Evaluation of the Performance of FLEX-L and FLEX-Topo Hydrological Models in Surface Runoff Simulation in the Tamer Loess Watershed

Background and Objectives: Accurate simulation of surface runoff in loess deposits, particularly at the watershed scale such as the Tamer Loess Watershed, plays a fundamental and pivotal role in sustainable water resource management, flood risk mitigation, erosion control, soil conservation, and enhanced environmental planning, thereby serving as an essential foundation for scientific, executive, and managerial decision making. In recent years, the application of conceptual and semi-distributed hydrological models has garnered increasing attention among researchers as an efficient and robust approach to better understand the complex hydrological response of such watersheds and to analyse the intricate interplay of climatic, geological, and topographical factors, particularly under changing environmental conditions. In this context, a rigorous and comprehensive evaluation of the performance of the FLEX-L and FLEX-Topo models, representing an integrated structure and an advanced structure explicitly based on topographic heterogeneity respectively, is of paramount importance. Accordingly, the present study was conducted to comprehensively and systematically assess the performance of the FLEX-L and FLEX-Topo hydrological models in simulating surface runoff in the Tamer Loess Watershed. The primary objective was to compare the capabilities of these two models and to quantitatively and qualitatively determine their efficiency and reliability in accurately representing the governing processes of runoff generation, water transmission, infiltration, and runoff storage in loess environments. Furthermore, both models were employed, utilizing their specific conceptual structures, to simulate the hydrological response of the watershed and to conduct a detailed analysis of surface and subsurface flow behavior. This integrated approach effectively and realistically incorporated the role of spatial heterogeneity, topographical characteristics, hydrodynamic differences in soils, and the interactive effects of climatic factors.
Materials and Methods: To simulate the rainfall-runoff process in the Tamer Loess Watershed, comprehensive climatic and hydrometric data, including daily precipitation, temperature, and discharge, were systematically collected to accurately analyze the hydrological behavior of the basin and provide a suitable and reliable basis for modeling. Additionally, various spatial layers including a Digital Elevation Model (DEM), slope, stream network, land use, and the Height Above the Nearest Drainage (HAND) index were utilized to effectively delineate landscape units and prepare essential model inputs. In this study, two conceptual models from the FLEX family, namely FLEX-L and FLEX-Topo, were employed. Due to their flexible and conceptual structure, FLEX family models enable the simulation and analysis of various hydrological behaviors across different spatial and temporal scales and are capable of modeling complex rainfall-runoff processes in diverse watersheds. The FLEX-L model is a lumped model that considers the entire watershed as a single integrated unit and does not utilize elevation data, whereas the FLEX-Topo model is semi-distributed, explicitly incorporates topographical information, and comprises four parallel components to represent the hydrological functioning of different landscape elements. Daily outlet runoff simulation was performed for a ten-year period from 2013 to 2023 using both models. Both models consist of a combination of reservoirs, lag functions, and connection elements, which are flexibly linked to simulate various hydrological processes, including surface and subsurface flow, infiltration, and runoff storage. The structure of these models was developed using the SUPERFLEX modeling framework, and their coding and implementation were carefully carried out in the Python programming environment to enable more detailed analysis and greater control over model components. To evaluate the accuracy and reliability of the simulated outlet runoff, two well-established statistical indices, namely the Nash–Sutcliffe Efficiency (NSE) and the Kling–Gupta Efficiency (KGE), were employed. These indices effectively determine the degree of agreement between simulations and observed data, providing a comprehensive and robust criterion for comparing the performance of the two models and analyzing their capability and reliability in reproducing the watershed's hydrological processes.
Results: Comparison of the performance evaluation indices for the two models clearly indicated that the FLEX-Topo model exhibits higher accuracy and superior performance in simulating runoff compared to the FLEX-L model. The Nash–Sutcliffe Efficiency (NSE) value for FLEX-L was 0.62, while for FLEX-Topo it was 0.69; the Kling–Gupta Efficiency (KGE) values were 0.67 and 0.72, respectively, and the Root Mean Square Error (RMSE) values were 2.9 and 1.8, respectively. These results collectively and consistently demonstrate the superiority of the FLEX-Topo model in reproducing streamflow dynamics and substantially reducing simulation error relative to the observed data. From a behavioral perspective, although the FLEX-L model successfully simulated the overall trend of flow variations, the simulated flow values on some peak runoff days were noticeably lower than the observed values, indicating its limitation in capturing extreme events. In contrast, by effectively utilizing topographical information and disaggregating the watershed into distinct hydrological landscape units, the FLEX-Topo model was able to more accurately represent the spatial patterns of runoff generation and the inherent spatial heterogeneity of hydrological processes. On most days with high precipitation, this model simulated runoff values considerably closer to the actual measurements and demonstrated more precise and reliable performance. Furthermore, detailed analysis of the one-to-one scatter plot of observed versus simulated flow reveals that in the FLEX-Topo model, the data points are more consistently concentrated around the 1:1 line, indicating stronger agreement and lower bias between observed and simulated flows. In the FLEX-L model, however, the data points exhibit greater dispersion and less concentration around the 1:1 line, particularly at higher flow values, reflecting this model's lower sensitivity to spatial heterogeneities and its tendency toward systematic underestimation. These findings conclusively confirm that incorporating topographical characteristics and landscape structure in the FLEX-Topo model plays a significant and determinative role in improving runoff simulation accuracy and hydrological process representation in the Tamer Loess Watershed.
Conclusion: The results of this study indicate that both the FLEX-L and FLEX-Topo models possess acceptable efficiency in simulating surface runoff in the Tamer Loess Watershed and are capable of representing the general flow pattern and its temporal variations with reasonable accuracy. However, the FLEX-Topo model demonstrated more accurate and reliable performance compared to the FLEX-L model due to its disaggregation of the watershed into hydrologically homogeneous units and its incorporation of spatial heterogeneity. While FLEX-L represents the general flow trend, its simple structure and insufficient attention to spatial heterogeneity limit its capability in simulating peak discharges. In the FLEX-Topo model, different landscapes were identified and simulated based on dominant hydrological processes, such as surface runoff on steep slopes, subsurface flow in forested and rangeland plains, and saturated excess runoff in low-lying areas. Overall, the FLEX-Topo model is introduced as a more suitable and reliable option for hydrological analyses in the Tamer Loess Watershed; nevertheless, parameter optimization in both models can still play an important role in enhancing accuracy and reducing simulation uncertainty.