A multimodal data fusion and deep learning framework for large-scale wildfire surface fuel mapping

Mohamad Alipour; Inga P. La Puma; Joshua J. Picotte; Kasra Shamsei; Eric Rowell; Adam Watts; Branko Kosovic; Hamed Ebrahimian; Erugrul Taciroglu

doi:10.3390/fire6020036

USGS · 70248758

A multimodal data fusion and deep learning framework for large-scale wildfire surface fuel mapping

Mohamad Alipour·Inga P. La Puma·Joshua J. Picotte·Kasra Shamsei·Eric Rowell·Adam Watts·Branko Kosovic·Hamed Ebrahimian·Erugrul Taciroglu

View original source ↗Export metadata ↓

Abstract

Accurate estimation of fuels is essential for wildland fire simulations as well as decision-making related to land management. Numerous research efforts have leveraged remote sensing and machine learning for classifying land cover and mapping forest vegetation species. In most cases that focused on surface fuel mapping, the spatial scale of interest was smaller than a few hundred square kilometers; thus, many small-scale site-specific models had to be created to cover the landscape at the national scale. The present work aims to develop a large-scale surface fuel identification model using a custom deep learning framework that can ingest multimodal data. Specifically, we use deep learning to extract information from multispectral signatures, high-resolution imagery, and biophysical climate and terrain data in a way that facilitates their end-to-end training on labeled data. A multi-layer neural network is used with spectral and biophysical data, and a convolutional neural network backbone is used to extract the visual features from high-resolution imagery. A Monte Carlo dropout mechanism was also devised to create a stochastic ensemble of models that can capture classification uncertainties while boosting the prediction performance. To train the system as a proof-of-concept, fuel pseudo-labels were created by a random geospatial sampling of existing fuel maps across California. Application results on independent test sets showed promising fuel identification performance with an overall accuracy ranging from 55% to 75%, depending on the level of granularity of the included fuel types. As expected, including the rare—and possibly less consequential—fuel types reduced the accuracy. On the other hand, the addition of high-resolution imagery improved classification performance at all levels.

Explore related subjects

Fire Article

Source-reported bounding extent: 32.534156° to 42.009247° latitude; -124.409591° to -114.134768° longitude. This indicates report coverage, not an exact sampling location. View area on OpenStreetMap.

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Mohamad Alipour, Inga P. La Puma, Joshua J. Picotte, Kasra Shamsei, Eric Rowell, Adam Watts, Branko Kosovic, Hamed Ebrahimian, Erugrul Taciroglu. 2023-01-17. A multimodal data fusion and deep learning framework for large-scale wildfire surface fuel mapping. https://doi.org/10.3390/fire6020036

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related USGS reports

USGS2024

The spatially adaptable filter for error reduction (SAFER) process: Remote sensing-based LANDFIRE disturbance mapping updates

LANDFIRE (LF) has been producing periodic spatially explicit vegetation change maps (i.e., LF disturbance products) across the entire United States since 1999 at a 30 m spatial resolution. These disturbance products include data products produced by various fire programs, field-mapped vegetation and fuel treatment activity (i.e., events) submissions from various agencies, and disturbances detected by the U.S. Geological Survey Earth Resources Observation and Science (EROS)-based Remote Sensing of Landscape Change (RSLC) process. The RSLC process applies a bi-temporal change detection algorithm to Landsat satellite-based seasonal composites to generate the interim disturbances that are subsequently reviewed by analysts to reduce omission and commission errors before ingestion them into LF’s disturbance products. The latency of the disturbance product is contingent on timely data availability and analyst review. This work describes the development and integration of the Spatially Adaptable Filter for Error Reduction (SAFER) process and other error and latency reduction improvements to the RSLC process. SAFER is a random forest-based supervised classifier and uses predictor variables that are derived from multiple years of pre- and post-disturbance Landsat band observations. Predictor variables include reflectance, indices, and spatial contextual information. Spatial contextual information that is unique to each contiguous disturbance region is parameterized as Z scores using differential observations of the disturbed regions with its undisturbed neighbors. The SAFER process was prototyped for inclusion in the RSLC process over five regions within the conterminous United States (CONUS) and regional model performance, evaluated using 2016 data. Results show that the inclusion of the SAFER process increased the accuracies of the interim disturbance detections and thus has potential to reduce the time needed for analyst review. LF does not track the time taken by each analyst for each tile, and hence, the relative effort saved was parameterized as the percentage of 30 m pixels that are correctly classified in the SAFER outputs to the total number of pixels that are incorrectly classified in the interim disturbance and are presented. The SAFER prototype outputs showed that the relative analysts’ effort saved could be over 95%. The regional model performance evaluation showed that SAFER’s performance depended on the nature of disturbances and availability of cloud-free images relative to the time of disturbances. The accuracy estimates for CONUS were inferred by comparing the 2017 SAFER outputs to the 2017 analyst-reviewed data. As expected, the SAFER outputs had higher accuracies compared to the interim disturbances, and CONUS-wide relative effort saved was over 92%. The regional variation in the accuracies and effort saved are discussed in relation to the vegetation and disturbance type in each region. SAFER is now operationally integrated into the RSLC process, and LANDFIRE is well poised for annual updates, contingent on the availability of data.

Fire↗

USGS2023

Unrecorded tundra fires of the Arctic Slope, Alaska USA

Few fires are known to have burned the tundra of the Arctic Slope north of the Brooks Range in Alaska, USA. A total of 90 fires between 1969 and 2022 are known. Because fire has been rare, old burns can be detected by the traces of thermokarst and distinct vegetation they leave in otherwise uniform tundra, which are visible in aerial photograph archives. Several prehistoric tundra burns have been found in this way. Detection of tundra fires in this sparsely populated and remote area has been historically inconsistent and opportunistic, relying on reports by aircraft pilots. Fire reports have been logged into an administrative database which, out of necessity, has been used to scientifically evaluate changes in the fire regime. To improve the consistency of the record, we completed a systematic search of Landsat Collection 2 for the Brooks Range Foothills ecoregion over the period 1972–2022. We found 57 unrecorded tundra burns, about 41% of the total, which now numbers 138. Only 15% and 33% of all fires appear in MODIS and VIIRS satellite-borne thermal anomaly products, respectively. The fire frequency in the first 37 years of the record is 0.89 y −1 for natural ignitions that spread ≥10 ha. Frequency in the last 13 years is 2.5 y −1 , indicating a nearly three-fold increase in fire frequency.

Alaska↗

USGS2022

What makes wildfires destructive in California?

As human impacts from wildfires mount, there is a pressing need to understand why structures are lost in destructive fires. Despite growing research on factors contributing to structure loss, fewer studies have focused on why some fires are destructive and others are not. We characterized overall differences between fires that resulted in structure loss (“destructive fires”) and those that did not (“non-destructive wildfires”) across three California regions. Then, we performed statistical analyses on large fires only (≥100 ha) to distinguish the primary differences between large destructive large fires and large non-destructive fires. Overall, destructive fires were at least an order of magnitude larger than non-destructive fires, with the largest area burned varying by season in different regions. Fire severity was also significantly higher in destructive than non-destructive fires. The statistical analysis showed that, in the San Francisco Bay Area and the northern Sierra Nevada foothills, proximity to the Wildland Urban Interface (WUI) was by far the most important factor differentiating destructive and non-destructive wildfires, followed by different combinations of short-term weather, seasonal climate, topography, and vegetation productivity. In Southern California, wind velocity on the day of the fire ignition was the top factor, which is consistent with previous assumptions that wind-driven fires tend to be most destructive and most of the destruction occurs within the first 24 h. Additionally, Southern California’s high population density increases the odds that a human-caused wildfire may occur during a severe fire-weather event. The geographical differences among regions and the variation of factors explaining the differences between large destructive and large non-destructive fires reflects the complexity inherent in decision-making for reducing wildfire risk. Land use planning to reduce future exposure of housing development to fire and increased focus on wildfire ignition prevention emerge as two approaches with substantial potential.

California↗