Abstract
Forest ecosystems in the Mediterranean basin are significantly affected by summer wildfires. Drought, extreme temperatures, and strong winds increase the fire risk in Greece. This study explores the potential of NDVI for assessing and forecasting post-fire regeneration in burnt areas of the Peloponnese (2007) and Evros (2011). NDVI data from Landsat 7 and 9 were analyzed to identify the stages of the regeneration process and the dominant vegetation species at each stage. Comparing pre-fire and post-fire values highlighted the recovery rate, while the trendline slope indicated the regeneration rate. This combined analysis forms a methodology that allows drawing conclusions about the vegetation type that prevails after the fire. Validation was conducted using photointerpretation techniques and CORINE land cover data. The findings suggest that sclerophyllous species regenerate faster, while fir forests recover slowly and may be replaced by sclerophylls. To predict vegetation regrowth, two time series models (ARMA, VARIMA) and two machine learning-based ones (random forest, XGBoost) were tested. Their performance was evaluated by comparing the predicted and actual numerical values, calculating error metrics (RMSE, MAPE), and analyzing how the predicted patterns align with the observed ones. The results showed the overperformance of multivariate models and the need to introduce additional variables, such as soil characteristics and the effect of climate change on weather parameters, to improve predictions.
Add the full text or supplementary notes for the publication here using Markdown formatting.
Paraskevi Oikonomou
Research Assistant
Ioannis Papoutsis
Head of Orion Lab
Assistant Professor of Artificial Intelligence for Earth Observation @ NTUA
Adjunct Researcher @ NOA
Earth Observation, Machine Learning, Natural Hazard Monitoring