The Big Data for Drought Monitoring in Georgia

The dangerous hydrometeorological phenomenon—drought is frequent in Georgia. The SPI and SPEI 3, 6 and 12 month drought indices were used to analyze drought frequency and intensity on the territory of Georgia for 1991–2020 year period. The structured data that is the part of big data, of ground hydrometeorological observation network of Georgia have been used to conduct research. The following statistical parameters were calculated: Pearson correlation coefficient (PCC), determination coefficient (R2), and root mean square error (RMSE) both for the entire period and for months. The correlation coefficient is in a good agreement for all cases, and the absolute deviation shows data scattering, which should be related to the complex relief of Georgia, as well as the heterogeneity of precipitation data distribution. The calculated Standardized Precipitation Index (SPI) for 3 months of Kakheti region was subjected to Machine Learning. A Support Vector Machine (SVM) was selected in the Matlab space—the algorithm for Supervised Machine Learning method. The tenth model showed the best result; using of mentioned model it became possible to determine the drought probability by months at each point. Despite of obtained good parameters, it was necessary to add additional stations, because there was not enough information in the Kakheti region for the correct analysis of Machine Learning avoiding overfitting. The study is important for climate change assessment and hydrometeorological disaster early warning system implementation.