Effects of digital elevation model data resolution on hydrological modelling using SWAT: a case study of the Swartkops river catchment
Year: 2025
Type: article
Abstract: Digital Elevation Models (DEMs) are widely used as input to model hydrological processes. Using the appropriate resolution of DEMs is critical in producing accurate simulations of hydrological processes using hydrological models such as the Soil and Water Assessment Tool (SWAT) model. As a result of the availability of DEMs in recent decades and land use/cover (LULC) change impacts on hydrological regimes, the aim of this study was to investigate the effect of DEM data resolution on SWAT model performance for analysing the hydrological response to LULC changes in the Swartkops River catchment. This included using the SWAT model to analyse runoff conditions in the catchment. The main objectives of the study were: (1) to generate LULC maps and analyse LULC changes using Multi-Temporal Landsat TM and OLI images from 1990 to 2022 within the Swartkops catchment using IDRISI Terrset and ArcMap 10.7; (2) to assess the influence of different DEM data resolutions on SWAT model performance via the quantification of model results; and (3) to utilise the SWAT model with optimal performance to analyse the impact of LULC change on the hydrological response (i.e., runoff generation) of the Swartkops catchment using ArcSWAT. To do this, 1990 and 2022 Landsat images were acquired, pre-processed, and classified using supervised image classification via the Maximum Likelihood algorithm. Idrisi Terrset software was used process the Landsat images by generating natural colour composite images which were then used to sample training sites. A stratified random sampling approach was employed which produced spectral signature files which were the basis for employing the Maximum Likelihood classifier to classify the images. Classified images were exported to ArcMap 10.7 where they underwent classification accuracy assessments. An error matrix approach was implemented, and the accuracy of the maps was assessed using the producer’s, user’s, overall and Kappa accuracy. Thereafter, LULC changes were quantified using ArcMap 10.7 by determining the area difference of each LULC type between 1990 and 2022. The SWAT models for the different resolution DEMs (i.e., 20m, 30m, and 90m) were set up on ArcSWAT, and the Swartkops catchment was delineated, subdivided into Hydrological Response Units, and run by inserting all the relevant input data such as LULC data (i.e., 1990, and 2022 classified maps), topographic data (i.e., the DEMs), soil data (i.e., FAO map), and weather data. Statistical evaluations for the performance rating of the models were conducted and incorporated the use of statistical parameters such as Nash-Sutcliffe Efficiency (NSE), Percentage of Bias (PBIAS), Coefficient of determination ((R2), and RMSE-Observation standard deviation ratio (RSR) for assessing the performance of each SWAT model. The model producing the most satisfactory performance was used to compute runoff conditions for the Swartkops catchment. Computations were conducted using the Soil Conservation Service (SCS) curve number method. Fieldwork, which incorporated measuring iii runoff for different LULC types using Gerlach troughs at hillslope scale, was conducted to validate the results of the SWAT model. Results indicated that the 1990 and 2022 LULC maps had an overall accuracy of 87.11% and 94.89%, respectively which was highly satisfactory. The Kappa statistics reflected similar results where the 1990 map had a value of 0.84 while the 2022 map had a value of 0.94, indicating high statistical agreement between the classified maps and the reference data. The study also found that bare areas exuberated the greatest area change with a decline of 9.49%, while built-up areas manifested the highest increase of 5.97% from 1990 to 2022. Agricultural land increased by 4.79% in conjunction with a slight increase of 0.44% by green vegetation while water declined by 1.71%. To determine the ideal DEM to analyse hydrological response, the LULC maps in conjunction with other input data were inserted into the SWAT model to assess model performance. It was found that the 30m produced satisfactory model performance (NSE = 0.49 at calibration and 0.67 at validation) and was eligible to assess runoff conditions as a function of LULC change. Other statistics indicated satisfactory model performance at calibration (PBIAS = 48.80, R2 = 0.51, RSR = 1.02) and validation stage (PBIAS = -38.1, R2 = 0.47, RSR = 1.08). The 90m DEM possessed the poorest model performance (NSE = 0.18 at calibration and 0.35 at validation) followed by the 20m DEM (NSE = 0.41 at calibration and 0.32 at validation). Runoff simulations using the 30m SWAT model showed that surface runoff was highest in areas where there were built-up areas. Built-up areas and bare areas yielded 36% of the total surface runoff individually, while vegetation produced only 28% in 1990. Similar readings were obtained for 2022, which included agricultural land that produced 29% of the surface runoff. Runoff volume measured at hillslope scale was highest in bare areas (6.6 L) and built-up areas (9.3 L). Therefore, this study shows that the resolution of DEMs must concur with the relevant scale of the study to produce optimal results. Additionally, this study showed that a change in LULC heavily affects the amount of surface runoff generated, which has several implications in terms of flooding. Therefore, this study is useful as it will inform sustainable catchment management decisions, and water resource management, and enhance our understanding of the relationships between DEM spatial resolution, LULC change, SWAT modelling, and water quantity calculations.,Thesis (MSc) -- Faculty of Science, School of Environmental Sciences, 2025
Author: Loni, Litha
Language: English
Subfield: Water Science and Technology
Field: Environmental Science
Domain: Physical Sciences
Keywords: digital elevation model resolution, DEM resolution, hydrological modeling, land use and land cover change, runoff generation, Landsat image classification, maximum likelihood classification, multi-temporal Landsat imagery, model calibration and validation, ArcSWAT, hydrological response units, watershed delineation
Open Access status: green
OpenAlex ID: W7220795289
