A machine learning based approach for forecasting remotely sensed vegetation health in Italy.

Machine learning approaches are increasingly popular for monitoring and predicting vegetation health and activities. Satellite-based vegetation health indices such as Normalised Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) have been extensively utilized to monitor vegetation activities. These indices are quantitative assessments of the vegetation that indicate a monthly pattern. In this study, remotely sensed (Sentinel-2) and in situ (Tree Talker) data were used to extract NDVI and EVI values. Trend analysis was performed using Mann-Kendall (MK) test and Sen’s slope to understand the vegetation activities pattern. Four machine learning algorithms Linear and Polynomial Regression, Random Forest (RF) and Support Vector Regression (SVR) were tested for forecasting the vegetation indices. Results demonstrate good performance of the models in reproducing the spatiotemporal variability of NDVI and EVI, even under extreme conditions. These models show great potential for monitoring vegetation dynamics and as an alternate approach in predicting vegetation health.