Predicting the variability of dam water levels with land-use and climatic factors using random forest and vector autoregression models

Abstract

Dams play a significant role in the storage, supply, and capitalization of water resources. This study analyzes the influences of land-use and climate factors on two interconnected dams, Gaborone and Bokaa dams, in the semi-arid Botswana from 2001 to 2019. Using Random Forest regression (RFR) and Vector AutoRegression (VAR) models, the monthly dam water levels were predicted based on the variabilities of rainfall and temperature, climate indices (DSLP, Aridity Index (AI), SOI and Niño 3.4) and land-use land-cover (LULC) information comprising of built-up, cropland, water, forest, shrubland, grassland and bare-land. The prediction results using the climate factors and climate indices show that for both dams, RFR was able to detect the correlations between the dam water levels with R2 of between 0.805 and 0.845 with min, average and max temperatures as the best combined predictors. Using differenced stationary datasets, VAR identified the climate indices as the suitable predictors for water levels in Gaborone and Bokaa dams with R2 of 0.929 and 0.916 respectively. VAR also detected LULC to be strongly correlated to the dam water levels. Nevertheless, LULC was considered as more significant when combined with the climate-based predictor variables. Comparatively, VAR was able to detect the interdependence between the two dams and with the other conjunctive water sources as the water levels in both dams were not significantly correlated with rainfall trends, while RFR relied on the seasonal temperature variabilities to accurately predict the fluctuations in the dam water levels.