Abstract:Agriculture, meteorological, and hydrological drought is a natural hazard which affects ecosystems in the central India of Maharashtra state. Due to limited historical data for drought monitoring and forecasting available in the central India of Maharashtra state, implementing machine learning (ML) algorithms could allow for the prediction of future drought events. In this paper, we have focused on the prediction accuracy of meteorological drought in the semi-arid region based on the standardized precipitation index (SPI) using the random forest (RF), random tree (RT), and Gaussian process regression (GPR-PUK kernel) models. A different combination of machine learning models and variables has been performed for the forecasting of metrological drought based on the SPI-6 and 12 months. Models were developed using monthly rainfall data for the period of 2000–2019 at two meteorological stations, namely, Karanjali and Gangawdi, each representing a geographical region of Upper Godavari river basin area in the central India of Maharashtra state which frequently experiences droughts. Historical data from the SPI from 2000 to 2013 was processed to train the model into machine learning model, and the rest of the 2014 to 2019-year data were used for testing to forecast the SPI and metrological drought. The mean square error (MSE), root mean square error (RMSE), adjusted R2, Mallows’ (Cp), Akaike’s (AIC), Schwarz’s (SBC), and Amemiya’s PC were used to identify the best combination input model and best subregression analysis for both stations of SPI-6 and 12. The correlation coefficient (rdocumentclass[12pt]{minimal}usepackage{amsmath}usepackage{wasysym}usepackage{amsfonts}usepackage{amssymb}usepackage{amsbsy}usepackage{mathrsfs}usepackage{upgreek}setlength{oddsidemargin}{-69pt}egin{document}$$r$$end{document}), mean absolute error (MAE), root mean square error (RMSE), relative absolute error (RAE), and root relative squared error (RRSE) were used to perform evaluation for SPI-6 and 12 months of both stations with RF, RT, and GPR-PUK kernel models during the training and testing scenarios. The results during testing phase revealed that the RF was found as the best model in forecasting droughts with values of rdocumentclass[12pt]{minimal}usepackage{amsmath}usepackage{wasysym}usepackage{amsfonts}usepackage{amssymb}usepackage{amsbsy}usepackage{mathrsfs}usepackage{upgreek}setlength{oddsidemargin}{-69pt}egin{document}$$r$$end{document}, MAE, RMSE, RAE (%), and RRSE (%) being 0.856, 0.551, 0.718, 74.778, and 54.019, respectively, for SPI-6 while 0.961, 0.361, 0.538, 34.926, and 28.262, respectively, for SPI-12 scales at Gangawdi station. Further, the respective values of evaluators at Karanjali station were 0.913 and 0.966, 0.541 and 0.386, 0.604 and 0.589, 52.592 and 36.959, and 42.315 and 31.394 for PUK kernel and RT models, respectively, during SPI-6 and SPI-12. Machine learning models are potential drought warning techniques because they take less time, have fewer inputs, and are less sophisticated than dynamic or scientific models.

Determining the most appropriate drought index using the random forest algorithm with an emphasis on agricultural drought

Assessment of drought risk using multi-sensor drought indices and vulnerability factors: a case study of semi-arid region in Iran

Using the multiple linear regression based on the relative importance metric and data visualization models for assessing the ability of drought indices

Modified version for SPEI to evaluate and modeling the agricultural drought severity

Comparison of different drought monitoring indices in different climatic conditions in Iran

Drought monitoring and prediction using SPI, SPEI, and random forest model in various climates of Iran

Effect of different timescales of drought on water productivity of rain-fed winter wheat in arid and semi-arid regions

Using the Fuzzy Clustering and Principle Component Analysis for Assessing the Impact of Potential Evapotranspiration Calculation Method On the Modified RDI Index

Meteorological Drought Prediction Based on Evaluating the Efficacy of Several Prediction Models

Prediction of meteorological drought and standardized precipitation index based on the random forest (RF), random tree (RT), and Gaussian process regression (GPR) models

Short-Term Spatio-Temporal Drought Forecasting Using Random Forests Model at New South Wales, Australia

Meteorological and hydrological drought monitoring using several drought indices

Evaluation of hydro-meteorological drought indices for characterizing historical droughts in the Mediterranean climate of Algeria

Assessing drought impacts on groundwater and agriculture in Iran using high-resolution precipitation and evapotranspiration products

Index-based Assessment of Agricultural Drought in a Semi-Arid Region of Inner Mongolia, China

Long-term monitoring and evaluation of drought and determining the accuracy of its indicators in western Iran

A New Drought Monitoring Index on the Tibetan Plateau Based on Multisource Data and Machine Learning Methods

Evaluation of meteorological drought effects on underground water level fluctuations using data mining methods (case study: semi-deep wells of Golestan province)

Drought prediction using artificial intelligence models based on climate data and soil moisture

Drought assessment and monitoring using a combination of the ADI index and SWAT model

Prediction of short and long-term droughts using artificial neural networks and hydro-meteorological variables