Streamflow simulation across the tropics is limited by the lack of data to calibrate and validate large-scale hydrological models. Here, we applied the process-based, conceptual HYPE (Hydrological Predictions for the Environment) model to quantitatively assess Costa Rica's water resources at a national scale. Data scarcity was compensated for by using adjusted global topography and remotely sensed climate products to force, calibrate, and independently evaluate the model. We used a global temperature product and bias-corrected precipitation from Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) as model forcings. Daily streamflow from 13 gauges for the period 1990–2003 and monthly Moderate Resolution Imaging Spectroradiometer (MODIS) potential evapotranspiration (PET) and actual evapotranspiration (AET) for the period 2000–2014 were used to calibrate and evaluate the model applying four different model configurations (M1, M2, M3, M4). The calibration consisted of step-wise parameter constraints preserving the best parameter sets from previous simulations in an attempt to balance the variable data availability and time periods. The model configurations were independently evaluated using hydrological signatures such as the baseflow index, runoff coefficient, and aridity index, among others. Results suggested that a two-step calibration using monthly and daily streamflow (M2) was a better option than calibrating only with daily streamflow (M1), with similar mean Kling–Gupta efficiency (KGE
Tropical regions differ from temperate regions by their larger energy inputs, more intense atmospheric dynamics, higher precipitation rates, larger streamflow, and greater sediment yields (Dehaspe et al., 2018; Esquivel-Hernández et al., 2017; Wohl et al., 2012). Moreover, tropical regions are among the fastest-changing environments, with a hydrological cycle pressurized by population growth (Wohl et al., 2012; Ziegler et al., 2007), land use/cover modifications (Gibbs et al., 2010), and altered precipitation and runoff patterns (Esquivel-Hernández et al., 2017) due to climate change. Central America, the northern boundary of the humid tropics, was identified by Giorgi (2006) as the most sensitive tropical region to climate change due to the location between two major water bodies, the Pacific Ocean and the Caribbean Sea.
Increasing concerns about the effects of human activities and climate change on tropical catchments demand accurate quantification of the water balance components in space and time to guarantee future water resources availability for ecosystems and socioeconomic activities (Esquivel-Hernández et al., 2017; Wohl et al., 2012). Hydrological models have been widely used to assess the spatiotemporal variability of water resources and to provide insights into potential future climate and management decisions (Andersson et al., 2015; Xiong and Zeng, 2019).
However, models also implicitly include many uncertainties (Beven, 2012). For example, Birkel et al. (2020) and Dehaspe et al. (2018) highlighted those hydrological models that are useful for predicting streamflow but showed limitations to assess water partitioning and storage changes required for water management in the humid tropics. Modeling in the tropics is further hampered by the lack of good quality hydrometric data used to drive models and for calibration (Westerberg and Birkel, 2015; Westerberg et al., 2014). Moreover, a decrease in hydrological measurements and monitoring networks in many tropical regions has occurred during the last 3 decades (Wohl et al., 2012), limiting the applicability of hydrological models or reducing their performance in simulating streamflow in Central America (Westerberg et al., 2014) and South America (Guimberteau et al., 2012). Model calibration mostly leads to several combinations of parameters with similar streamflow response, i.e., equifinality (Beven, 2012; Xiong and Zeng, 2019), and it is therefore desirable to reduce or constrain the uncertainty of model parameters. Moreover, some case studies around the world have found that soil model parameters can be relatively insensitive to streamflow simulations (Massari et al., 2015; Rajib et al., 2018b; Silvestro et al., 2015).
Opportunities to provide more realistic internal hydrological partitioning exist in the form of including additional variables to streamflow, such as, e.g., tracers and remotely sensed variables of evapotranspiration and soil moisture (Dal Molin et al., 2020; Rakovec et al., 2016; Xiong and Zeng, 2019; Massari et al., 2015). The latter, however, may come at the expense of increased complexity for model calibration and evaluation (Arheimer et al., 2020; Her and Seong, 2018; Massari et al., 2015; Xiong and Zeng, 2019; Zhang et al., 2018) since non-linearity increases complexity in data assimilation (Massari et al., 2015; Rajib et al., 2018a, b). In addition, hydrological signatures can improve model realism through the synthesis of many simultaneous catchment processes at different scales (Arheimer et al., 2020; Sawicz et al., 2011). Hydrological signatures can be used to increase our understanding of water balance partitioning and hydrological similarity across different scales (e.g., Arciniega-Esparza et al., 2017; Beck et al., 2015; Kirchner, 2009; Troch et al., 2009) and have been applied to improve model evaluation (e.g., Andersson et al., 2015; Arheimer et al., 2020; Dal Molin et al., 2020; Raphael Tshimanga and Hughes, 2014; Westerberg et al., 2014). Despite uncertainties in observed hydrological signatures (Westerberg and McMillan, 2015), there is potential to identify model weaknesses and to ultimately produce a more well-balanced catchment representation.
Most hydrological models have been developed since the 1970s to solve different needs at catchment scales (Pechlivanidis and Arheimer, 2015; Todini, 2007). Nevertheless, water management increasingly requires detailed hydrological information over larger, aggregated spatial domains instead of a single catchment (Arheimer et al., 2020; Rojas-Serna et al., 2016). Global hydrological models can serve this purpose but suffer from rather coarse spatial resolution and increased computational cost (Kumar et al., 2013; Sood and Smakhtin, 2015). Distributed landscape characteristics at large scales such as soil, topography, and land cover can result in complex hydrological models with many calibrated model parameters (Gurtz et al., 1999) and result in greater uncertainty. However, distributed model parameterization based on landscape characteristics also promises the advantage of predicting the hydrological response of ungauged basins (Hrachowitz et al., 2013; Pechlivanidis and Arheimer, 2015). Therefore, the question as to how complex or simple a hydrological model should be remains an open scientific debate considering that simpler models can lead to similar results in comparison with more complex and more highly parameterized models (Archfield et al., 2015; Rojas-Serna et al., 2016).
An alternative to simulate the hydrology at large spatial scales is by means of
semi-distributed, conceptual hydrological models together with global data
of precipitation, evapotranspiration, and soil moisture (Andersson et al.,
2015; Brocca et al., 2020). Conceptual models fall in the category between
very simple bucket models and physically based, distributed models, limiting the numbers of parameters while still being able to gain insights into the hydrological processes governing a set of
neighboring catchments (e.g., Beven, 2012 for a model classification). Moreover, recent hydrological studies have implemented data assimilation from
remote sensing and global products of soil moisture (Kwon et al., 2020;
Massari et al., 2015; Silvestro et al., 2015), snow depth (Infante-Corona et
al., 2014), evapotranspiration (Lin et al., 2018; Rajib et al., 2018a, b), and terrestrial water storage (Getirana et al., 2020; Reager et al., 2015), often in combination with conceptual models in order to reduce or constrain the model parameter uncertainty and to help with model evaluation (e.g., Sheffield et al., 2018). Such an approach needs testing in tropical regions such as Central America, located on the narrow continental bridge (
Therefore, this paper aims to test the use of the large-scale conceptual but process-based semi-distributed HYPE model (Lindström et al., 2010), exploring strategies to improve regional modeling of tropical data-scarce regions, and incorporating different time steps and global gridded products for the complex topographical regions of Costa Rica. We, therefore, additionally used the potential evapotranspiration (PET) and actual evapotranspiration (AET) products, respectively, from MODIS (Moderate Resolution Imaging Spectroradiometer) in order to streamflow time series to calibrate the model followed by a posteriori independent evaluation of hydrological signatures calculated from these global data sets. The model was calibrated using a step-wise procedure tracking the most effective strategy to constrain the parameter space and to reduce the model uncertainty.
Our specific objectives are the following.
Adjust the open-source, conceptual rainfall–runoff model HYPE to simulate Costa Rica's catchment hydrology at the national scale using remotely sensed global climate data and landscape products to drive and evaluate the model under four different step-wise calibration strategies. Analyze the effect of remotely sensed PET and AET data on model calibration and their capability to improve the simulated water balance and matching hydrological signatures.
The study area corresponds to Costa Rica, located on the Central American
Isthmus, between 8 and 11
Figure 1a shows the study area boundaries, the precipitation gauges (blue dots), the monitored catchments (red polygons), and their respective streamflow gauges (black squares), as well as the catchments used within the HYPE model (gray polygons). In situ data consisted of 75 precipitation stations obtained from the National Meteorological Service (IMN in Spanish)
containing a minimum length of 10 years of data overlapping the period from 1981 to 2017. This period was selected to compare ground precipitation
records with precipitation from global remote sensing products. Moreover, 13 streamflow gauges with daily records from 1990 to 2003 were obtained from
the Costa Rican Electricity Institute (ICE). The attributes and climate
properties of monitored catchments are shown in Table 1, with catchment
areas ranging from 74 to 4772 km
Study area
Physical and climatological properties of the monitored catchments. Streamflow gauges were grouped according to their location in the Caribbean and the Pacific basins. AI stands for aridity index and EI stands for evaporative index.
Regarding model simulations, more than 600 nested catchments covering the
whole country were delineated using the 30 m Shuttle Radar Topography Mission (SRTM) elevation model (Bamler, 1999) and the terrain analysis
toolset from SAGA GIS v.6.4 (Conrad et al., 2015), where the fill sinks algorithm by Wang and Liu (2006) was applied with a minimum slope parameter of 0.0001
Figure 1b and c shows the soil types and land uses across Costa Rica, respectively. Soil types were derived from SoilGrids (Hengl et al., 2017; see dataset description in Table 2) and compared to national-scale soil maps. Sand content and clay content at 1 m depth were used to classify the
soil types from the USDA classification criteria in SAGA GIS tools. Furthermore, in order to reduce the number of model parameters, only the
four most frequent soil types were considered (Fig. 1b). The predominant soil texture is clay loam covering an area of
Remote sensing and global products used in this study.
The climatological space–time series were obtained from remote sensing and
global products, described in Table 2. The precipitation grid was obtained
from the Climate Hazards Group InfraRed Precipitation with Satellite data (CHIRPS) version 2 (Funk et al., 2015), and the mean daily temperature was obtained from the CPC Global Daily Temperature product provided by the
NOAA/OAR/ESRL PSL (
The yearly cycle of wet and dry deviations in the ocean–atmosphere is linked to changes in the sea surface temperature of both the Pacific Ocean and the Caribbean Sea, where the El Niño Southern Oscillation (ENSO) is associated with a decrease of the mean annual precipitation across the Pacific Basin, and an increase of precipitation in the Caribbean Basin (Muñoz et al., 2008).
Moreover, the cold-phase La Niña is the cause of an increase in precipitation in the Pacific Basin and a decrease in the Caribbean (Waylen
et al., 1996). Overall, the mean annual precipitation averaged
The rainfall patterns across Costa Rica are reflected in the streamflow
responses of catchments on the Pacific and Caribbean sides. The daily streamflow tends to be higher in the Caribbean Basin (9.2 mm d
Potential evapotranspiration and actual evapotranspiration were obtained from the MODIS product (Mu et al., 2011) distributed by the Numerical Terradynamic Simulation Group at the University of Montana, USA, which compared well to the few available ground stations in Costa Rica, with errors from
Figure 1f shows the mean annual AET from MODIS, which spatially ranges from 547 to 1612 mm. The highest AET values were observed at the coast (Caribbean and Pacific). Moreover, the lower AET values overlap with low humidity zones and sparse vegetation areas (northwestern Costa Rica), as well as higher elevation cloud cover that decreases soil evaporation (Caribbean slope mountain region).
We used Hydrological Predictions for the Environment (HYPE) version 5.9, a semi-distributed hydrological model, for the assessment of water resources and water quality at small and large scales (Lindström et al., 2010) in order to simulate the hydrological response of Costa Rican catchments. The HYPE model can be considered as the evolution of the distributed Hydrologiska Byråns Vattenbalansavdelning (HBV) model (Lindström et al., 1997). HYPE was developed by the Swedish Meteorological and Hydrological Institute (SMHI) as the operative model for drought and flood forecasting across Sweden (Pechlivanidis et al., 2014). Moreover, HYPE was recently applied to other climatic regions (Andersson et al., 2017; Arheimer et al., 2018; Berg et al., 2018; Lindström, 2016; Pugliese et al., 2018; Tanouchi et al., 2019), including a global-scale application (Arheimer et al., 2020).
The HYPE model allows simulating the water balance and nutrient fluxes at a
daily or sub-daily scale using precipitation and temperature as forcings
(SMHI, 2018). The model structure (Fig. 2a) describes the major water
pathways and fluxes, ensuring mass conservation at the catchment and sub-catchment scale. Furthermore, each sub-catchment is divided into the
most fundamental spatial soil and land use classes (SLCs) depending on the
classification of soil types, land cover, climate, and elevation, as shown in Fig. 2b. The SLCs in HYPE provide the capability to predict streamflows in ungauged basins since the parameters that regulate the fluxes and storages are linked to each SLC, with a maximum of three layers of different soil thickness, as shown in Fig. 2b. Water bodies such as lakes and rivers may be considered as an SLC, where lakes can be defined as natural lakes or regulated dams with multiple water outputs. For full details of the HYPE model, see the description by Lindström et al. (2010) and the open-access code references located at
Schematic representation of the HYPE model
For this study, Costa Rica was divided into 605 catchments (Fig. 1a) with
12 SLCs obtained from the spatial combination of soil types and land cover
maps shown in Fig. 1b and c, respectively. Outlet lakes (which discharge to downstream catchments) and internal lakes (which discharge into the main river or tributaries) were set up as different SLCs to consider the water bodies that regulate the streamflow. The largest water body in Costa Rica is the Arenal reservoir, located in the San Carlos River catchment (Fig. 1b). The Arenal reservoir is an artificial lake used for hydropower purposes with an average surface area of 87.8 km
Soil thickness varied for different SLCs, with a maximum soil thickness of 3 m under forests and a minimum of 2 m for bare soil cover, following Arheimer et al. (2020). Furthermore, delimited catchments were classified according to their elevation and location (Pacific Basin and Caribbean Basin), applying regional factors to correct the hydrological behavior of lowland and mountainous catchments with similar SLCs, resulting in six defined regions.
Daily time series of precipitation from CHIRPS and temperature from NOAA for
the period 2000–2014 were extracted for each catchment using GRASS GIS (Neteler et al., 2012), where datasets were resampled to 1 km using the
nearest-neighbor criteria and spatially averaged for each catchment. The
climatological forcings were resampled due to the small size of some catchments (area of
Rainfall estimations from satellites are subject to systematic errors that
may produce uncertainty in hydrological simulations (Goshime et al., 2019;
Grillakis et al., 2018; Infante-Corona et al., 2014; Wörner et al., 2019). The CHIRPS product already incorporates a bias correction procedure
but uses only a few concentrated ground stations in Costa Rica. The performance of
CHIRPS estimating annual water balances is shown in Fig. S1a and c in the Supplement, with frequent underestimation in the monitored catchments. Therefore, we applied a linear scaling to further correct for the bias between the product and ground precipitation from 75 available stations across Costa Rica (Fig. 1a). The corrected precipitation was estimated as
Some monitored catchments exhibiting higher annual streamflow than annual
precipitation could not be corrected due to groundwater contributions from
neighboring catchments (Genereux and Jordan, 2006; Genereux et al., 2002), under-catch at rainfall gauges (Frumau et al., 2011), and the insufficient number of precipitation stations to correct the CHIRPS database at a national scale. Nevertheless, errors in climatological data have been found to be the most common issue for water balance modeling in Central America (Westerberg et al., 2014; Birkel et al., 2012). In that sense, an additional approach was implemented to reduce the unrealistic relationship between streamflow and precipitation, which consisted of the creation of virtual stations at the catchment centroid, where the new bias factor was computed as
Finally, BF points from precipitation stations and BF
HYPE incorporates four methods for PET estimation (SMHI, 2018). After initial tests, we found that the monthly PET signal from MODIS in Costa Rica can be reproduced by only using temperature as forcing, where PET is computed as
Parameter names and initial ranges of the step-wise parameter estimation for each model configuration. Columns M1, M2, M3, and M4 correspond to parameters optimized during each step of the configuration,
and
Figure 3 shows the workflow adopted for model calibration, which involves a qualitative parameter sensitivity analysis to find the most suitable range of values for the automatic calibration. The initial parameter ranges were obtained from manual iterations of one parameter at a time to facilitate automatic calibration (Infante-Corona et al., 2014).
Schematic representation of the HYPE model calibration strategy considering a step-wise procedure to constrain parameters. Four model configurations (M1, M2, M3, M4) were established using different data sets and/or different timescales. From each calibration step, the 10th and 90th values of the best-fit parameters were used to constrain the parameters of the next step.
We considered four model configurations to analyze the effect of including
PET and AET in model calibration:
model configuration 1 (M1), calibrated using only daily streamflow ( model configuration 2 (M2), calibrated using monthly streamflow followed by daily streamflow; model configuration 3 (M3), incorporating a calibration using monthly PET and AET, followed by daily streamflow; model configuration 4 (M4), similar to M3, additionally using monthly streamflow before daily streamflow calibration.
For comparison purposes, M1 was chosen as the baseline model configuration, which usually is standard in hydrological practice. The steps are described in Fig. 3. The common period between
The streamflow records were divided into the period from 1991 to 1999 for calibration and from 2000 to 2003 for validation. The PET and AET calibration period was established as 2002 to 2010 and the validation period as 2011 to 2014. In both cases, for model warm-up we ran it for 2 years prior to calibration since our modeling tests showed that using 2 years was enough to stabilize the effects of initial conditions of water content in soil layers, rivers, and reservoirs. The 13 monitored catchments were used for streamflow calibration. For PET and AET calibration steps, only the 130 catchments within the 13 monitored catchments were used since our tests showed that using the 605 catchments did not significantly increase the model performance but it did increase the calibration time by a factor of 5. The simulations of the 605 catchments were used to compute the metrics for the calibration and validations periods.
A total of 86 parameters were used to build the HYPE model structure consisting of 36 parameters linked to four soil types, 24 parameters linked to four land cover classes, 6 parameters for the general structure, 12 parameters for the regional correction of PET and temperature, and 8 parameters for lake discharge. The Monte Carlo (MC) routine for parameter sampling and sensitivity analysis included in HYPE was used for calibration, and the model configurations were run 10 000 times for each step, except for M1, which used 20 000 runs to cover more parameter combinations since this configuration only used daily streamflow. Despite the lower computational efficiency of the MC with respect to other optimization schemes (such as gradient-based methods), the MC routines are more flexible in accounting for multiple parameters sets in complex models (Beven, 2006). The 10th and 90th percentiles of the resulting parameters from the best 100 runs were used to constrain the parameters for the next calibration step.
The CHIRPS product was evaluated with ground records using the false alarm
rate (FAR, computed with Eq. 7), probability of detection (PD, computed
with Eq. 8), and threat score (TS, computed with Eq. 9):
Hydrological signatures used as independent performance evaluation criteria.
The model performance was evaluated using the Kling–Gupta efficiency (KGE; Kling and Gupta, 2009), computed as
Finally, the non-parametric Kruskal–Wallis (Kruskal and Wallis, 1952) and Mann–Whitney (Mann and Whitney, 1947) tests were used to detect statistically different performance.
Comparing precipitation from CHIRPS with annual streamflow and streamflow
plus evapotranspiration (assuming long-term balance
Performance of the CHIRPS precipitation product in representing observed rainfall in Costa Rica.
For modeling purposes, we evaluated the temporal synchronicity of rainfall
versus streamflow (Fig. 4b) using cross-correlation between daily streamflow and catchment-scale daily precipitation from CHIRPS, where the
The bias correction improved the annual precipitation where CHIRPSc was
consistent with annual streamflow, and the long-term water balance was mostly preserved, as observed in Fig. S1b and d. Figure S2 shows the MAE normalized by mean precipitation for CHIRPS and bias-corrected CHIRPS (CHIRPSc), both with respect to the 75 precipitation stations, where boxplots correspond to the variability of normalized MAE estimated by each point. The average normalized MAE at a daily scale was estimated at
Figure S2d shows the probability of success and failure of CHIRPSc at detecting rainy or dry days with respect to ground stations, where the probability was computed from a single time series merged from the 75 station records. Furthermore, Fig. S2e to g show the false alarm
rate (FAR), probability of detection (PD), and threat score (TS), respectively. Results indicated that CHIRPSc detected true rainy and dry days with a similar probability (0.31 to 0.34) as in situ observed rainfall, whereas the FAR ranged from 0.15 to 0.38 with the greater values (i.e., incorrect detection of dry days as rainy days by CHIRPS) to the southeast, and the PD showed greater values (i.e., better performance of CHIRPS at detecting rainy days) for the Pacific Basin (median of
Figure 5 shows the comparison of the model configurations' performance for
the calibration (dark blue) and validation (light blue) periods.
Simulated daily streamflow for the 13 gauged catchments was similar for
baseline configuration (M1) and M2 during the calibration period (1991–1999; Fig. 5a) with a mean KGE of
The range of KGE values for the calibration (dark blue) and validation periods (light blue).
The configuration M2 best reproduced monthly streamflow for the calibration
period (Fig. 5b), with a mean KGE of
Figure 5c and d show the effect of including AET and PET in the calibration steps, and the KGE was computed by aggregating the complete domain (605 nested catchments). The calibration consisted of 130 nested catchments within the monitored catchments. Furthermore, M1 and M2 were only plotted for comparison purposes since these configurations were calibrated with streamflow. From Fig. 5c, we observed that simulated monthly AET for the calibration period (2002–2010) improved for M3 and M4 with a mean KGE of
The results from Fig. 6 suggested that the best performance of daily and
monthly streamflow for the calibration period (2001–2009) was obtained for
catchments in the southeast of Costa Rice, such as the Palmar, Caracucho, El Rey, and Guapinol catchments, with KGEs higher than 0.55 (NSE
Matrix of spatially distributed KGE results for the calibration period (streamflow from 1991 to 1999, PET and AET from 2001 to 2010), where green and blue reflect better performance. Configurations M1 and M2 were calibrated only with streamflow. Nevertheless, PET and AET panels are compared to show the effect of including such variables in the calibration procedure. The mean
The spatially distributed KGE on the last two panels of Fig. 6 shows the improvement by including AET and PET in the calibration steps (Fig. 6k,
l, o and p), unlike the case of daily and monthly
Figure 7 shows the model parameter ranges from the 100 best-fit simulations resulting from the last calibration step for each configuration. The red dots from Fig. 7 correspond to the optimal parameters used for modeling, where multiple red dots and boxplots for each model are shown by soil type and land use.
A posteriori parameter distribution for the 100 best-fit simulations from the last calibration step for each configuration, where red dots correspond to the optimal parameters. Multiple boxplots for each model configuration correspond to the parameters of different soil and land classes. The top two rows of parameter panels correspond to streamflow components, the third row to water content parameters, and the fourth row to PET and AET processes (see Table 3 for reference). For comparison purposes, the bottom row shows the spatial variability of the best-fit calibrated srrate (–) parameter for each configuration.
A large dispersion with a coefficient of variation (CV
The soil type and land use coverage influence the calibrations' parametrization. M2 and M4 showed constrained distributions of parameters srrate and rrcs1 for clay-loam soil (third class), the most frequent soil type in the monitored catchments (Fig. 1b). The bottom panel in Fig. 7 shows the spatial distribution of the srrate parameter, with similar values for M2, M3, and M4 and the most frequent soil classes (clay and clay-loam).
The soil parameters that regulate the soil water content (wcwp, wcep) showed
similar distributions with the median value of the fraction of soil water
available for evapotranspiration (wcfc). The effective porosity (wcep) was
slightly higher for configurations M1 and M2, but the final parameters (red
dots) differed between the models. Furthermore, for M3 and M4, the parameters lp
and cevpam exhibited constrained distributions with a CV of 0.12 and 0.11,
respectively. In comparison, M1 and M2 showed CV values of
The step-wise calibration improved model performance in different aspects. Figure 8 shows the comparison of the hydrological simulations for two monitored catchments contrasting the best simulation with the highest KGE performance (Palmar catchment) and the worst simulation with the lowest KGE performance (Rancho Rey catchment).
Simulated versus observed time series for catchments with the best streamflow KGE performance (Palmar watershed) and the worst streamflow simulation (Rancho Rey watershed). Black lines represent streamflow observations from 1990 to 2003. The panels on the right
The Palmar catchment exhibited acceptable performance (KGE
At a monthly scale, streamflow was preserved by the model configurations in
several catchments, except for Rancho Rey, where simulated streamflow was on
average 2 times larger than the observed streamflow during the rainy season
(Fig. 8l). Such overestimation indicated that the bias factor was insufficient to correct the global precipitation product or for large discharge measurement errors. Furthermore, all configurations reproduced the seasonality of AET and PET from MODIS (not shown), but M3 and M4 underestimated the AET and PET in Palmar while showing good performance for AET in Rancho Rey. Moreover, simulated monthly soil moisture (SM) content was independently compared with the catchment average soil moisture content from the Land Parameter Retrieval Model (LPRM; Owe et al., 2008) product for the period 2012 to 2016. The simulated SM for M1 followed the seasonal behavior of the LPRM product in the Palmar catchment, matching the absolute LPRM % SM content. The LPRM product uses SM from the upper 5 cm against the 50 cm of the upper layer defined for all model configurations. However, all model configurations show a high correlation (CC
A comparison of observed flow duration curves (FDCs) as a hydrological signature and the matching simulated FDC for each model configuration. The simulated period was from 1991 to 2003.
The observed and simulated flow duration curves for all monitored catchments are shown for the period 1991–2003 in Fig. 9. The baseline (M1, dashed red line) underestimated the median and low flows in several catchments (Guardia, Rancho Rey, Guatuso, Terron Colorado, Caracucho, El Rey, Guapinol), with a median RMSE of 1.15 considering all monitored catchments, 2 times larger than other model configurations. M2 (dashed blue line) exhibited the best performance for median and low flows (with a median RMSE of 0.42), whereas M3 (orange line) and M4 (blue line) showed similar results to M2. Higher efficiencies for median and low flows were obtained for catchments that exhibited higher cross-correlation with precipitation (Fig. 4b), as was the case for Palmar, Caracucho, and El Rey, among others.
The simulated and observed hydrological signatures are shown in Fig. 10, where simulations covered the period 1991–2014, and observations covered different periods depending on available records. The comparison of hydrological signatures by monitored watershed is shown in Tables S1 and S2 in the Supplement. The simulated long-term mean annual water balance (
Observed and simulated spatial distribution of
hydrological signatures. BFI: baseflow index (baseflow/streamflow), EI: evaporative index (AET/
The spatial distribution of baseflow indices (BFI) derived from M2, M3, and M4 exhibited similarities with respect to the observations. Simulated BFI showed an overwhelming groundwater contribution to streamflow with relatively similar average values of 0.70, 0.69, 0.68, and 0.74 for model configurations M1, M2, M3, and M4, respectively.
Larger differences were observed in the northwest and southwest when comparing the BFI of M1 with respect to other configurations, whereas M4 resulted in larger contributions of baseflow to streamflow in coastal areas of the Caribbean. Similar spatial patterns were obtained for the streamflow coefficient (
The EI and AI were similar for M3 and M4 due to their similar model parameters. The spatial distribution of observed EI from MODIS was reproduced by M3 and M4, whereas AI spatial patterns were preserved by M1. In addition, M1 and M3–M4 showed similar spatial patterns for EI and AI across the north, but differences were observed in the south, where M1 indicated lower water availability attributed to higher evaporative ratios (higher EI). M2 simulated the driest catchments, with an average value of EI and AI of 0.50 and 0.63, respectively, whereas M1 showed median values of 0.47 and 0.59, and M3–M4 values of 0.42 and 0.51.
Daily precipitation from CHIRPS was preferred over other global precipitation products because of a relatively higher spatial resolution and good performance across different climates and biomes (e.g., Bayissa et al., 2017; Ullah et al., 2019; Zambrano-Bigiarini et al., 2017). Nevertheless, CHIRPS showed a large bias and more rainy days with respect to ground precipitation across Costa Rica (Fig. 4a). The results suggested that at large scales, the precipitation bias was compensated for since the mean bias factor (BF) was
Our simple, linear bias correction of CHIRPS showed better performance at monthly and annual scales and solved the water balance inconsistencies of most catchments (Fig. 10). However, the cross-correlation between daily precipitation and streamflow remains unchanged by the bias correction. Not surprisingly, our results showed that catchments with highly correlated streamflow and daily precipitation exhibited better performance than catchments with low correlations. Several studies highlighted those meteorological forcings are the largest source of uncertainty in hydrological modeling (e.g., Arheimer et al., 2020; Dal Molin et al., 2020; Lin et al., 2018; Wörner et al., 2019), whereas more complex bias correction techniques (e.g., quantile mapping) may improve the results (e.g., Goshime et al., 2019; Wörner et al., 2019). However, the lack of matching daily streamflows with precipitation inputs and intense rain events might persist. Nonetheless, Infante-Corona et al. (2014) suggested that global products can achieve better streamflow simulation results than sparse ground precipitation data, whereas Westerberg and Birkel (2015) found that in situ precipitation in Costa Rica may require corrections to achieve better model results.
The global CPC temperature dataset used was not bias-corrected due to the lack of sufficient in situ measurements. Temperature can introduce large
errors in hydrological simulations if used for the estimation of potential
evapotranspiration and actual evapotranspiration (Andersson et al., 2015).
We corrected the temperature data set using elevation and a lapse rate
parameter (Eq. 6) during the calibration steps. The corrected temperature
closely followed the environmental lapse rate of 6
Simulated daily streamflow showed reasonable performance using the four
model configurations, where M2 (calibrated
Results from the Kruskal–Wallis test suggested median KGE in AET and PET from M3–M4 were statistically different compared to the baseline configuration M1 (the statistic
Such multi-objective calibration trade-offs were previously observed by, e.g., Zhang et al. (2018). Larger improvements were obtained for AET simulation of M3 and M4, whereas M1 (only daily
Despite the generally reasonable performance of our model configurations, we
found some issues when comparing our results with previous efforts, such as those of
Birkel et al. (2012), who modeled the streamflow in the Sarapiqui River basin (data not used in this study for calibration) with the HBVlight model
(Seibert, 2005) for the period from 1983 to 1991 and obtained an NSE of 0.74 after rainfall correction of the underestimated observed precipitation.
Our configuration M4 resulted in an NSE
A hydrological model useful for water management should be able to mimic streamflow seasonality and to realistically represent the large-scale physical processes of the water partitioned by vegetation interception and
the soil matrix into evapotranspiration and discharge (Arheimer et al., 2020; Kwon et al., 2020; Pechlivanidis and Arheimer, 2015; Rajib et al., 2018b; Rakovec et al., 2016; Xiong and Zeng, 2019). We, therefore, independently evaluated the four configurations using a range of hydrological signatures (Table 4) following Westerberg and McMillan (2015) in an attempt to single out the sought-after well-balanced model for use in decision-making. However, using multiple signatures also complicated the interpretation of simulations since daily streamflow (
Significant spatial variations in hydrological signatures were observed between M1–M2 and M3–M4 since implementing a spatial calibration of AET improved the representativeness of the more complex large-scale climate gradient. Similar results were found in catchments in the United States (Lin et al., 2018; Rajib et al., 2018a) and worldwide (Arheimer et al., 2020). The model configurations M3 and M4 better reproduced the spatial variability between the Pacific and Caribbean basins and the north–south gradient of the AI and EI (Esquivel-Hernandez et al., 2017). Furthermore, the resulting hydrological signatures of M3 and M4 were consistent with previous small catchment-scale studies that showed that runoff coefficients tend to be larger than the evaporative index (Dehaspe et al., 2018; Gómez-Delgado et al., 2011). Results also suggested that the event streamflow response is dominated by quick near-surface soil water discharge (Dehaspe et al., 2018), with streamflow being fed by groundwater during dry periods resulting in BFI values exceeding 0.7 (Birkel et al., 2012). In contrast to Westerberg et al. (2014), who calibrated Central American catchments using FDC information, we used the observed FDCs as an independent hydrological signature (Fig. 9). The configurations M2 to M4 outperformed the baseline (M1), supporting the notion that only streamflow used for calibration is not enough to produce a well-balanced model.
This study is the first attempt to apply the process-based, conceptual rainfall–runoff HYPE model at the national scale of Costa Rica ( Bias was observed in the precipitation from CHIRPS, with underestimation in mountainous regions and overestimation in the driest region with around 1000 mm of annual rainfall in Costa Rica. CHIRPS showed Our bias correction procedure using the linear-scaling technique reduced annual water balance inconsistencies ( The temperature could efficiently be used with an elevation correction; nevertheless, a higher-resolution temperature product or downscaling approach would improve the many micro-climates across the complex topography in Costa Rica. HYPE successfully reproduced major processes (evapotranspiration, runoff, baseflow discharge) of tropical catchments in Costa Rica, where we obtained acceptable performance for daily streamflow (median KGE from 0.4 to 0.6) and good performance for monthly streamflow (median KGE from 0.6 to 0.9, and median NSE from 0.4 to 0.55), with best-fit results for PET and AET of KGE Model calibration using monthly and daily streamflow (M2) improved the performance of the low flows in comparison to only daily streamflow (M1) calibration, where the average RMSLE of FDC was computed as Remotely sensed PET and AET constrained the soil type and land cover parameters associated with the evapotranspiration process. Statistical differences in AET and PET performance was observed for M3–M4 with respect to M1 and M2, but not for daily and monthly (KGE) streamflow simulations. Including PET and AET in calibration (M3 and M4) slightly decreased the overall streamflow performance (average KGE of Simulated hydrological signatures (aridity index, evaporative index, baseflow index, streamflow coefficient, flow duration curve) differed for each calibrated model, but configurations M3 and M4 more realistically mimicked the spatial distribution of all tested hydrological signatures.
We conclude that M3 and M4 are promising model configurations for the quantitative assessment of water resources in Costa Rica and that PET-AET and daily streamflow (M3) and PET-AET, and daily and monthly streamflow (M4) represent an
appropriate calibration sequence for regional modeling. Improvements to
these models could be achieved by incorporating more independent data into
the calibration process, such as soil moisture and groundwater level and
storage data. However, all global products crucially depend on evaluation
and even correction, which require observational in situ data. Nonetheless, we
hope to have provided a way forward towards a large-sale operational
hydrological model for the humid tropics of Costa Rica and potentially for
other humid regions of the world.
The hydrological simulations of model configuration 4 (M4) represent the HYPE for Costa Rica version 1.0 dataset (HYPE CR 1.0), and are freely available online at
The supplement related to this article is available online at:
All authors contributed to the study's conception and design. The data collection was made by SAE and CB. The data processing and the model construction were made by SAE and ACP. Results were analyzed by SAE and CB. BA and ABN contributed to the methodology review and discussion of the model results. CB and ABN critically revised the work, and all authors approved the final manuscript.
The contact author has declared that neither they nor their co-authors have any competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Saúl Arciniega-Esparza was supported by the CONACYT graduate scholarship program and the Programa de Maestría y Doctorado en Ingeniería at UNAM. Christian Birkel would like to acknowledge funding from the Observatorio del Agua y Cambio Global (OACG) under UCR grant ED-3319. The authors are also grateful for the Center for Geophysical Research (CIGEFI) and thank Ana Maria Duran at UCR for sharing meteorological station data under IMN contract IMN-DIM-CM-117-0917. We thank Alejandra González Hernández for their support in editing and translating the manuscript.
The authors are grateful to the research centers that provide free access to their databases and models, such as the Climate Hazards Center at the University of California, Santa Barbara, the Numerical Terradynamic Simulation Group at the University of Montana, the National Oceanic and Atmospheric Administration, and the Swedish Meteorological and Hydrological Institute, among others. The authors also thank the two anonymous reviewers that improved the content of this paper.
This paper was edited by Yi He and reviewed by two anonymous referees.