A fusion-based methodology for meteorological drought estimation using remote sensing data

An effective planning and management to deal with potential impacts of drought requires accurate estimation and analysis of this natural complex phenomenon. Application of new fusion approaches using high-resolution satellite-based products, unlike ground-based observations, can provide accurate drought analysis. This study examines three advanced fusion-based methodologies including Ordered Weighted Averaged (OWA) approach based on ORNESS weighting method (ORNESS-OWA) and ORLIKE weighting method (ORLIKE-OWA) as well as K-nearest neighbors algorithm (KNN) to fuse estimations by five individual estimator models using different remotely sensed data products. The precipitation data from Global Precipitation Climatology Project (GPCP), CPC Merged Analysis of Precipitation (CMAP), CICS High-Resolution Optimal Interpolation Microwave Precipitation from Satellites (CHOMPS), Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks-Climate Data Record (PERSIANN-CDR), Tropical Rainfall Measuring Mission (TRMM), The second Modern-Era Retrospective analysis for Research and Applications (MERRA-2) and Global Land Data Assimilation System Version-2 (GLDAS-2) products is utilized in estimating nonparametric-SPI as a meteorological drought index versus ground-based observations analysis. To achieve more accurate drought estimation, ground-based observations are classified in different clusters based on K-means clustering algorithm. Five individual Artificial Intelligence (AI) models including Multi-Layer Perceptron (MLP), Adaptive Neuro-Fuzzy Inference System (ANFIS), M5P model tree, Group Method of Data Handling (GMDH) and Support Vector Regression (SVR) are developed for each cluster and their best results are used in fusion process. In addition, the Genetic Algorithm (GA) optimization model is utilized to determine optimal weights in weighting methods. Estimation performance of all models are evaluated using statistical error indices of Mean Absolute Relative Error (MARE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and coefficient of determination (R²). Application of proposed methodology is verified over Fars province in Iran and the results are compared. Results showed that ORNESS-OWA method with lowest estimation error (MARE of 2.51% and R² of 95%) had the superb performance in comparison with all other individual AI and fusion-based models. Also, the proposed framework based on remotely sensed precipitation data and fusion-based models demonstrated an effective proficiency in drought estimation.