Abu Ghraib Vegetation Forecasting: NDVI Analysis Using Google Earth Engine Landsat 8 Data and MATLAB-Based Machine Learning

Abstract

This study investigates vegetation dynamics and forecasts future vegetation conditions in Abu Ghraib, Iraq, using a monthly Normalized Difference Vegetation Index (NDVI) time series derived from Landsat 8 imagery through the Google Earth Engine (GEE) platform. The study area represents an irrigated, semi-arid agricultural environment affected by seasonal water availability, climatic variability, and land-use pressure. Landsat 8 data from January 2015 to December 2023 were processed to generate monthly NDVI values using spatial and temporal filtering and cloud masking. The resulting time series was used as input for a Random Forest Regression model implemented in MATLAB. A 12-month sliding-window approach transformed the NDVI sequence into supervised learning samples, enabling the model to capture annual seasonality and temporal dependence. The trained model forecast NDVI trends over a five-year horizon. The model produced low absolute prediction errors, with a test RMSE of 0.0149, MAE of 0.0126, and MAPE of 18.5%. However, the negative test R² value of −0.35 indicates limited ability to explain variance in unseen data, despite capturing the general seasonal NDVI pattern. This limitation may be attributed to the relatively short time series and environmental variability. Residual analysis, out-of-bag error behavior, autocorrelation assessment, and feature-importance ranking supported the model’s ability to reproduce the dominant temporal structure. The forecasted outputs can support agricultural planning, irrigation monitoring, environmental assessment, and sustainable resource management in semi-arid irrigated regions.