Scholarlink

Background <p>Prescribed fire is an effective method to control woody encroachment into sagebrush steppe, which covers 40 million hectares of the Western United States. Medium resolution remote sensing products (e.g., Landfire) are widely available but do not adequately meet the needs of rangeland prescribed fire planners and fuels managers who require fine-scale, spatial depictions of fuel type (vegetation) composition and burn severity outcomes to ensure resource conservation and effective fire treatments. We compared the accuracy of pre-fire and post-fire datasets at different spatial resolutions and assessed the tradeoffs of using the data for machine learning modeling of burn severity.</p> Results <p>Our study focused on a prescribed fire that took place in a sagebrush (<i>Artemisia</i> spp.) dominated watershed in Southwestern Idaho for juniper control on 6 October 2023. We found that high resolution 0.5&#xa0;m WorldView-2 pre-fire fuel maps were 83.0% accurate compared to 45.1% with the 30-m Landfire product. We also found that a post-fire 0.5&#xa0;m WorldView-3 modeled ash map more accurately mapped burned area (90.4% accuracy) than two Sentinel-2 derived relative differenced normalized burn ratio (76.0% accuracy) and differenced normalized burn ratio (68.7% accuracy). Finally, we predicted burn severity with machine learning models to determine the tradeoffs of each combination of pre-fire and post-fire datasets. We used datasets of fuel type, topographic variables, and derivatives of both to predict severity. Our models demonstrate that high resolution fuel maps and burn severity maps are needed to develop predictive models of burn severity, especially for differentiating low and moderate severities (73.7–75.2% accurate). When coarse resolution burn severity products were used, the models collapsed and failed to converge.</p> Conclusions <p>High resolution burn severity and fuels datasets are required by rangeland prescribed fire planners, fuels managers, and wildland firefighters of the mountainous Western United States to better predict burn severity. High-resolution satellite imagery like WorldView-2 and WorldView-3 are well suited to this task, and when paired with topographic datasets, can accurately predict burn severity across diverse topography.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Assessment of pre and post fire remotely sensed data products across spatial and temporal scales for prescribed fire operations in rangelands

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月4日

查看全文
  • Craig D. Woodruff,
  • Patrick E. Clark,
  • Peter J. Olsoy,
  • Josh Enterkine,
  • Akira B. Byrne,
  • Michelle Sclafani,
  • David P. Huber,
  • John Y. Park

摘要

Background

Prescribed fire is an effective method to control woody encroachment into sagebrush steppe, which covers 40 million hectares of the Western United States. Medium resolution remote sensing products (e.g., Landfire) are widely available but do not adequately meet the needs of rangeland prescribed fire planners and fuels managers who require fine-scale, spatial depictions of fuel type (vegetation) composition and burn severity outcomes to ensure resource conservation and effective fire treatments. We compared the accuracy of pre-fire and post-fire datasets at different spatial resolutions and assessed the tradeoffs of using the data for machine learning modeling of burn severity.

Results

Our study focused on a prescribed fire that took place in a sagebrush (Artemisia spp.) dominated watershed in Southwestern Idaho for juniper control on 6 October 2023. We found that high resolution 0.5 m WorldView-2 pre-fire fuel maps were 83.0% accurate compared to 45.1% with the 30-m Landfire product. We also found that a post-fire 0.5 m WorldView-3 modeled ash map more accurately mapped burned area (90.4% accuracy) than two Sentinel-2 derived relative differenced normalized burn ratio (76.0% accuracy) and differenced normalized burn ratio (68.7% accuracy). Finally, we predicted burn severity with machine learning models to determine the tradeoffs of each combination of pre-fire and post-fire datasets. We used datasets of fuel type, topographic variables, and derivatives of both to predict severity. Our models demonstrate that high resolution fuel maps and burn severity maps are needed to develop predictive models of burn severity, especially for differentiating low and moderate severities (73.7–75.2% accurate). When coarse resolution burn severity products were used, the models collapsed and failed to converge.

Conclusions

High resolution burn severity and fuels datasets are required by rangeland prescribed fire planners, fuels managers, and wildland firefighters of the mountainous Western United States to better predict burn severity. High-resolution satellite imagery like WorldView-2 and WorldView-3 are well suited to this task, and when paired with topographic datasets, can accurately predict burn severity across diverse topography.