An integrated GEE and machine learning framework for detecting ecological stability under land use/land cover changes

Abstract

Ecological stability (ES) is recognized as a crucial factor for sustainable development at global and regional scales. However, the importance of this factor was not considered significant. Hence, the main aim of this study was to introduce a new approach that focuses on detecting ES over the Maharloo watershed in Iran. To achieve this goal, we extracted land use and land cover (LULC) data from the Google Earth Engine (GEE) platform by applying the random forest (RF) machine learning method, which obtained Kappa statistics of 0.85, 0.86, and 0.87 for the years 2002, 2013, and 2023, respectively. We identified both stable and unstable regions based on LULC changes and employed them using machine learning to forecast the ES. The most important predictors of ES were elevation, soil organic carbon index, precipitation, and salinity. The results of this research revealed that certain areas within the Maharloo watershed have experienced ecological instability in recent years, with gardens showing the highest percentage (60.65 %) of instability among all land-use categories. The performance and validation of our model suggest that the study results are reliable (AUC = 0.86). This study offers detailed maps of ES and trends, offering valuable insights for decision makers to support landscape conservation and restoration efforts. Overall, the findings contribute to a more comprehensive understanding of the ecological dynamics of the Maharloo watershed and provide valuable insights for sustainable development and conservation efforts in other regions.

Keywords

• Ecological stability
• Random Forest
• Google Earth Engine
• Sustainable development