Rain4pe

MARCH 2022 FERNANDEZ-PALOMINO ET AL.
309
A Novel High-Resolution Gridded Precipitation Dataset for Peruvian and

Ecuadorian Watersheds: Development and Hydrological Evaluation
CARLOS ANTONIO FERNANDEZ-PALOMINO,a,b FRED F. HATTERMANN,a VALENTINA KRYSANOVA,a

ANASTASIA LOBANOVA,a FIORELLA VEGA-JÁCOME,c WALDO LAVADO,c WILLIAM SANTINI,d CESAR AYBAR,e AND
AXEL BRONSTERTb
a
Research Department II–Climate Resilience, Potsdam Institute for Climate Impact Research, Potsdam, Germany
b
Institute of Environmental Science and Geography, University of Potsdam, Potsdam, Germany
c
Hidrologı́a–Estudios e Investigaciones Hidrológicas, Servicio Nacional de Meteorologı́a e Hidrologı́a del Perú, Lima, Peru
d
Institut de Recherche pour le Développement, Laboratoire GET (IRD, CNRS, UPS, CNES), Toulouse, France
e
Department of Geoinformatics–Z_GIS, University of Salzburg, Salzburg, Austria
(Manuscript received 2 December 2020, in final form 20 November 2021)
ABSTRACT: A novel approach for estimating precipitation patterns is developed here and applied to generate a new hydro-
logically corrected daily precipitation dataset, called RAIN4PE (Rain for Peru and Ecuador), at 0.18 spatial resolution for the
period 1981–2015 covering Peru and Ecuador. It is based on the application of 1) the random forest method to merge multi-
source precipitation estimates (gauge, satellite, and reanalysis) with terrain elevation, and 2) observed and modeled streamflow
data to first detect biases and second further adjust gridded precipitation by inversely applying the simulated results of the eco-
hydrological model SWAT (Soil and Water Assessment Tool). Hydrological results using RAIN4PE as input for the Peruvian
and Ecuadorian catchments were compared against the ones when feeding other uncorrected (CHIRP and ERA5) and gauge-
corrected (CHIRPS, MSWEP, and PISCO) precipitation datasets into the model. For that, SWAT was calibrated and validated
at 72 river sections for each dataset using a range of performance metrics, including hydrograph goodness of fit and flow dura-
tion curve signatures. Results showed that gauge-corrected precipitation datasets outperformed uncorrected ones for streamflow
simulation. However, CHIRPS, MSWEP, and PISCO showed limitations for streamflow simulation in several catchments drain-
ing into the Pacific Ocean and the Amazon River. RAIN4PE provided the best overall performance for streamflow simulation,
including flow variability (low, high, and peak flows) and water budget closure. The overall good performance of RAIN4PE as
input for hydrological modeling provides a valuable criterion of its applicability for robust countrywide hydrometeorological
applications, including hydroclimatic extremes such as droughts and floods.
SIGNIFICANCE STATEMENT: We developed a novel precipitation dataset RAIN4PE for Peru and Ecuador by merg-
ing multisource precipitation data (satellite, reanalysis, and ground-based precipitation) with terrain elevation using the ran-
dom forest method. Furthermore, RAIN4PE was hydrologically corrected using streamflow data in watersheds with
precipitation underestimation through reverse hydrology. The results of a comprehensive hydrological evaluation showed
that RAIN4PE outperformed state-of-the-art precipitation datasets such as CHIRP, ERA5, CHIRPS, MSWEP, and PISCO
in terms of daily and monthly streamflow simulations, including extremely low and high flows in almost all Peruvian and
Ecuadorian catchments. This underlines the suitability of RAIN4PE for hydrometeorological applications in this region.
Furthermore, our approach for the generation of RAIN4PE can be used in other data-scarce regions.
KEYWORDS: Amazon region; Complex terrain; South America; Streamflow; Precipitation; Hydrology; Water budget/balance;
Inverse methods; Mountain meteorology; Machine learning
1. Introduction and operational applications. Rain gauge observations are the

most used and}on a local scale}direct and accurate precipi-
Precipitation is an essential component of the water cycle
tation data sources. In addition, precipitation data can be
and reliable and accurate information about its spatiotempo-
derived from other sources such as rainfall radar stations, sat-
ral distribution is decisive for a multitude of scientific studies
ellites, reanalysis products, or based on merging procedures
(Sun et al. 2018). In many developing countries like Peru and
Denotes content that is immediately available upon publica- Ecuador, rain gauges are unevenly and sparsely distributed
tion as open access. (Scheel et al. 2011; Ochoa et al. 2014; Manz et al. 2016;
Hunziker et al. 2017; Aybar et al. 2020). These features limit
Supplemental information related to this paper is available at the precise estimation of spatial and temporal variability of
the Journals Online website: https://doi.org/10.1175/JHM-D-20- precipitation using only gauge-based measurements in the
0285.s1. tropical Andes.
Precipitation information derived from satellite data, e.g.,
Corresponding author: Carlos Antonio Fernandez-Palomino, CMORPH (Joyce et al. 2004), TMPA (Huffman et al. 2007),
cafpxl@gmail.com CHIRP (Funk et al. 2015a), and IMERG (Huffman et al.
DOI: 10.1175/JHM-D-20-0285.1
Ó 2022 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright
Policy (www.ametsoc.org/PUBSReuseLicenses).
Unauthenticated | Downloaded 03/30/22 07:57 PM UTC
310 JOURNAL OF HYDROMETEOROLOGY VOLUME 23
2019), with a high spatiotemporal resolution, near-global cov- hydrological modeling of the Andean Vilcanota River catch-
erage, and near-real-time availability have been produced in ment (Fernandez-Palomino et al. 2020), catchments draining
the last decades (see appendix A for abbreviations). These into the Pacific Ocean (Asurza-Véliz and Lavado-Casimiro
satellite-based precipitation products are promising alterna- 2020), and Peruvian catchments (Llauca et al. 2021). As the
tive sources for regions with sparse observations. However, method used to generate PISCO mainly corrects the biases of
previous studies for the Andes domain (Scheel et al. 2011; CHIRP using in situ precipitation data, the higher accuracy of
Kneis et al. 2014; Mantas et al. 2014; Ochoa et al. 2014; Zulka- precipitation estimates is constrained to gauged regions such
fli et al. 2014; Satgé et al. 2016; Chavez and Takahashi 2017; as the Pacific coast and the eastern and western slopes of the
Manz et al. 2017; Baez-Villanueva et al. 2018; Erazo et al. Andes of Peru (Aybar et al. 2020; Llauca et al. 2021). Hence,
2018) have reported that precipitation estimates from satel- the application of PISCO for Peruvian Amazon and trans-
lites can be erroneous or biased, and that ground-based data boundary river catchments is limited. This motivated us to
are often needed to reduce their bias. Furthermore, the cur- generate a new rainfall dataset for hydrometeorological appli-
rent short length of satellite records in this region constitutes cations at the national scale of Peru and Ecuador, exploiting
an important restriction for the use of most of these products the lessons learned from precipitation estimates derived not
for long-term applications. only from gauges and satellites but also from the state-of-the-
Reanalysis precipitation data, such as CFSR (Saha et al. art reanalysis ERA5. Indeed, ERA5 and CHIRP, which has
2010), JRA-55 (Kobayashi et al. 2015), and MERRA (Reichle long-term daily precipitation data available (1981–present)
et al. 2017), rely on uncertain parameterizations, and their spa- and hence appropriate for long-term hydrological applica-
tial resolution is too coarse to represent orographic precipita- tions, were used for the precipitation merging procedure in
tion (Beck et al. 2020b). Recently, the state-of-the-art climate this study. Moreover, terrain elevation which was reported to
reanalysis ERA5 (Hersbach et al. 2020) was released, which has be a key physical variable with a strong influence on precipita-
been shown to outperform previous reanalyses for precipitation tion patterns in mountainous regions (Chavez and Takahashi
estimation (Beck et al. 2019a; Tall et al. 2019; Xu et al. 2019a; 2017; Bhuiyan et al. 2019; Beck et al. 2020b) was considered
Fallah et al. 2020; Gleixner et al. 2020), and has shown accept- as an additional predictor variable.
able performance for hydrological modeling over North Amer- Besides sparseness and uncertainty of rainfall observations
ica (Tarek et al. 2020), the Amazon River basin (Towner et al. in complex tropical mountain ranges, in some of those regions
2019), and at the global scale (Alfieri et al. 2020). depositing fog and clouds may contribute significantly to pre-
In recent years, global merged precipitation products that cipitation, but cannot be recorded with conventional measure-
incorporate satellite and reanalysis information with gauge- ments. In páramos (grassland ecosystems extending from
based datasets such as CHIRPS (Funk et al. 2015a) and northern Peru to Venezuela and occurring between the tree
MSWEP (Beck et al. 2017, 2019b) have been published and are line and glaciers) and tropical montane cloud forest (TMCF)
available. Many studies worldwide have shown that these prod- such precipitation plays a key role in the water cycle as the
ucts have higher accuracy than precipitation estimates based on cloud/fog interception by the páramos/forest constitutes an
one source only (e.g., either satellite- or reanalysis-based precip- important water source to the system (Gomez-Peralta et al.
itation products) and have significant potential for hydrometeo- 2008; Bruijnzeel et al. 2011; Clark et al. 2014; Cárdenas et al.
rological studies (Bai and Liu 2018; Beck et al. 2019a; Wu et al. 2017; Strauch et al. 2017). Modeled contributions of cloud
2019; Xu et al. 2019b). CHIRPS has been used successfully to water varying from less than 5% of total precipitation in wet
understand the precipitation variability over the Andes (Segura areas to more than 75% in low-rainfall areas in TMCF were
et al. 2019) and Amazonia (Paccini et al. 2018; Espinoza et al. reported by Bruijnzeel et al. (2011). Fog water contribution of
2019; da Motta Paca et al. 2020). In South America, the accu- up to 30% of bulk precipitation (rainfall plus fog water) was
racy of merged precipitation products has been tested only in a estimated in tropical montane forests in the eastern Andes of
few studies using ground-based precipitation (Zambrano- Central Peru using fog gauges (Gomez-Peralta et al. 2008).
Bigiarini et al. 2017; Baez-Villanueva et al. 2018; Satgé et al. Cloud water contribution of up to 15% of streamflow was
2019) and hydrological modeling (Wongchuig Correa et al. reported for the montane Kosñypata catchment in the eastern
2017; Satgé et al. 2019). CHIRPS and MSWEP showed good Peruvian Andes using an isotopic mixing model (Clark et al.
performance for streamflow simulation in the Amazon River 2014). Fog water contribution of up to 28% of the total pre-
basin (Wongchuig Correa et al. 2017) and in catchments drain- cipitation to páramos in the Colombian Andes was measured
ing into Titicaca Lake (Satgé et al. 2019, 2020). To the best of using fog gauges (Cárdenas et al. 2017). To correct the under-
our knowledge, there are no case studies in the literature on estimation of precipitation by gridded precipitation products,
hydrological evaluation of CHIRPS and MSWEP in Peruvian adjustment of precipitation data for regions covered by cloud
and Ecuadorian watersheds, which is addressed in this study. forests has been proposed (Strauch et al. 2017) with reported
At the regional scale, recently a high-resolution (0.18) daily increases of up to 50% of the precipitation values in the
gridded precipitation dataset for Peru as part of PISCO data- WFDEI dataset (Weedon et al. 2014) required to improve
sets was developed (PISCO hereafter) (Aybar et al. 2020). streamflow simulation in the tropical montane watersheds.
PISCO is based on the merging of satellite estimates However, the cloud/fog water component is not repre-
(CHIRP) and ground-based observations. It is used by sented in the aforementioned precipitation data sources. This
SENAMHI for operational purposes in Peru for droughts and lacuna, together with the dearth of precipitation gauges could
floods monitoring at the national scale, and was applied for explain some of the poor hydrologic model performances and

MARCH 2022 FERNANDEZ-PALOMINO ET AL. 311
problems with water budget closure reported in previous 2. Study area and data
studies in páramo and/or montane catchments draining into
a. Study area
the Amazon River (Zulkafli et al. 2014; Zubieta et al. 2015;
Manz et al. 2016; Zubieta et al. 2017; Strauch et al. 2017; The study area covers Peru and Ecuador with elevation
Aybar et al. 2020). Thus, for reliable and accurate estimation ranging from 0 to 6518 m MSL (Fig. 1). The new precipitation
of precipitation in regions such as the TMCF and páramos, it dataset [Rain for Peru and Ecuador (RAIN4PE)] is generated
is important to consider the contribution of cloud/fog water to for the terrestrial land surface between 198S–28N and
the terrestrial hydrological system. 828–678W. The study area has complex hydroclimatic condi-
Correcting potential errors in gridded precipitation data- tions related to its variable climate zones and the Andes Cor-
sets for these areas requires application of other types of dillera, which acts as a topographic barrier between the cold
observations and estimates. Corrected estimates of precipi- and dry eastern Pacific and the warm and moist Amazon
tation using satellite soil moisture products have been region. The Andes divides the study area into three natural
derived in recent years (Brocca et al. 2013; Román-Cascón drainage basins (Fig. 1): (i) the Pacific basin (watersheds
et al. 2017; Brocca et al. 2019). However, the utility of located on the western side of the Andes that convey water to
these products could be limited due to their low accuracy the Pacific Ocean); (ii) Amazon basin (watersheds located on
in regions with dense forests (Brocca et al. 2020), such as the eastern side of the Andes that drain to Amazon River);
TMCF and rain forest areas. Streamflow observations, and (iii) Titicaca Lake basin (catchments draining into
which are spatially integrative and could be another source Titicaca Lake).
of data supplementing information from sparse rain In the region, the great spatial variability of precipitation
gauges, offer an additional method to infer precipitation patterns is modulated by the interplay among large-scale
patterns and evaluate precipitation datasets (Le Moine (e.g., latitudinal migration of Atlantic intertropical conver-
et al. 2015; Henn et al. 2018). In this study, we applied gence zone, South American monsoon systems, Hadley,
regional streamflow observations inversely to infer or cor- Walker cell, marine currents, Bolivian high) and local circula-
rect the precipitation input for the corresponding regional tion patterns (e.g., upslope and downslope moisture trans-
hydrological simulations. This approach has been termed port) and the complex Andean orography (Laraque et al.
“hydrology backwards” or “reverse hydrology” by Kirch- 2007; Tobar and Wyseure 2018; Segura et al. 2019; Espinoza
ner (2009) and has so far been applied in mountainous
et al. 2020). Furthermore, El Niño–Southern Oscillation
catchments like Rietholzbach in Switzerland (Teuling et al.
(ENSO) is a major modulator of hydroclimatology at interan-
2010), Alzette in Luxembourg (Krier et al. 2012), Schliefau
nual time scales along the Andes (Poveda et al. 2020). The
and Krems in Austria (Herrnegger et al. 2015), and the
study area hosts a diversity of ecosystems such as deserts,
Sierra Nevada mountain range of California (Henn et al.
punas (high mountain grasslands), páramos, glaciers, moun-
2015, 2018). These studies used a simple lumped hydrologi-
tain forests, TMCFs, and rain forests. From these, páramo
cal model to do reverse hydrology. In our case, we applied
and TMCF (Fig. 1) are ecosystems where an important cloud/
a process-based hydrological model to correct precipita-
fog water input to the system was reported (Gomez-Peralta
tion biases using streamflow data. We hypothesize that cor-
et al. 2008; Bruijnzeel et al. 2011; Clark et al. 2014; Cárdenas
rection of precipitation using streamflow data can improve
et al. 2017). This is an important precipitation source to con-
closing the observed water budget gap over complex tropi-
sider in hydrological modeling of páramo and montane water-
cal mountainous catchments such as páramo and montane
sheds, as it was carried out herein.
watersheds.
This study is the first attempt to generate a precipitation
dataset for Peru and Ecuador by merging different sources of b. Data
precipitation and correcting precipitation estimates through 1) GROUND-BASED PRECIPITATION DATA
reverse hydrology. Furthermore, we evaluate the applicability
of the precipitation dataset generated in this study, uncor- The precipitation data of a total of 804 precipitation gauges
rected precipitation datasets used for merging procedure with record length greater than ten years for the 1981–2015
(CHIRP and ERA5), and current state-of-the-art local period were used for this study (Fig. 1), out of which 587
(PISCO) and global (CHIRPS and MSWEP) merged precipi- (217) gauges have daily (only monthly) precipitation data.
tation products for hydrological modeling of Peruvian and The data were collected from different sources such as
Ecuadorian river catchments. This will demonstrate the effec- national hydrometeorological institutions and previous stud-
tiveness of the new methods combined here, and will help ies in the region. The data for Peru were obtained from the
illustrate the appropriateness of multiple precipitation data- Peruvian ANA (Autoridad Nacional del Agua) and Aybar
sets for the countrywide hydrometeorological applications et al. (2020); for Ecuador from Morán-Tejeda et al. (2016),
both in Peru and Ecuador. The objectives of this study are 1) Tamayo (2017), Tobar and Wyseure (2018); for Brazil from
to generate a high-spatial-resolution and hydrologically Xavier et al. (2016, 2017); and for Colombia from IDEAM
adjusted precipitation dataset for Peru and Ecuador, and 2) to (Instituto de Hidrologı́a, Meteorologı́a y Estudios Ambien-
assess and compare the applicability of this precipitation data tales). We used 587 (804) precipitation gauges with daily
and the current state-of-the-art uncorrected and merged pre- (monthly) data for the merging of precipitation datasets at
cipitation products for hydrological modeling. the daily (monthly) time step.

FIG. 1. (left) Study area and spatial distribution of precipitation gauges with record length greater than 10 years for
the 1981–2015 period used for the merging procedure. (right) Drainage systems, river networks, and streamflow sta-
tions used for hydrological model calibration based on the cascading calibration approach. Red polygons show the
gauged catchments with water budget imbalance where gridded precipitation datasets are corrected using streamflow
data through reverse hydrology. Nueva Loja station gauges the catchment “A”, San Sebastian (B), Francisco De Orel-
lana (C), Santiago (D), Borja (E), Shanao (F), Chazuta (G), Puerto Inca (H), and Lagarto (I). Boundaries of the
páramo and tropical montane cloud forest (TMCF) ecosystems were obtained from Helmer et al. (2019).
2) DISCHARGE DATA generated for Funk et al. (2015b). We selected CHIRP since it
has high spatial resolution and long-term (from 1981 onward)
Discharge data of 72 streamflow stations (Fig. 1) with
daily precipitation data, which is appropriate for long-term
record lengths ranging from one to 33 years for 1983–2015
hydrometeorological applications. ERA5 (Hersbach et al. 2020)
were obtained from different sources, such as the Peruvian
is the latest climate reanalysis dataset produced by the European
ANA and SENAMHI for catchments draining into the Pacific
Centre for Medium Weather Forecasts (ECMWF). Compared
Ocean and located in the Andes. For the Amazon lowland,
with its predecessor ERA-Interim (Dee et al. 2011) that became
data were obtained from the Critical Zone Observatory
operational in 2006, ERA5 is based on the ECMWF’s Integrated
HYBAM (Hydrogéochimie du Bassin Amazonien, www.so-
Forecasting System Cycle 41r2 which was operational in 2016.
hybam.org). This hydrological network has been operated by
ERA5 thus benefits from a decade worth of numerical weather
an international team from IRD (Institut de Recherche pour
prediction developments in model physics, core dynamics, and
le Développement; France), SENAMHI (Peru), INAMHI
data assimilation relative to ERA-Interim. Moreover, ERA5 has
(Instituto Nacional de Meteorologı́a e Hidrologı́a; Ecuador),
a much higher temporal and spatial resolution than previous
and the Brazilian ANA (Agência Nacional de Águas; Brazil)
global reanalyses. The hourly ERA5 precipitation data were
since 2003 (Armijos et al. 2013; Santini et al. 2019).
downloaded and aggregated to obtain daily time step records
matching the local gauge observations (from 0700 to 0700 local
3) GRIDDED PRECIPITATION DATA
time).
Table 1 presents the five precipitation datasets used in this To compare RAIN4PE against other gauge-corrected pre-
study. We used the non-gauge-corrected datasets (CHIRP and cipitation datasets besides the uncorrected ones (CHIRP and
ERA5) for the merging procedure to generate RAIN4PE data- ERA5), we selected three merged products (CHIRPS,
set. The satellite-based CHIRP precipitation dataset (Funk et al. MSWEP, and PISCO) widely used in data evaluation and
2015a) is obtained by considering infrared-based precipitation hydrometeorological applications in the region (Wongchuig
estimates and corresponding monthly precipitation climatology Correa et al. 2017; Paccini et al. 2018; Bhuiyan et al. 2019;

TABLE 1. List of gridded precipitation datasets used in this study. In uncorrected datasets, their temporal dynamics depend
entirely on satellite (S) or reanalysis (R) data, while in gauge-corrected datasets, their temporal dynamics depend at least partly on
gauge (G) data. In the spatial coverage column, “Global” means fully global coverage including oceans, while “Land” indicates that
the coverage is limited to the terrestrial land surface.
Dataset Data Spatial Spatial Temporal Temporal

(version) source(s) resolution coverage resolution coverage Reference
Non-gauge-corrected datasets
CHIRP (V2.0) S 0.058 Land, 508N/S Daily 1981–present Funk et al. (2015a)
ERA5 R 0.258 Global Hourly 1950–present Hersbach et al.
(2020)
Gauge-corrected datasets
CHIRPS (V2.0) G, S 0.058 Land, 508N/S Daily 1981–present Funk et al. (2015a)
MSWEP (V2.2) G, S, R 0.18 Global 3-hourly 1979–present Beck et al. (2017,
2019b)
PISCO (V2.1) G, S 0.18 Peru Daily 1981–2016 Aybar et al. (2020)
Espinoza et al. 2019; Satgé et al. 2019, 2020; Asurza-Véliz and 4) ADDITIONAL DATA
Lavado-Casimiro 2020; Baez-Villanueva et al. 2020; da Motta
In addition to various precipitation products, Table 2 pre-
Paca et al. 2020; Fernandez-Palomino et al. 2020; Llauca et al.
sents other datasets that were used for the hydrological model-
2021). CHIRPS (Funk et al. 2015a) and PISCO (Aybar et al.
ing process. The surface elevation data were used both for the
2020) are obtained by merging CHIRP and gauge estimates
merging procedure and setting up the hydrological model.
through deterministic and geostatistical interpolation meth-
ods. Finally, MSWEP is derived by optimally merging a range
of gauge, satellite, and reanalysis precipitation estimates,
3. Methods
where satellite and reanalysis datasets are merged using
weights for each one based on the coefficient of determination The framework of this study involves three main steps (Fig. 2):
between 3-day mean gauge- and grid-based precipitation time (i) merging procedure through a machine learning technique at
series (Beck et al. 2017, 2019b). The daily MSWEP precipita- the daily and monthly scales; (ii) calibration of model parameters
tion data were provided for this study. and hydrological adjustment through the reverse hydrology
TABLE 2. Data used for hydrological modeling.
Data type Resolution Description/source

Elevation 90 m Surface elevation (m MSL) from Multi-Error-Removed
Improved Terrain (MERIT; Yamazaki et al. 2017)
Land use 100 m Land use classification representative for the year 2015
obtained from Copernicus Global Land Service
(Buchhorn et al. 2019)
Soil 1000 m Soil parameters for SWAT based on the Harmonized World
Soil Database version 1.21 soil data (Abbaspour and
Ashraf Vaghefi 2019)
Soil thickness 1000 m Soil thickness data (Pelletier et al. 2016) were used to
implement variable soil thicknesses at hydrological
response units (HRUs)
Groundwater table depth 1000 m Groundwater table depth data (Fan et al. 2013) were used to
constrain soil thickness in shallow water tables across the
rain forest region
Temperature Daily/10 km (1981–2016) Gridded temperature (maximum and minimum) dataset for
Peru (Huerta et al. 2018) as provided by SENAMHI
(ftp://publi_dgh2:123456@ftp.senamhi.gob.pe/)
Solar radiation 3-hourly/10 km (1983–2018) Long-term monthly averages of solar radiation based on the
global surface solar radiation data (Tang et al. 2019; Tang
2019) were used
Evapotranspiration Daily/0.258 (1980–2020) Evapotranspiration data from the Global Land Evaporation
Amsterdam Model (GLEAM v3.5a; Miralles et al. 2011;
Martens et al. 2017)
Evapotranspiration 8-day/1 km (2000–14) Evapotranspiration data from the Moderate Resolution
Imaging Spectroradiometer Global Evaporation (MOD16;
Mu et al. 2011)

FIG. 2. Flowchart for (i) the generation of gridded precipitation dataset, (ii) hydrological model calibration and adjust-
ment of precipitation datasets, and (iii) hydrological evaluation. Here d (m) indicates the daily (monthly) time step,
BD(1), … ,(n) are buffer distances (distance from any point to all precipitation gauges), BCF is the bias correction factor,
OFs are the objective functions for hydrological model calibration, and GOFs are the goodness of fit measures. BCF is
optimized only over catchments with water budget imbalance. Note that for hydrological evaluation (step iii), the model
was rerun using the respective corrected precipitation data and optimum model parameters values with BCF set to 1.
concept; and (iii) evaluation of all precipitation products through to 0.18 spatial resolution applying the bilinear interpolation
hydrological modeling. (nearest neighbor) method.
a. Merging procedure 2) RANDOM FOREST MODELING TO COMBINE DIFFERENT
In this section, the merging procedure to obtain RAIN4PE DATA SOURCES
at 0.18 spatial resolution for the 1981–2015 period is described; In this study, the RF method (Breiman 2001) was applied
see Fig. 2 for a scheme. to produce a gridded precipitation dataset by merging multi-
ple precipitation sources (gauge, satellite, and reanalysis). RF
1) COVARIATES
has been used and proved recently to have similar or superior
For the merging procedure at the daily (monthly) scale, we performance in the interpolation of environmental variables
used daily (monthly) precipitation estimates of CHIRP and such as precipitation, temperature, and evapotranspiration
reanalysis ERA5, surface elevation (Yamazaki et al. 2017), compared to traditional spatial interpolation techniques, e.g.,
and buffer distances from observation points as covariates. regression kriging and inverse distance weighting (Hengl et al.
The latter is to account for geographical proximity effects in 2018; da Silva Júnior et al. 2019; Sekulić et al. 2020). Last, RF-
the prediction process using the random forest (RF) method based methodologies (Bhuiyan et al. 2019; Baez-Villanueva
as suggested by Hengl et al. (2018). The elevation is taken et al. 2020) to merge precipitation products with ground-
into account because it is a key physical variable with a strong based measurements were developed and applied successfully
influence on precipitation patterns (Chavez and Takahashi in data-scarce and complex terrain regions such as the
2017; Beck et al. 2020b). We selected these covariates: satellite Peruvian and Colombian Andes (Bhuiyan et al. 2019) and
precipitation, reanalysis precipitation, and elevation, all of them Chilean territory (Baez-Villanueva et al. 2020).
based on recent studies (Bhuiyan et al. 2019; Baez-Villanueva RF is a multivariate and nonparametric machine learning
et al. 2020; Beck et al. 2020b; Hong et al. 2021). To match the algorithm, in which the prediction is generated as an ensem-
0.18 spatial resolution of the final precipitation product, the ble estimate from a number of regression trees (Breiman
covariates with grid cell size , 0.18 (.0.18) were regridded 2001) as shown in Eq. (1):

2017). SWAT is one of the most widely used ecohydrological

1 N
f̂ (x) 5 Tb (x), (1) models in the world (Gassman et al. 2014; Tan et al. 2020),
N b51
and had been applied successfully already for ecohydrological
modeling of an Andean basin of Peru (Fernandez-Palomino
where f̂ (x) is the final prediction, b is the individual bootstrap
et al. 2020). SWAT is a process-oriented, semidistributed and
sample, N is the total number of trees, and Tb is the individual
time-continuous river basin model used to simulate hydrologi-
regression tree.
In RF, each tree is constructed from the random selection cal processes as well as vegetation dynamics, nutrients, pesti-
of covariates which ensures that trees are decorrelated with cides, and sediment loads within a basin (Arnold et al. 1998;
each other and a bootstrap sample of the observations (Brei- Neitsch et al. 2011). SWAT divides a basin into subbasins,
man 2001). The unsampled data, called out-of-bag, can be which are then further subdivided into hydrological response
used to test the prediction accuracy and the importance of units (HRUs) representing unique combinations of land use,
input variables, and thus no extra independent validation soil type, and slope classes (Neitsch et al. 2011). The water
dataset is needed (Breiman 2001). balance computation is performed at the HRU level consider-
We implemented RF using the R package randomForest ing four water storage types (snow, soil profile, and shallow
(Liaw and Wiener 2002) and the following RF parameters: 1) and deep aquifers), as follows:
the number of trees (set at 1000); 2) the number of predictor

N
variables randomly selected at each node (set at one-third of the DS 5 (BCF 3 P 2 WYLD 2 ET 2 GWL), (2)
number of variables, default value); 3) the minimum number of i51
observation in a tree’s terminal node (set at 5, default value);

where DS is the change in water storage (mm); N is the time
and 4) the out-of-bag portion to test the accuracy of the predic-
in days; and P, WYLD, ET, and GWL are the amount of pre-
tions (set at one-third of the total number of observations).
cipitation (mm), water yield (mm), evapotranspiration (mm),
These parameter values were successfully used in other studies
and groundwater losses (mm), respectively. BCF introduced
(Baez-Villanueva et al. 2020; Fox et al. 2020; Sekulić et al. 2020).
herein is the bias correction factor to infer the precipitation
In the merging procedure (see Fig. 2), an RF model was
trained using ground-based observations as the dependent fields from observed streamflow data.
variable and the selected covariates as predictor variables for In SWAT, flow routing in river channels can be computed
each day and month in the 1981–2015 period. The trained RF using the Muskingum or the variable storage method, consid-
models were then applied to covariates, yielding preliminary ering the flow velocity to be the same across the channel and
daily precipitation data (Pd0) and monthly precipitation data floodplain section (Neitsch et al. 2011). This approach has
(Pm1). Finally, Pd0 was corrected to match Pm1. For that, the been shown to be inefficient for flow routing in Amazon rivers
ratio of Pm1 over the monthly precipitation derived from Pd0 (Santini 2020), where flows are largely affected by floodplains
was computed on each grid cell for each month, and this ratio that act as reservoirs, causing significant flood peak delay and
was then applied to multiply the Pd0 values on the grid for attenuation (Paiva et al. 2011; Yamazaki et al. 2011; Santini
the month to generate the RF-based RAIN4PE dataset. This et al. 2015; Santini 2020). To exclude this limitation, Santini
correction was because the interpolation of precipitation pat- (2020) has implemented a new flow routing method for
terns at a monthly scale is more reliable and accurate than the SWAT to consider the river–floodplain dynamics, where the
daily interpolation (Aybar et al. 2020; He et al. 2020). associated floodplain of each river reach was treated as a sim-
ple storage model, as in other hydrological models (Paiva et al.
b. Hydrological modeling and adjustment of 2011; Yamazaki et al. 2011). This approach was used in our
precipitation datasets study.
This section describes the approaches applied for hydrologi-
cal model calibration and validation and hydrological adjust-
2) SWAT MODEL SETUP, CALIBRATION, AND VALIDATION
ment of precipitation datasets using streamflow data through
the reverse hydrology concept. The hydrological correction is The SWAT model was set up for Peruvian and Ecuadorian
applied only for nine catchments (Fig. 1) having a water budget catchments (total of 1 638 793 km2) based on the input data
imbalance due to underestimated streamflow in the simulations listed in Table 2. The model includes 2675 subcatchments and
with uncorrected precipitation inputs, as reported in previous 6843 HRUs. Channel cross-section parameters such as the
studies (Zulkafli et al. 2014; Zubieta et al. 2015, 2017; Strauch bankfull width (B) and channel depth (CHD) were estimated
et al. 2017). We applied the reverse hydrology using the Soil using geomorphologic equations based on upstream drainage
and Water Assessment Tool (SWAT; Arnold et al. 1998) model areas derived for Amazon rivers (Paiva et al. 2011). Floodplain
in which both the bias correction factor (BCF) for precipitation width is estimated by multiplying the bankfull width by a fac-
fields and model parameters were calibrated jointly (Fig. 2). tor (set at 5, default value). We assigned Manning’s n values of
0.03 (0.10) for channels (floodplains). The modified Soil Con-
1) SWAT MODEL
servation Service curve number, the Priestley–Taylor equa-
Hydrological simulations were performed with the SWAT tion, and the variable storage methods were used to simulate
2012 model (Arnold et al. 1998), updated for improved repre- surface runoff and infiltration, potential evapotranspiration,
sentation of tropical vegetation dynamics (Alemayehu et al. and river flow routing, respectively.

TABLE 3. Parameters and their ranges for model calibration for evapotranspiration (ET), streamflow (Q), and precipitation (P). In
the “Change type” column, R (V) refers to a relative (absolute) change of parameter values during the calibration. Parameter set 1
was applied for Andean catchments draining into the Pacific Ocean and Titicaca Lake and for Andean catchments upstream the
montane watersheds. Montane watersheds having a water budget imbalance were calibrated using parameter set 2. Catchments
downstream the montane watersheds were calibrated using parameter set 3. Note that BCF is applied only for catchments with water
budget closure problems to infer precipitation from streamflow data. See Neitsch et al. (2011) for detailed parameter definitions.
Set
Calibrated Change
Parameter Description (unit) output Range type 1 2 3
SOL_AWC Soil available water capacity (mm H2O/mm soil) ET [20.8, 0.8] R X
GW_REVAP Groundwater “revap” coefficient ET [0, 0.2] V X
SURLAG Surface runoff delay coefficient Q [0.1, 2] V X X X
GW_DELAY Groundwater delay time (days) Q [1, 100] V X X X
RCHRG_DP Deep aquifer percolation fraction Q [0, 1] V X X X
GWQMN Threshold for return flow from shallow aquifer (mm) Q [500, 1000] V X X X
ALPHA_BF Baseflow recession constant Q [0.01, 1] V X X X
CH_K2 Hydraulic conductivity of main channel (mm h21) Q [0, 50] V X
CHD Main channel depth (m) Q [20.1, 0.5] R X
FP_W_F Ratio of floodplain width over bankfull width Q [1, 5] V X
BCF Bias correction factor P, Q [0, 1] R X
The simulation period was from 1981 to 2015. The first two averaged to obtain the FDCsign to take into consideration the
years were considered for the model spinup. For the model cali- hydrological signatures for model calibration. The respective
bration, all flow data were used for stations with a record lower FDC segmentation represents peak flow events occurring
than 10 years, and for those with longer, two-third of the data rarely, quick runoff (due to snowmelt and/or rainfall), the
were used. In the latter case, the remaining flow data were used flashiness of a basin’s response, and the streamflow’s baseflow
for model validation (53 out of 72 streamflow stations). The components. The Borg multiobjective evolutionary algorithm
model calibration for each precipitation product was performed (Borg MOEA; Hadka and Reed 2013) was used to optimize
applying the multisite cascading calibration approach (Xue et al. the objective functions (maximization of lNSE and minimiza-
2016) in nine sequences (Fig. 1), where the calibrated discharge tion of FDCsign) with 1000 iterations as maximum. The Borg
from the upstream catchments was used as input for the down- MOEA parameterization was the same as in Fernandez-Palo-
stream. The model parameters and BCFs for each (sub)catch- mino et al. (2020). The parameters for ungauged catchments
ment were calibrated using the respective set of parameters (at HRU level) were obtained applying the spatial proximity
defined in Table 3 for Andean, montane, and lower Amazon approach (Guo et al. 2021) using the inverse distance weight-
catchments. Moreover, plant parameters were adopted from ing (Shepard 1968). For regionalization of parameters, donor
our previous study (Fernandez-Palomino et al. 2020). catchments (gauged) within a radius of 150 km were used to
The optimum values of model parameters and BCFs were avoid the influence of Amazonian catchments in the estima-
obtained through multiobjective calibration. For that, the tion of parameters for Andean basins draining into the Pacific
model was calibrated against observed discharge using the Ocean and Titicaca Lake.
Nash–Sutcliffe efficiency log (lNSE) and aggregated flow
duration curve signature (FDCsign) as objective functions (see 3) HYDROLOGICAL ADJUSTMENT OF PRECIPITATION
Table 4). We selected lNSE and FDCsign since these have DATASETS
been shown sufficient to test the model for simulating all
hydrograph aspects in the calibration (Fernandez-Palomino The optimum BCFs obtained for each catchment with
et al. 2020). Moreover, the application of FDC-based signa- water budget imbalance (Fig. 1) in the calibration procedure
tures provides more information about the hydrological were applied to the respective daily gridded precipitation data
behavior of the modeled basin (Yilmaz et al. 2008; Hracho- to obtain the hydrologically corrected daily precipitation data-
witz et al. 2014) and leads to better parameter identifiability, set (Fig. 2). For that, a continuous BCF map at 0.18 spatial res-
more accurate discharge simulation, and reduction of predic- olution was produced where grid cells within the respective
tive uncertainty (Yilmaz et al. 2008; Pokhrel and Yilmaz 2012; catchment retained the respective BCF, and for cells on the
Hrachowitz et al. 2014; Pfannerstill et al. 2014, 2017; Chilkoti boundary, the area-weighted BCFs were estimated. It is note-
et al. 2018; Fernandez-Palomino et al. 2020; Sahraei et al. worthy that applying the resulting BCF map to gridded pre-
2020). Following Chilkoti et al. (2018) and Fernandez-Palo- cipitation data can result in spatial discontinuities of
mino et al. (2020), we estimated percent bias for four seg- precipitation patterns at the border of the catchments. To
ments of the FDC [peak flow (0%–2%), high flow (2%–20%), reduce such discontinuities, we further applied a 5 3 5 mean
midsegment (20%–70%), and low flow (70%–100%)], and filter to the BCF map. Finally, the corrected precipitation
then the absolute values of the bias percentages were data were used as input to SWAT to run the model with the

TABLE 4. Mathematical formulation of the goodness of fit metrics and hydrological signatures. Here, O and S are observed and
simulated flow (m3 s21), respectively; EP is exceedance probability; P, H, and L are the indices of the minimum flow of the peak
flow, high flow, and low flow segments, respectively. In the optimization process for hydrological model calibration, lNSE was
maximized, whereas FDCsign was minimized.
Criterion (reference) Equation Description

Discharge-related performance measures
n
(Si 2 Oi )2
Nash–Sutcliffe efficiency (Nash and NSE 5 1 2 i51
n Oa is the average of the observed flow
Sutcliffe 1970) (Oi 2 Oa )2 and n is the number of observations
i51 under evaluation
n 2
ln(Si ) 2 ln(Oi )
Nash–Sutcliffe efficiency log (Krause lNSE 5 1 2 i51
n
2
et al. 2005) ln(Oi ) 2 ln Op
i51

n
(Si 2 Oi )
Percent bias (Gupta et al. 1999) PBIAS 5 i51

n 3 100
Oi
i51

2
Kling–Gupta efficiency (Gupta et al. KGE 5 (r 2 1) 1 (b 2 1)2 1 (g 2 1)2 r is the Pearson product-moment
2009; Kling et al. 2012) correlation coefficient and beta
(gamma) indicates the bias (relative
dispersion) between observed and
simulated flows
Signature measures based on flow duration curve (FDC)

P

Sp 2 Op 3 100
p51
Percent bias in FDC peak segment Speak 5 p 5 1, 2, … , P are flow indices located

P
volume (Yilmaz et al. 2008) OP within the FDC peak flow segment
p51 (EP lower than 2%)

H
(Sh 2 Oh ) 3 100
Percent bias in FDC high segment Shigh 5 h51
h 5 1, 2, … , H are flow indices located

H
volume (Yilmaz et al. 2008) Oh within the high flow segment
h51 (2%–20% flow EP)

(Sm1 2 Sm2 ) 2 (Om1 2 Om2 ) 3 100
Percent bias in FDC midsegment Smid 5 m1 and m2 are the lowest and highest
(Om1 2 Om2 )
slope (Yilmaz et al. 2008; flow EP, respectively, within the
van Werkhoven et al. 2009) midsegment (20%–70%)

L
(Sl 2 Ol ) 3 100
Percent bias in FDC low segment Slow 5 l51
l 5 1, 2, … , L are flow indices located

L
volume (Yilmaz et al. 2008) Ol within the low flow segment
l51 (70%–100% flow EP)
1
FDC signature (Chilkoti et al. 2018) FDCsign 5 Speak 1 Shigh 1 |Smid | 1 |Slow | FDCsign is the aggregated FDC signature
4
respective optimum parameters for the simulation period to absolute error (MAE) and determination coefficient (R2)
compute the model performance measures for the hydrologi- based on the out-of-bag sample.
cal evaluation of precipitation datasets.
2) HYDROLOGICAL EVALUATION
c. Evaluation methods
We evaluated the accuracy of precipitation estimates through
1) EVALUATION USING OUT-OF-BAG SAMPLE
hydrological modeling for the three drainage systems in the
The prediction accuracy of preliminary daily precipitation study area. It is an adequate approach evaluating gauge-cor-
data (Pd0) and monthly precipitation data (Pm1) produced by rected precipitation datasets since streamflow observations are
the RF method (see Fig. 2) was assessed using the mean independent from ground precipitation observations that are

R2 R2
0.00 0.25 0.50 0.78 0.48 0.65 0.86
2015 2015
2010 2010
2005 2005
Year
2000 2000
1995 1995
1990 1990
1985 1985
1981 1981
1.00 1.00
0.75 0.75
R2
0.50 0.50
0.25 0.25
0.00 0.00
6 60
MAE [mm]
4 40
2 20
0 0
1 61 121 181 241 301 366 1 2 3 4 5 6 7 8 9 10 11 12
DOY Month
FIG. 3. Performance of the random forest algorithm for spatial interpolation of (left) daily and (right) monthly pre-
cipitations. Here, R2 is the coefficient of determination, and MAE is the mean absolute error. The middle and bottom
graphs show the performance measures averaged for each day or month in the 1981–2015 period.
used in these datasets (Beck et al. 2020a; Brocca et al. 2020; Furthermore, in this study, we analyzed the distribution of
Satgé et al. 2020). model parameters and compared the evapotranspiration (ET)
For hydrological evaluation, a multicriteria evaluation of simulated by SWAT with remotely sensed ET from Global
SWAT-simulated streamflow using all precipitation products Land Evaporation Amsterdam Model (GLEAM) and Mod-
was carried out. For that, both hydrograph goodness of fit erate Resolution Imaging Spectroradiometer Global Evapo-
metrics and hydrological signatures (Table 4) were considered ration (MOD16). The ET estimates from MOD16 and
for both calibration and validation periods. The modified GLEAM are based on the Penman–Monteith and Priestly–
Kling–Gupta efficiency (KGE) and percent bias (PBIAS) Taylor equations, respectively. This comparison is to verify
were used for assessing model skills in representing general the plausibility of ET estimates which is one of the largest
discharge dynamics and over or underestimation tendencies, components of the water budget besides precipitation and
respectively; lNSE and percent bias in FDC low segment vol- difficult to estimate over complex terrain. Results of the anal-
ume (Slow) for low flows; Nash–Sutcliffe efficiency (NSE) and ysis of parameter distribution and ET estimates are described
percent bias in FDC high segment volume (Shigh) for high in appendices B and C.
flows; and percent bias in FDC peak segment volume (Speak)
for extremely peak flow conditions. This multicriteria evalua-
tion aims to assess model skills representing all aspects of the 4. Results
observed FDC and hydrographs, which is important for a. Performance of the merging algorithm
assessing the reliability of precipitation products for hydrome-
teorological applications such as the analysis of water budget The skill of the RF method for predicting daily and
and hydroclimatic extremes (floods and droughts). The hydro- monthly precipitation patterns was evaluated using perfor-
logical model performance was ranked based on the rating mance measures (R2 and MAE) based on the out-of-bag sam-
performance criteria of Moriasi et al. (2007). Thus, for simple. Figure 3 shows that based on the temporal distribution of
plicity, the absolute values of PBIAS, Slow, Shigh, and Speak , R2, the RF performance does not have a seasonal pattern for
10 were considered as very good, (10–15) good, (15–25) satis- the daily precipitation prediction, whereas it exhibits better
factory, and (.25) unsatisfactory, and KGE, NSE, and performance in the period from April to December for
lNSE . 0.75 were considered very good, (0.65–0.75) good, monthly prediction. Furthermore, R2 shows that prediction is
(0.50–0.65) satisfactory, and (,0.50) unsatisfactory. better for the monthly (mean R2 5 0.72) than the daily (mean

FIG. 4. (top) Bias correction factors (BCFs) for six precipitation datasets and (bottom) long-term mean seasonal streamflow (Q) dynam-
ics in the period 1983–2015 after SWAT model calibration over nine catchments with underestimation of precipitation amounts in compar-
ison with the observed mean seasonal discharge. The mean BCF was computed using the catchment areas as weights. Note that both
observed and seasonal streamflow were computed only for the months with available streamflow data.
R2 5 0.25) precipitation. This result supports the correction a unique correction factor to WFDEI (Weedon et al. 2014)
of daily-predicted precipitation values to match the monthly dataset for all montane regions. The lower values of BCFs for
predictions performed in our study as described in the meth- ERA5 are related to significant precipitation overestimation
ods. MAE is much lower in the period June–September for along the Andes by ERA5 (Figs. 5 and 6). The results for the
both daily and monthly precipitation prediction, indicating other datasets (Fig. 4) show that higher BCFs were the result
that precipitation is more easily predictable when most of the for MSWEP (mean BCF 5 1.66) and lower for RAIN4PE
study area experiences lower precipitation during the dry sea- (mean BCF 5 1.38). For a BCF of 1.38, on average, 28% of
son. It is important to mention that satellite precipitation total precipitation is the precipitation underpredicted in
(CHIRP) was often the most important covariate in the merg- páramo and montane watersheds in the study area which falls
ing procedure both at daily and monthly scale, followed by in the range (0%–30%) of cloud/fog water contribution to
reanalysis precipitation (ERA5) and terrain elevation, while total precipitation reported in previous studies of the region
buffer distances were negligible (Fig. S1 in the online (Gomez-Peralta et al. 2008; Cárdenas et al. 2017). Figure 4
supplemental material). also shows that significant benefits of precipitation correction
made for RAIN4PE are obvious in a good representation of
b. Hydrological correction of the gridded
streamflow seasonality for all nine catchments. The correction
precipitation datasets
of CHIRPS also works relatively well in most of the catch-
The spatial variation of the obtained bias correction factors ments in terms of seasonal streamflow prediction, although it
(BCFs) for six precipitation datasets is shown in Fig. 4. This fails over the southern Ecuadorian Amazon (at Santiago sta-
differs from the method of Strauch et al. (2017), who applied tion). The hydrological correction of the other datasets

FIG. 5. The spatial patterns of average annual precipitation for the period 1981–2015 based on (top) raw and (middle) hydrologically
adjusted precipitation data of ERA5, CHIRP, CHIRPS, MSWEP, PISCO, and RAIN4PE. (bottom) The underestimated precipitation
fields for each precipitation dataset. The numbers in brackets represent the precipitation ranges. In the case of ERA5, precipitation values
exceeding 8000 mm are in purple (distributed over the Ecuadorian Andes mainly).
(CHIRP, ERA5, MSWEP, and PISCO) performs well for In addition, a comparison of the unadjusted precipitation
southern catchments (from Borja to Lagarto station) but not data with gauge observation was done (Fig. 6) to assess pre-
in Ecuadorian catchments (from Nueva Loja to Santiago sta- cipitation datasets’ reliability or critical shortcomings. It
tion) since the streamflow seasonality change is underesti- shows that ERA5 overestimates precipitation significantly
mated, indicating a serious drawback of these datasets. over the Andes. CHIRP, CHIRPS, MSWEP, and PISCO
(CHIRP and CHIRPS) underestimate (overestimate) pre-
c. Spatial patterns of precipitation cipitation over the northern (arid southern) Pacific coastal
In general, the spatial variability of the long-term average areas. Furthermore, ERA5, CHIRP, CHIRPS, MSWEP,
annual precipitation (1981–2015) portrayed by all precipitation and PISCO have inconsistent temporal distribution of pre-
datasets looks quite similar (Fig. 5), although PISCO shows dis- cipitation over the northern Amazon, which is confirmed by
tinct precipitation patterns and magnitudes in the rain forest low values of correlation and determination coefficients that
regions. Figure 5 also shows the spatial patterns of the estimated result from comparing these products with gauge observations
precipitation underestimates for each precipitation dataset. As at a monthly scale (Fig. 6) and SWAT-simulated seasonal
can be seen, these patterns look quite similar over the Peruvian streamflow using these datasets (Fig. 4). Therefore, these data-
Amazon for five datasets (CHIRP, CHIRPS, MSWEP, PISCO, sets are less suitable for characterizing spatiotemporal variabil-
and RAIN4PE) but vary over the northern Amazon basin in ity of precipitation over the Ecuadorian Amazon than
Ecuador. The substantial precipitation underestimation (rang- RAIN4PE. However, it should be kept in mind that the com-
ing from 0 to 3369 mm, Fig. 5) found here suggests that precipi- parison measures in Fig. 6 could be biased toward datasets
tation correction was necessary to achieve the closure of the (CHIRPS, MSWEP, PISCO, and RAIN4PE) that used data
water budget and appropriate hydrological modeling of the from the assimilated precipitation gauges in their production
páramo and montane watersheds. (see Table 1).

FIG. 6. Performance of the unadjusted precipitation datasets in comparison with gauge observations: MAP is the mean annual precipita-
tion, ME is the mean error, r is the Pearson’s correlation coefficient, and R2 is the coefficient of determination. The comparison measures
(ME, r, and R2) were computed using monthly precipitation time series for 1981–2015.
d. Hydrological evaluation scale since CHIRP, CHIRPS, and MSWEP were delivered using
different daily time window aggregation than the local one (from
In this section, we evaluate the performance of the SWAT
model driven by the hydrologically adjusted CHIRP (CHIRP- 0700 to 0700 local time). Furthermore, our analyses are based on
SWAT), ERA5 (ERA5-SWAT), CHIRPS (CHIRPS-SWAT), the results of the only one hydrological model, SWAT, and the
MSWEP (MSWEP-SWAT), PISCO (PISCO-SWAT), and application of other hydrological models could be done in future
RAIN4PE (RAIN4PE-SWAT) for calibration and validation to verify and refine the obtained results.
periods. For that, we used multiple performance measures to
assess the model skills in representing discharge dynamics includ- 1) PERFORMANCE EVALUATION FOR DAILY STREAM-
FLOW AND EXTREMES
ing all flow conditions (low, high, and peak flows). It is important
to mention that temporal mismatches in the daily precipitation We investigated the spatial variability of hydrological
accumulation may influence the model performance at the daily model performance for streamflow simulation forced by six

FIG. 7. Hydrological model performance metrics for daily streamflow simulations by SWAT driven by six precipitation datasets in the cali-
bration period: (top) spatial patterns of KGE and (bottom) boxplots showing seven criteria for all streamflow (Q) stations and stations located
in catchments draining into the Amazon River, Pacific Ocean, and Titicaca Lake. The datasets are sorted in ascending order of the median
KGE for all Q stations. Values exceeding 0.5 (between 625%) for KGE, lNSE, and NSE (PBIAS, Slow, Shigh, and Speak) are considered skillful
(marked by light gray background in boxplots). Black points in the upper part represent negative values of KGE. Note that the x axis starts at
0 for KGE, NSE, and lNSE to improve visualization, whereas PBIAS, Slow, Shigh, and Speak were constrained between 650%.
precipitation products in calibration (Fig. 7, Table S1) and precipitation product for the simulation period (1981–2015).
validation (Fig. S2, Table S2) periods. These figures present The results described in this section are based on the outputs
the Kling–Gupta efficiency spatial distribution and show for calibration period (Fig. 7) but they are also valid for the
results in terms of seven criteria for all streamflow stations validation period (Fig. S2), as results for both periods are
and catchments draining into the Titicaca Lake, the Pacific similar.
Ocean, and the Amazon River as boxplots. Table 5 shows Results for catchments draining into Titicaca Lake show
each criterion’s median values for each drainage system and that SWAT driven by gauge-corrected precipitation datasets

TABLE 5. Median values of each performance measure for daily streamflow simulation for the period 1983–2015 (without the
spinup period). Values in bold denote the best performing product in each drainage system and the study area according to the
specific score on the left.
Basin Product KGE NSE lNSE |PBIAS| |Slow| |Shigh| |Speak|

Titicaca Lake ERA5 21.1 23.04 20.07 205.9 47.8 197.5 44.5
Titicaca Lake CHIRP 0.62 0.29 0.67 17.5 31.4 18.2 7.2
Titicaca Lake CHIRPS 0.81 0.64 0.76 6.1 16 9 7.3
Titicaca Lake MSWEP 0.79 0.67 0.8 14.6 20 15.2 8.8
Titicaca Lake PISCO 0.84 0.74 0.82 7.2 18.5 7.6 13.9
Titicaca Lake RAIN4PE 0.86 0.77 0.86 6.8 14.3 6.7 14.1
Pacific ERA5 22.29 27.2 21.47 325.6 238 293.9 117.9
Pacific CHIRP 0.55 0.44 0.62 28.9 14.1 35.8 43.5
Pacific CHIRPS 0.66 0.5 0.66 15.2 11.1 12.1 18.7
Pacific MSWEP 0.68 0.55 0.66 17.4 14.8 21 27.5
Pacific PISCO 0.74 0.57 0.73 10.2 11.4 10.7 17.3
Pacific RAIN4PE 0.78 0.67 0.74 5.2 6 7.5 7.7
Amazon ERA5 0.63 0.46 0.6 12.5 19.5 17.2 18.8
Amazon CHIRP 0.63 0.4 0.45 7.7 12 9.6 12.6
Amazon CHIRPS 0.67 0.31 0.38 13.7 15.3 14.5 14.6
Amazon MSWEP 0.69 0.47 0.48 15.4 15.2 11.8 12.2
Amazon PISCO 0.49 0 20.11 21.9 23.9 18 18.9
Amazon RAIN4PE 0.8 0.7 0.73 6.2 10.8 7.3 10.7
All watersheds ERA5 20.58 22.26 20.3 148.4 55 115.1 44.2
All watersheds CHIRP 0.58 0.43 0.57 13.2 13.9 16.3 21.4
All watersheds CHIRPS 0.67 0.41 0.58 13.6 15 13 15.8
All watersheds MSWEP 0.7 0.53 0.63 15.3 16.5 15.2 15.1
All watersheds PISCO 0.7 0.51 0.66 12.8 18.6 12.7 17.7
All watersheds RAIN4PE 0.8 0.7 0.74 5.9 9.7 7.3 9.6
performs satisfactorily to very good for daily streamflow simu- 0.49) provided the worse measures, particularly over the
lation (median KGE $ 0.79), including all flow conditions. lower Amazon catchments which is consistent with previous
The good performance of MSWEP and CHIRPS for hydro- studies (Aybar et al. 2020; Llauca et al. 2021). Despite the fact
logical modeling in the Titicaca Lake basin shown here that median KGE (.0.5) is satisfactory for CHIRP, CHIRPS,
coheres with the performance demonstrated in Satgé et al. ERA5, and MSWEP, the other measures such as the lNSE
(2019, 2020). However, RAIN4PE (median KGE 5 0.86) was and NSE show that they tend to perform unsatisfactorily for
shown in our simulation to be the best choice for this drainage the simulation of low- and high-flow dynamics. However,
system. Regarding two non-gauge-corrected datasets, CHIRP- KGE patterns (Fig. 7) show unsatisfactory scores over the
SWAT has unsatisfactory performances for high-flow dynam- Ecuadorian Amazon catchments, showing the limitations of
ics, and ERA5-SWAT significantly overestimates streamflow all products (including RAIN4PE) in portraying the actual
(Fig. S4). daily precipitation variability there.
In the Pacific basin, CHIRPS-SWAT, MSWEP-SWAT, and In general, SWAT performance for all streamflow stations
even PISCO-SWAT have low KGE (#0.5), high biases, and (Fig. 7 and Fig. S2 and Table 5, Tables S1 and S2) suggests
poor performance for high and peak flows for some stations. that RAIN4PE (e.g., median KGE 5 0.80) is the most appro-
The outcome for MSWEP and PISCO aligns with the findings priate product for daily streamflow simulation, including all
of previous studies (Bhuiyan et al. 2019; Derin et al. 2019; flow conditions in the study area.
Asurza-Véliz and Lavado-Casimiro 2020). CHIRP-SWAT
has more skill than ERA5-SWAT, which shows a significant 2) PERFORMANCE EVALUATION FOR MONTHLY
overestimation of streamflow; however, they both are outper- STREAMFLOW
formed by the gauge-corrected precipitation datasets. The
overall good performance of RAIN4PE-SWAT (median Fig. 8 and Fig. S3 display the spatial distribution of KGE,
KGE 5 0.78) allowed us to conclude that RAIN4PE is the NSE, lNSE, and PBIAS to assess the SWAT model skill
most suitable precipitation product for daily streamflow simu- for the monthly streamflow simulation in the calibration and
lation (including all flow conditions and water budget closure) validation periods. These figures show that results in both
in the catchments draining into the Pacific Ocean. periods are quite similar, although the overall performance of
In the Amazon basin, among the six precipitation products PISCO-SWAT and MSWEP-SWAT is a bit lower in the vali-
driving SWAT, RAIN4PE (median KGE 5 0.80) provided dation period. Based on results of model performance in the
the best performance measures for daily streamflow simula- validation period (Fig. 8), among the six precipitation prod-
tion (including all flow conditions). PISCO (median KGE 5 ucts driving SWAT, overall RAIN4PE (median KGE 5 0.86,

FIG. 8. Hydrological model performance metrics KGE, NSE, lNSE, and PBIAS for monthly streamflow simulations by SWAT driven by
six precipitation datasets in the validation period. Black points represent negative values of KGE, NSE, and lNSE.
NSE 5 0.82, lNSE 5 0.82, and |PBIAS| 5 5.4%) provided the Ocean, which is in agreement with the results for the daily
best performance measures for monthly streamflow simula- outputs. Otherwise, ERA5-SWAT was found to perform
tion in all evaluated catchments. Despite the median KGE, unsatisfactorily for Andean basins, although its performance
NSE, and lNSE were satisfactory (.0.5, Fig. 8) for CHIRP, improved for larger catchments in the Amazon basin. The
CHIRPS, MSWEP, and PISCO, the spatial patterns of these overall very good performance in accordance with criteria by
measures show the limitation (e.g., NSE , 0.5) of these prod- Moriasi et al. (2007) obtained by RAIN4PE-SWAT highlights
ucts for hydrological modeling over the Ecuadorian Amazon, the increased utility of RAIN4PE for countrywide hydrome-
lower Amazon, and some catchments draining into the Pacific teorological applications in Peru and Ecuador.

5. Discussion parameter values to counterbalance precipitation uncertainty

can lead to misrepresentation of the basinwide water budget
a. Advantages of the merging methodology
(see more details in appendix B). To overcome these deficien-
This study demonstrates a successful method for merging cies, we used streamflow data to adjust precipitation biases.
multiple precipitation sources (based on gauge, satellite, Our results show that the hydrological correction of precip-
and reanalysis data) with surface elevation using the RF itation datasets was more efficient over the regions with the
method to generate a spatially gridded precipitation dataset strongest rainfall seasonality such as the Peruvian catchments
RAIN4PE. This is supported by the significant improvement (Espinoza Villar et al. 2009; Segura et al. 2019). This suggests
of RAIN4PE for hydrological simulations compared to the that actual spatiotemporal precipitation fields over these
non-gauge-corrected datasets (CHIRP and ERA5) used for regions are well depicted by the assessed datasets, whereas
the merging procedure. Furthermore, the superiority of the correction efficiency over the Ecuadorian Amazon catch-
RAIN4PE regarding the gauge-corrected datasets (CHIRPS, ments, which experience precipitation throughout the year
MSWEP, and PISCO) for hydrological simulations suggests with high spatial regime variability (Laraque et al. 2007;
that the methodology applied herein to generate RAIN4PE is Tobar and Wyseure 2018), is more variable and depends
much more robust than that of the other merged precipitation more strongly on the precipitation product. For instance, the
products. This means that the RF method is more effective in correction was not feasible for CHIRP, ERA5, MSWEP,
merging multiple precipitation data sources than deterministic PISCO, and even CHIRPS (at southern Ecuadorian Amazon)
and geostatistical interpolation methods (Funk et al. 2015a; which led to the underrepresentation of the seasonal stream-
Aybar et al. 2020) and merging approaches that use weights flow patterns and hence the true seasonal precipitation pat-
for each source (Beck et al. 2017, 2019b). Compared to the terns as well (Figs. 4 and 6). This is a critical drawback of
aforementioned merging approaches, RF has the flexibility to these products, and our findings here could be helpful for
include multiple precipitation sources and environmental var- their revision and improvement. Even though the hydrologi-
iables (e.g., surface elevation) that explain precipitation pat- cal correction of CHIRPS resulted in the improvement of the
terns. Besides this advantage, RF can capture nonlinear seasonal streamflow simulations for the northern Ecuadorian
dependencies and interactions of variables, such as the nonlin- Amazon, CHIRPS-SWAT still performed unsatisfactorily for
ear interactions among the precipitation and terrain elevation the daily and monthly discharge dynamics (Figs. 7, 8 and Figs.
due to the complex Andes morphology (Figs. 1 and 5; Chavez S2 and S3), indicating that CHIRPS does not represent well
and Takahashi 2017), which could be challenging to model the actual precipitation patterns over the Ecuadorian Amazon
using geostatistical techniques. However, it is important to catchments. In these catchments, other datasets such as
keep in mind the RF limitation for predicting value beyond gauge-based ORE HYBAM (Guimberteau et al. 2012),
the range in the training data (Hengl et al. 2018). Overall, the gauge-corrected WFDEI, reanalysis ERA-Interim, and satel-
results of our study provide a reference for merging multi- lite-based PERSIANN (Hsu et al. 1997), TMPA, CMORPH,
source precipitation data and environmental variables using and IMERG have also been reported to perform unsatisfacto-
RF in complex data-scarce regions. rily for streamflow simulation (Zulkafli et al. 2014; Zubieta
et al. 2015; Strauch et al. 2017; Zubieta et al. 2017; Towner
b. Hydrological correction of the gridded
et al. 2019). Overall, when comparing RAIN4PE to CHIRP,
precipitation datasets
ERA5, MSWEP, PISCO, CHIRPS, and other datasets men-
The high BCF values (Fig. 4) obtained to correct gridded pre- tioned above, we can see that it shows satisfactory perfor-
cipitation biases make evident that most of the datasets evaluated mance for monthly (Fig. 8) and seasonal (Fig. 4) streamflow
(CHIRP, CHIRPS, MSWEP, PISCO, and RAIN4PE) often simulations with SWAT over the Ecuadorian Amazon. How-
have precipitation underestimation over the páramo and montane ever, its performance for daily simulation is still unsatisfactory
watersheds in the Amazon (Fig. 5). This underestimation, espe- (Fig. 7), which highlights that estimation of precipitation at a
cially by gauge-corrected datasets, could be caused by the low daily resolution over data-scarce regions such as the equatorial
number of precipitation gauges available (Fig. 1), which is further Amazon region is very challenging. The exposed shortcomings
amplified by the fact that the gauges do not account for the of precipitation datasets suggest the urgent implementation and
important cloud/fog water input into the system (Gomez-Peralta densification of precipitation and cloud/fog gauge networks over
et al. 2008; Clark et al. 2014; Cárdenas et al. 2017). the Ecuadorian Amazon and Peruvian montane watersheds.
A substantial precipitation underestimation over the These could help to improve the depiction of rainfall amounts
páramo and montane watersheds is critical since it might even and their spatiotemporal distribution and hence could be useful
lead to physically unrealistic runoff ratios above 1 in water for improving streamflow simulations. It is important to keep in
budget estimates as reported in previous studies (Zulkafli et al. mind that the correction of the proposed precipitation product
2014; Zubieta et al. 2015; Manz et al. 2016; Strauch et al. 2017; through the reverse hydrology concept was performed using the
Builes-Jaramillo and Poveda 2018; Aybar et al. 2020). Further- SWAT hydrological model, and therefore the performance of
more, precipitation errors in the upstream catchments can neg- the RAIN4PE dataset may change if another hydrological model
atively affect simulation results for downstream river is used. Though, as SWAT is a widely used comprehensively ver-
catchments. For instance, the assignment of unrealistic model ified model, we expect only minor deviation.

c. Implications for hydrological modeling routing that considers river–floodplain dynamics can be
observed in the good representation of discharge dynamics of
The results of the hydrological evaluation clearly show the
the Amazonian rivers in this study. For instance, in the Ucayali
advantages and shortcomings of each evaluated precipitation
River (a tributary of the Amazon River), the significant
dataset for streamflow simulation, including low, high, and
observed flood peak delay (on a scale of months) from Lagarto
peak flows. Moreover, we presented the comparison of
to Requena station is well reproduced by SWAT (see Fig. S6),
SWAT-simulated seasonal streamflow using all evaluated
which is consistent with the findings of Santini (2020).
datasets against observed seasonal streamflow for the three
It is also important to highlight that this study is the first
drainage systems (Titicaca Lake basin, Pacific basin, and
applying SWAT at the country-level of Peru and performing a
Amazon basin) in Figs. S4–S6. These figures can assist practi-
multiobjective calibration and validation using hydrograph
tioners in selecting the appropriate precipitation product for
goodness of fit and FDC signatures for large-domain modeling
hydrological applications. In general, the hydrological evalua-
(1.6 million km2) in a region with complex hydroclimatic condi-
tion highlighted RAIN4PE as the best precipitation dataset tions. Our results show the robustness of signature-based cali-
for hydrological modeling of the Peruvian and Ecuadorian bration guiding the model to reproduce not only one common
watersheds. RAIN4PE is the only gridded precipitation prod- objective function (e.g., high flows given by NSE) but all aspects
uct for Peru and Ecuador, which benefits from maximum of the hydrograph and FDC as supported by RAIN4PE-SWAT
available in situ observations, multiple precipitation sources, good performances reproducing all flow conditions. This is cru-
environmental variable (elevation data), and is supplemented cial for robust hydrometeorological applications including
by streamflow data to correct the precipitation underestima- extremes such as droughts and floods as well as for the assess-
tion over páramos and montane catchments. The exploitation ment of precipitation dataset reliability. Furthermore, our
of all these variables using state-of-the-practice methods to results reinforce previous study findings (Shafii and Tolson
generate RAIN4PE proved that RAIN4PE-SWAT was capa- 2015; Chilkoti et al. 2018; Fernandez-Palomino et al. 2020),
ble of closing the (hitherto) observed water budget imbalance which proved the robustness of a signature-based calibration
over Peruvian and Ecuadorian catchments which, eventually, approach in the hydrological modeling of small watersheds. We
makes the RAIN4PE a good candidate for hydrological appli- consider that our approaches can be helpful for future studies
cations in the region. Despite this, we consider that RAIN4PE related to precipitation estimates as well as to hydrological
is still subject to uncertainties, especially in regions where pre- model calibration, evaluation, and application.
cipitation was inferred from the observed streamflow data.
For these regions, precipitation estimates should be viewed d. Future development and application
with some care due to uncertainties in streamflow data, Based on the experiences we gained, our future investigations
inferred evapotranspiration, gridded precipitation data, and will focus on applying RAIN4PE-SWAT to analyze the water
hydrological model structure. budget at the national scale of Peru, as well as climate change
In this study, besides evaluating precipitation datasets for impacts on water resources using RAIN4PE as the basis for
streamflow simulation, we show that uncertainties associated bias adjustment, and trends in frequency and intensity of mete-
with precipitation estimates have implications in estimating orological and hydrological droughts. The current RAIN4PE
hydrological model parameters (see appendix B) and water bud- data availability (1981–2015) is planned to be extended in the
get components (e.g., evapotranspiration, see appendix C). This future. Moreover, the methodology presented in the paper will
is critical for the regionalization of parameters and reliable esti- also be extended to the entire Amazon basin.
mation of the water budget for water resources management.
Furthermore, an aftermath verification of RAIN4PE-SWAT-
simulated evapotranspiration with GLEAM and MOD16 esti- 6. Summary and conclusions
mates (appendix C) shows that GLEAM and MOD16 return We developed a new hydrologically adjusted daily precipi-
higher estimated values of evapotranspiration which would not tation dataset (1981–2015, 0.18 resolution) called RAIN4PE
allow the water budget closure and bring inconsistencies in the by merging three existing datasets for a domain covering Peru
temporal evapotranspiration distribution over northern Amazon and Ecuador. This dataset takes advantages of ground-, satel-
in Ecuador. This suggests that evapotranspiration estimation is lite-, and reanalysis-based precipitation datasets, including
still a challenge for remotely sensed based evapotranspiration CHIRP and ERA5, which are merged with terrain elevation
products in the region. using the random forest (RF) method to provide precipita-
It is important to highlight that this study is the first applying tion estimates. Furthermore, streamflow data was used to
SWAT updated for improved representation of tropical vegeta- correct precipitation estimates over catchments with water
tion dynamics (Alemayehu et al. 2017) and river–floodplain budget closure problems (e.g., the páramo and montane
dynamics. These improvements are crucial to model the hydro- watersheds) through the reverse hydrology methods, for
logical processes of Andean and Amazonian river catchments which the SWAT model was applied for the first time
appropriately. The benefits of appropriate representation of herein. Moreover, a comprehensive hydrological evaluation
tropical vegetation dynamics were demonstrated in previous of RAIN4PE, CHIRP, ERA5, and the existing state-of-
studies (Strauch and Volk 2013; Alemayehu et al. 2017; Fernan- the-art gauge-corrected precipitation datasets}CHIRPS,
dez-Palomino et al. 2020), while the benefit of flow water MSWEP, and PISCO}in the Peruvian and Ecuadorian

river catchments using a range of performance metrics was spatiotemporal variations of droughts and floods, and explor-
performed. For that, SWAT was calibrated and validated ing spatial variations and regimes of precipitation. We con-
with each precipitation dataset in a number of catchments. sider that RAIN4PE and our RAIN4PE-SWAT model can be
We summarize our findings as follows. adopted as a benchmark to evaluate precipitation datasets in
Peru and Ecuador.
• The good RAIN4PE-SWAT performance for streamflow
simulation suggests the effectiveness of the RF method to
Acknowledgments. The authors thank the EPICC project
merge multisource precipitation estimates with terrain ele-
that is part of the International Climate Initiative (IKI).
vation to develop a reliable spatially gridded precipitation
The Federal Ministry for the Environment, Nature Conser-
dataset. As all datasets (CHIRP, ERA5, and terrain eleva- vation and Nuclear Safety (BMU) supports this initiative on
tion) used to develop RAIN4PE are freely available, this the basis of a decision adopted by the German Bundestag.
approach can be used in other data-scarce regions. We also thank SENAMHI, Peruvian ANA, observatory
• The utility of streamflow data to improve both precipitation HYBAM, Enrique Morán-Tejeda, Guido G. Tamayo, Juan
and streamflow simulations over the páramo and montane J. Nieto, Vladimiro Tobar, and Kevin J. Perez for providing
watersheds with precipitation underestimation was demon- the hydrometeorological dataset. The authors are thankful
strated herein. This highlights that the reverse hydrology to CHIRP, CHIRPS, ERA5, MSWEP, and PISCO data
approach offers a new effective way of understanding the generation teams for providing the precipitation data at
free of cost. We are thankful to Dr. David Hadka and
hydrological processes of the Andean–Amazon catchments,
Dr. Patrick M. Reed for making their software “BORG:
which have a key role in the hydrological variability of the Many-Objective Evolutionary Computing Framework”
entire Amazon basin. available for this study. We are grateful to the Editor, Liz
• The hydrological evaluation results from uncorrected pre- Stephens, Oscar M. Baez-Villanueva, and one anonymous
cipitation datasets forcing SWAT for streamflow simulation reviewer for their constructive comments.
revealed that CHIRP outperformed ERA5, which signifi-
cantly overestimate precipitation along the Andes. How- Data availability statement. The RAIN4PE data record is
ever, these products were outperformed by the gauge- freely available at https://doi.org/10.5880/pik.2020.010 (Fer-
based precipitation datasets. nandez-Palomino et al. 2021).
• Among the gauge-corrected precipitation datasets forcing
SWAT for streamflow simulation, all products performed
well in the catchments draining into the Titicaca Lake. For APPENDIX A
catchments draining into the Pacific Ocean and Amazon Glossary
River, CHIRPS, MSWEP, and PISCO performed unsatis-
factorily in several catchments, indicating the limitations of CHIRP Climate Hazards Group Infrared Precipitation
these products for hydrological modeling over these drain- CHIRPS CHIRP with Station data
age systems. In contrast, RAIN4PE was the only product CMORPH Climate Prediction Center morphing technique
that provided consistently good performance for the daily IMERG Global Precipitation Measurement (GPM)
and monthly streamflow simulations, including all discharge Integrated Multisatellite Retrievals
conditions (low, high, and peak flows) and water budget MSWEP Multi-Source Weighted-Ensemble Precipitation
closure in almost all Peruvian and Ecuadorian river PISCO Peruvian Interpolated data of SENAMHI’s
catchments. Climatological and Hydrological
• We found that CHIRP, CHIRPS, ERA5, MSWEP, and Observations
PISCO cannot represent the seasonal distribution of precipita- SENAMHI Servicio Nacional de Meteorologı́a e
tion and hence the seasonal streamflow over the Ecuadorian Hidrologı́a del Perú
Amazon. This is a critical drawback that can have implications TMPA Tropical Rainfall Measuring Mission (TRMM)
in hydrometeorological applications in the Amazon basin. Multisatellite Precipitation Analysis
• We found that uncertainties in precipitation data in existing WFDEI WATCH Forcing Data methodology applied
datasets affect the estimation of model parameters and to ERA-Interim data
water budget components, suggesting the importance of
developing high-quality meteorological forcing datasets in APPENDIX B
mountainous regions. Our contribution is in line with this
and marks progress in developing precipitation datasets in Evaluating the Distribution of Model Parameters
the region.
In this section, we analyze the distribution of calibrated
The overall good performance of the RAIN4PE highlights model parameters to see the regional parameter behavior
its utility as an important new gridded precipitation dataset, and to elucidate potential input errors as they were identi-
which opens new possibilities for numerous hydrometeorolog- fied to achieve the water budget closure using different pre-
ical applications throughout Peru and Ecuador. Examples are cipitation datasets. Thus, unrealistic parameter values could
streamflow simulations, estimation of the water budget and its be linked to input error. We advise readers to see Table 3
evolution, water resources management, understanding for the description of parameters and Neitsch et al. (2011)

FIG. B1. Calibrated parameter values for the soil available water capacity (SOL_AWC) for topsoil (1) and subsoil (2), the surface runoff
delay coefficient (SURLAG), and the groundwater delay time (GW_DELAY). The HWSD map shows SOL_AWC values derived from
the Harmonized World Soil Database, which were used for setting up the SWAT model.
for detailed parameter definitions. Among the calibrated Figure B1 shows that the SOL_AWC, which constrains
SWAT parameters, only two (SOL_AWC, GW_REVAP) the maximum amount of plant available water a soil can
can alter the water budget since they influence evapotrans- provide and was derived from the Harmonized World Soil
piration and, subsequently, runoff estimation. The remain- Database (HWSD; Abbaspour et al. 2019), was adjusted
ing parameters influence the surface runoff (SURLAG), mainly for the Andean catchments. This is a critical param-
groundwater (GW_DELAY, RCHRG_DP, GWQMN, eter since higher values can lead to higher evapotranspiration
ALPHA_BF), and flow routing (CH_K2, CHD, FP_W_F) and vice versa. The results show that high SOL_AWC values
not affecting water loss from the system. We illustrate in were identified for compensating the ERA5 precipitation over-
Figs. B1–B3 the spatial patterns of the calibrated parame- estimation (see positive errors in Fig. 6). However, despite this
ters related to six precipitation datasets. trade-off, discharge overestimation by ERA5-SWAT remains

FIG. B2. Calibrated parameter values for the deep aquifer percolation fraction (RCHRG_DP), the threshold for return flow from the shal-
low aquifer (GWQMN), the groundwater “revap” coefficient (GW_REVAP), and the baseflow recession constant (ALPHA_BF).
(see PBIAS in Fig. 8), suggesting that ERA5 precipitation esti- Figure B1 shows overall low values (ranging from 0.05 to 0.5)
mates must be bias-corrected for the Andean regions prior to for SURLAG in the study area, which is important for smooth-
hydrological applications. Otherwise, unrealistic low SOL_- ing the simulated hydrograph due to the delay in surface runoff
AWC values (≈0) and the prevalence of discharge underesti- released from the HRUs (Neitsch et al. 2011) to match the
mation (Fig. 8 and Fig. S5) over the northern pacific coastal peaks in the observed hydrograph. The GW_DELAY values
catchments suggest that precipitation could be underestimated (ranging from 1 to 50 days) reflect the lag in time that water in
there, particularly by CHIRP, CHIRPS, MSWEP, and even soil profile needs to enter shallow aquifer; high (low) values are
PISCO (see negative errors in Fig. 6). usual for most of the Andean (Amazonian) catchments.
For the remaining parameters, we describe each one briefly Figure B2 shows spatial distribution of the calibrated
based on the calibrated parameters for RAIN4PE-SWAT. groundwater-related parameters. The RCHRG_DP parameter

ERA5 CHIRP CHIRPS MSWEP PISCO RAIN4PE
CHD [m]
30
15
10
5
3
FP_W_F
5
CH_K2
[mm/h]
44
20
0
FIG. B3. Calibrated parameter values for the main channel depth (CHD), the ratio of floodplain width over bankfull width (FP_W_F), and
the hydraulic conductivity of main channel (CH_K2). Reaches in gray indicate that the parameter was not important in these reaches.
reflects the water volume percolated into the deep aquifer rel- (e.g., lower Amazon catchments), whereas the low values (∼0)
ative to the total recharge entering aquifers (both shallow and show those with slow contributions (e.g., most of the Andean
deep). Therefore, the calibrated RCHRG_DP values provide catchments draining into the Pacific Ocean).
an insight into the important recharge entering deep aquifers Figure B3 shows the calibrated reach and floodplain
in Peruvian Andean catchments, which subsequently sustain parameters (CHD, FP_W_F, and CH_K2). Among these
the prolonged dry season flow in these catchments (Clark et al. parameters, the FP_W_F values can reflect the occurrence
2014; Fernandez-Palomino et al. 2020). The GW_REVAP val- of flow over floodplains during the high discharge season in
ues greater than zero reflect the areas (lower Amazon) where the lower Amazon rivers. The CH_K2 values greater than
water is re-evaporated from the shallow aquifer (water entering zero show reaches where water is infiltrated at the flood-
the soil for evaporation and transpiration). In these areas, deep- plain surface from floodplain flow or ponded water during
rooted evergreen forests can draw water from the shallow aqui- overbank flood events. Then water stored at floodplain allu-
fer to meet their demands if available water in the soil profile is vium flows back to the channel when flood wave has passed
insufficient. All calibrated GWQMN values favor the return and water levels in the channel have dropped, and the
flow from aquifers and the re-evaporation from the shallow aqui- hydraulic gradient is reversed. This interaction between
fer in areas (lower Amazon) where GWQMN values are greater floodplains and reaches can explain the significant observed
than 750 mm (default water depth threshold in the shallow aqui- flood peak delay (on a scale of months) from Lagarto to
fer to allow re-evaporation). The high ALPHA_BF values (∼1) Requena station (see Fig. S6), which is consistent with the
show shallow aquifers quickly contributing return flow to streams findings of Santini (2020).

FIG. C1. Comparison of evapotranspiration (ET) estimates from the calibrated SWAT model using different precipitation datasets as
input with the remotely sensed based ET estimates from GLEAM and MOD16: (top) average annual ET for the period 2000–14 and
Pearson’s correlation coefficient (r) between SWAT-simulated ET and ET estimates from (middle) GLEAM and (bottom) MOD16. The
comparison measure (r) was computed using monthly ET time series at the subcatchment scale for 2000–14.
APPENDIX C Andean catchments. The correlation coefficient (Fig. C1,

middle panel) shows better agreement between GLEAM
Comparison of SWAT and Remotely Sensed Based and SWAT-simulated ET, which are based on the same
Evapotranspiration equation (Priestley–Taylor) for potential evapotranspiration
Figure C1 compares evapotranspiration (ET) estimates from estimation. Spatially, both GLEAM and MOD16 agree well
the calibrated SWAT model driven by different precipitation with the SWAT-simulated ET in areas with strong seasonal
datasets with the GLEAM and MOD16 estimates. All ET esti- precipitation variability, as the Peruvian Andes and south-
mates show similar spatial patterns with increasing ET gra- ern region of the Peruvian Amazon. However, negative cor-
dients from west to east. The differences in SWAT-simulated relation values over the northern Amazon basin areas with
ET volumes can be attributed to inappropriate parameter esti- a bimodal rainfall regime (Laraque et al. 2007) can indicate
mation due to precipitation biases and uncertainties. The inconsistency in the temporal distribution in GLEAM and
higher ET values for CHIRPS-SWAT are likely due to the MOD16 ET estimates there. This is in line with the findings
prevalence of dry conditions in CHIRPS which affect the esti- of Dile et al. (2020), who reported that remotely sensed
mation of daily relative humidity and subsequently potential based ET did not respond well to the rainfall in areas with
evapotranspiration, vapor stress on plant growth, and ET. a bimodal rainfall pattern in Ethiopia. Furthermore, the ET
We compared GLEAM and MOD16 against the simu- estimates could be affected by the inherent uncertainties in
lated ET by RAIN4PE-SWAT since it represents the water methods and input data. Our results demonstrate that the
budget well (see PBIAS in Fig. 8). Figure C1 shows a gen- ET estimation by the remotely sensed ET products is still a
eral tendency for GLEAM and MOD16 to overestimate challenge in the region, and the ground-based measure-
ET in the study area, and even their estimates are greater ments are required for better understanding the ET spatio-
than precipitation along the Andes (see Fig. C1 and Fig. 6), temporal patterns and for a more reliable evaluation of the
which would not allow the water budget closure in the ET estimates.

REFERENCES Levizzani et al., Eds., Advances in Global Change Research,

Vol. 69, Springer, 625–653, https://doi.org/10.1007/978-3-030-
Abbaspour, K. C., and S. A. Ashraf Vaghefi, 2019: Harmonized 35798-6_9.
World Soil Database in SWAT Format. PANGAEA, accessed }}, E. F. Wood, T. R. McVicar, M. Zambrano-Bigiarini, C.
30 March 2020, https://doi.org/10.1594/PANGAEA.901309. Alvarez-Garreton, O. M. Baez-Villanueva, J. Sheffield, and
}}, S. A. Vaghefi, H. Yang, and R. Srinivasan, 2019: Global D. N. Karger, 2020b: Bias correction of global high-resolution
soil, landuse, evapotranspiration, historical and future precipitation climatologies using streamflow observations
weather databases for SWAT Applications. Sci. Data, 6, 263, from 9372 catchments. J. Climate, 33, 1299–1315, https://doi.
https://doi.org/10.1038/s41597-019-0282-4. org/10.1175/JCLI-D-19-0332.1.
Alemayehu, T., A. van Griensven, B. T. Woldegiorgis, and W. Bhuiyan, M. A. E., E. I. Nikolopoulos, and E. N. Anagnostou,
Bauwens, 2017: An improved SWAT vegetation growth mod- 2019: Machine learning–based blending of satellite and rean-
ule and its evaluation for four tropical ecosystems. Hydrol. alysis precipitation datasets: A multiregional tropical complex
Earth Syst. Sci., 21, 4449–4467, https://doi.org/10.5194/hess-21- terrain evaluation. J. Hydrometeor., 20, 2147–2161, https://doi.
4449-2017. org/10.1175/JHM-D-19-0073.1.
Alfieri, L., V. Lorini, F. A. Hirpa, S. Harrigan, E. Zsoter, C. Breiman, L., 2001: Random forests. Mach. Learn., 45, 5–32,
Prudhomme, and P. Salamon, 2020: A global streamflow https://doi.org/10.1023/A:1010933404324.
reanalysis for 1980–2018. J. Hydrol. X, 6, 100049, https://doi. Brocca, L., T. Moramarco, F. Melone, and W. Wagner, 2013: A
org/10.1016/j.hydroa.2019.100049. new method for rainfall estimation through soil moisture
Armijos, E., and Coauthors, 2013: Suspended sediment dynamics observations. Geophys. Res. Lett., 40, 853–858, https://doi.org/
in the Amazon River of Peru. J. S. Amer. Earth Sci., 44, 75– 10.1002/grl.50173.
84, https://doi.org/10.1016/j.jsames.2012.09.002. }}, and Coauthors, 2019: SM2RAIN–ASCAT (2007–2018):
Arnold, J. G., R. Srinivasan, R. S. Muttiah, and J. R. Williams, Global daily satellite rainfall data from ASCAT soil moisture
1998: Large area hydrologic modeling and assessment part I: observations. Earth Syst. Sci. Data, 11, 1583–1601, https://doi.
Model development. J. Amer. Water Resour. Assoc., 34, 73– org/10.5194/essd-11-1583-2019.
89, https://doi.org/10.1111/j.1752-1688.1998.tb05961.x. }}, C. Massari, T. Pellarin, P. Filippucci, L. Ciabatta, S. Camici,
Asurza-Véliz, F. A., and W. S. Lavado-Casimiro, 2020: Regional Y. H. Kerr, and D. Fernández-Prieto, 2020: River flow pre-
parameter estimation of the SWAT model: Methodology and diction in data scarce regions: Soil moisture integrated satel-
application to river basins in the Peruvian Pacific drainage. lite rainfall products outperform rain gauge observations in
Water, 12, 3198, https://doi.org/10.3390/w12113198. West Africa. Sci. Rep., 10, 12517, https://doi.org/10.1038/
Aybar, C., C. Fernández, A. Huerta, W. Lavado, F. Vega, and O. s41598-020-69343-x.
Felipe-Obando, 2020: Construction of a high-resolution Bruijnzeel, L. A., M. Mulligan, and F. N. Scatena, 2011: Hydrome-
gridded rainfall dataset for Peru from 1981 to the present teorology of tropical montane cloud forests: Emerging pat-
day. Hydrol. Sci. J., 65, 770–785, https://doi.org/10.1080/ terns. Hydrol. Processes, 25, 465–498, https://doi.org/10.1002/
02626667.2019.1649411. hyp.7974.
Baez-Villanueva, O. M., M. Zambrano-Bigiarini, L. Ribbe, A. Buchhorn, M., B. Smets, L. Bertels, M. Lesiv, N.-E. Tsendbazar,
Nauditt, J. D. Giraldo-Osorio, and N. X. Thinh, 2018: Tem- M. Herold, and S. Fritz, 2019: Copernicus global land service:
poral and spatial evaluation of satellite rainfall estimates over Land cover 100 m: Epoch 2015: Globe (V2.0.2). Zenodo,
different regions in Latin-America. Atmos. Res., 213, 34–50, accessed 30 March 2020, https://doi.org/10.5281/zenodo.
https://doi.org/10.1016/j.atmosres.2018.05.011. 3243509.
}}, and Coauthors, 2020: RF-MEP: A novel random Forest Builes-Jaramillo, A., and G. Poveda, 2018: Conjoint analysis of
method for merging gridded precipitation products and surface and atmospheric water balances in the Andes-Ama-
ground-based measurements. Remote Sens. Environ., 239, zon system. Water Resour. Res., 54, 3472–3489, https://doi.org/
111606, https://doi.org/10.1016/j.rse.2019.111606. 10.1029/2017WR021338.
Bai, P., and X. Liu, 2018: Evaluation of five satellite-based precipita- Cárdenas, M. F., C. Tobón, and W. Buytaert, 2017: Contribution
tion products in two gauge-scarce basins on the Tibetan Pla- of occult precipitation to the water balance of páramo ecosys-
teau. Remote Sens., 10, 1316, https://doi.org/10.3390/rs10081316. tems in the Colombian Andes. Hydrol. Processes, 31, 4440–
Beck, H. E., A. I. J. M. van Dijk, V. Levizzani, J. Schellekens, 4449, https://doi.org/10.1002/hyp.11374.
D. G. Miralles, B. Martens, and A. de Roo, 2017: MSWEP: 3- Chavez, S. P., and K. Takahashi, 2017: Orographic rainfall hot
hourly 0.258 global gridded precipitation (1979–2015) by merg- spots in the Andes-Amazon transition according to the
ing gauge, satellite, and reanalysis data. Hydrol. Earth Syst. TRMM precipitation radar and in situ data. J. Geophys. Res.
Sci., 21, 589–615, https://doi.org/10.5194/hess-21-589-2017. Atmos., 122, 5870–5882, https://doi.org/10.1002/2016JD026282.
}}, and Coauthors, 2019a: Daily evaluation of 26 precipitation Chilkoti, V., T. Bolisetti, and R. Balachandar, 2018: Multi-objec-
datasets using stage-IV gauge-radar data for the CONUS. tive autocalibration of SWAT model for improved low flow
Hydrol. Earth Syst. Sci., 23, 207–224, https://doi.org/10.5194/ performance for a small snowfed catchment. Hydrol. Sci. J.,
hess-23-207-2019. 63, 1482–1501, https://doi.org/10.1080/02626667.2018.1505047.
}}, E. F. Wood, M. Pan, C. K. Fisher, D. G. Miralles, A. I. J. Clark, K. E., and Coauthors, 2014: The hydrological regime of a
M. van Dijk, T. R. McVicar, and R. F. Adler, 2019b: forested tropical Andean catchment. Hydrol. Earth Syst. Sci.,
MSWEP V2 global 3-hourly 0.18 precipitation: Methodology 18, 5377–5397, https://doi.org/10.5194/hess-18-5377-2014.
and quantitative assessment. Bull. Amer. Meteor. Soc., 100, da Motta Paca, V. H., G. E. Espinoza-Dávalos, D. M. Moreira,
473–500, https://doi.org/10.1175/BAMS-D-17-0138.1. and G. Comair, 2020: Variability of trends in precipitation
}}, and Coauthors, 2020a: Global-scale evaluation of 22 precipi- across the Amazon River Basin determined from the
tation datasets using gauge observations and hydrological CHIRPS precipitation product and from station records.
modeling. Satellite Precipitation Measurement, Vol. 2, V. Water, 12, 1244, https://doi.org/10.3390/w12051244.

da Silva Júnior, J. C., V. Medeiros, C. Garrozi, A. Montenegro, }}, A. Verdin, J. Michaelsen, P. Peterson, D. Pedreros, and G.
and G. E. Gonçalves, 2019: Random forest techniques for Husak, 2015b: A global satellite-assisted precipitation clima-
spatial interpolation of evapotranspiration data from Brazil- tology. Earth Syst. Sci. Data, 7, 275–287, https://doi.org/10.
ian’s Northeast. Comput. Electron. Agric., 166, 105017, 5194/essd-7-275-2015.
https://doi.org/10.1016/j.compag.2019.105017. Gassman, P. W., A. M. Sadeghi, and R. Srinivasan, 2014: Applica-
Dee, D. P., and Coauthors, 2011: The ERA-Interim reanalysis: tions of the SWAT model special section: Overview and
Configuration and performance of the data assimilation sys- insights. J. Environ. Qual., 43, 1–8, https://doi.org/10.2134/
tem. Quart. J. Roy. Meteor. Soc., 137, 553–597, https://doi.org/ jeq2013.11.0466.
10.1002/qj.828. Gleixner, S., T. Demissie, and G. T. Diro, 2020: Did ERA5
Derin, Y., and Coauthors, 2019: Evaluation of GPM-era global improve temperature and precipitation reanalysis over East
satellite precipitation products over multiple complex terrain Africa? Atmosphere, 11, 996, https://doi.org/10.3390/
regions. Remote Sens., 11, 2936, https://doi.org/10.3390/ atmos11090996.
rs11242936. Gomez-Peralta, D., S. F. Oberbauer, M. E. McClain, and T. E.
Dile, Y. T., E. K. Ayana, A. W. Worqlul, H. Xie, R. Srinivasan, Philippi, 2008: Rainfall and cloud-water interception in tropi-
N. Lefore, L. You, and N. Clarke, 2020: Evaluating satellite- cal montane forests in the eastern Andes of Central Peru.
based evapotranspiration estimates for hydrological applica- For. Ecol. Manage., 255, 1315–1325, https://doi.org/10.1016/j.
tions in data-scarce regions: A case in Ethiopia. Sci. Total foreco.2007.10.058.
Environ., 743, 140702, https://doi.org/10.1016/j.scitotenv.2020. Guimberteau, M., and Coauthors, 2012: Discharge simulation in
140702. the sub-basins of the Amazon using ORCHIDEE forced by
Erazo, B., L. Bourrel, F. Frappart, O. Chimborazo, D. Labat, L. new datasets. Hydrol. Earth Syst. Sci., 16, 911–935, https://doi.
Dominguez-Granda, D. Matamoros, and R. Mejia, 2018: Val- org/10.5194/hess-16-911-2012.
idation of satellite estimates (Tropical Rainfall Measuring Guo, Y., Y. Zhang, L. Zhang, and Z. Wang, 2021: Regionalization
Mission, TRMM) for rainfall variability over the Pacific slope of hydrological modeling for predicting streamflow in unga-
uged catchments: A comprehensive review. Wiley Interdiscip.
and coast of Ecuador. Water, 10, 213, https://doi.org/10.3390/
Rev. Water, 8, https://doi.org/10.1002/wat2.1487.
w10020213.
Gupta, H. V., S. Sorooshian, and P. O. Yapo, 1999: Status of auto-
Espinoza, J. C., J. Ronchail, J. A. Marengo, and H. Segura, 2019:
matic calibration for hydrologic models: Comparison with
Contrasting North–South changes in Amazon wet-day and
multilevel expert calibration. J. Hydrol. Eng., 4, 135–143,
dry-day frequency and related atmospheric features
https://doi.org/10.1061/(ASCE)1084-0699(1999)4:2(135).
(1981–2017). Climate Dyn., 52, 5413–5430, https://doi.org/10.
}}, H. Kling, K. K. Yilmaz, and G. F. Martinez, 2009: Decom-
1007/s00382-018-4462-2.
position of the mean squared error and NSE performance
}}, R. Garreaud, G. Poveda, P. A. Arias, J. Molina-Carpio, M.
criteria: Implications for improving hydrological modelling.
Masiokas, M. Viale, and L. Scaff, 2020: Hydroclimate of the
J. Hydrol., 377, 80–91, https://doi.org/10.1016/j.jhydrol.2009.
Andes. Part I: Main climatic features. Front. Earth Sci., 8, 64,
08.003.
https://doi.org/10.3389/feart.2020.00064.
Hadka, D., and P. Reed, 2013: Borg: An auto-adaptive many-
Espinoza Villar, J. C., and Coauthors, 2009: Spatio-temporal rain-
objective evolutionary computing framework. Evol. Comput.,
fall variability in the Amazon basin countries (Brazil, Peru,
21, 231–259, https://doi.org/10.1162/EVCO_a_00075.
Bolivia, Colombia, and Ecuador). Int. J. Climatol., 29, He, J., K. Yang, W. Tang, H. Lu, J. Qin, Y. Chen, and X. Li,
1574–1594, https://doi.org/10.1002/joc.1791. 2020: The first high-resolution meteorological forcing dataset
Fallah, A., G. R. Rakhshandehroo, P. Berg, O. Sungmin, and R. for land process studies over China. Sci. Data, 7, 25, https://
Orth, 2020: Evaluation of precipitation datasets against local doi.org/10.1038/s41597-020-0369-y.
observations in southwestern Iran. Int. J. Climatol., 40, Helmer, E. H., E. A. Gerson, L. S. Baggett, B. J. Bird, T. S.
4102–4116, https://doi.org/10.1002/joc.6445. Ruzycki, and S. M. Voggesser, 2019: Neotropical cloud for-
Fan, Y., H. Li, and G. Miguez-Macho, 2013: Global patterns of ests and páramo to contract and dry from declines in cloud
groundwater table depth. Science, 339, 940–943, https://doi. immersion and frost. PLOS ONE, 14, e0213155, https://doi.
org/10.1126/science.1229881. org/10.1371/journal.pone.0213155.
Fernandez-Palomino, C. A., F. F. Hattermann, V. Krysanova, F. Hengl, T., M. Nussbaum, M. N. Wright, G. B. M. Heuvelink, and
Vega-Jácome, and A. Bronstert, 2020: Towards a more con- B. Gräler, 2018: Random forest as a generic framework for
sistent eco-hydrological modelling through multi-objective predictive modeling of spatial and spatio-temporal variables.
calibration: a case study in the Andean Vilcanota River PeerJ, 6, e5518, https://doi.org/10.7717/peerj.5518.
basin, Peru. Hydrol. Sci. J., 66, 59–74, https://doi.org/10.1080/ Henn, B., M. P. Clark, D. Kavetski, and J. D. Lundquist, 2015:
02626667.2020.1846740. Estimating mountain basin-mean precipitation from stream-
}}, and Coauthors, 2021: Rain for Peru and Ecuador flow using Bayesian inference. Water Resour. Res., 51, 8012–
(RAIN4PE). V. 1.0. GFZ Data Services, accessed 14 January 8033, https://doi.org/10.1002/2014WR016736.
2022, https://doi.org/10.5880/pik.2020.010. }}, }}, }}, A. J. Newman, M. Hughes, B. McGurk, and
Fox, E. W., J. M. Ver Hoef, and A. R. Olsen, 2020: Comparing J. D. Lundquist, 2018: Spatiotemporal patterns of precipita-
spatial regression to random forests for large environmental tion inferred from streamflow observations across the Sierra
data sets. PLOS ONE, 15, e0229509, https://doi.org/10.1371/ Nevada mountain range. J. Hydrol., 556, 993–1012, https://
journal.pone.0229509. doi.org/10.1016/j.jhydrol.2016.08.009.
Funk, C., and Coauthors, 2015a: The climate hazards infrared pre- Herrnegger, M., H. P. Nachtnebel, and K. Schulz, 2015: From run-
cipitation with stations - A new environmental record for off to rainfall: Inverse rainfall–runoff modelling in a high
monitoring extremes. Sci. Data, 2, 150066, https://doi.org/10. temporal resolution. Hydrol. Earth Syst. Sci., 19, 4619–4639,
1038/sdata.2015.66. https://doi.org/10.5194/hess-19-4619-2015.

Hersbach, H., and coauthors, 2020: The ERA5 global reanalysis. Laraque, A., J. Ronchail, G. Cochonneau, R. Pombosa, and J. L.
Quart. J. Roy. Meteor. Soc., 146, 1999–2049, https://doi.org/10. Guyot, 2007: Heterogeneous distribution of rainfall and dis-
1002/qj.3803. charge regimes in the Ecuadorian Amazon basin. J. Hydro-
Hong, Z., Z. Han, X. Li, D. Long, G. Tang, and J. Wang, 2021: meteor., 8, 1364–1381, https://doi.org/10.1175/2007JHM784.1.
Generation of an improved precipitation data set from multi- Le Moine, N., F. Hendrickx, J. Gailhard, R. Garçon, and F.
source information over the Tibetan Plateau. J. Hydrome- Gottardi, 2015: Hydrologically aided interpolation of daily
teor., 22, 1275–1295, https://doi.org/10.1175/JHM-D-20-0252.1. precipitation and temperature fields in a mesoscale Alpine
Hrachowitz, M., and Coauthors, 2014: Process consistency in mod- catchment. J. Hydrometeor., 16, 2595–2618, https://doi.org/10.
els: The importance of system signatures, expert knowledge, 1175/JHM-D-14-0162.1.
and process complexity. Water Resour. Res., 50, 7445–7469, Liaw, A., and M. Wiener, 2002: Classification and regression by
https://doi.org/10.1002/2014WR015484. randomForest. R News, 2, 18–22.
Hsu, K., X. Gao, S. Sorooshian, H. V. Gupta, K. Hsu, X. Gao, S. Llauca, H., W. Lavado-Casimiro, C. Montesinos, W. Santini, and
Sorooshian, and H. V. Gupta, 1997: Precipitation estimation P. Rau, 2021: PISCO_HyM_GR2M: A model of monthly
from remotely sensed information using artificial neural net- water balance in Peru (1981–2020). Water, 13, 1048, https://
works. J. Appl. Meteor., 36, 1176–1190, https://doi.org/10. doi.org/10.3390/w13081048.
1175/1520-0450(1997)036,1176:PEFRSI.2.0.CO;2. Mantas, V. M., Z. Liu, C. Caro, and A. J. S. C. Pereira, 2014: Val-
Huerta, A., C. Aybar, and W. Lavado-Casimiro, 2018: PISCO idation of TRMM Multi-satellite Precipitation Analysis
temperatura v.1.1. SENAMHI -DHI-2018, 16 pp. (TMPA) products in the Peruvian Andes. Atmos. Res., 163,
Huffman, G. J., and Coauthors, 2007: The TRMM Multisatellite 132–145, https://doi.org/10.1016/j.atmosres.2014.11.012.
Precipitation Analysis (TMPA): Quasi-global, multiyear, Manz, B., W. Buytaert, Z. Zulkafli, W. Lavado, B. Willems, L. A.
combined-sensor precipitation estimates at fine scales. Robles, and J.-P. Rodrı́guez-Sánchez, 2016: High-resolution
J. Hydrometeor., 8, 38–55, https://doi.org/10.1175/JHM560.1. satellite-gauge merged precipitation climatologies of the trop-
}}, and Coauthors, 2019: NASA Global Precipitation Measure- ical Andes. J. Geophys. Res. Atmos., 121, 1190–1207, https://
ment (GPM) Integrated Multi-satellitE Retrievals for GPM doi.org/10.1002/2015JD023788.
(IMERG). Algorithm Theoretical Basis Doc., version 06, }}, and Coauthors, 2017: Comparative ground validation of
IMERG and TMPA at variable spatiotemporal scales in the
32 pp., https://pmm.nasa.gov/sites/default/files/document_files/
tropical Andes. J. Hydrometeor., 18, 2469–2489, https://doi.
IMERG_ATBD_V06.pdf.
org/10.1175/JHM-D-16-0277.1.
Hunziker, S., and Coauthors, 2017: Identifying, attributing, and
Martens, B., and Coauthors, 2017: GLEAM v3: Satellite-based
overcoming common data quality issues of manned station
land evaporation and root-zone soil moisture. Geosci. Model
observations. Int. J. Climatol., 37, 4131–4145, https://doi.org/
Dev., 10, 1903–1925, https://doi.org/10.5194/gmd-10-1903-2017.
10.1002/joc.5037.
Miralles, D. G., T. R. H. Holmes, R. A. M. De Jeu, J. H. Gash,
Joyce, R. J., J. E. Janowiak, P. A. Arkin, P. Xie, R. J. Joyce, J. E.
A. G. C. A. Meesters, and A. J. Dolman, 2011: Global land-
Janowiak, P. A. Arkin, and P. Xie, 2004: CMORPH: A
surface evaporation estimated from satellite-based observa-
method that produces global precipitation estimates from
tions. Hydrol. Earth Syst. Sci., 15, 453–469, https://doi.org/10.
passive microwave and infrared data at high spatial and tem-
5194/hess-15-453-2011.
poral resolution. J. Hydrometeor., 5, 487–503, https://doi.org/
Morán-Tejeda, E., and Coauthors, 2016: Climate trends and vari-
10.1175/1525-7541(2004)005,0487:CAMTPG.2.0.CO;2.
ability in Ecuador (1966–2011). Int. J. Climatol., 36, 3839–
Kirchner, J. W., 2009: Catchments as simple dynamical systems:
3855, https://doi.org/10.1002/joc.4597.
Catchment characterization, rainfall-runoff modeling, and Moriasi, D. N., J. G. Arnold, M. W. Van Liew, R. L. Binger,
doing hydrology backward. Water Resour. Res., 45, W02429, R. D. Harmel, and T. L. Veith, 2007: Model evaluation
https://doi.org/10.1029/2008WR006912. guidelines for systematic quantification of accuracy in water-
Kling, H., M. Fuchs, and M. Paulin, 2012: Runoff conditions in shed simulations. Trans. ASABE, 50, 885–900, https://doi.org/
the upper Danube basin under an ensemble of climate 10.13031/2013.23153.
change scenarios. J. Hydrol., 424–425, 264–277, https://doi. Mu, Q., M. Zhao, and S. W. Running, 2011: Improvements to a
org/10.1016/j.jhydrol.2012.01.011. MODIS global terrestrial evapotranspiration algorithm.
Kneis, D., C. Chatterjee, and R. Singh, 2014: Evaluation of Remote Sens. Environ., 115, 1781–1800, https://doi.org/10.
TRMM rainfall estimates over a large Indian River Basin 1016/j.rse.2011.02.019.
(Mahanadi). Hydrol. Earth Syst. Sci., 18, 2493–2502, https:// Nash, J. E., and J. V. Sutcliffe, 1970: River flow forecasting
doi.org/10.5194/hess-18-2493-2014. through conceptual models part I }A discussion of princi-
Kobayashi, S., and Coauthors, 2015: The JRA-55 reanalysis: Gen- ples. J. Hydrol., 10, 282–290, https://doi.org/10.1016/0022-
eral specifications and basic characteristics. Kisho Shushi. 1694(70)90255-6.
Dai2shu, 93, 5–48, https://doi.org/10.2151/jmsj.2015-001. Neitsch, S. L., J. G. Arnold, J. R. Kiniry, and J. R. Williams, 2011:
Krause, P., D. P. Boyle, and F. Bäse, 2005: Comparison of differ- Soil & Water Assessment Tool Theoretical Documentation
ent efficiency criteria for hydrological model assessment. Version 2009. Texas Water Resource Institute, 647 pp.,
Adv. Geosci., 5, 89–97, https://doi.org/10.5194/adgeo-5- https://swat.tamu.edu/media/99192/swat2009-theory.pdf.
89-2005. Ochoa, A., L. Pineda, P. Crespo, and P. Willems, 2014: Evaluation
Krier, R., P. Matgen, K. Goergen, L. Pfister, L. Hoffmann, J. W. of TRMM 3B42 precipitation estimates and WRF retrospec-
Kirchner, S. Uhlenbrook, and H. H. G. Savenije, 2012: Infer- tive precipitation simulation over the Pacific–Andean region
ring catchment precipitation by doing hydrology backward: of Ecuador and Peru. Hydrol. Earth Syst. Sci., 18, 3179–3193,
A test in 24 small and mesoscale catchments in Luxembourg. https://doi.org/10.5194/hess-18-3179-2014.
Water Resour. Res., 48, W10525, https://doi.org/10.1029/ Paccini, L., J. C. Espinoza, J. Ronchail, and H. Segura, 2018:
2011WR010657. Intra-seasonal rainfall variability in the Amazon basin related

to large-scale circulation patterns: A focus on western Ama- }}, D. Ruelland, M.-P. Bonnet, J. Molina, and R. Pillco, 2019:
zon-Andes transition region. Int. J. Climatol., 38, 2386–2399, Consistency of satellite-based precipitation products in space
https://doi.org/10.1002/joc.5341. and over time compared with gauge observations and snow-
Paiva, R. C. D., W. Collischonn, and C. E. M. Tucci, 2011: Large hydrological modelling in the Lake Titicaca region. Hydrol.
scale hydrologic and hydrodynamic modeling using limited Earth Syst. Sci., 23, 595–619, https://doi.org/10.5194/hess-23-
data and a GIS based approach. J. Hydrol., 406, 170–181, 595-2019.
https://doi.org/10.1016/j.jhydrol.2011.06.007. }}, Y. Hussain, J. Molina-Carpio, R. Pillco, C. Laugner, G.
Pelletier, J. D., and Coauthors, 2016: A gridded global data set of Akhter, and M. Bonnet, 2020: Reliability of SM2RAIN pre-
soil, intact regolith, and sedimentary deposit thicknesses for cipitation datasets in comparison to gauge observations and
regional and global land surface modeling. J. Adv. Model. hydrological modelling over arid regions. Int. J. Climatol., 41,
Earth Syst., 8, 41–65, https://doi.org/10.1002/2015MS000526. E517–E536, https://doi.org/10.1002/joc.6704.
Pfannerstill, M., B. Guse, and N. Fohrer, 2014: Smart low flow sig- Scheel, M. L. M., M. Rohrer, C. Huggel, D. Santos Villar, E.
nature metrics for an improved overall performance evalua- Silvestre, and G. J. Huffman, 2011: Evaluation of TRMM
tion of hydrological models. J. Hydrol., 510, 447–458, https:// Multi-satellite Precipitation Analysis (TMPA) performance
doi.org/10.1016/j.jhydrol.2013.12.044. in the Central Andes region and its dependency on spatial
}}, K. Bieger, B. Guse, D. D. Bosch, N. Fohrer, and J. G. and temporal resolution. Hydrol. Earth Syst. Sci., 15, 2649–
Arnold, 2017: How to constrain multi-objective calibrations 2663, https://doi.org/10.5194/hess-15-2649-2011.
of the SWAT model using water balance components. J. Segura, H., C. Junquas, J. C. Espinoza, M. Vuille, Y. R. Jauregui,
Amer. Water Resour. Assoc., 53, 532–546, https://doi.org/10. A. Rabatel, T. Condom, and T. Lebel, 2019: New insights
1111/1752-1688.12524. into the rainfall variability in the tropical Andes on seasonal
Pokhrel, P., and K. K. Yilmaz, 2012: Multiple-criteria calibration and interannual time scales. Climate Dyn., 53, 405–426,
of a distributed watershed model using spatial regularization https://doi.org/10.1007/s00382-018-4590-8.
and response signatures. J. Hydrol., 418–419, 49–60, https:// Sekulić, A., M. Kilibarda, G. B. M. Heuvelink, M. Nikolić, and B.
doi.org/10.1016/j.jhydrol.2008.12.004. Bajat, 2020: Random Forest spatial interpolation. Remote
Poveda, G., J. C. Espinoza, M. D. Zuluaga, S. A. Solman, R. Sens., 12, 1687, https://doi.org/10.3390/rs12101687.
Garreaud, and P. J. van Oevelen, 2020: High impact weather Shafii, M., and B. A. Tolson, 2015: Optimizing hydrological consis-
events in the Andes. Front. Earth Sci., 8, 162, https://doi.org/ tency by incorporating hydrological signatures into model cal-
10.3389/feart.2020.00162. ibration objectives. Water Resour. Res., 51, 3796–3814, https://
Reichle, R. H., Q. Liu, R. D. Koster, C. S. Draper, S. P. P. doi.org/10.1002/2014WR016520.
Mahanama, and G. S. Partyka, 2017: Land surface precipita- Shepard, D., 1968: A two-dimensional interpolation function for
tion in MERRA-2. J. Climate, 30, 1643–1664, https://doi.org/ irregularly-spaced data. ACM 68: Proceedings of the 1968
10.1175/JCLI-D-16-0570.1. 23rd ACM National Conference, Association for Computing
Román-Cascón, C., and Coauthors, 2017: Correcting satellite- Machinery, 517–524, https://doi.org/10.1145/800186.810616.
based precipitation products through SMOS soil moisture Strauch, M., and M. Volk, 2013: SWAT plant growth modification
data assimilation in two land-surface models of different com- for improved modeling of perennial vegetation in the tropics.
plexity: API and SURFEX. Remote Sens. Environ., 200, 295– Ecol. Modell., 269, 98–112, https://doi.org/10.1016/j.ecolmodel.
310, https://doi.org/10.1016/j.rse.2017.08.022. 2013.08.013.
Saha, S., and Coauthors, 2010: The NCEP Climate Forecast Sys- }}, R. Kumar, S. Eisner, M. Mulligan, J. Reinhardt, W. Santini,
tem Reanalysis. Bull. Amer. Meteor. Soc., 91, 1015–1057, T. Vetter, and J. Friesen, 2017: Adjustment of global precipi-
https://doi.org/10.1175/2010BAMS3001.1. tation data for enhanced hydrologic modeling of tropical
Sahraei, S., M. Asadzadeh, and F. Unduche, 2020: Signature- Andean watersheds. Climatic Change, 141, 547–560, https://
based multi-modelling and multi-objective calibration of doi.org/10.1007/s10584-016-1706-1.
hydrologic models: Application in flood forecasting for Cana- Sun, Q., C. Miao, Q. Duan, H. Ashouri, S. Sorooshian, and K.
dian Prairies. J. Hydrol., 588, 125095, https://doi.org/10.1016/j. Hsu, 2018: A review of global precipitation data sets: Data
jhydrol.2020.125095. sources, estimation, and intercomparisons. Rev. Geophys., 56,
Santini, W., 2020: Caractérisation de la dynamique hydro- 79–107, https://doi.org/10.1002/2017RG000574.
sédimentaire du bassin de l’Ucayali (Pérou), par une Tall, M., and Coauthors, 2019: Towards a long-term reanalysis of
approche intégrant réseau de mesures, télédétection et modé- land surface variables over western Africa: LDAS-Monde
lisation hydrologique. Université Toulouse III - Paul Saba- applied over Burkina Faso from 2001 to 2018. Remote Sens.,
tier, 479 pp., https://hal.archives-ouvertes.fr/tel-03276962/. 11, 735, https://doi.org/10.3390/rs11060735.
}}, and Coauthors, 2015: Sediment budget in the Ucayali River Tamayo, G. G., 2017: Evaluación de los caudales lı́quidos y de
basin, an Andean tributary of the Amazon River. Proc. Int. producción de sedimentos estimados con el modelo Soil
Assoc. Hydrol. Sci., 367, 320–325, https://doi.org/10.5194/ Water Assessment Tool (SWAT) y su relación con los cau-
piahs-367-320-2015. dales lı́quidos y sólidos observados. Escuela Politécnica
}}, and Coauthors, 2019: An index concentration method for Nacional Quito, accessed 22 October 2020, http://bibdigital.
suspended load monitoring in large rivers of the Amazonian epn.edu.ec/handle/15000/17067.
foreland. Earth Surf. Dyn., 7, 515–536, https://doi.org/10.5194/ Tan, M. L., P. W. Gassman, X. Yang, and J. Haywood, 2020: A
esurf-7-515-2019. review of SWAT applications, performance and future needs
Satgé, F., M.-P. Bonnet, M. Gosset, J. Molina, W. Hernan Yuque for simulation of hydro-climatic extremes. Adv. Water
Lima, R. Pillco Zolá, F. Timouk, and J. Garnier, 2016: Resour., 143, 103662, https://doi.org/10.1016/j.advwatres.2020.
Assessment of satellite rainfall products over the Andean pla- 103662.
teau. Atmos. Res., 167, 1–14, https://doi.org/10.1016/j. Tang, W., 2019: Dataset of high-resolution (3 hour, 10 km) global
atmosres.2015.07.012. surface solar radiation (1983–2018). National Tibetan Plateau

Data Center, accessed 17 January 2021, https://doi.org/10. Remoto, São Paulo, Brazil, INPE, 28–31, https://proceedings.
11888/Meteoro.tpdc.270112. science/sbsr/papers/an-update-of-xavier–king-and-scanlon–2016–
}}, K. Yang, J. Qin, X. Li, and X. Niu, 2019: A 16-year dataset daily-precipitation-gridded-data-set-for-the-brazil.
(2000–2015) of high-resolution (3 h, 10 km) global surface Xu, X., S. K. Frey, A. Boluwade, A. R. Erler, O. Khader, D. R.
solar radiation. Earth Syst. Sci. Data, 11, 1905–1915, https:// Lapen, and E. Sudicky, 2019a: Evaluation of variability
doi.org/10.5194/essd-11-1905-2019. among different precipitation products in the northern Great
Tarek, M., F. P. Brissette, and R. Arsenault, 2020: Evaluation of Plains. J. Hydrol. Reg. Stud., 24, 100608, https://doi.org/10.
the ERA5 reanalysis as a potential reference dataset for 1016/j.ejrh.2019.100608.
hydrological modelling over North America. Hydrol. Earth Xu, Z., Z. Wu, H. He, X. Wu, J. Zhou, Y. Zhang, and X. Guo,
Syst. Sci., 24, 2527–2544, https://doi.org/10.5194/hess-24-2527- 2019b: Evaluating the accuracy of MSWEP V2.1 and its per-
2020. formance for drought monitoring over mainland China.
Teuling, A. J., I. Lehner, J. W. Kirchner, and S. I. Seneviratne, Atmos. Res., 226, 17–31, https://doi.org/10.1016/j.atmosres.
2010: Catchments as simple dynamical systems: Experience 2019.04.008.
from a Swiss prealpine catchment. Water Resour. Res., 46, Xue, X., K. Zhang, Y. Hong, J. J. Gourley, W. Kellogg, R. A.
W10502, https://doi.org/10.1029/2009WR008777. McPherson, Z. Wan, and B. N. Austin, 2016: New multisite
Tobar, V., and G. Wyseure, 2018: Seasonal rainfall patterns classi- cascading calibration approach for hydrological models: Case
fication, relationship to ENSO and rainfall trends in Ecuador. study in the Red River Basin using the VIC model. J. Hydrol.
Int. J. Climatol., 38, 1808–1819, https://doi.org/10.1002/joc. Eng., 21, 05015019, https://doi.org/10.1061/(ASCE)HE.1943-
5297. 5584.0001282.
Towner, J., H. L. Cloke, E. Zsoter, Z. Flamig, J. M. Hoch, J. Yamazaki, D., S. Kanae, H. Kim, and T. Oki, 2011: A physically
Bazo, E. Coughlan de Perez, and E. M. Stephens, 2019: based description of floodplain inundation dynamics in a
Assessing the performance of global hydrological models for global river routing model. Water Resour. Res., 47, 4501,
capturing peak river flows in the Amazon basin. Hydrol. https://doi.org/10.1029/2010WR009726.
Earth Syst. Sci., 23, 3057–3080, https://doi.org/10.5194/hess-23- }}, and Coauthors, 2017: A high-accuracy map of global terrain
3057-2019. elevations. Geophys. Res. Lett., 44, 5844–5853, https://doi.org/
van Werkhoven, K., T. Wagener, P. Reed, and Y. Tang, 2009: 10.1002/2017GL072874.
Sensitivity-guided reduction of parametric dimensionality for Yilmaz, K. K., H. V. Gupta, and T. Wagener, 2008: A process-
multi-objective calibration of watershed models. Adv. Water based diagnostic approach to model evaluation: Application
Resour., 32, 1154–1169, https://doi.org/10.1016/j.advwatres. to the NWS distributed hydrologic model. Water Resour.
2009.03.002. Res., 44, W09417, https://doi.org/10.1029/2007WR006716.
Weedon, G. P., G. Balsamo, N. Bellouin, S. Gomes, M. J. Best, Zambrano-Bigiarini, M., A. Nauditt, C. Birkel, K. Verbist, and L.
and P. Viterbo, 2014: The WFDEI meteorological forcing Ribbe, 2017: Temporal and spatial evaluation of satellite-based
data set: WATCH Forcing data methodology applied to rainfall estimates across the complex topographical and climatic
ERA-Interim reanalysis data. Water Resour. Res., 50, 7505– gradients of Chile. Hydrol. Earth Syst. Sci., 21, 1295–1320,
7514, https://doi.org/10.1002/2014WR015638. https://doi.org/10.5194/hess-21-1295-2017.
Wongchuig Correa, S., R. C. D. de Paiva, J. C. Espinoza, and W. Zubieta, R., A. Getirana, J. C. Espinoza, and W. Lavado, 2015:
Collischonn, 2017: Multi-decadal hydrological retrospective: Impacts of satellite-based precipitation datasets on rainfall–-
Case study of Amazon floods and droughts. J. Hydrol., 549, runoff modeling of the Western Amazon basin of Peru and
667–684, https://doi.org/10.1016/j.jhydrol.2017.04.019. Ecuador. J. Hydrol., 528, 599–612, https://doi.org/10.1016/j.
Wu, W., Y. Li, X. Luo, Y. Zhang, X. Ji, and X. Li, 2019: Perfor- jhydrol.2015.06.064.
mance evaluation of the CHIRPS precipitation dataset and }}, }}, J. C. Espinoza, W. Lavado-Casimiro, and L. Aragon,
its utility in drought monitoring over Yunnan Province, 2017: Hydrological modeling of the Peruvian–Ecuadorian
China. Geomatics Nat. Hazards Risk, 10, 2145–2162, https:// Amazon Basin using GPM-IMERG satellite-based precipita-
doi.org/10.1080/19475705.2019.1683082. tion dataset. Hydrol. Earth Syst. Sci., 21, 3543–3555, https://
Xavier, A. C., C. W. King, and B. R. Scanlon, 2016: Daily gridded doi.org/10.5194/hess-21-3543-2017.
meteorological variables in Brazil (1980–2013). Int. J. Clima- Zulkafli, Z., and Coauthors, 2014: A comparative performance
tol., 36, 2644–2659, https://doi.org/10.1002/joc.4518. analysis of TRMM 3B42 (TMPA) versions 6 and 7 for hydro-
}}, }}, and B. R. Scanlon, 2017: An update of Xavier, King logical applications over Andean–Amazon River Basins.
and Scanlon (2016) daily precipitation gridded data set for J. Hydrometeor., 15, 581–592, https://doi.org/10.1175/JHM-D-
the Brazil. Proc. XVIII Simpósio Brasileiro de Sensoriamento 13-094.1.

Rain4pe

Uploaded by

Copyright:

Available Formats

Rain4pe

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Rain4pe

Uploaded by

Copyright:

Available Formats

MARCH 2022 FERNANDEZ-PALOMINO ET AL.

A Novel High-Resolution Gridded Precipitation Dataset for Peruvian and

CARLOS ANTONIO FERNANDEZ-PALOMINO,a,b FRED F. HATTERMANN,a VALENTINA KRYSANOVA,a

(Manuscript received 2 December 2020, in ﬁnal form 20 November 2021)

1. Introduction and operational applications. Rain gauge observations are the

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Dataset Data Spatial Spatial Temporal Temporal

TABLE 2. Data used for hydrological modeling.

Data type Resolution Description/source

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

2017). SWAT is one of the most widely used ecohydrological

observation in a tree’s terminal node (set at 5, default value);

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Criterion (reference) Equation Description

Signature measures based on ﬂow duration curve (FDC)

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Basin Product KGE NSE lNSE |PBIAS| |Slow| |Shigh| |Speak|

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

5. Discussion parameter values to counterbalance precipitation uncertainty

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

ERA5 CHIRP CHIRPS MSWEP PISCO RAIN4PE

ERA5 CHIRP CHIRPS MSWEP PISCO RAIN4PE

ERA5 CHIRP CHIRPS MSWEP PISCO RAIN4PE

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

APPENDIX C Andean catchments. The correlation coefﬁcient (Fig. C1,

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

REFERENCES Levizzani et al., Eds., Advances in Global Change Research,

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

Unauthenticated | Downloaded 03/30/22 07:57 PM UTC

You might also like