Dam Water Level Prediction Using Vector AutoRegression, Random Forest Regression and MLP-ANN Models Based on Land-Use and Climate Factors

To predict the variability of dam water levels, parametric Multivariate Linear Regression (MLR), stochastic Vector AutoRegressive (VAR), Random Forest (RFR) and Multilayer Perceptron (MLP) Artificial Neural Network (ANN) models were compared based on influences climate factors (rainfall temperature), indices (DSLP, Aridity Index (AI), SOI Niño 3.4) land-use land-cover (LULC) as predictor variables. For case study Gaborone Bokaa in semi-arid Botswana, from 2001 to 2019, prediction results showed that linear MLR is not robust for predicting complex non-linear variabilities levels with The VAR detected relationship between LULC R2 > 0.95; however, it was unable sufficiently capture influence levels. RFR MLP-ANN significant correlations indices, a higher value 0.890 0.926, dam, 0.704–0.865 dam. Using predictions, performed better than MLP-ANN, accuracy Based provided best both dams. improve results, VAR-ANN hybrid model found be more suitable integrating conditions time-series components