GEE-based environmental monitoring and phenology correlation investigation using Support Vector Regression

Abstract

Environmental changes over time and across different regions profoundly affect agriculture, forestry, water management, public health, and ecosystems. Therefore, monitoring these fluctuations is crucial for informing decision-making and developing strategies for long-term sustainability. While ground-based methods provide valuable insights into environmental dynamics, they are inherently limited in scope and coverage. Consequently, satellite-based techniques have become essential for comprehensive ecological monitoring over extensive spatial and temporal scales. This study investigates spatio-temporal patterns of environmental factors and their correlation with phenology in Ilam Province, Iran, from 2014 to 2021, utilizing remote sensing data and Google Earth Engine (GEE). Landsat 8 satellite data was used to generate time series maps and timelines for land cover, temperature, and soil moisture, using the Soil-Adjusted Vegetation Index (SAVI), Land Surface Temperature (LST) anomaly, and Soil Moisture Index (SMI). Subsequently, the Temporal Soil-Adjusted Vegetation Phenology Index (TSPI) was calculated to track annual vegetation variations and analyze its correlation with the specified parameters using Support Vector Regression (SVR). Our results revealed significant trends in environmental factors, highlighting robust correlations with the TSPI. Soil moisture peaked in late winter and early spring, declining during the summer, with the highest levels recorded in 2018. Vegetation reached its maximum density in mid-spring and its minimum in winter, with a notable greening surge observed in 2019. Temperatures were highest in summer and lowest in winter, showing minimal year-to-year variation. Spatial analysis indicated a consistent increase in land surface temperature from the northeast toward the southwest, corresponding to declines in vegetation and soil moisture levels. Regression analysis specified strong associations between the TSPI and environmental variables, with R-squared values of 0.83 for LST, 0.86 for SAVI, and 0.79 for SMI. These findings emphasize the effectiveness of remote sensing methods, such as time series satellite imagery and streamlined indices, for large-scale ecological analyses using the GEE platform and underscore the potential of TSPI as a proper indicator for future environmental management research.