Drought Monitoring and Analysis Using Multi-Source Data and Machine Learning Abdullah Sukkar, Ozan Ozturk, Dursun Zafer Seker
ITU, Istanbul Technical University
Abstract
Drought is a major environmental challenge with widespread implications for ecosystems, economies, and societies. Despite its critical importance, drought remains one of the most challenging natural disasters to monitor due to its gradual onset and complex interactions with climatic and environmental factors. In a conflict-affected region such as northeast Syria, understanding the drought patterns and trends is a crucial step in post-conflict rehabilitation, particularly because this area relies heavily on agriculture. To enhance our knowledge of the drought phenomenon in northeast Syria, a comprehensive big dataset was created, including a variety of meteorological, vegetation, and soil parameters. Then, a machine learning model based on the XGBoost algorithm was employed to assess the most important features affecting the drought. The Standardized Precipitation Evapotranspiration Index, Vegetation Health Index, and Soil Moisture anomaly were selected as targets to represent the meteorological, vegetation, and soil data. For more reliable results, when selecting a target, the data of this target was left out of the training process, which enables the detection of how other parameters affect that target. The results showed that the most important parameter affecting the drought is the temperature.
Keywords: Drought monitoring- Multi-Source data integration- XGBoost- Drought indices
