Modeling hydrological dynamics in a dryland catchment with dense reservoir networks

Abstract

Drylands represent some of the most vulnerable ecosystems to climate change and anthropogenic activities. Hydrological modeling, particularly through tools like the Soil and Water Assessment Tool (SWAT), serves as an essential instrument for addressing practical challenges such as supporting decision-making in water resource management and hydrological forecasting for drylands. However, there remains substantial scope for evaluating the performance of such models in dryland watersheds, especially those characterized by dense reservoir networks, and assessing their behavior during consecutive years of drought as well as during years with extreme rainfall events. Furthermore, during periods of extreme aridity, particularly severe droughts, the models face significant challenges in accurately representing crucial processes, including runoff transmission losses, evapotranspiration rates, and anthropogenic influences. This thesis aims to advance the application of the SWAT model in dryland catchments by addressing its ability to simulate the hydrological processes of these regions, particularly in the context of dense reservoir networks and their impact on water retention across interannual drought and flood scenarios. Additionally, the thesis evaluates the model's performance during an extended drought and their hydrological recovery period in a more arid and warmer climate decade, subsequent to a previously calibrated time series. The study was conducted in the Conceição River Catchment (3,347 km2), located in Ceará, northeastern Brazil. This catchment contains 230 reservoirs (0.068 reservoirs per km2), with storage capacities ranging from less than 0.01 hm3 to 52 hm3. The analysis was based on a 40- year dataset of climate and observed runoff, with 30 years used for model calibration and an additional 10 years for extending the time series to include an extreme drought period. The key findings of this research are as follows: (i) The model's daily performance was found to be acceptable (Nash-Sutcliffe Efficiency = 0.63, Kling-Gupta Efficiency = 0.81 and Percent BIAS = 0.53%), demonstrating reliability in representing peak flows during wet years, periods of no flow, and the rising limb of the hydrograph. While small reservoirs (less than 0.01 hm3) were found to reduce streamflow, their overall impact on catchment retention was minimal, with water retention rates of 2% in wet years and 9% in dry years; (ii) The reduction in runoff due to the increased number of small reservoirs was more pronounced during dry years (up to 30%) compared to wet years (up to 8%). This reduction also escalated over consecutive years of drought, a result obtained in the scenario analysis, considering reservoir densities ranging from 0 to 3 reservoirs per km2 within the catchment, potentially exacerbating the effects of prolonged droughts; (iii) Despite the model's strong performance over a 30-year climatic series, which included periods of severe drought, it struggled to accurately predict streamflow during the extreme drought and the subsequent recovery phase, showing a marked decline in performance during these critical periods. Even after adjusting the most sensitive parameters related to key hydrological processes, the model's performance did not significantly improve. The SWAT model proves to be an acceptable model in dryland modeling for non-extreme drought events, and the methodology proposed in this study is highly transferable for different catchments worldwide. Overall, this thesis provides valuable insights into enhancing dryland modeling, bridging the gap between scientific research, modeling, and water resource management. These advancements are critical for improving the accuracy of scenario predictions and, consequently, for enhancing strategies to manage water resources in regions vulnerable to scarcity.