Short-Term and Medium-Term Drought Forecasting Using Generalized Additive Models

• Main Authors: Fhumulani Mathivha, Caston Sigauke, Hector Chikoore, John Odiyo
• Format: Article
• Language: English
• Published: MDPI AG 2020-05-01
• Series: Sustainability
• Subjects: drought; forecasting; generalized additive models; hydrological extremes; SPEI; water resources
• Online Access: https://www.mdpi.com/2071-1050/12/10/4006

Forecasting extreme hydrological events is critical for drought risk and efficient water resource management in semi-arid environments that are prone to natural hazards. This study aimed at forecasting drought conditions in a semi-arid region in north-eastern South Africa. The Standardized Precipitation Evaporation Index (SPEI) was used as a drought-quantifying parameter. Data for SPEI formulation for eight weather stations were obtained from South Africa Weather Services. Forecasting of the SPEI was achieved by using Generalized Additive Models (GAMs) at 1, 6, and 12 month timescales. Time series decomposition was done to reduce time series complexities, and variable selection was done using Lasso. Mild drought conditions were found to be more prevalent in the study area compared to other drought categories. Four models were developed to forecast drought in the Luvuvhu River Catchment (i.e., GAM, Ensemble Empirical Mode Decomposition (EEMD)-GAM, EEMD-Autoregressive Integrated Moving Average (ARIMA)-GAM, and Forecast Quantile Regression Averaging (fQRA)). At the first two timescales, fQRA forecasted the test data better than the other models, while GAMs were best at the 12 month timescale. Root Mean Square Error values of 0.0599, 0.2609, and 0.1809 were shown by fQRA and GAM at the 1, 6, and 12 month timescales, respectively. The study findings demonstrated the strength of GAMs in short- and medium-term drought forecasting.