Assessing the impacts of urbanization and climate variability on land use dynamics using LSTM networks and satellite remote sensing data

Majeed, Mokhalad Azzat; Sammen, Saad; Ahmed, Kaywan Othman

doi:10.22098/mmws.2026.19554.1797

Assessing the impacts of urbanization and climate variability on land use dynamics using LSTM networks and satellite remote sensing data

مقالات آماده انتشار

نوع مقاله : Special Issue: New Approaches to Water and Soil Management and Modeling

نویسندگان

¹ Department of Civil Engineering, College of Engineering, University of Diyala, Diyala Governorate, Baqubah 32001, Iraq

² Civil Engineering Department, Tishk International University, Sulaimani, Iraq

10.22098/mmws.2026.19554.1797

چکیده

This study investigates the impacts of climate variability and urbanization on vegetation dynamics in Diyala City, Iraq, using remote sensing data and deep learning techniques. Multi-source satellite and climate datasets covering the period from 2014 to 2024 were processed and analyzed using Google Earth Engine (GEE). Sentinel-2, Landsat-8/9, MODIS, and CHIRPS datasets were utilized to derive Land Surface Temperature (LST), Normalized Difference Vegetation Index (NDVI), and Normalized Difference Built-up Index (NDBI). In addition, Dynamic World land use/land cover (LULC) data were employed to quantify temporal changes in vegetation cover, croplands, and built-up areas. A Long Short-Term Memory (LSTM) deep learning model was developed to forecast NDVI variations based on historical environmental data. The results revealed a substantial expansion of built-up areas by approximately 61.3% between 2017 and 2024, accompanied by a 15.4% increasing in crop-cover areas and a 72.7% decrease in high-vegetation-density classes. Furthermore, low and very low NDVI classes are significantly higher in central urbanised regions, while NDBI and LST analyses indicated increasing urban density and surface temperature patterns. The LSTM model demonstrated strong predictive capability, achieving validation RMSE and MAE values of 0.047 and 0.034, respectively, indicating reliable performance in forecasting vegetation dynamics. The findings demonstrate the effectiveness of integrating remote sensing technologies, Google Earth Engine, and deep learning models for environmental monitoring and vegetation prediction. This study provides valuable insights for sustainable urban planning, environmental management, and long-term land-use monitoring in semi-arid regions.

کلیدواژه‌ها