A machine learning model for drought tracking and forecasting using remote precipitation data and a standardized precipitation index from arid regions. (June 2021)

• Record Type: Journal Article
• Authors: Bouaziz, Moncef; Medhioub, Emna; Csaplovisc, Elmar
• Is Part Of: Journal of arid environments. Volume 189(2021)
• Publication Date: 2021-06
• Subjects: Drought -- Standardized precipitation index -- CHIRPS -- Drought forecast -- Extreme learning machine; Arid regions ecology -- Periodicals; Arid regions -- Periodicals; Écologie des régions arides -- Périodiques; Régions arides -- Périodiques; 577.54
• DOI: 10.1016/j.jaridenv.2021.104478
• Languages: English
• ISSNs: 0140-1963

Abstract

Abstract: Drought is a catastrophe that impacts agriculture and causes economic and social damage. An effective monitoring and forecasting system is needed to assess the extent of droughts and to mitigate their effects at both spatial and temporal levels. To this end, we used a Standardized Precipitation Index (SPI) in various timescales to classify and track drought events based on CHIRPS rainfall data for the period between 1981 and 2019. Three models (M1, M2, M3) were then tested for annual drought prediction (SPI_12) using precipitation data and the lagged SPI as input variables. Extreme Learning Machine algorithms displayed rapid drought prediction, with high accuracy on different timescales (0.7–0.8 R 2 ). Highlights: Advanced techniques of machine learning algorithm and statistical tools were used to rapidly assess the potential of remote-precipitation data (CHRIPS) to extract features and patterns of drought in southern Tunisia. The proposed research work confirms the high performance of using the Extreme Learning Machine to forecast meteorological drought and detect temporal patterns of drought with a high coefficient of determination (R 2 = 0.71) and a good residual predictive deviation value. Our study demonstrated the consistency of using SPI as input for the Extreme Learning Machine to predict drought events and their severity for the near future. The forecasting of droughts events in such arid region can be helpful for local communities to adapt their land use accordingly and to plan new strategies of dealing with upcoming droughts in advance.