[Paper Review] Monitoring the Impact of Wildfires on Tree Species with Deep Learning

Zhou Wang, Levente J. Klein|arXiv (Cornell University)|Nov 4, 2020

Fire effects on ecosystems25 references4 citations

TL;DR

This paper proposes a deep learning model to classify tree species from four-band aerial imagery before and after wildfires, achieving 92.2% accuracy. It reveals long-term impacts such as reduced conifer regeneration and shrub encroachment in burned areas, demonstrating the method's ability to track ecological changes across large regions over time.

ABSTRACT

One of the impacts of climate change is the difficulty of tree regrowth after wildfires over areas that traditionally were covered by certain tree species. Here a deep learning model is customized to classify land covers from four-band aerial imagery before and after wildfires to study the prolonged consequences of wildfires on tree species. The tree species labels are generated from manually delineated maps for five land cover classes: Conifer, Hardwood, Shrub, ReforestedTree and Barren land. With an accuracy of $92\%$ on the test split, the model is applied to three wildfires on data from 2009 to 2018. The model accurately delineates areas damaged by wildfires, changes in tree species and rebound of burned areas. The result shows clear evidence of wildfires impacting the local ecosystem and the outlined approach can help monitor reforested areas, observe changes in forest composition and track wildfire impact on tree species.

Motivation & Objective

• To address the challenge of monitoring long-term changes in tree species composition after wildfires due to sparse and irregular field surveys.
• To develop a scalable, automated method for classifying tree species using high-resolution aerial imagery to support large-scale forest monitoring.
• To quantify the impact of repeated wildfires on tree regeneration and ecosystem shifts, particularly in climate-vulnerable regions like California’s Sierra Nevada.
• To enable forest and environmental agencies to track reforestation success and ecological recovery over time using remote sensing and deep learning.

Proposed method

• Utilizes four-band aerial imagery (Red, Green, Blue, Near-Infrared) from the NAIP dataset collected between 2009 and 2018, resampled to 0.6 m resolution.
• Leverages manually delineated land cover maps from 2011 to generate training labels for five classes: Conifer, Hardwood, Shrub, ReforestedTree, and Barren.
• Applies a deep learning model (specific architecture not detailed) to classify 32×32 pixel image patches based on spectral signatures and vegetation indices.
• Filters low-NDVI samples to exclude non-vegetated or low-vegetation areas, improving model generalization and reducing noise.
• Trains the model on 93,849 samples, with 5,000 each for validation and testing, using a curated dataset to ensure class balance and data quality.
• Applies the trained model to three historical wildfire regions (2013 Swedes Fire, 2017 Wall Fire, 2007 Fletcher Fire) to generate time-series classification maps.

Experimental results

Research questions

• RQ1 How accurately can a deep learning model classify tree species from pre- and post-wildfire aerial imagery across multiple years?
• RQ2 What are the long-term changes in tree species composition following repeated wildfires in fire-prone regions?
• RQ3 To what extent do wildfires impede the regeneration of key tree species such as conifers and hardwoods in the Sierra Nevada?
• RQ4 Can deep learning models detect and quantify the transition from forested to shrub-dominated or barren land after wildfires?
• RQ5 How does the presence of reforestation efforts influence the recovery trajectory of burned areas over time?

Key findings

• The deep learning model achieved a test accuracy of 92.2%, demonstrating high reliability in classifying tree species from aerial imagery.
• After the 2013 Swedes Fire, 18% of the affected area converted to Barren land, with significant loss of Hardwood and Conifer coverage.
• In the 2017 Wall Fire, most Hardwood and ReforestedTree areas were lost, and Shrub coverage tripled, indicating ecosystem shift toward flammable vegetation.
• Conifer coverage declined consistently over time in the affected regions, suggesting long-term vulnerability to climate and fire stress.
• In areas without repeated fires, such as the 2007 Fletcher Fire site, Barren land decreased steadily, and ReforestedTree coverage increased, indicating potential for natural recovery.
• The model successfully captured the transition from forest to shrub-dominated landscapes in frequently burned areas, highlighting the risk of irreversible ecosystem change.