting years, as early as 1972 [17]. The development
of these updated IDF curves was based on data of
mass curves of precipitation for selected events at 1-min
resolution, covering either 24 h or one week, depending on
the type of pluviograph strip chart available. We compared
maximum intensities for selected durations and return
periods coming from IDEAM with those estimated from
IMERG data. Perhaps, different periods for comparison
can induce some differences in the curves that were not
considered in the analysis. However, UNAL [17] reports
that when updating the IDF curves up to the year 2010,
the differences in rainfall intensities for the new period
(updated) and the ones derived using the old period are
almost negligible, except for the short 15-minute duration.
Due to the IMERG temporal resolution (30 minutes), the 15
minutes duration was not investigated; then, we assumed
here that there are no significant differences in IDF
curves, associated with the different periods used in their
construction.
4. Methodology
4.1 Ground Validation
Similar studies where satellite products such as IMERG,
TMPA, APHRODITE, and PERSIANN-CDR are verified, have
adopted validation metrics and detection rates such as
Pearson’s correlation coefficient (r), mean absolute error
(MAE), root-mean-square error (RMSE), Nash–Sutcliffe
efficiency (NSE), relative bias (bias), relative error (RE),
and probability of detection (POD), false alarm ratio (FAR),
equitable threat score (ETS), frequency bias index (FBI),
critical success index (CSI), respectively [26, 55–59]. These
metrics have been key for the verification of the potential
of satellite precipitation products. Validation of the IMERG
V06B FINAL product was performed on a pixel-point
basis by comparing the IMERG gridded data with in-situ
information (IDEAM) at three different time resolutions:
daily, monthly, and yearly. Although several studies
35
E. Rodríguez et al., Revista Facultad de Ingeniería, Universidad de Antioquia, No. 110, pp. 31-47, 2024
have reported on the use of the pixel-point approach
[37, 45, 47], it is clear that a comparison between the
satellite gridded information and the rain gauge data is not
direct, as the properties of rainfall change with the spatial
scale investigated. In this regard, a study in the USA [26]
proposes a downscaling procedure to tackle the problem
of scale disparity between the two sources of information.
Equations 1 to 5 show the five statistical metrics used
in the validation process to investigate the magnitude
of the error, the correlation, and the skill of the IMERG
product to reproduce the observed rainfall information
in each rain gauge station. The metrics include the
normalized root mean square error (nRMSE), used to
describe the error, with an optimum value equal to zero;
the dimensionless Pearson correlation coefficient (r) that
establishes the relationship between the covariance and
variance of the IMERG estimation and the IDEAM data,
with an optimum value equal to 1; the normalized mean
absolute error (nMAE), that describes the magnitude of
the overestimation or underestimation of the IMERG data,
with a perfect value equal to 0; the bias, in percentage,
that represents the magnitude of the underestimation
or overestimation of the satellite data, with an optimum
value equal to 0; and the dimensionless Nash-Sutcliffe
efficiency (NSE) with a perfect value equal to the unity,
used to determine the relative magnitude of the residual
variance compared to the measured data variance. We
have used the normalized statistics for RMSE and MAE
to facilitate comparisons between stations with large
precipitation variability.
To support the analysis, we produced maps for each
temporal resolution and statistical criteria. Additionally,
for each rain gauge and both the IMERG and the IDEAM
datasets, we made comparisons for the mean rainfall at
daily and monthly intervals, during the period 2001-2019.
Besides, to study the variability of the metrics among
regions, we also created boxplots.
nRM SE =
√
1
N
∑N
i=1
(
IM ERGi − I DEAMi)2
∑N
i=1 IDEAMi
(1)
r = N ∑N
i=1 (IM ERGi ∗ IDEAMi) −
√
N ∑N
i=1 IM ERG2
i −
(∑N
i=1 IM ERGi
)2
∑N
i=1 IM ERGi ∗ ∑N
i=1 IDEAMi
√
N ∑N
i=1 IDEAM 2
i −
(∑N
i=1 IDEAMi
)2
(2)
nM AE =
1
N
∑N
i=1 |IM ERGi − IDEAMi|
∑N
i=1 IDEA
(3)
bias(%) =
1
N
∑N
i=1 (IM ERGi − IDEAMi)
∑N
i=1 IDEAMi
∗ 100 (4)
N SE = 1 −
∑N
i=1 (IM ERGi − IDEAMi)2
∑N
i=1
(IDEAMi − IDEAM )2 (5)
At daily resolution, we calculated four dimensionless
contingency metrics to identify the ability of the IMERG
product to detect the occurrence of rainfall. Description
and application of these categorical or contingency
statistics for comparing satellite-based precipitation with
observed data have been given in many references [37,
42, 43, 45]. These criteria, originally proposed in two
studies [60, 61], are shown in Equations 6 to 9 and
include the Probability of Detection (POD), the Probability
of False Detection (POFD), the False Alarm Ratio (FAR)
and the Critical Success Index (CSI). In Equations 6 to
9, hits represent the occurrence of daily rainfall in both,
IMERG and IDEAM datasets; misses represent daily rainfall
observed by IDEAM but not detected by the IMERG dataset;
false alarm represents daily rainfall detected by IMERG but
not observed by IDEAM, and correct negatives are cases
where neither of the two products detects rainfall. The
results for the four contingency criteria are dimensionless
and vary between 0 and 1. For POD and CSI, the perfect
score is 1 and for POFD and FAR, it is 0. For each
contingency criterion, we constructed a map, and boxplots
classified by regions to facilitate the analysis.
P OD = hits
hits + misses (6)
P OF D = false alarm
correct negatives + false alarm (7)
F AR = false alarm
hits + false alarm (8)
CSI = hits
hits + misses + false alarm =
1
1
(1−F AR) + 1
(1−P OD) − 1
(9)
We also produced quantile-quantile plots for the IMERG
and the IDEAM datasets to investigate, for each of the
76 rain gauges, how the two rainfall distributions are
compared at a daily scale and how the extreme daily
events are captured by the IMERG product in each rain
gauge station.
Finally, to investigate the explained variance regarding the
36
have reported on the use of the pixel-point approach
[37, 45, 47], it is clear that a comparison between the
satellite gridded information and the rain gauge data is not
direct, as the properties of rainfall change with the spatial
scale investigated. In this regard, a study in the USA [26]
proposes a downscaling procedure to tackle the problem
of scale disparity between the two sources of information.
Equations 1 to 5 show the five statistical metrics used
in the validation process to investigate the magnitude
of the error, the correlation, and the skill of the IMERG
product to reproduce the observed rainfall information
in each rain gauge station. The metrics include the
normalized root mean square error (nRMSE), used to
describe the error, with an optimum value equal to zero;
the dimensionless Pearson correlation coefficient (r) that
establishes the relationship between the covariance and
variance of the IMERG estimation and the IDEAM data,
with an optimum value equal to 1; the normalized mean
absolute error (nMAE), that describes the magnitude of
the overestimation or underestimation of the IMERG data,
with a perfect value equal to 0; the bias, in percentage,
that represents the magnitude of the underestimation
or overestimation of the satellite data, with an optimum
value equal to 0; and the dimensionless Nash-Sutcliffe
efficiency (NSE) with a perfect value equal to the unity,
used to determine the relative magnitude of the residual
variance compared to the measured data variance. We
have used the normalized statistics for RMSE and MAE
to facilitate comparisons between stations with large
precipitation variability.
To support the analysis, we produced maps for each
temporal resolution and statistical criteria. Additionally,
for each rain gauge and both the IMERG and the IDEAM
datasets, we made comparisons for the mean rainfall at
daily and monthly intervals, during the period 2001-2019.
Besides, to study the variability of the metrics among
regions, we also created boxplots.
nRM SE =
√
1
N
∑N
i=1
(
IM ERGi − I DEAMi)2
∑N
i=1 IDEAMi
(1)
r = N ∑N
i=1 (IM ERGi ∗ IDEAMi) −
√
N ∑N
i=1 IM ERG2
i −
(∑N
i=1 IM ERGi
)2
∑N
i=1 IM ERGi ∗ ∑N
i=1 IDEAMi
√
N ∑N
i=1 IDEAM 2
i −
(∑N
i=1 IDEAMi
)2
(2)
nM AE =
1
N
∑N
i=1 |IM ERGi − IDEAMi|
∑N
i=1 IDEA
(3)
bias(%) =
1
N
∑N
i=1 (IM ERGi − IDEAMi)
∑N
i=1 IDEAMi
∗ 100 (4)
N SE = 1 −
∑N
i=1 (IM ERGi − IDEAMi)2
∑N
i=1
(IDEAMi − IDEAM )2 (5)
At daily resolution, we calculated four dimensionless
contingency metrics to identify the ability of the IMERG
product to detect the occurrence of rainfall. Description
and application of these categorical or contingency
statistics for comparing satellite-based precipitation with
observed data have been given in many references [37,
42, 43, 45]. These criteria, originally proposed in two
studies [60, 61], are shown in Equations 6 to 9 and
include the Probability of Detection (POD), the Probability
of False Detection (POFD), the False Alarm Ratio (FAR)
and the Critical Success Index (CSI). In Equations 6 to
9, hits represent the occurrence of daily rainfall in both,
IMERG and IDEAM datasets; misses represent daily rainfall
observed by IDEAM but not detected by the IMERG dataset;
false alarm represents daily rainfall detected by IMERG but
not observed by IDEAM, and correct negatives are cases
where neither of the two products detects rainfall. The
results for the four contingency criteria are dimensionless
and vary between 0 and 1. For POD and CSI, the perfect
score is 1 and for POFD and FAR, it is 0. For each
contingency criterion, we constructed a map, and boxplots
classified by regions to facilitate the analysis.
P OD = hits
hits + misses (6)
P OF D = false alarm
correct negatives + false alarm (7)
F AR = false alarm
hits + false alarm (8)
CSI = hits
hits + misses + false alarm =
1
1
(1−F AR) + 1
(1−P OD) − 1
(9)
We also produced quantile-quantile plots for the IMERG
and the IDEAM datasets to investigate, for each of the
76 rain gauges, how the two rainfall distributions are
compared at a daily scale and how the extreme daily
events are captured by the IMERG product in each rain
gauge station.
Finally, to investigate the explained variance regarding the
36
E. Rodríguez et al., Revista Facultad de Ingeniería, Universidad de Antioquia, No. 110, pp. 31-47, 2024
statistical, contingency and IDFs error metrics, and reduce
the dimensionality of the data, we implemented a Principal
Component Analysis (PCA). We also explored relationships
of the error metrics with five different proxies, including
rain gauge latitude, altitude, geographic region, mean
annual rainfall, and precipitation regime.
For this purpose, in the principal components space,
we plotted values for the Objective Function (OF) shown
in Equation 10 (optimum value 0), which includes each of
the metrics described above and the four Relative Errors
(ERR REL DUR), calculated as the ratio of the difference
between the IMERG intensity and the IDEAM intensity, and
the IDEAM intensity, for durations DUR of 30, 60, 120 and
360 min. This approach helped us to identify common
physiographic and climatological characteristics of the
stations with similar performance levels of the IMERG
product, both at validation and at the reproduction of
intensities of IDF curves.
4.2 Intensity estimation using IMERG
For each of the 110 rain gauge stations, and during
the period 2001-2019, we aggregated the 0.5 h IMERG
intensity series at 1, 2, and 6 h. We used no other duration
to allow the direct comparison of rainfall intensities from
the IDEAM IDF curves constructed at 15, 30, 60, 120, and
360 minutes. Unfortunately, the resolution of the IMERG
product allowed no comparisons at 15 minutes. After
aggregation of the data, for each rain gauge station and
the four durations (0.5, 1, 2, and 6 h), we constructed the
annual maximum rainfall intensity time series. Afterward,
we performed a frequency analysis using the lmoms
R-library by fitting the Gumbel probability distribution
and obtaining intensities for the selected return periods
(2, 5, 10, 25, and 50 years). We used this probability
distribution as it was the one used to construct the IDEAM
IDF curves, and it has been reported as the one with the
best fit to annual maximum intensities from IMERG data.
We calculated no intensities for 100 years as results for
this return period are highly uncertain due to the IMERG
time-series length of only 19 years (2001-2019).
Rainfall intensities obtained for the selected durations
and return periods were compared with the corresponding
values from the IDEAM IDF curves using a graphical
methodology. Relative errors in (%) for each duration
and return period were also calculated using Equation 11.
The relative errors for each duration were averaged over
all the return periods considered and plotted in maps.
To analyze the variability of the relative errors and the
other metrics among natural regions, boxplots were also
created.
5. Results
5.1 Validation process
Figure 2 shows the spatial distribution of statistical
measurements, including nRMSE, r, and nMAE, for the
three temporal resolutions investigated (yearly, monthly,
and daily). Despite the low number of stations in the
Orinoco and Amazon regions, maps in Figure 2 indicate that
better results are obtained for stations in these two zones.
These regions are characterized by flat terrain. Results
for all the statistical criteria but r suggest that most of
the stations obtained better results at the yearly scale,
followed by monthly and daily resolutions. For r, better
results are obtained at the monthly scale, followed by the
yearly and daily time resolutions (see box plots in Figure
3). This may be a consequence of the global calibration
of the IMERG product at monthly resolution. The bottom
panels in Figure 2 depict the contingency measurements
POD, FAR, and POFD on a daily scale. In general, POD is
high in all stations, FAR is quite variable and PFOD is even
more variable. The summary of the results, consolidated
in the boxplots in Figure 3, shows mixed tendencies. Some
stations located in the Pacific region, such as Aeropuerto El
Caraño, obtained an adequate performance of the IMERG
product for detecting the occurrence of daily rainfall (POD
= 0.94, FAR = 0.06, CSI = 0.88).
OF = (1 − P OD) + P OF D + F AR + (1 − CSI)+
RM SE+ | bias | +M AE+
+ (1 − r) + (1 − N SE)+ | ERR REL 30 | +
13
+ | ERR REL 60| + | ERR REL 120 | +
13 (10)
RELATIVE ERROR = ((IIMERG − IIDEAM ) ∗ 100/IIDEAM )
(11)
Others such as Aeropuerto Benito Salas, located in the
Andes Region, obtained acceptable performances (POD =
0.87, FAR = 0.40, CSI = 0.55), and others, such as Manaure,
located in the Caribbean Region, had inferior results (POD
= 0.74, FAR = 0.70, CSI = 0.27). Overall, results suggest that
the occurrence of rainfall is better detected in stations with
high precipitation (> 2,400 mm), reasonably well detected
in stations with medium precipitation (1,200 - 2,400 mm),
and poorly detected in stations with low precipitation (<
1,200 mm).
The supplementary material includes detailed results
for each rain gauge station, aggregation scale, and
statistical and contingency criteria. Throughout the
spatial domain investigated, results emphasize the mixed
tendencies. Yet considering the uneven distribution of rain
37
statistical, contingency and IDFs error metrics, and reduce
the dimensionality of the data, we implemented a Principal
Component Analysis (PCA). We also explored relationships
of the error metrics with five different proxies, including
rain gauge latitude, altitude, geographic region, mean
annual rainfall, and precipitation regime.
For this purpose, in the principal components space,
we plotted values for the Objective Function (OF) shown
in Equation 10 (optimum value 0), which includes each of
the metrics described above and the four Relative Errors
(ERR REL DUR), calculated as the ratio of the difference
between the IMERG intensity and the IDEAM intensity, and
the IDEAM intensity, for durations DUR of 30, 60, 120 and
360 min. This approach helped us to identify common
physiographic and climatological characteristics of the
stations with similar performance levels of the IMERG
product, both at validation and at the reproduction of
intensities of IDF curves.
4.2 Intensity estimation using IMERG
For each of the 110 rain gauge stations, and during
the period 2001-2019, we aggregated the 0.5 h IMERG
intensity series at 1, 2, and 6 h. We used no other duration
to allow the direct comparison of rainfall intensities from
the IDEAM IDF curves constructed at 15, 30, 60, 120, and
360 minutes. Unfortunately, the resolution of the IMERG
product allowed no comparisons at 15 minutes. After
aggregation of the data, for each rain gauge station and
the four durations (0.5, 1, 2, and 6 h), we constructed the
annual maximum rainfall intensity time series. Afterward,
we performed a frequency analysis using the lmoms
R-library by fitting the Gumbel probability distribution
and obtaining intensities for the selected return periods
(2, 5, 10, 25, and 50 years). We used this probability
distribution as it was the one used to construct the IDEAM
IDF curves, and it has been reported as the one with the
best fit to annual maximum intensities from IMERG data.
We calculated no intensities for 100 years as results for
this return period are highly uncertain due to the IMERG
time-series length of only 19 years (2001-2019).
Rainfall intensities obtained for the selected durations
and return periods were compared with the corresponding
values from the IDEAM IDF curves using a graphical
methodology. Relative errors in (%) for each duration
and return period were also calculated using Equation 11.
The relative errors for each duration were averaged over
all the return periods considered and plotted in maps.
To analyze the variability of the relative errors and the
other metrics among natural regions, boxplots were also
created.
5. Results
5.1 Validation process
Figure 2 shows the spatial distribution of statistical
measurements, including nRMSE, r, and nMAE, for the
three temporal resolutions investigated (yearly, monthly,
and daily). Despite the low number of stations in the
Orinoco and Amazon regions, maps in Figure 2 indicate that
better results are obtained for stations in these two zones.
These regions are characterized by flat terrain. Results
for all the statistical criteria but r suggest that most of
the stations obtained better results at the yearly scale,
followed by monthly and daily resolutions. For r, better
results are obtained at the monthly scale, followed by the
yearly and daily time resolutions (see box plots in Figure
3). This may be a consequence of the global calibration
of the IMERG product at monthly resolution. The bottom
panels in Figure 2 depict the contingency measurements
POD, FAR, and POFD on a daily scale. In general, POD is
high in all stations, FAR is quite variable and PFOD is even
more variable. The summary of the results, consolidated
in the boxplots in Figure 3, shows mixed tendencies. Some
stations located in the Pacific region, such as Aeropuerto El
Caraño, obtained an adequate performance of the IMERG
product for detecting the occurrence of daily rainfall (POD
= 0.94, FAR = 0.06, CSI = 0.88).
OF = (1 − P OD) + P OF D + F AR + (1 − CSI)+
RM SE+ | bias | +M AE+
+ (1 − r) + (1 − N SE)+ | ERR REL 30 | +
13
+ | ERR REL 60| + | ERR REL 120 | +
13 (10)
RELATIVE ERROR = ((IIMERG − IIDEAM ) ∗ 100/IIDEAM )
(11)
Others such as Aeropuerto Benito Salas, located in the
Andes Region, obtained acceptable performances (POD =
0.87, FAR = 0.40, CSI = 0.55), and others, such as Manaure,
located in the Caribbean Region, had inferior results (POD
= 0.74, FAR = 0.70, CSI = 0.27). Overall, results suggest that
the occurrence of rainfall is better detected in stations with
high precipitation (> 2,400 mm), reasonably well detected
in stations with medium precipitation (1,200 - 2,400 mm),
and poorly detected in stations with low precipitation (<
1,200 mm).
The supplementary material includes detailed results
for each rain gauge station, aggregation scale, and
statistical and contingency criteria. Throughout the
spatial domain investigated, results emphasize the mixed
tendencies. Yet considering the uneven distribution of rain
37
E. Rodríguez et al., Revista Facultad de Ingeniería, Universidad de Antioquia, No. 110, pp. 31-47, 2024
Figure 2 Maps showing the spatial distribution of statistical measurements nRMSE (-), r (-), nMAE (-) at yearly, monthly and daily
scales, and contingency metrics POD (-), FAR (-) and POFD (-) calculated at daily time resolution.
38
Figure 2 Maps showing the spatial distribution of statistical measurements nRMSE (-), r (-), nMAE (-) at yearly, monthly and daily
scales, and contingency metrics POD (-), FAR (-) and POFD (-) calculated at daily time resolution.
38
E. Rodríguez et al., Revista Facultad de Ingeniería, Universidad de Antioquia, No. 110, pp. 31-47, 2024
Figure 3 Box plots for the statistical metrics nRMSE, bias, nMAE and r were used to investigate IMERG performance at yearly,
monthly and daily time resolutions, and contingency metrics POD, FAR, POFD and CSI calculated at a daily time resolution
Table 2 Summary of the performance of the IMERG product in capturing daily extreme rainfall events in the investigated stations as
a result of the quantile-quantile analysis
Region No. stations Underestimation % Overestimation % Correct %
Caribbean 11 45 36 19
Andes 53 26 48 26
Orinoco 7 29 57 14
Amazon 3 33 67 -
Pacific 2 - 50 50
All 76 29 48 23
Figure 4 a) Comparison of the mean multiannual daily precipitation, b) the mean monthly precipitation, and c) the yearly
precipitation in the Apto Pto Carreño - Orinoco Region rain gauge station for the IMERG and IDEAM datasets
gauge stations in the five regions, in general, the Pearson
correlations calculated depend on the time scale of
analysis. For stations located in the Orinoco and Amazon
regions, Pearson correlation shows high values when the
monthly time scale is evaluated; reasonable results are
obtained with annual and monthly evaluations for stations
located in the Pacific and Caribbean Regions. In the case
of the Andes region, results are mixed between low and
high. From the analysis of the r metric at the regional
scale (see box plots in the supplementary material) results
39
Figure 3 Box plots for the statistical metrics nRMSE, bias, nMAE and r were used to investigate IMERG performance at yearly,
monthly and daily time resolutions, and contingency metrics POD, FAR, POFD and CSI calculated at a daily time resolution
Table 2 Summary of the performance of the IMERG product in capturing daily extreme rainfall events in the investigated stations as
a result of the quantile-quantile analysis
Region No. stations Underestimation % Overestimation % Correct %
Caribbean 11 45 36 19
Andes 53 26 48 26
Orinoco 7 29 57 14
Amazon 3 33 67 -
Pacific 2 - 50 50
All 76 29 48 23
Figure 4 a) Comparison of the mean multiannual daily precipitation, b) the mean monthly precipitation, and c) the yearly
precipitation in the Apto Pto Carreño - Orinoco Region rain gauge station for the IMERG and IDEAM datasets
gauge stations in the five regions, in general, the Pearson
correlations calculated depend on the time scale of
analysis. For stations located in the Orinoco and Amazon
regions, Pearson correlation shows high values when the
monthly time scale is evaluated; reasonable results are
obtained with annual and monthly evaluations for stations
located in the Pacific and Caribbean Regions. In the case
of the Andes region, results are mixed between low and
high. From the analysis of the r metric at the regional
scale (see box plots in the supplementary material) results
39
E. Rodríguez et al., Revista Facultad de Ingeniería, Universidad de Antioquia, No. 110, pp. 31-47, 2024
Figure 5 Maps showing the nRMSE (left panels), percent bias (%) (center panels) and the Pearson correlation coefficient (r) (-)
(right panels), for the mean multiannual daily precipitation (top panels) and the mean monthly precipitation (bottom panels), as
estimated using the IDEAM dataset as ground-validation
show that at the daily time resolution, higher median
values are obtained for the Orinoco (r=0.65), followed
by the Pacific (0.60) and Amazon (r=0.55) regions. This
is associated with the low number of stations in these
three regions, compared to the number of stations in the
Caribbean (r=0.45) and Andean (r=0.50) regions.
Regarding the multiannual analysis conducted for
validation of the IMERG product, Figure 4 shows an
example (rain gauge station Apto. Pto Carreño - Orinoco)
of the mean daily multiannual precipitation, the mean
monthly multiannual precipitation, and the yearly rainfall,
as calculated using the IMERG and IDEAM datasets. For
the mean daily precipitation, 30-day moving averages
were also plotted, and showed a very good agreement
between both datasets. A detailed observation of
Figure 4a shows that for this station, extreme daily
rainfall is underestimated, and overall, there is a small
overestimation of the multiannual mean daily and monthly
rainfall, which is larger at the annual scale. Error metrics
are nRMSE = 0.50, bias = + 13.47%, and r = 0.88 for the
mean daily data, and nRMSE = 0.18, bias = + 13.47%, and
r = 0.99 for the mean monthly data. These results show
that the IMERG product well captures the seasonality of
daily and monthly average rainfall. Results for other rain
gauge stations are included in the supplementary material
and spatialized in Figure 5, where the percent bias and
the Pearson correlation coefficient are plotted for each
station and temporal resolutions, showing again mixing
tendencies. The best results were obtained for stations in
the Orinoco, Amazon, and Pacific Regions. Better results
are obtained for the multiannual monthly precipitation,
than for the multiannual daily rainfall, thus reinforcing the
findings described before during the analysis of the whole
time series.
To complement the validation of the performance of
the IMERG product at a daily scale, we constructed
quantile-quantile plots for all 76 rain gauge stations.
Examples of the results are presented in Figure 6 and
summarized for all stations in Table 2. Results showed
mixed tendencies for the daily extreme rainfall events,
with overestimations in most of the cases (48% overall),
followed by underestimations (29% overall) and good
agreements (23% overall).
40
Figure 5 Maps showing the nRMSE (left panels), percent bias (%) (center panels) and the Pearson correlation coefficient (r) (-)
(right panels), for the mean multiannual daily precipitation (top panels) and the mean monthly precipitation (bottom panels), as
estimated using the IDEAM dataset as ground-validation
show that at the daily time resolution, higher median
values are obtained for the Orinoco (r=0.65), followed
by the Pacific (0.60) and Amazon (r=0.55) regions. This
is associated with the low number of stations in these
three regions, compared to the number of stations in the
Caribbean (r=0.45) and Andean (r=0.50) regions.
Regarding the multiannual analysis conducted for
validation of the IMERG product, Figure 4 shows an
example (rain gauge station Apto. Pto Carreño - Orinoco)
of the mean daily multiannual precipitation, the mean
monthly multiannual precipitation, and the yearly rainfall,
as calculated using the IMERG and IDEAM datasets. For
the mean daily precipitation, 30-day moving averages
were also plotted, and showed a very good agreement
between both datasets. A detailed observation of
Figure 4a shows that for this station, extreme daily
rainfall is underestimated, and overall, there is a small
overestimation of the multiannual mean daily and monthly
rainfall, which is larger at the annual scale. Error metrics
are nRMSE = 0.50, bias = + 13.47%, and r = 0.88 for the
mean daily data, and nRMSE = 0.18, bias = + 13.47%, and
r = 0.99 for the mean monthly data. These results show
that the IMERG product well captures the seasonality of
daily and monthly average rainfall. Results for other rain
gauge stations are included in the supplementary material
and spatialized in Figure 5, where the percent bias and
the Pearson correlation coefficient are plotted for each
station and temporal resolutions, showing again mixing
tendencies. The best results were obtained for stations in
the Orinoco, Amazon, and Pacific Regions. Better results
are obtained for the multiannual monthly precipitation,
than for the multiannual daily rainfall, thus reinforcing the
findings described before during the analysis of the whole
time series.
To complement the validation of the performance of
the IMERG product at a daily scale, we constructed
quantile-quantile plots for all 76 rain gauge stations.
Examples of the results are presented in Figure 6 and
summarized for all stations in Table 2. Results showed
mixed tendencies for the daily extreme rainfall events,
with overestimations in most of the cases (48% overall),
followed by underestimations (29% overall) and good
agreements (23% overall).
40
E. Rodríguez et al., Revista Facultad de Ingeniería, Universidad de Antioquia, No. 110, pp. 31-47, 2024
Figure 6 Quantile-quantile plots of daily rainfall (IMERG and IDEAM) for the selected stations. The arrows and numbers show the
location of quantiles
As an illustration, in Figure 6, results for station Apto
El Caraño - Pacific indicate that daily low precipitation (<
25 mm) is well captured, but there are underestimations
for the moderate rain (25 -180 mm) and overestimations
of the heavy rains (> 180 mm). For station Manaure -
Caribbean, results indicate that light and moderate daily
rains are slightly overestimated, while heavy rainfall (> 70
mm) has mixed patterns with under and overestimations.
For station Araracuara - Amazon, light and moderate rains
(< 70 mm) are well captured, and heavy rains (> 70 mm) are
overestimated. For station Apto Pto. Carreño - Orinoco,
results indicate that low and moderate rains (< 70 mm) are
well captured, and heavy rainfall are underestimated. For
station Santa Cecilia in the Andes Region, results show
that light rains are well captured, and moderate and heavy
rains have mixed patterns.
For the San Francisco rain gauge, also in the Andes
region, results show that there is an underestimation
of the daily rainfall events for all types of rainfall.
Quantile-quantile plots for all 76 rain gauge stations are
included in the supplementary material. Results highlight
the limitations of the IMERG product in capturing daily
extreme rainfall events in the majority of the cases (only
a fifth of them with good agreement). These extreme
events are the most used to derive IDF curves and, in this
sense, limitations in capturing extreme daily rainfall could
be transferred into weaknesses of the IMERG product
for reproducing maximum intensities in IDF curves in
Colombia.
The selected results of the Principal Component Analysis
implemented are shown in Figure 7. Additional results are
included in the supplementary material. For the 13-error
metrics calculated at a daily scale, the three principal
components (PC1, PC2, and PC3) explain in total 83.4% of
the data variance, with individual values of 41.3%, 29.5%,
and 12.6%, respectively. In the PC1-PC2 space, besides
the trivial correlations (CSI is inversely correlated with FAR
and POFD; ERR_REL_30, ERR_REL_60, ERR_REL_120,
ERR_REL_360 are highly correlated; normalized statistical
measurements 1-NSE, nMAE, nRMSE and bias are highly
41
Figure 6 Quantile-quantile plots of daily rainfall (IMERG and IDEAM) for the selected stations. The arrows and numbers show the
location of quantiles
As an illustration, in Figure 6, results for station Apto
El Caraño - Pacific indicate that daily low precipitation (<
25 mm) is well captured, but there are underestimations
for the moderate rain (25 -180 mm) and overestimations
of the heavy rains (> 180 mm). For station Manaure -
Caribbean, results indicate that light and moderate daily
rains are slightly overestimated, while heavy rainfall (> 70
mm) has mixed patterns with under and overestimations.
For station Araracuara - Amazon, light and moderate rains
(< 70 mm) are well captured, and heavy rains (> 70 mm) are
overestimated. For station Apto Pto. Carreño - Orinoco,
results indicate that low and moderate rains (< 70 mm) are
well captured, and heavy rainfall are underestimated. For
station Santa Cecilia in the Andes Region, results show
that light rains are well captured, and moderate and heavy
rains have mixed patterns.
For the San Francisco rain gauge, also in the Andes
region, results show that there is an underestimation
of the daily rainfall events for all types of rainfall.
Quantile-quantile plots for all 76 rain gauge stations are
included in the supplementary material. Results highlight
the limitations of the IMERG product in capturing daily
extreme rainfall events in the majority of the cases (only
a fifth of them with good agreement). These extreme
events are the most used to derive IDF curves and, in this
sense, limitations in capturing extreme daily rainfall could
be transferred into weaknesses of the IMERG product
for reproducing maximum intensities in IDF curves in
Colombia.
The selected results of the Principal Component Analysis
implemented are shown in Figure 7. Additional results are
included in the supplementary material. For the 13-error
metrics calculated at a daily scale, the three principal
components (PC1, PC2, and PC3) explain in total 83.4% of
the data variance, with individual values of 41.3%, 29.5%,
and 12.6%, respectively. In the PC1-PC2 space, besides
the trivial correlations (CSI is inversely correlated with FAR
and POFD; ERR_REL_30, ERR_REL_60, ERR_REL_120,
ERR_REL_360 are highly correlated; normalized statistical
measurements 1-NSE, nMAE, nRMSE and bias are highly
41
E. Rodríguez et al., Revista Facultad de Ingeniería, Universidad de Antioquia, No. 110, pp. 31-47, 2024
Figure 7 Selected results of the Principal Component Analysis. For each rain gauge station, values in colors indicate the results of
the objective function as formulated in Equation 10 for the corresponding proxy investigated (mean annual precipitation, rainfall
regime, natural region, latitude, and altitude)
correlated), we found an inverse relationship between POD
and the average relative errors in the estimated intensities
of IDF curves (ERR_REL 30, ERR_REL_60, ERR_REL_120,
ERR_REL 360). This means that, in general, the best
results in the estimation of intensities of IDF curves are
obtained for stations with a high probability of detection of
daily rainfall (POD), such as El Rancho - Andes (POD=0.97),
and poorer results are obtained for rain gauges with
lower POD values, such as station Manaure - Caribbean
(POD=0.74). In the PC1-PC3 and PC2-PC3 spaces, it
is clear that statistical metrics and errors in IDF curve
estimations are highly correlated and that CSI and RMSE
are inversely correlated. Although counterintuitive, in
these two spaces, we found that POD and POFD are highly
correlated for stations located in the Orinoco, Amazonas,
and Pacific Regions. However, these results were obtained
with very few stations in these three areas, so they cannot
be generalized for the whole regions. Additionally, the
best performances of the IMERG FINAL product occur for
plain areas with annual precipitation larger than 2,400
mm, experiencing a monomodal rainfall regime. We found
no correlation between IMERG performance and altitude.
Numbers presented in the PCA spaces in Figure 7 show the
values of the objective function formulated in Equation 10,
which helped us to investigate relationships between the
IMERG performance and physiographic and climatological
characteristics. In general, we found that better results
are obtained for stations located mainly in the Orinoco,
Amazon, and Pacific Regions with a monomodal rainfall
regime, and mean annual precipitation above 2,400 mm.
Regarding latitude, it seems that IMERG results improve
42
Figure 7 Selected results of the Principal Component Analysis. For each rain gauge station, values in colors indicate the results of
the objective function as formulated in Equation 10 for the corresponding proxy investigated (mean annual precipitation, rainfall
regime, natural region, latitude, and altitude)
correlated), we found an inverse relationship between POD
and the average relative errors in the estimated intensities
of IDF curves (ERR_REL 30, ERR_REL_60, ERR_REL_120,
ERR_REL 360). This means that, in general, the best
results in the estimation of intensities of IDF curves are
obtained for stations with a high probability of detection of
daily rainfall (POD), such as El Rancho - Andes (POD=0.97),
and poorer results are obtained for rain gauges with
lower POD values, such as station Manaure - Caribbean
(POD=0.74). In the PC1-PC3 and PC2-PC3 spaces, it
is clear that statistical metrics and errors in IDF curve
estimations are highly correlated and that CSI and RMSE
are inversely correlated. Although counterintuitive, in
these two spaces, we found that POD and POFD are highly
correlated for stations located in the Orinoco, Amazonas,
and Pacific Regions. However, these results were obtained
with very few stations in these three areas, so they cannot
be generalized for the whole regions. Additionally, the
best performances of the IMERG FINAL product occur for
plain areas with annual precipitation larger than 2,400
mm, experiencing a monomodal rainfall regime. We found
no correlation between IMERG performance and altitude.
Numbers presented in the PCA spaces in Figure 7 show the
values of the objective function formulated in Equation 10,
which helped us to investigate relationships between the
IMERG performance and physiographic and climatological
characteristics. In general, we found that better results
are obtained for stations located mainly in the Orinoco,
Amazon, and Pacific Regions with a monomodal rainfall
regime, and mean annual precipitation above 2,400 mm.
Regarding latitude, it seems that IMERG results improve
42
E. Rodríguez et al., Revista Facultad de Ingeniería, Universidad de Antioquia, No. 110, pp. 31-47, 2024
Figure 8 Average relative error (%) ((IMERG- IDEAM)*100/ IDEAM) in rainfall intensities for all return periods (Tr) and for durations
0.5h (top left panel), 1h (top right panel), 2h (bottom left panel), 6h bottom right panel)
towards the south. No clear relationship was evidenced
between the error metrics and the altitude.
5.2 Comparison of intensities of IDF curves
Figure 8 shows for the four durations investigated (0.5,
1, 2, 6 h) the magnitude and sign of the average relative
errors for all return periods in the 110 stations. Maps
for each duration and return period are included in
the supplementary material. Results show that, for
short durations (0.5 and 1 h), there is a systematic
underestimation of the rainfall intensities in almost all
stations and that for large durations (2 and 6 h) the
opposite occurs, with a general overestimation of the
rainfall intensities. Figure 9 compares the intensities of
IDF curves for return periods (Tr) of 2 and 50 years, for the
selected stations in each of the five regions in Colombia.
In general, results show that there are underestimations
of the intensities for short durations (0.5 and 1 h) and
overestimations for large durations (2 and 6 h). For
rain gauge stations El Rancho (Andes) and Mercaderes
(Pacific), the IDF curves are quite similar, with low errors
that become larger for the 2 and 6 h durations. For
the Aeropuerto El Caraño rain gauge (Pacific), results
show that there is an underestimation of the intensities
(maximum of -22%) for the 0.5 h duration, while there is
an overestimation of the intensities for the other durations
and return periods, which is smaller for the lower return
periods (+25%) and larger for the others (+31%).
The Manaure rain gauge station, in the Caribbean
region, shows one of the poorest results overall. There
are important underestimations of the intensities for
short durations (0.5 and 1 h) with a maximum of -55%.
For large durations, the observed and estimated IDFs
are similar, with maximum relative errors of ± 20%. For
the Araracuara rain gauge station (Orinoquia), results
show maximum underestimations of -41% and maximum
43
Figure 8 Average relative error (%) ((IMERG- IDEAM)*100/ IDEAM) in rainfall intensities for all return periods (Tr) and for durations
0.5h (top left panel), 1h (top right panel), 2h (bottom left panel), 6h bottom right panel)
towards the south. No clear relationship was evidenced
between the error metrics and the altitude.
5.2 Comparison of intensities of IDF curves
Figure 8 shows for the four durations investigated (0.5,
1, 2, 6 h) the magnitude and sign of the average relative
errors for all return periods in the 110 stations. Maps
for each duration and return period are included in
the supplementary material. Results show that, for
short durations (0.5 and 1 h), there is a systematic
underestimation of the rainfall intensities in almost all
stations and that for large durations (2 and 6 h) the
opposite occurs, with a general overestimation of the
rainfall intensities. Figure 9 compares the intensities of
IDF curves for return periods (Tr) of 2 and 50 years, for the
selected stations in each of the five regions in Colombia.
In general, results show that there are underestimations
of the intensities for short durations (0.5 and 1 h) and
overestimations for large durations (2 and 6 h). For
rain gauge stations El Rancho (Andes) and Mercaderes
(Pacific), the IDF curves are quite similar, with low errors
that become larger for the 2 and 6 h durations. For
the Aeropuerto El Caraño rain gauge (Pacific), results
show that there is an underestimation of the intensities
(maximum of -22%) for the 0.5 h duration, while there is
an overestimation of the intensities for the other durations
and return periods, which is smaller for the lower return
periods (+25%) and larger for the others (+31%).
The Manaure rain gauge station, in the Caribbean
region, shows one of the poorest results overall. There
are important underestimations of the intensities for
short durations (0.5 and 1 h) with a maximum of -55%.
For large durations, the observed and estimated IDFs
are similar, with maximum relative errors of ± 20%. For
the Araracuara rain gauge station (Orinoquia), results
show maximum underestimations of -41% and maximum
43
E. Rodríguez et al., Revista Facultad de Ingeniería, Universidad de Antioquia, No. 110, pp. 31-47, 2024
Figure 9 Comparison between IDEAM and IMERG intensities from IDF curves for selected stations. To facilitate the comparison,
results are only shown for the minimum (2 years) and maximum (50 years) return periods (Tr).
overestimations of +17%. The comparisons of intensities
for all IDF curves are included in the supplementary
material and show mixed results, similar to the ones
obtained for the six stations here analyzed.
6. Conclusions
In this study, we validated and explored the capabilities of
the IMERG V06B FINAL product for reproducing intensities
of IDF curves in 110 rain gauges in Colombia using in-situ
and satellite data for the period 2001-2019. Based on the
methods and analysis implemented, we found the following
conclusions:
• The performance of the IMERG FINAL product in
capturing rainfall distribution at yearly, monthly, and
daily scales in the period 2001-2019 is quite variable
and depends on the physiographic and climatological
characteristics of the places where the stations are
located. However, aggregated analysis for all stations
in the five natural regions in Colombia shows that the
best results are obtained yearly, followed by monthly,
and daily time resolutions. The regional analysis
of the errors and contingency metrics highlights
that the best results are obtained for the Orinoco,
Amazon, and Pacific regions. Although this could
be associated with the low number of stations in
these three regions, the comparisons between IDEAM
and IMERG maximum intensities reinforce these
findings. Nevertheless, caution is recommended
when extrapolating these results.
• The best agreements between intensities of IDF
curves derived from IMERG and those available from
IDEAM occur for the rain gauge stations located in
the Orinoco, Amazon, and Pacific regions with relative
errors in the rainfall intensities in the range +/- 2%.
These rain gauges represent around a fifth of the
total. For other stations, results are inferior, with
larger relative errors in rainfall intensities in the
range -69, +17%.
• In the majority of the stations for the IMERG IDF
curves, there are underestimations of the rainfall
intensities for short durations (0.5 and 1 h) and
overestimations for the large durations (2 and 6 h),
with relative errors in the ranges [-69, +5%] and [-61,
+171%], respectively.
• In general terms, the best results for the intensities
of IMERG IDF curves were obtained for stations
with a high probability of detection of daily rainfall
(POD). However, results also show that the IMERG
product possibly reaches a high POD at the expense
44
Figure 9 Comparison between IDEAM and IMERG intensities from IDF curves for selected stations. To facilitate the comparison,
results are only shown for the minimum (2 years) and maximum (50 years) return periods (Tr).
overestimations of +17%. The comparisons of intensities
for all IDF curves are included in the supplementary
material and show mixed results, similar to the ones
obtained for the six stations here analyzed.
6. Conclusions
In this study, we validated and explored the capabilities of
the IMERG V06B FINAL product for reproducing intensities
of IDF curves in 110 rain gauges in Colombia using in-situ
and satellite data for the period 2001-2019. Based on the
methods and analysis implemented, we found the following
conclusions:
• The performance of the IMERG FINAL product in
capturing rainfall distribution at yearly, monthly, and
daily scales in the period 2001-2019 is quite variable
and depends on the physiographic and climatological
characteristics of the places where the stations are
located. However, aggregated analysis for all stations
in the five natural regions in Colombia shows that the
best results are obtained yearly, followed by monthly,
and daily time resolutions. The regional analysis
of the errors and contingency metrics highlights
that the best results are obtained for the Orinoco,
Amazon, and Pacific regions. Although this could
be associated with the low number of stations in
these three regions, the comparisons between IDEAM
and IMERG maximum intensities reinforce these
findings. Nevertheless, caution is recommended
when extrapolating these results.
• The best agreements between intensities of IDF
curves derived from IMERG and those available from
IDEAM occur for the rain gauge stations located in
the Orinoco, Amazon, and Pacific regions with relative
errors in the rainfall intensities in the range +/- 2%.
These rain gauges represent around a fifth of the
total. For other stations, results are inferior, with
larger relative errors in rainfall intensities in the
range -69, +17%.
• In the majority of the stations for the IMERG IDF
curves, there are underestimations of the rainfall
intensities for short durations (0.5 and 1 h) and
overestimations for the large durations (2 and 6 h),
with relative errors in the ranges [-69, +5%] and [-61,
+171%], respectively.
• In general terms, the best results for the intensities
of IMERG IDF curves were obtained for stations
with a high probability of detection of daily rainfall
(POD). However, results also show that the IMERG
product possibly reaches a high POD at the expense
44
E. Rodríguez et al., Revista Facultad de Ingeniería, Universidad de Antioquia, No. 110, pp. 31-47, 2024
of also a high probability of false detection (POFD).
This represents an opportunity to improve the IMERG
product by its developers.
• The results reported in this study used all the 110
rain gauge stations in Colombia with available IDF
curves, updated to the year 2010. In this sense, our
results are limited and may be considered cautiously.
The inclusion of more stations in the analysis and
the continuous update of the existing IDF curves are
necessary to improve the results and enlarge the
period of comparison, which was short in this study
(2001-2019). An invitation to IDEAM was made to
support these research activities.
7. Declaration of competing interest
We declare that we have no significant competing interests,
including financial or non-financial, professional, or
personal interests interfering with the full and objective
presentation of the work described in this manuscript.
8. Acknowledgements
We acknowledge IDEAM for supporting the development of
the IDF curves. Also, many thanks to the two anonymous
reviewers for their thoughtful comments that allowed us to
improve the quality of the manuscript.
9. Funding
This work was supported by Universidad Nacional de
Colombia.
10. Author contributions
ER provided the idea for the study and helped write the
first version of the article. CG, AG, JS, MP, and DA
compiled the data and developed the codes for processing
the information, producing the figures, maps, and tables.
They also contributed to the writing of the final version of
the paper.
11. Data availability Statement
The authors confirm that all the data, codes
and results supporting the findings of this study
are available in the supplementary material.
https://data.mendeley.com/datasets/bvbxpm4fsm/2
References
[1] D. Koutsoyiannis, D. Kozonis, and A. Manetas, “A mathematical
framework for studying rainfall intensity-duration-frequency
relationships,” Journal of Hydrology, vol. 206, no. 1-2, apr. 1998.
[Online]. Available: https://doi.org/10.1016/S0022-1694(98)00097-3
[2] Z. Şen, “Annual daily maximum rainfall-based idf curve derivation
methodology,” Earth Systems and Environment, vol. 3, Sep. 24, 2019.
[Online]. Available: https://doi.org/10.1007/s41748-019-00124-x
[3] N. Bezak, M. Šraj, and M. Mikoš, “Copula-based idf curves and
empirical rainfall thresholds for flash floods and rainfall-induced
landslides,” Journal of Hydrology, vol. 541, Mar. 08, 2016. [Online].
Available: https://doi.org/10.1016/j.jhydrol.2016.02.058
[4] E. V. Aristizábal-Giraldo, E. F. García-Aristizábal, R. J.
Marín-Sánchez, F. Gómez-Cardona, and J. C. Guzmán-Martínez,
“Rainfall-intensity effect on landslide hazard assessment due
to climate change in north-western colombian andes,” B.A.
thesis, Universidad de Antioquia, Facultad de Ingeniería, Medellín,
Colombia, 2022.
[5] B. D. Avila-Rangel and H. F. Avila-Rangel, “Spatial and temporal
estimation of the erosivity factor r based on daily rainfall data for
the department of atlántico, colombia,” B.A. thesis, Universidad
Nacional, Ingeniería e investigación, Medellín, Colombia, 2015.
[6] K. Kirby and J. Stratton, “Hidfun una herramienta para la extracción
y análisis de pluviogramas,” XXVIII Congreso latinoamericano de
hidráulica, Buenos Aires, Argentina, 2018.
[7] S. M. Papalexiou and D. Koutsoyiannis, “Battle of extreme value
distributions: A global survey on extreme daily rainfall,” Water
resources Research, vol. 49, no. 1, Jan. 15, 2013. [Online]. Available:
https://doi.org/10.1029/2012WR012557
[8] F. Serinaldi and C. G. Kilsby, “Rainfall extremes: Toward
reconciliation after the battle of distributions,” Water resources
Research, vol. 50, no. 1, Jan. 04, 2014. [Online]. Available:
https://doi.org/10.1002/2013WR014211
[9] S. Emmanouil, A. Langousis, E. I. Nikolopoulos, and E. N.
Anagnostou, “Quantitative assessment of annual maxima,
peaks-over-threshold and multifractal parametric approaches in
estimating intensity-duration-frequency curves from short rainfall
records,” Journal of Hydrology, vol. 589, May. 01, 2020. [Online].
Available: https://doi.org/10.1016/j.jhydrol.2020.125151
[10] D. Grajales-Cardona and L. F. Carvajal-Serna, “Nonstationary
intesity-duration-frequency curves for medellin river basin,” DYNA,
vol. 86, no. 208, Jan-Mar. 2019. [Online]. Available: DOI:https:
//doi.org/10.15446/dyna.v86n208.69300
[11] A. Langousis and D. Veneziano, “Intensity-duration-frequency
curves from scaling representations of rainfall,” Water resources
Research, vol. 43, no. 2, Feb. 21, 2007. [Online]. Available:
https://doi.org/10.1029/2006WR005245
[12] H. V. de Vyver and G. R. Demarée, “Construction of
intensity–duration–frequency (idf) curves for precipitation at
lubumbashi, congo, under the hypothesis of inadequate data,”
Hydrological Sciences Journal, vol. 55, no. 4, Nov. 02, 2009. [Online].
Available: https://doi.org/10.1080/02626661003747390
[13] H. Tyralis and A. Langousis, “Estimation of
intensity–duration–frequency curves using max-stable processes,”
Stochastic Environmental Research and Risk Assessment, vol. 33,
Jul. 06, 2018. [Online]. Available: https://doi.org/10.1007/
s00477-018-1577-2
[14] C. Cheng-lung, “Rainfall intensity duration frequency formulas,”
Journal of Hydraulic Engineering, vol. 109, no. 12, Dec. 01, 1983.
[Online]. Available: https://doi.org/10.1061/(ASCE)0733-9429(1983)
109:12(1603)
[15] S. Urbanek, M. Vargas, and R. Díaz-Granados, “Curvas sintéticas
regionalizadas de intensidad-duración-frecuencia para colombia.”
in Seminario de Hidráulica e Hidrología, Sociedad Colombiana de
Ingenieros, Cali, Colombia, 1998, pp. 119–128.
[16] E. G. Pulgarín-Dávila, “Fórmulas regionales para la estimación
de curvas intensidad - frecuencia - duración basadas en las
propiedades de escala de la lluvia (región andina colombiana),”
45
of also a high probability of false detection (POFD).
This represents an opportunity to improve the IMERG
product by its developers.
• The results reported in this study used all the 110
rain gauge stations in Colombia with available IDF
curves, updated to the year 2010. In this sense, our
results are limited and may be considered cautiously.
The inclusion of more stations in the analysis and
the continuous update of the existing IDF curves are
necessary to improve the results and enlarge the
period of comparison, which was short in this study
(2001-2019). An invitation to IDEAM was made to
support these research activities.
7. Declaration of competing interest
We declare that we have no significant competing interests,
including financial or non-financial, professional, or
personal interests interfering with the full and objective
presentation of the work described in this manuscript.
8. Acknowledgements
We acknowledge IDEAM for supporting the development of
the IDF curves. Also, many thanks to the two anonymous
reviewers for their thoughtful comments that allowed us to
improve the quality of the manuscript.
9. Funding
This work was supported by Universidad Nacional de
Colombia.
10. Author contributions
ER provided the idea for the study and helped write the
first version of the article. CG, AG, JS, MP, and DA
compiled the data and developed the codes for processing
the information, producing the figures, maps, and tables.
They also contributed to the writing of the final version of
the paper.
11. Data availability Statement
The authors confirm that all the data, codes
and results supporting the findings of this study
are available in the supplementary material.
https://data.mendeley.com/datasets/bvbxpm4fsm/2
References
[1] D. Koutsoyiannis, D. Kozonis, and A. Manetas, “A mathematical
framework for studying rainfall intensity-duration-frequency
relationships,” Journal of Hydrology, vol. 206, no. 1-2, apr. 1998.
[Online]. Available: https://doi.org/10.1016/S0022-1694(98)00097-3
[2] Z. Şen, “Annual daily maximum rainfall-based idf curve derivation
methodology,” Earth Systems and Environment, vol. 3, Sep. 24, 2019.
[Online]. Available: https://doi.org/10.1007/s41748-019-00124-x
[3] N. Bezak, M. Šraj, and M. Mikoš, “Copula-based idf curves and
empirical rainfall thresholds for flash floods and rainfall-induced
landslides,” Journal of Hydrology, vol. 541, Mar. 08, 2016. [Online].
Available: https://doi.org/10.1016/j.jhydrol.2016.02.058
[4] E. V. Aristizábal-Giraldo, E. F. García-Aristizábal, R. J.
Marín-Sánchez, F. Gómez-Cardona, and J. C. Guzmán-Martínez,
“Rainfall-intensity effect on landslide hazard assessment due
to climate change in north-western colombian andes,” B.A.
thesis, Universidad de Antioquia, Facultad de Ingeniería, Medellín,
Colombia, 2022.
[5] B. D. Avila-Rangel and H. F. Avila-Rangel, “Spatial and temporal
estimation of the erosivity factor r based on daily rainfall data for
the department of atlántico, colombia,” B.A. thesis, Universidad
Nacional, Ingeniería e investigación, Medellín, Colombia, 2015.
[6] K. Kirby and J. Stratton, “Hidfun una herramienta para la extracción
y análisis de pluviogramas,” XXVIII Congreso latinoamericano de
hidráulica, Buenos Aires, Argentina, 2018.
[7] S. M. Papalexiou and D. Koutsoyiannis, “Battle of extreme value
distributions: A global survey on extreme daily rainfall,” Water
resources Research, vol. 49, no. 1, Jan. 15, 2013. [Online]. Available:
https://doi.org/10.1029/2012WR012557
[8] F. Serinaldi and C. G. Kilsby, “Rainfall extremes: Toward
reconciliation after the battle of distributions,” Water resources
Research, vol. 50, no. 1, Jan. 04, 2014. [Online]. Available:
https://doi.org/10.1002/2013WR014211
[9] S. Emmanouil, A. Langousis, E. I. Nikolopoulos, and E. N.
Anagnostou, “Quantitative assessment of annual maxima,
peaks-over-threshold and multifractal parametric approaches in
estimating intensity-duration-frequency curves from short rainfall
records,” Journal of Hydrology, vol. 589, May. 01, 2020. [Online].
Available: https://doi.org/10.1016/j.jhydrol.2020.125151
[10] D. Grajales-Cardona and L. F. Carvajal-Serna, “Nonstationary
intesity-duration-frequency curves for medellin river basin,” DYNA,
vol. 86, no. 208, Jan-Mar. 2019. [Online]. Available: DOI:https:
//doi.org/10.15446/dyna.v86n208.69300
[11] A. Langousis and D. Veneziano, “Intensity-duration-frequency
curves from scaling representations of rainfall,” Water resources
Research, vol. 43, no. 2, Feb. 21, 2007. [Online]. Available:
https://doi.org/10.1029/2006WR005245
[12] H. V. de Vyver and G. R. Demarée, “Construction of
intensity–duration–frequency (idf) curves for precipitation at
lubumbashi, congo, under the hypothesis of inadequate data,”
Hydrological Sciences Journal, vol. 55, no. 4, Nov. 02, 2009. [Online].
Available: https://doi.org/10.1080/02626661003747390
[13] H. Tyralis and A. Langousis, “Estimation of
intensity–duration–frequency curves using max-stable processes,”
Stochastic Environmental Research and Risk Assessment, vol. 33,
Jul. 06, 2018. [Online]. Available: https://doi.org/10.1007/
s00477-018-1577-2
[14] C. Cheng-lung, “Rainfall intensity duration frequency formulas,”
Journal of Hydraulic Engineering, vol. 109, no. 12, Dec. 01, 1983.
[Online]. Available: https://doi.org/10.1061/(ASCE)0733-9429(1983)
109:12(1603)
[15] S. Urbanek, M. Vargas, and R. Díaz-Granados, “Curvas sintéticas
regionalizadas de intensidad-duración-frecuencia para colombia.”
in Seminario de Hidráulica e Hidrología, Sociedad Colombiana de
Ingenieros, Cali, Colombia, 1998, pp. 119–128.
[16] E. G. Pulgarín-Dávila, “Fórmulas regionales para la estimación
de curvas intensidad - frecuencia - duración basadas en las
propiedades de escala de la lluvia (región andina colombiana),”
45
E. Rodríguez et al., Revista Facultad de Ingeniería, Universidad de Antioquia, No. 110, pp. 31-47, 2024
M.S. thesis, Departamento de Geociencias y Medio Ambiente,
Universidad Nacional, Medellín, Colombia, 2009.
[17] Estudio de Actualización y Análisis Comparativo de las Curvas
Intensidad, Duración y Frecuencia Disponibles en el IDEAM, Convenio
Interadministrativo No. 113 de 2016, IDEAM-UNAL, Bogotá, D.C,
2016.
[18] T. Dinku, F. Ruiz, S. J. Connor, and P. Ceccato, “Validation and
intercomparison of satellite rainfall estimates over colombia,”
Journal of Applied Meteorology and Climatology, vol. 49, no. 5, May. 01,
2010. [Online]. Available: https://doi.org/10.1175/2009JAMC2260.1
[19] M. Ma, H. Wang, P. Jia, G. Tang, D. Wang, and et al., “Application
of the gpm-imerg products in flash flood warning: A case study
in yunnan, china,” Remote Sensing, vol. 12, no. 12, Jun. 15, 2020.
[Online]. Available: https://doi.org/10.3390/rs12121954
[20] W. Guachamín, S. Páez-Bimos, and N. Horna, “Evaluation of imerg
v03 and tmpa v7 products in the detection of flood events in cañar
river basin,” Revista Politécnica, vol. 42, no. 2, Nov-Jan. 2019.
[Online]. Available: http:https://tinyurl.com/mvm8ubst
[21] R. Zubieta1, A. Getirana, J. C. Espinoza, W. Lavado-Casimiro, and
L. Aragon, “Hydrological modeling of the peruvian–ecuadorian
amazon basin using gpm-imerg satellite-based precipitation
dataset,” Communications Week, vol. 21, no. 7, Jul. 14, 2017.
[Online]. Available: https://doi.org/10.5194/hess-21-3543-2017
[22] R. Junqueira, M. R. Viola, J. da S. Amorim, C. Camargos, and
C. R. de Mello, “Hydrological modeling using remote sensing
precipitation data in a brazilian savanna basin,” Journal of South
American Earth Sciences, vol. 115, Mar. 13, 2022. [Online]. Available:
https://doi.org/10.1016/j.jsames.2022.103773
[23] F. Satgé, D. Ruelland, M. P. Bonnet, J. Molina, and R. Pillco,
“Consistency of satellite-based precipitation products in space
and over time compared with gauge observations and snow-
hydrological modelling in the lake titicaca region,” Hydrology and
Earth System Sciences, vol. 23, no. 1, Jan. 14, 2019. [Online].
Available: https://doi.org/10.5194/hess-23-595-2019
[24] E. Rodríguez, I. Sánchez, N. Duque, P. Arboleda, C. Vega, and et
al., “Combined use of local and global hydro meteorological data
with hydrological models for water resources management in the
magdalena - cauca macro basin – colombia,” Communications Week,
vol. 34, Mar. 19, 2019. [Online]. Available: https://doi.org/10.1007/
s11269-019-02236-5
[25] P. López-López, W. W. Immerzeel, E. A. Rodríguez-Sandoval,
G. Sterk, and J. Schellekens, “Spatial downscaling of satellite-based
precipitation and its impact on discharge simulations in the
magdalena river basin in colombia,” Communications Week, vol. 6,
Jun. 08, 2018. [Online]. Available: https://doi.org/10.3389/feart.
2018.00068
[26] E. Zorzetto and M. Marani, “Downscaling of rainfall extremes
from satellite observations,” Water Resources Research, vol. 5,
no. 1, Nov. 26, 2018. [Online]. Available: https://doi.org/10.1029/
2018WR022950
[27] M. Zambrano-Bigiarini, C. Soto-Escobar, and O. M. Baez-Villanueva,
“Spatially-distributed idf curves for center-southern chile using
imerg,” in EGU General Assembly 2020, Munich, Alemania, 2020.
[28] M. Noor, T. Ismail, S. Shahid, M. Asaduzzaman, and
A. Dewan, “Evaluating intensity-duration-frequency (idf) curves
of satellite-based precipitation datasets in peninsular malaysia,”
Atmospheric Research, vol. 248, no. 19, Aug. 18, 2020. [Online].
Available: https://doi.org/10.1016/j.atmosres.2020.105203
[29] K. Venkatesh, R. Maheswaran, and J. Devacharan, “Framework for
developing idf curves using satellite precipitation: a case study using
gpm-imerg v6 data,” Earth Science Informatics, vol. 15, Oct. 02, 2021.
[Online]. Available: https://doi.org/10.1007/s12145-021-00708-0
[30] M. Ombadi, P. Nguyen, S. Sorooshian, and K. Hsu, “Developing
intensity-duration-frequency (idf) curves from satellite-based
precipitation: Methodology and evaluation,” Water Resources
Research, vol. 54, no. 10, Sep. 04, 2018. [Online]. Available:
https://doi.org/10.1029/2018WR022929
[31] G. J. Huffman, E. F. Stocker, D. T. Bolvin, E. J. Nelkin, and
J. Tan. (2019) Gpm imerg final precipitation l3 half hourly 0.1
degree x 0.1 degree v06 (gpm-3imerghh). [Online]. Available:
https://doi.org/10.5067/GPM/IMERG/3B-HH/06
[32] R. K. Pradhan, Y. Markonis, M. R. Vargas-Godoy, A. Villalba-Pradas,
K. M. Andreadis, and et al., “Review of gpm imerg performance: A
global perspective,” Remote Sensing of Environment, vol. 268, Oct. 12,
2021. [Online]. Available: https://doi.org/10.1016/j.rse.2021.112754
[33] A. Almagro, P. T. Sanches-Oliveira, and L. Brocca, “Assessment of
bottom-up satellite rainfall products on estimating river discharge
and hydrologic signatures in brazilian catchments,” Journal of
Hydrology, vol. 603, Agu. 28, 2021. [Online]. Available: https:
//doi.org/10.1016/j.jhydrol.2021.126897
[34] J. R. Rozante, D. A. Vila, J. Barboza-Chiquetto, A. D. A. Fernandes,
and D. Souza-Alvim, “Evaluation of trmm/gpm blended daily
products over brazil,” Remote Sensing, vol. 10, no. 6, Apr. 11, 2018.
[Online]. Available: https://doi.org/10.3390/rs10060882
[35] E. da S. Freitas, V. H. R. Coelho, Y. Xuan, D. de C. D. Melo, A. N.
Gadelha, and et al., “The performance of the imerg satellite-based
product in identifying sub-daily rainfall events and their properties,”
Journal of Hydrology, vol. 589, May. 28, 2020. [Online]. Available:
https://doi.org/10.1016/j.jhydrol.2020.125128
[36] Y. Rojas, J. R. Minder, L. S. Campbell, A. Massmann, and
R. Garreaud, “Assessment of gpm imerg satellite precipitation
estimation and its dependence on microphysical rain regimes
over the mountains of south-central chile,” Atmospheric Research,
vol. 253, no. 19, Jan. 04, 2021. [Online]. Available: https:
//doi.org/10.1016/j.atmosres.2021.105454
[37] S. Palomino-Ángel, J. A. Anaya-Acevedo, and B. A. Botero,
“Evaluation of 3b42v7 and imerg daily-precipitation products for
a very high-precipitation region in northwestern south america,”
Communications Week, vol. 217, Oct. 23, 2018. [Online]. Available:
https://doi.org/10.1016/j.atmosres.2018.10.012
[38] D. Delgado, M. Sadaoui, W. Ludwig, and W. Méndez,
“Spatio-temporal assessment of rainfall erosivity in ecuador
based on rusle using satellite-based high frequency gpm-imerg
precipitation data,” Communications Week, vol. 219, Aug. 21, 2022.
[Online]. Available: https://doi.org/10.1016/j.catena.2022.106597
[39] B. Manz, S. Páez-Bimos, N. Horna, W. Buytaert, B. Ochoa-Tocachi,
and et al., “Comparative ground validation of imerg and tmpa at
variable spatiotemporal scales in the tropical andes,” Journal of
Hydrometeorology, vol. 18, no. 9, Sep. 01, 2017. [Online]. Available:
https://doi.org/10.1175/JHM-D-16-0277.1
[40] A. Kyaw-Kyaw, S. Shahid, and X. Wang, “Remote sensing for
development of rainfall intensity–duration–frequency curves at
ungauged locations of yangon, myanmar,” Water, vol. 14, no. 11, May.
23, 2022. [Online]. Available: https://doi.org/10.3390/w14111699
[41] S. OH, U. Foelsche, G. Kirchengast, J. Fuchsberger, J. Tan, and et
al., “Evaluation of gpm imerg early, late, and final rainfall estimates
using wegenernet gauge data in southeastern austria,” Hydrology
and Earth System Sciences, vol. 21, no. 12, Nov. 13, 2017. [Online].
Available: https://doi.org/10.5194/hess-21-6559-2017
[42] F. F. Maghsood, H. Hashemi, S. Hasan-Hosseini, and R. Berndtsson,
“Ground validation of gpm imerg precipitation products over iran,”
Remote Sensing, vol. 12, no. 1, Dec. 18, 2019. [Online]. Available:
https://doi.org/10.3390/rs12010048
[43] I. W. Andi-Yuda1, R. Prasetia1, A. R. As-syakur, T. Osawa, and
Masahiko-Naga, “An assessment of imerg rainfall products over
bali at multiple time scale,” Environmental Sciences, vol. 153, no. 12,
Feb. 17, 2020. [Online]. Available: https://doi.org/10.1051/e3sconf/
202015302001
[44] A. Navarro, E. García-Ortega, A. Merino, J. L. Sánchez, and F. J.
Tapiador, “Orographic biases in imerg precipitation estimates in
the ebro river basin (spain): The effects of rain gauge density and
altitude,” Atmospheric Research, vol. 244, no. 19, May. 22, 2020.
[Online]. Available: https://doi.org/10.1016/j.atmosres.2020.105068
[45] S. Moazami and M. R. Najafi, “A comprehensive evaluation of
gpm-imerg v06 and mrms with hourly ground-based precipitation
observations across canada,” Journal of Hydrology, vol. 594, Dec.
20, 2020. [Online]. Available: https://doi.org/10.1016/j.jhydrol.2020.
125929
[46] M. Arshad, X. Ma, J. Yin, W. Ullah, G. Ali, and et al., “Evaluation of
gpm-imerg and trmm-3b42 precipitation products over pakistan,”
46
M.S. thesis, Departamento de Geociencias y Medio Ambiente,
Universidad Nacional, Medellín, Colombia, 2009.
[17] Estudio de Actualización y Análisis Comparativo de las Curvas
Intensidad, Duración y Frecuencia Disponibles en el IDEAM, Convenio
Interadministrativo No. 113 de 2016, IDEAM-UNAL, Bogotá, D.C,
2016.
[18] T. Dinku, F. Ruiz, S. J. Connor, and P. Ceccato, “Validation and
intercomparison of satellite rainfall estimates over colombia,”
Journal of Applied Meteorology and Climatology, vol. 49, no. 5, May. 01,
2010. [Online]. Available: https://doi.org/10.1175/2009JAMC2260.1
[19] M. Ma, H. Wang, P. Jia, G. Tang, D. Wang, and et al., “Application
of the gpm-imerg products in flash flood warning: A case study
in yunnan, china,” Remote Sensing, vol. 12, no. 12, Jun. 15, 2020.
[Online]. Available: https://doi.org/10.3390/rs12121954
[20] W. Guachamín, S. Páez-Bimos, and N. Horna, “Evaluation of imerg
v03 and tmpa v7 products in the detection of flood events in cañar
river basin,” Revista Politécnica, vol. 42, no. 2, Nov-Jan. 2019.
[Online]. Available: http:https://tinyurl.com/mvm8ubst
[21] R. Zubieta1, A. Getirana, J. C. Espinoza, W. Lavado-Casimiro, and
L. Aragon, “Hydrological modeling of the peruvian–ecuadorian
amazon basin using gpm-imerg satellite-based precipitation
dataset,” Communications Week, vol. 21, no. 7, Jul. 14, 2017.
[Online]. Available: https://doi.org/10.5194/hess-21-3543-2017
[22] R. Junqueira, M. R. Viola, J. da S. Amorim, C. Camargos, and
C. R. de Mello, “Hydrological modeling using remote sensing
precipitation data in a brazilian savanna basin,” Journal of South
American Earth Sciences, vol. 115, Mar. 13, 2022. [Online]. Available:
https://doi.org/10.1016/j.jsames.2022.103773
[23] F. Satgé, D. Ruelland, M. P. Bonnet, J. Molina, and R. Pillco,
“Consistency of satellite-based precipitation products in space
and over time compared with gauge observations and snow-
hydrological modelling in the lake titicaca region,” Hydrology and
Earth System Sciences, vol. 23, no. 1, Jan. 14, 2019. [Online].
Available: https://doi.org/10.5194/hess-23-595-2019
[24] E. Rodríguez, I. Sánchez, N. Duque, P. Arboleda, C. Vega, and et
al., “Combined use of local and global hydro meteorological data
with hydrological models for water resources management in the
magdalena - cauca macro basin – colombia,” Communications Week,
vol. 34, Mar. 19, 2019. [Online]. Available: https://doi.org/10.1007/
s11269-019-02236-5
[25] P. López-López, W. W. Immerzeel, E. A. Rodríguez-Sandoval,
G. Sterk, and J. Schellekens, “Spatial downscaling of satellite-based
precipitation and its impact on discharge simulations in the
magdalena river basin in colombia,” Communications Week, vol. 6,
Jun. 08, 2018. [Online]. Available: https://doi.org/10.3389/feart.
2018.00068
[26] E. Zorzetto and M. Marani, “Downscaling of rainfall extremes
from satellite observations,” Water Resources Research, vol. 5,
no. 1, Nov. 26, 2018. [Online]. Available: https://doi.org/10.1029/
2018WR022950
[27] M. Zambrano-Bigiarini, C. Soto-Escobar, and O. M. Baez-Villanueva,
“Spatially-distributed idf curves for center-southern chile using
imerg,” in EGU General Assembly 2020, Munich, Alemania, 2020.
[28] M. Noor, T. Ismail, S. Shahid, M. Asaduzzaman, and
A. Dewan, “Evaluating intensity-duration-frequency (idf) curves
of satellite-based precipitation datasets in peninsular malaysia,”
Atmospheric Research, vol. 248, no. 19, Aug. 18, 2020. [Online].
Available: https://doi.org/10.1016/j.atmosres.2020.105203
[29] K. Venkatesh, R. Maheswaran, and J. Devacharan, “Framework for
developing idf curves using satellite precipitation: a case study using
gpm-imerg v6 data,” Earth Science Informatics, vol. 15, Oct. 02, 2021.
[Online]. Available: https://doi.org/10.1007/s12145-021-00708-0
[30] M. Ombadi, P. Nguyen, S. Sorooshian, and K. Hsu, “Developing
intensity-duration-frequency (idf) curves from satellite-based
precipitation: Methodology and evaluation,” Water Resources
Research, vol. 54, no. 10, Sep. 04, 2018. [Online]. Available:
https://doi.org/10.1029/2018WR022929
[31] G. J. Huffman, E. F. Stocker, D. T. Bolvin, E. J. Nelkin, and
J. Tan. (2019) Gpm imerg final precipitation l3 half hourly 0.1
degree x 0.1 degree v06 (gpm-3imerghh). [Online]. Available:
https://doi.org/10.5067/GPM/IMERG/3B-HH/06
[32] R. K. Pradhan, Y. Markonis, M. R. Vargas-Godoy, A. Villalba-Pradas,
K. M. Andreadis, and et al., “Review of gpm imerg performance: A
global perspective,” Remote Sensing of Environment, vol. 268, Oct. 12,
2021. [Online]. Available: https://doi.org/10.1016/j.rse.2021.112754
[33] A. Almagro, P. T. Sanches-Oliveira, and L. Brocca, “Assessment of
bottom-up satellite rainfall products on estimating river discharge
and hydrologic signatures in brazilian catchments,” Journal of
Hydrology, vol. 603, Agu. 28, 2021. [Online]. Available: https:
//doi.org/10.1016/j.jhydrol.2021.126897
[34] J. R. Rozante, D. A. Vila, J. Barboza-Chiquetto, A. D. A. Fernandes,
and D. Souza-Alvim, “Evaluation of trmm/gpm blended daily
products over brazil,” Remote Sensing, vol. 10, no. 6, Apr. 11, 2018.
[Online]. Available: https://doi.org/10.3390/rs10060882
[35] E. da S. Freitas, V. H. R. Coelho, Y. Xuan, D. de C. D. Melo, A. N.
Gadelha, and et al., “The performance of the imerg satellite-based
product in identifying sub-daily rainfall events and their properties,”
Journal of Hydrology, vol. 589, May. 28, 2020. [Online]. Available:
https://doi.org/10.1016/j.jhydrol.2020.125128
[36] Y. Rojas, J. R. Minder, L. S. Campbell, A. Massmann, and
R. Garreaud, “Assessment of gpm imerg satellite precipitation
estimation and its dependence on microphysical rain regimes
over the mountains of south-central chile,” Atmospheric Research,
vol. 253, no. 19, Jan. 04, 2021. [Online]. Available: https:
//doi.org/10.1016/j.atmosres.2021.105454
[37] S. Palomino-Ángel, J. A. Anaya-Acevedo, and B. A. Botero,
“Evaluation of 3b42v7 and imerg daily-precipitation products for
a very high-precipitation region in northwestern south america,”
Communications Week, vol. 217, Oct. 23, 2018. [Online]. Available:
https://doi.org/10.1016/j.atmosres.2018.10.012
[38] D. Delgado, M. Sadaoui, W. Ludwig, and W. Méndez,
“Spatio-temporal assessment of rainfall erosivity in ecuador
based on rusle using satellite-based high frequency gpm-imerg
precipitation data,” Communications Week, vol. 219, Aug. 21, 2022.
[Online]. Available: https://doi.org/10.1016/j.catena.2022.106597
[39] B. Manz, S. Páez-Bimos, N. Horna, W. Buytaert, B. Ochoa-Tocachi,
and et al., “Comparative ground validation of imerg and tmpa at
variable spatiotemporal scales in the tropical andes,” Journal of
Hydrometeorology, vol. 18, no. 9, Sep. 01, 2017. [Online]. Available:
https://doi.org/10.1175/JHM-D-16-0277.1
[40] A. Kyaw-Kyaw, S. Shahid, and X. Wang, “Remote sensing for
development of rainfall intensity–duration–frequency curves at
ungauged locations of yangon, myanmar,” Water, vol. 14, no. 11, May.
23, 2022. [Online]. Available: https://doi.org/10.3390/w14111699
[41] S. OH, U. Foelsche, G. Kirchengast, J. Fuchsberger, J. Tan, and et
al., “Evaluation of gpm imerg early, late, and final rainfall estimates
using wegenernet gauge data in southeastern austria,” Hydrology
and Earth System Sciences, vol. 21, no. 12, Nov. 13, 2017. [Online].
Available: https://doi.org/10.5194/hess-21-6559-2017
[42] F. F. Maghsood, H. Hashemi, S. Hasan-Hosseini, and R. Berndtsson,
“Ground validation of gpm imerg precipitation products over iran,”
Remote Sensing, vol. 12, no. 1, Dec. 18, 2019. [Online]. Available:
https://doi.org/10.3390/rs12010048
[43] I. W. Andi-Yuda1, R. Prasetia1, A. R. As-syakur, T. Osawa, and
Masahiko-Naga, “An assessment of imerg rainfall products over
bali at multiple time scale,” Environmental Sciences, vol. 153, no. 12,
Feb. 17, 2020. [Online]. Available: https://doi.org/10.1051/e3sconf/
202015302001
[44] A. Navarro, E. García-Ortega, A. Merino, J. L. Sánchez, and F. J.
Tapiador, “Orographic biases in imerg precipitation estimates in
the ebro river basin (spain): The effects of rain gauge density and
altitude,” Atmospheric Research, vol. 244, no. 19, May. 22, 2020.
[Online]. Available: https://doi.org/10.1016/j.atmosres.2020.105068
[45] S. Moazami and M. R. Najafi, “A comprehensive evaluation of
gpm-imerg v06 and mrms with hourly ground-based precipitation
observations across canada,” Journal of Hydrology, vol. 594, Dec.
20, 2020. [Online]. Available: https://doi.org/10.1016/j.jhydrol.2020.
125929
[46] M. Arshad, X. Ma, J. Yin, W. Ullah, G. Ali, and et al., “Evaluation of
gpm-imerg and trmm-3b42 precipitation products over pakistan,”
46
E. Rodríguez et al., Revista Facultad de Ingeniería, Universidad de Antioquia, No. 110, pp. 31-47, 2024
Atmospheric Research, vol. 249, Oct. 24, 2020. [Online]. Available:
https://doi.org/10.1016/j.atmosres.2020.105341
[47] S. M. Vallejo-Bernal, V. Urrea, J. M. Bedoya-Soto, A. O. D. Posada,
and et al., “Ground validation of trmm 3b43 v7 precipitation
estimates over colombia. part i: Monthly and seasonal timescales,”
International Jornal of Climatology, vol. 41, no. 1, May. 14, 2020.
[Online]. Available: https://doi.org/10.1002/joc.6640
[48] S. Blenkinsop, H. J. Fowler, R. Barbero, S. C. Chan, S. B.
Guerreiro, E. Kendon, G. Lenderink, and et al., “The intense
project: using observations and models to understand the past,
present and future of sub-daily rainfall extremes,” Advances in
Science and Research, vol. 15, Jun. 05, 2018. [Online]. Available:
https://doi.org/10.5194/asr-15-117-2018
[49] O. D. Álvarez Villa, J. I. Vélez, and G. Poveda, “Improved long-term
mean annual rainfall fields for colombia,” International Journal
of Climatology, vol. 31, no. 14, Oct. 07, 2010. [Online]. Available:
https://doi.org/10.1002/joc.2232
[50] G. Poveda, “La hidroclimatología de colombia: una síntesis desde la
escala inter-decadal hasta la escala diurna,” Revista de la Academia
Colombiana de Ciencias Exactas, Fìsicas y Naturales, vol. 28, no. 107,
Jun. 2004. [Online]. Available: https://tinyurl.com/4wsv9fec
[51] O. Mesa, L. F. Carvajal, and G. Poveda, Introducción al clima de
Colombia. Medellín, Colombia: Universidad Nacional de Colombia,
Facultad de Minas, 1997.
[52] J. Pabón, J. Eslava, and R. Gómez, “Características de gran escala
del clima de la américa tropical, 4, 47-53,” Meteorol, vol. 4, pp. 47–53,
2001.
[53] DHIME. (2021) Consulta y descarga de datos hidrometeorológicos.
IDEAM. [Online]. Available: https://dhime.ideam.gov.co/webgis/
home/
[54] G. N. repository. (2021) Imerg v06b final time series of rainfall
intensity at 0.5 hintervals in (mm/hr) and at a resolution of 0.10°.
NASA Official. [Online]. Available: https://giovanni.gsfc.nasa.gov/
giovanni/
[55] Z. Li, D. Yang, B. Gao, Y. Jiao, Y. Hong, and et al., “Multiscale
hydrologic applications of the latest satellite precipitation products
in the yangtze river basin using a distributed hydrologic model,”
Journal of Hydrometeorology, vol. 16, no. 1, Feb. 01, 2015. [Online].
Available: https://doi.org/10.1175/JHM-D-14-0105.1
[56] B. Peng, J. Shi, W. Ni-Meister, T. Zhao, and D. Ji, “Evaluation
of trmm multisatellite precipitation analysis (tmpa) products and
their potential hydrological application at an arid and semiarid
basin in china,” IEEE Journal of Selected Topics in Applied Earth
Observations and Remote, vol. 7, no. 9, Sep. 2014. [Online]. Available:
https://doi.org/10.1109/JSTARS.2014.2320756
[57] X. Guan, J. Zhang, Q. Yang, X. Tang, C. Liu, and et al., “Evaluation
of precipitation products by using multiple hydrological models
over the upper yellow river basin, china,” Remote Sensing, vol. 12,
no. 24, Dec. 06, 2020. [Online]. Available: https://doi.org/10.3390/
rs12244023
[58] Z. Wang, J. Chen, C. Lai, R. Zhong, X. Chen, and et al., “Hydrologic
assessment of the tmpa 3b42-v7 product in a typical alpine
and gorge region: the lancang river basin, china,” Hydrology
Research, vol. 49, no. 6, Dec. 01, 2018. [Online]. Available:
https://doi.org/10.2166/nh.2018.024
[59] B. Zhu, Y. Huang, Z. Zhang, R. Kong, J. Tian, and et al., “Evaluation
of tmpa satellite precipitation in driving vic hydrological model over
the upper yangtze river basin,” Water, vol. 12, no. 11, Nov. 16, 2020.
[Online]. Available: https://doi.org/10.3390/w12113230
[60] R. J. Donaldson, R. M. Dyer, and M. J. Kraus, “An objective evaluator
of techniques for predicting severe weather events,” in Preprints
Ninth Conference on Severe Local Storms, Norman, Amer, 1975.
[61] C. A. D. III, R. Davis-Jones, and D. L. Keller, “On summary measures
of skill in rare event forecasting based on contingency tables,”
Weather and Forecasting, vol. 5, no. 4, 1990. [Online]. Available: https:
//doi.org/10.1175/1520-0434(1990)005<0576:OSMOSI>2.0.CO;2
47
Atmospheric Research, vol. 249, Oct. 24, 2020. [Online]. Available:
https://doi.org/10.1016/j.atmosres.2020.105341
[47] S. M. Vallejo-Bernal, V. Urrea, J. M. Bedoya-Soto, A. O. D. Posada,
and et al., “Ground validation of trmm 3b43 v7 precipitation
estimates over colombia. part i: Monthly and seasonal timescales,”
International Jornal of Climatology, vol. 41, no. 1, May. 14, 2020.
[Online]. Available: https://doi.org/10.1002/joc.6640
[48] S. Blenkinsop, H. J. Fowler, R. Barbero, S. C. Chan, S. B.
Guerreiro, E. Kendon, G. Lenderink, and et al., “The intense
project: using observations and models to understand the past,
present and future of sub-daily rainfall extremes,” Advances in
Science and Research, vol. 15, Jun. 05, 2018. [Online]. Available:
https://doi.org/10.5194/asr-15-117-2018
[49] O. D. Álvarez Villa, J. I. Vélez, and G. Poveda, “Improved long-term
mean annual rainfall fields for colombia,” International Journal
of Climatology, vol. 31, no. 14, Oct. 07, 2010. [Online]. Available:
https://doi.org/10.1002/joc.2232
[50] G. Poveda, “La hidroclimatología de colombia: una síntesis desde la
escala inter-decadal hasta la escala diurna,” Revista de la Academia
Colombiana de Ciencias Exactas, Fìsicas y Naturales, vol. 28, no. 107,
Jun. 2004. [Online]. Available: https://tinyurl.com/4wsv9fec
[51] O. Mesa, L. F. Carvajal, and G. Poveda, Introducción al clima de
Colombia. Medellín, Colombia: Universidad Nacional de Colombia,
Facultad de Minas, 1997.
[52] J. Pabón, J. Eslava, and R. Gómez, “Características de gran escala
del clima de la américa tropical, 4, 47-53,” Meteorol, vol. 4, pp. 47–53,
2001.
[53] DHIME. (2021) Consulta y descarga de datos hidrometeorológicos.
IDEAM. [Online]. Available: https://dhime.ideam.gov.co/webgis/
home/
[54] G. N. repository. (2021) Imerg v06b final time series of rainfall
intensity at 0.5 hintervals in (mm/hr) and at a resolution of 0.10°.
NASA Official. [Online]. Available: https://giovanni.gsfc.nasa.gov/
giovanni/
[55] Z. Li, D. Yang, B. Gao, Y. Jiao, Y. Hong, and et al., “Multiscale
hydrologic applications of the latest satellite precipitation products
in the yangtze river basin using a distributed hydrologic model,”
Journal of Hydrometeorology, vol. 16, no. 1, Feb. 01, 2015. [Online].
Available: https://doi.org/10.1175/JHM-D-14-0105.1
[56] B. Peng, J. Shi, W. Ni-Meister, T. Zhao, and D. Ji, “Evaluation
of trmm multisatellite precipitation analysis (tmpa) products and
their potential hydrological application at an arid and semiarid
basin in china,” IEEE Journal of Selected Topics in Applied Earth
Observations and Remote, vol. 7, no. 9, Sep. 2014. [Online]. Available:
https://doi.org/10.1109/JSTARS.2014.2320756
[57] X. Guan, J. Zhang, Q. Yang, X. Tang, C. Liu, and et al., “Evaluation
of precipitation products by using multiple hydrological models
over the upper yellow river basin, china,” Remote Sensing, vol. 12,
no. 24, Dec. 06, 2020. [Online]. Available: https://doi.org/10.3390/
rs12244023
[58] Z. Wang, J. Chen, C. Lai, R. Zhong, X. Chen, and et al., “Hydrologic
assessment of the tmpa 3b42-v7 product in a typical alpine
and gorge region: the lancang river basin, china,” Hydrology
Research, vol. 49, no. 6, Dec. 01, 2018. [Online]. Available:
https://doi.org/10.2166/nh.2018.024
[59] B. Zhu, Y. Huang, Z. Zhang, R. Kong, J. Tian, and et al., “Evaluation
of tmpa satellite precipitation in driving vic hydrological model over
the upper yangtze river basin,” Water, vol. 12, no. 11, Nov. 16, 2020.
[Online]. Available: https://doi.org/10.3390/w12113230
[60] R. J. Donaldson, R. M. Dyer, and M. J. Kraus, “An objective evaluator
of techniques for predicting severe weather events,” in Preprints
Ninth Conference on Severe Local Storms, Norman, Amer, 1975.
[61] C. A. D. III, R. Davis-Jones, and D. L. Keller, “On summary measures
of skill in rare event forecasting based on contingency tables,”
Weather and Forecasting, vol. 5, no. 4, 1990. [Online]. Available: https:
//doi.org/10.1175/1520-0434(1990)005<0576:OSMOSI>2.0.CO;2
47