Drought monitoring and forecasting based on artificial intelligence and remote sensing: The case study of Thrace Region

This research focuses on drought analysis and forecast in the Thrace Region. In the study, monthly total precipitation data from nine meteorological stations from 1966 to 2021 and MODIS satellite images from 2000 to 2021 were used. The aim of the study is to determine the trend of precipitation data in the Thrace Region, to understand the drought characteristics and to predict future droughts. Mann-Kendall and Innovative Trend Analysis (ITA) methods were applied using monthly precipitation data between 1966-2021. According to the Mann-Kendall test, there is a decreasing trend in total precipitation during the wheat growing season at six stations and an increasing trend in total precipitation during the sunflower growing season at eight stations. Statistically significant trends of 1.75% were found in the Mann-Kendall test and 80.7% in the ITA method. Standard Precipitation Index (SPI) for 1, 3, 6 and 12 months was calculated for all stations. Agricultural drought analysis was performed with the Vegetation Health Index (VHI). According to VHI, the most severe agricultural drought was experienced in 2001. A correlation test was performed between the average VHI values of the rainfed agricultural areas in 2001-2002 and the 1, 3, 6 and 12-month SPI values representing these areas and it was determined that the 12-month SPI indicated agricultural drought. According to the obtained SPI12 values, critical drought severity-duration-frequency curves were constructed at different return intervals. According to these curves, it was determined that Kırklareli and Lüleburgaz stations are more prone to severe drought in the long term. Finally, future drought severity was predicted using ARIMA, LSTM and CNN models. The ARIMA model underperformed in the 12-month long-term drought forecasts, while the LSTM and CNN models outperformed during this period. LSTM performed best for Kırklareli station (training RMSE:0.042 and test RMSE:0.105), while CNN performed best for Sarıyer station (training RMSE:0.029 and test RMSE:0.091). Information on drought forecasting and expected severity will help water resource managers, infrastructure planners, government policy makers and decision makers in future agricultural and water management planning, as well as in the design and construction of more resilient communities.