Modeling hydrologic response due to the impact of land use changes in the Upper Bernam River basin using machine learning

Land use changes in a watershed can affect the watershed hydrology in various ways. Some types of land development can be associated with increased impervious area causing an increase in surface runoff and a decrease in groundwater recharge. Both of these processes can have large-scale ramifications through time. Increased runoff results in higher flows during rainfall events, which in turn increases the number of times that a river floods the adjacent land areas. On the other hand, the groundwater recharge decreased due to the increase in the impervious surfaces and decrease in the soil infiltration rate. The main objectives of this study was to analyze and assess the impacts of land use changes on the watershed runoff in Upper Bernam River Basin by using Soil and Water Assessment Tool (SWAT) model, and to develop a Watershed Best Management Practice model using Machine Learning (WBMP-ML) to determine the best optimal locations, numbers and operations of ponds to control flood during high flow season, maintain river base flow and supply irrigation water demand during low dry season. The Bernam River is the main source of irrigation water for 20,000 ha rice granary area. Land use changes in the study area have experienced tremendous changes from 1984 to date. Eight land use of years 1984, 1990, 1998, 2000, 2002, 2004, 2006 and 2010 were used for investigation study and assessment analysis. Projected land use of the year 2020 with other scenarios of 40% and 50% of urban were used for flow prediction to assess the future impact of land use change. For forest as a form of land use, there was a percentage decrease from 56.3% in 1984 to 48.02% in 2010 and a further projected decrease to 45.81% in 2020. This decreasing trend is applicable to other forms of land use like orchard and rubber plantation, except urban area and oil palm which showed an increasing trend. The study was conducted using a 36 years flow record (1980-2016). Calibration was performed for the period of 1980 to 2004 with three land use of years 1984, 1990 and 1998 while the period of 2005 to 2016 was used for validation with the land use of years 2006 and 2010. The coefficient of determination (R²) and Nash-Sutcliffe coefficient (E) were used as evaluation criteria for model performance. The model showed a very good performance in simulating the runoff process. During calibration annual, monthly and daily results were 0.83, 0.83 and 0.77 for R² and 0.80, 0.81 and 0.76 for E respectively, while during validation, the results were 0.88, 0.89 and 0.79 for R² and 0.82, 0.86 and 0.76 for E respectively. Thus, in this study, watershed modeling was used to simulate and analyze the impact of land use changes on hydrology and stream stability. SWAT model was used to simulate and analyze the impact of land use change on hydrology runoff quantity. From SWAT application, it was found that the percentage change in runoff due to land use change in period 1998 to 2000 was small because the land use change in that period was not noticeable. However, the runoff increased significantly from 4.18% in 1984 to 22% in 2010 comparing with the scenario of 100% forest land. The model was then applied to simulate the runoff from future land development for the year 2020 (20% urban), scenarios of 40% and 50% urban, the predictions showed an increment of 32%, 45% and 59% due to land use change respectively. Analyses of three different annual rainfall amounts were carried out to identify the effects of land use change under different rainfall patterns. The results showed that the percentage of flow has increased as a result of rainfall amount change, where the watershed response is noticeably higher due to rainfall change than individual changes in land use. This study comes to address the challenges of tropical hydrology system. It deals with the application of modeling that is new and an important aspect of understanding the global hydrological system; Machine Learning was used for the purpose of flood and drought control. The number of ponds was reduced by machine learning from 12 ponds that suggested by WARM model to 7 ponds with total area 1942 ha to store maximum water of 98.5 x 10⁶ m³. This methodology can be applied for any future development plan to predict the hydrological impacts and mitigate the risk of flood occurrence and avoid the shortage of irrigation water. The developed methodology, therefore, would be useful in assisting policy and decision making tool when formulating land use policies. It can be a practical tool for hydrologists, engineers, and town and country planners.

Text FK 2019 156 - ir.pdf Download (1MB)

Actions (login required)

View Item