Investigating Drought Dynamics: A Machine Learning-Based Integration of Multiple Remote Sensing Data


Şeker D. Z., Sukkar A., Öztürk O.

American Geophysical Union (AGU) Fall Meeting 2024, Washington, Amerika Birleşik Devletleri, 9 - 13 Aralık 2024, ss.1, (Özet Bildiri)

  • Yayın Türü: Bildiri / Özet Bildiri
  • Basıldığı Şehir: Washington
  • Basıldığı Ülke: Amerika Birleşik Devletleri
  • Sayfa Sayıları: ss.1
  • Recep Tayyip Erdoğan Üniversitesi Adresli: Evet

Özet

Drought is an intricate climatic phenomenon that can affect ecosystems and society. This complex event has wide-ranging effects that can extend beyond agricultural productivity, threatening ecological and socio-economic systems. Drought events are mainly caused by an extended period of below-average precipitation and high temperatures, resulting in drier air and reduced humidity. The severity and duration of drought events are also highly interconnected with soil properties, which can influence the occurrence of droughts. The Middle East is extremely prone to drought, especially due to ongoing challenges like high temperatures, low precipitation, and extreme climate variability. Since the early 2000s, Northeast Syria has encountered fluctuations in climate patterns that have substantially impacted its agricultural productivity. Therefore, by utilizing the recent remote sensing and GIS technologies, this study addressed the drought phenomena in the Northeastern part of Syria over the last two decades, aiming to provide a better understanding of the drought patterns and trends. Multiple data types from different sources were utilized to achieve the goals and objectives of this work. The ERA5 - Land dataset was selected as the main source for the meteorological parameters, including variables such as air temperature, soil temperature, precipitation, evaporation, dewpoint temperature, SPI, and relative humidity at an hourly scale. The SoilGrids platform was used to create a soil profile for the study area, describing the soil's physical and chemical properties. Moreover, MODIS was used to access different remotely sensed vegetation indices, such as NDVI, EVI, NDWI, NDDI, NDMI, and LST. Powerful regression models are required to establish a relationship between drought severity and the multiple data sources mentioned above. Therefore, this study used the machine-learning algorithm XGBoost as the main technique. This work is ongoing, and the results are waiting to provide comprehensive and valuable insights into the variations in climate patterns and the alteration in the vegetation cover. Most importantly, it will answer the question of which factor/s are significantly affecting the drought events in northeastern Syria.