Introduction
Water resources, which are scarce in western Peru, are increasingly at risk due to the impacts of population growth, economic development, and climate change (
Chevallier et al. 2011). In the region of Arequipa, high-elevation mountains (up to 6,450 m above mean sea level) in the east receive more than 500 mm of precipitation annually, while the rest of the region, where most of the 1.3 million people live, is either semiarid or arid (
Stensrud 2016;
Censos Nacionales 2017). Between the Andean highlands and the Pacific Coast, precipitation is less than 10 mm per year (
Moraes et al. 2019), but it is in this region that irrigated agriculture is rapidly expanding (
Stensrud 2016). Extensive water management infrastructure has been constructed in the region since approximately 1958 to support irrigated agriculture, urban demand, and hydropower generation (diversions for hydropower first occurred around 1909). This includes four major reservoirs in the headwaters of the Quilca basin, and four additional reservoirs in the adjacent Camaná basin that provide water to the Quilca basin through approximately 200 km of canals, tunnels, and aqueducts. Such infrastructure has severely altered the natural hydrology of the region (
ANA 2014).
Hydrologic information is critical for effectively managing the complex water resources in this region. Despite the large population and the agricultural and industrial sectors that depend on water, there are currently only three active streamflow monitoring stations in nonregulated catchments in the Quilca and Camaná basins to assess the current status and change in the natural hydrology in this region (
Moraes et al. 2020). The active monitoring stations that are located on regulated catchments often have short, discontinuous records. The 29 active stations in these two basins have records that in theory span 3–96 years (24 years on average); however, due to the discontinuous nature of the records, only 10 of these stations have data for a 10-year period with greater than 50% of daily discharge available on a monthly basis (
Moraes et al. 2020). Limited knowledge of the hydrology of the region makes water regulation decisions challenging. Knowing the amount of available water to be managed and its distribution in the region would help to determine whether water is adequate for current and planned projects and better allocate water resources among different sectors to support sustainable water resources management.
Hydrologic models are valuable tools for quantifying and comprehending hydrological processes and can be used to evaluate hydrologic response to stressors as well as alternative water management scenarios (
Devia et al. 2015). In ungauged basins, where streamflow measurements that would provide an integrated measure of hydrologic cycle response to stressors are missing, models can provide hydrologic insights (
Sivapalan 2003). The Soil and Water Assessment Tool (SWAT) is a comprehensive watershed model that evaluates the impact of land use, land management, hydrologic alteration, and climate change on hydrology and water quality at the watershed scale (
Neitsch et al. 2011). It is open source and free for download and so has been commonly used around the world (
SWAT Literature Database 2020). Because SWAT was originally developed for the United States, its rich database, one of the advantages of using the SWAT model, is derived from data sets only available in the US. The use of SWAT in other parts of the world requires preparation of model inputs that capture local conditions, as well as measured outputs for model calibration. The dependability of model performance on the quality of data makes its use challenging in data-scarce regions.
Global databases such as the Food and Agriculture Organization (FAO)–UNESCO digital soil map of the world (
FAO/UNESCO 2003), and global land cover for SWAT (
George and Leon 2008) has been commonly in used in data-scarce regions including the South American Andes (i.e.,
Oñate-Valdivieso and Sendra 2014;
García Quijano 2018;
Daneshvar et al. 2018). But these coarse data sets (e.g., 5-km resolution of FAO/UNESCO digital soil map of the world) do not address spatial variations at the watershed scale. Landsat images or local land cover maps have been used as an alternative to the coarse-resolution global land cover map, but the properties of each land cover including those that control plant growth are not available through remote sensing, and therefore land cover properties from the SWAT database are often used (
Oñate-Valdivieso and Sendra 2014;
Jodar-Abellan et al. 2019). Similarly, local soil maps such as soil taxonomy may be used as an alternative to the coarse global soil but they do not include needed soil properties. Therefore, either equivalent soils from the SWAT database were used (
Ndomba et al. 2008) or pedotransfer functions were used to estimate remaining parameters based on limited local soil sample data (
Narasimhan et al. 2012;
Biru and Kumar 2018). However, these methods have high uncertainty in soil properties estimation because there is a wide range of properties for soils with a similar taxonomy (
Di and Kemp 1989;
Quesada et al. 2010;
Narasimhan et al. 2012). Global weather data for SWAT (
CFSR 2019) or remotely sensed satellite data (e.g.,
Ercan et al. 2015) are often used to overcome the weather data scarcity problem (
Dile and Srinivasan 2014;
Ahmed et al. 2020). But these coarse data sets cannot capture the extreme variability with topography in places like the Peruvian Andes.
Limited observed hydrologic data is another issue in data-scarce regions that makes model performance assessment challenging. Regionalization is the most commonly used method for model calibration in data-scarce regions in which ungauged basins are assumed to have a similar hydrologic response as the neighboring gauged stations with similar physical properties; therefore, the same set of calibration parameters were applied to them as well (
Gitau and Chaubey 2010;
Mengistu et al. 2019). But limited hydrologic data that exist for this region are obtained in regulated basins and do not help to understand the natural hydrology based on the regionalization approach.
Overall, limited streamflow measurements and the lack of regional soil and land cover data sets have made SWAT application challenging in this region. The goal of this study was to develop data sources and methods to have a credible SWAT model for a watershed in order to better understand the hydrology of the Peruvian Andes region. Specific objectives were to (1) develop regional soil and land cover databases for SWAT, (2) evaluate model performance and uncertainty using a range of hydrologic and weather input data, and (3) provide insights about the hydrologic cycle in this region. The new databases developed have been made publicly accessible and will provide a basis for future models, and the model performance evaluation using naturalized streamflow and remotely sensed satellite evapotranspiration estimates will provide new insights for modeling in the region.
Conclusion
The goal of this study was to increase our understanding of the hydrologic water balance of the El Frayle watershed in Arequipa, Peru, through the development of a credible SWAT model database despite limited data availability. The developed SWAT model provided predictions of streamflow, reservoir evaporation, and land evapotranspiration. The calibrated model provided satisfactory estimations of daily streamflow. Gridded weather data also had satisfactory performance with even lower bias compared to station data. However, gridded weather data generated less surface runoff due to the difference in precipitation distribution in the watershed. Comparison of simulated reservoir evaporation with pan measurement and land evapotranspiration with MODIS data showed that SWAT results are reasonable and able to capture the same seasonal dynamics as the observed data.
Overall, the hydrologic and water balance analysis of model simulations for this semiarid watershed showed that surface runoff is low and base flow is the main driver of streamflow. There is limited surface runoff generation and rapid transfer of water through near-surface layers to support local streamflow generation during the wet season, with little carryover of local groundwater storage to support base flow during the dry season. The uncertainty of estimated inputs resulted in an uncertainty range of 16% (for ET) to 46% (for base flow) estimation. Estimates of recharge to regional groundwater are less sensitive to model parameters and vary from 27 to on average. This suggests that when model calibration is constrained by streamflow observations, most of the uncertainty results in uncertainty in estimates of ET, which may be less critical for water allocation. Therefore, the proposed methodologies for model development and performance assessment can be extended to similar data-scarce watersheds. This study also provides a valuable baseline for future regional hydrologic modeling and water management studies in the Peruvian Andes.