Document Type : Original Research Paper
Authors
Department of Surveying Engineering, South Tehran Branch, Islamic Azad University, Tehran, Iran
Abstract
Background and Objectives: Forest fires are one of the most important factors in the destruction of natural ecosystems and cause the loss of a large part of vegetation cover, biodiversity loss, habitat destruction, and changes in ecological cycles every year. Therefore, monitoring the process of vegetation recovery after a fire and assessing the speed of ecosystem return to pre-fire conditions are important issues in natural resource management, environmental protection, and planning restoration measures. The development of remote sensing technologies and access to time-series satellite data have enabled long-term monitoring of vegetation changes. Also, the use of machine learning methods has become an efficient tool for predicting vegetation changes due to its high ability to model nonlinear relationships between environmental variables. The purpose of this study is to investigate the process of vegetation recovery after a fire and evaluate the performance of machine learning models in the Bastrop and Zaka fire areas.
Methods: In this study, time series data of various spectral indices, including NDVI, SAVI, SAVI2, NBR, EVI, EVI2, and NDWI, were extracted for the period from 2000 to 2022, and their temporal variations after wildfire events were investigated. After preprocessing steps, including the removal of incomplete data, feature selection, and data normalization, three machine learning algorithms, including Random Forest (RF), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoost), were trained to predict NDVI using a chronological data partitioning approach, with 70% of the data used for training and 30% for testing. The performance of the models was then evaluated, and the importance of input features was analyzed using feature importance and permutation importance methods. Finally, the vegetation recovery process and the vegetation greenness recovery rate were assessed.
Findings: The results showed that the occurrence of fires in both regions caused a sharp decrease in greening indices, indicating extensive destruction of vegetation. Although the vegetation recovery process began in the years after the fires, neither region returned to the pre-fire level by the end of 2022. The results of the model evaluation showed that all three algorithms had high accuracy, but the SVR model performed best in the test data, with the highest R² value and the lowest RMSE value. In addition, the vegetation recovery rate showed that the Bastrop region had a faster recovery process with a recovery rate of 0.0128 compared to the Zaka region with a rate of 0.0092, which could be due to differences in climatic conditions, vegetation characteristics, and fire severity.
Conclusion: The results of this study show that the integration of remote sensing data and machine learning models provides an accurate and efficient approach for monitoring and predicting the process of vegetation recovery after fire. The findings of this study can be used in assessing the long-term effects of fire, monitoring the health of forest ecosystems, planning vegetation restoration operations, and developing intelligent natural resource management systems, and pave the way for a wider use of machine learning methods in remote sensing studies and environmental monitoring.
Keywords
Main Subjects
COPYRIGHTS
© 2026 The Author(s). This is an open-access article distributed under the terms and conditions of the Creative Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)
)