Post-Fire Forest Pulse Recovery: Superiority of Generalized Additive Models (GAM) in Long-Term Landsat Time-Series...

Area of Science:

• Ecology
• Remote Sensing
• Forestry

Background:

• Global wildfires are increasing, posing challenges for assessing post-fire vegetation recovery and ecosystem resilience.
• Long-term satellite data analysis is crucial for understanding vegetation dynamics after disturbances.
• Existing models may not adequately capture the complexity of post-fire recovery patterns.

Purpose of the Study:

• To analyze and compare post-fire vegetation recovery in contrasting fire-prone ecosystems in the US and Australia using Landsat time series.
• To evaluate the performance of different statistical models (linear, logistic, LOESS, GAM) in estimating vegetation recovery.
• To identify key factors influencing vegetation recovery and establish a transferable method for monitoring forest resilience.

Main Methods:

• Utilized long-term Landsat satellite imagery to extract vegetation area via Enhanced Vegetation Index (EVI) and Otsu thresholding.
• Modeled vegetation recovery to pre-fire levels using linear, logistic, locally estimated scatterplot smoothing (LOESS), and generalized additive models (GAM).
• Compared model performance using metrics like Akaike Information Criterion (AIC) and cross-validated Root Mean Square Error (RMSE_cv).

Main Results:

• Australian forests exhibited rapid recovery to pre-fire levels, unlike forests in the United States.
• In Australia, temperature was a dominant climatic factor influencing recovery (Spearman ρ = 0.513, p < 10⁻⁸), while no climatic variable significantly affected recovery in California.
• Generalized Additive Models (GAM) consistently outperformed other models, accurately capturing nonlinear and heterogeneous recovery patterns (e.g., US AIC: 142.89, Australia AIC: 46.70).

Conclusions:

• Post-fire vegetation recovery is nonlinear and ecosystem-specific, necessitating advanced modeling techniques.
• Simple models like linear and logistic regression are insufficient for accurately assessing complex recovery dynamics.
• Generalized Additive Models (GAM) provide a robust and transferable approach for monitoring forest resilience using remote sensing data in the face of increasing wildfire regimes.