Author Contributions
Conceptualization, D.H. (Dale Hamilton); methodology, D.H. (Dale Hamilton); software, C.M., D.H. (Daniel Harris) and W.G.; validation, C.M., D.H. (Daniel Harris) and W.G.; formal analysis, C.M., D.H. (Daniel Harris) and W.G.; investigation, D.H. (Dale Hamilton), C.M., D.H. (Daniel Harris) and W.G.; resources, D.H. (Dale Hamilton), C.M., D.H. (Daniel Harris) and W.G.; data curation, C.M., D.H. and W.G.; writing—original draft preparation, D.H. (Dale Hamilton), C.M., D.H. (Daniel Harris) and W.G.; writing—review and editing, D.H. (Daniel Harris), W.G., C.M. and D.H. (Dale Hamilton); visualization, C.M., D.H. (Daniel Harris) and W.G.; supervision, D.H. (Dale Hamilton); project administration, D.H. (Dale Hamilton); funding acquisition, D.H. (Dale Hamilton). All authors have read and agreed to the published version of the manuscript.
Figure 1.
Locations of the three study areas: Mesa, Four Corners, and McFarland fires.
Figure 1.
Locations of the three study areas: Mesa, Four Corners, and McFarland fires.
Figure 2.
The decision tree created by Sci-kit Learn for the Mesa fire.
Figure 2.
The decision tree created by Sci-kit Learn for the Mesa fire.
Figure 3.
The decision tree created by Sci-kit Learn for the Four Corners fire.
Figure 3.
The decision tree created by Sci-kit Learn for the Four Corners fire.
Figure 4.
Four Corners fire RGB imagery compared with the imagery informed by the ID3. (a) RGB post-fire image. (b) ID3 informed imagery composed of blue, red, and NIR bands. Imagery © 2023 Planet Labs Inc.
Figure 4.
Four Corners fire RGB imagery compared with the imagery informed by the ID3. (a) RGB post-fire image. (b) ID3 informed imagery composed of blue, red, and NIR bands. Imagery © 2023 Planet Labs Inc.
Figure 5.
The decision tree created by Sci-kit Learn for the McFarland fire.
Figure 5.
The decision tree created by Sci-kit Learn for the McFarland fire.
Figure 6.
McFarland fire RGB imagery compared with the imagery transformed by PCA. (a) RGB Post-fire image. (b) PCA-transformed imagery. Imagery © 2023 Planet Labs Inc.
Figure 6.
McFarland fire RGB imagery compared with the imagery transformed by PCA. (a) RGB Post-fire image. (b) PCA-transformed imagery. Imagery © 2023 Planet Labs Inc.
Figure 7.
Examples of training data drawn from the Mesa fire. Imagery © 2023 Planet Labs Inc.
Figure 7.
Examples of training data drawn from the Mesa fire. Imagery © 2023 Planet Labs Inc.
Figure 8.
The classification created by the support vector machine for the Mesa Fire. Dark grey indicates burn, light grey indicates high-severity burn, green indicates vegetation, and pink indicates unburned surface.
Figure 8.
The classification created by the support vector machine for the Mesa Fire. Dark grey indicates burn, light grey indicates high-severity burn, green indicates vegetation, and pink indicates unburned surface.
Figure 9.
Demonstrates the difference in spatial resolution between the 5 cm per pixel drone imagery (a) and the 3 m per pixel satellite imagery (b) when creating validation polygons for the Mesa fire. Imagery © 2023 Planet Labs Inc.
Figure 9.
Demonstrates the difference in spatial resolution between the 5 cm per pixel drone imagery (a) and the 3 m per pixel satellite imagery (b) when creating validation polygons for the Mesa fire. Imagery © 2023 Planet Labs Inc.
Table 1.
The acquisition dates of all Planet Scope imagery used for each fire.
Table 1.
The acquisition dates of all Planet Scope imagery used for each fire.
Fires | Pre-Fire Acquisition | Post-Fire Acquisition | Post-Fire Drone Acquisition |
---|
Four Corners | 15 July 2022 | 12 October 2022 | 6 July 2023 |
McFarland | 27 July 2021 | 16 September 2021 | N/A |
Mesa | 15 July 2018 | 3 September 2018 | 8 September 2018 |
Table 2.
Interopability between PlanetScope SuperDove and Sentinel-2 bands [
14]. Note that PlanetScope imagery does not contain the short-wave infrared bands contained in Sentinel-2 imagery.
Table 2.
Interopability between PlanetScope SuperDove and Sentinel-2 bands [
14]. Note that PlanetScope imagery does not contain the short-wave infrared bands contained in Sentinel-2 imagery.
Band | Name | Wavelength (FWHM) | Interoperable with Sentinel-2 |
---|
1 | Coastal Blue | 443 (20) | Yes—Sentinel-2 Band 1 |
2 | Blue | 490 (50) | Yes—Sentinel-2 Band 2 |
3 | Green 1 | 531 (36) | No equivalent with Sentinel-2 |
4 | Green | 565 (36) | Yes—Sentinel-2 Band 3 |
5 | Yellow | 610 (20) | No equivalent with Sentinel-2 |
6 | Red | 665 (31) | Yes—Sentinel-2 Band 4 |
7 | Red Edge | 705 (15) | Yes—Sentinel-2 Band 5 |
8 | NIR | 865 (40) | Yes—Sentinel-2 Band 8a |
Table 3.
Comparison of band frequencies between the DJI Phantom and PlanetLabs PS2 and PSB.SD sensors.
Table 3.
Comparison of band frequencies between the DJI Phantom and PlanetLabs PS2 and PSB.SD sensors.
| Phantom 4 | PlanetScope PS2 (Dove) | PlanetScope PSB.SD (Super Dove) |
---|
Band Name | Band | Frequency | Band | Frequency | Band | Frequency |
Coastal Blue | | | | | 1 | 431–452 |
Blue | 1 | 434–466 | 1 | 455–515 | 2 | 465–515 |
Green 1 | | | | | 3 | 513–549 |
Green | 2 | 544–576 | 2 | 500–590 | 4 | 547–583 |
Yellow | | | | | 5 | 600–620 |
Red | 3 | 634–666 | 3 | 590–670 | 6 | 650–680 |
Red Edge | | | | | 7 | 679–713 |
Near-infrared | | | 4 | 780–860 | 8 | 845–885 |
Table 4.
Unburned subclasses for each fire.
Table 4.
Unburned subclasses for each fire.
| Unburned Vegetation | Unburned Surface | Misc. |
---|
Mesa | Yes | Yes | Yes, black pixel border |
Four Corners | Yes | Yes | No |
McFarland | Yes | Yes | No |
Table 5.
How input imagery for the support vector machine was obtained for the Mesa, Four Corners, and McFarland fires.
Table 5.
How input imagery for the support vector machine was obtained for the Mesa, Four Corners, and McFarland fires.
|
Eight-Band
|
Four-Band
|
Three-Band RGB
|
Three-Band PCA-Transformed
|
Three-Band Informed by Band Entropy
|
---|
Mesa
|
N/A
|
Planet Scope PS2
|
Extracted PS2
|
PCA on PS2
|
Decision Tree on PS2
|
Four Corners
|
PlanetScope PSB.SD
|
Extracted PSB.SD
|
Extracted PSB.SD
|
PCA on PSB.SD
|
Decision Tree on PSB.SD
|
McFarland
|
PlanetScope PSB.SD
|
Extracted PSB.SD
|
Extracted PSB.SD
|
PCA on PSB.SD
|
Decision Tree on PSB.SD
|
Table 6.
Confusion matrix evaluation metrics for the burn extent of each image classified using a support vector machine for the Mesa fire.
Table 6.
Confusion matrix evaluation metrics for the burn extent of each image classified using a support vector machine for the Mesa fire.
Input Layer | Accuracy | Sensitivity | Specificity |
---|
RGB Bands | 77.41% | 97.86% | 57.40% |
RGB–NIR Four-Band Planet Scope | 86.50% | 93.96% | 79.19% |
PCA-Transformed Bands | 88.93% | 89.07% | 88.79% |
ID3-Informed Bands | 88.36% | 92.57% | 84.24% |
Average | 85.30% | 93.37% | 77.41% |
Table 7.
Confusion matrix evaluation metrics for the burn extent of each image classified using a support vector machine for the Four Corners fire.
Table 7.
Confusion matrix evaluation metrics for the burn extent of each image classified using a support vector machine for the Four Corners fire.
Input Layer | Accuracy | Sensitivity | Specificity |
---|
RGB Bands | 86.05% | 80.98% | 88.50% |
RGB–NIR Four-Band Planet Scope | 93.46% | 98.36% | 91.21% |
All Eight-Band Planet Scope | 92.66% | 97.78% | 90.39% |
PCA-Transformed Bands | 94.93% | 100.00% | 92.49% |
ID3-Informed Bands | 94.27% | 98.40% | 92.27% |
Average | 92.27% | 95.10% | 90.97% |
Table 8.
Confusion matrix evaluation metrics for the burn extent of each image classified using a support vector machine for the McFarland fire.
Table 8.
Confusion matrix evaluation metrics for the burn extent of each image classified using a support vector machine for the McFarland fire.
Input Layer | Accuracy | Sensitivity | Specificity |
---|
RGB Bands | 91.24% | 82.83% | 96.84% |
RGB–NIR Four-Band Planet Scope | 96.22% | 98.22% | 94.94% |
All Eight-Band Planet Scope | 89.07% | 72.06% | 99.89% |
PCA-Transformed Bands | 88.04% | 73.79% | 97.11% |
ID3-Informed Bands | 84.49% | 71.33% | 92.87% |
Average | 89.81% | 79.65% | 96.33% |
Table 9.
Confusion matrix evaluation metrics for the burn severity of each image classified using a support vector machine for the Mesa fire.
Table 9.
Confusion matrix evaluation metrics for the burn severity of each image classified using a support vector machine for the Mesa fire.
Input Layer | Accuracy | Sensitivity (Classified White Ash Well) | Specificity (Classified Black Ash Well) |
---|
RGB Bands | 86.18% | 72.76% | 99.45% |
RGB–NIR Four-Band Planet Scope | 88.50% | 78.29% | 99.52% |
PCA-Transformed Bands | 85.10% | 73.72% | 98.92% |
ID3-Informed Bands | 83.75% | 71.82% | 97.67% |
Average | 85.88% | 74.15% | 98.89% |
Table 10.
Confusion matrix evaluation metrics for the burn severity of each image classified using a support vector machine for the Four Corners fire.
Table 10.
Confusion matrix evaluation metrics for the burn severity of each image classified using a support vector machine for the Four Corners fire.
Input Layer | Accuracy | Sensitivity (Classified White Ash Well) | Specificity (Classified Black Ash Well) |
---|
RGB Bands | 98.19% | 95.08% | 100.00% |
RGB–NIR Four-Band Planet Scope | 97.78% | 94.44% | 100.00% |
All Eight-Band Planet Scope | 95.45% | 88.73% | 100.00% |
PCA-Transformed Bands | 96.24% | 91.57% | 100.00% |
ID3-Informed Bands | 94.59% | 88.24% | 100.00% |
Average | 96.45% | 91.61% | 100.00% |
Table 11.
Confusion matrix evaluation metrics for the burn severity of each image classified using a support vector machine for the McFarland fire. These results quantify white ash against all other classes.
Table 11.
Confusion matrix evaluation metrics for the burn severity of each image classified using a support vector machine for the McFarland fire. These results quantify white ash against all other classes.
Input Layer | Accuracy | Sensitivity (White Ash That Was Correctly Classified) | Specificity |
---|
RGB Bands | 91.99% | 66.70% | 98.05% |
RGB–NIR Four-Band Planet Scope | 96.20% | 96.36% | 96.17% |
All Eight-Band Planet Scope | 89.07% | 43.83% | 99.92% |
PCA-Transformed Bands | 88.06% | 47.42% | 97.81% |
ID3-Informed Bands | 84.53% | 42.54% | 94.60% |
Average | 89.97% | 69.37% | 97.31% |
Table 12.
Confusion matrix evaluation metrics for the burn severity of each image classified using a support vector machine for the McFarland fire, ignoring all data that is either incorrectly not identified as or incorrectly identified as white ash.
Table 12.
Confusion matrix evaluation metrics for the burn severity of each image classified using a support vector machine for the McFarland fire, ignoring all data that is either incorrectly not identified as or incorrectly identified as white ash.
Input Layer | White Ash Correctly Classified as White Ash | Amount of Classified White Ash That Is Actually White Ash |
---|
RGB Bands | 66.70% | 89.14% |
RGB–NIR Four-Band Planet Scope | 96.36% | 85.78% |
All Eight-Band Planet Scope | 48.83% | 99.24% |
PCA-Transformed Bands | 47.42% | 83.85% |
ID3-Informed Bands | 42.54% | 65.37% |
Table 13.
Average confusion matrix evaluation metrics across all three of the studied fires.
Table 13.
Average confusion matrix evaluation metrics across all three of the studied fires.
Input Layer | Accuracy | Sensitivity | Specificity |
---|
RGB Bands | 84.90% | 87.22% | 80.91% |
RGB–NIR Four-Band Planet Scope | 92.06% | 96.85% | 88.45% |
All Eight-Band Planet Scope * | 90.87% | 84.92% | 95.14% |
PCA-Transformed Bands ** | 90.63% | 87.62% | 92.80% |
ID3-Informed Bands ** | 89.04% | 87.43% | 89.79% |
Average | 89.50% | 88.81% | 89.42% |
Table 14.
Average accuracy metrics for burn extent across all imagery layers used in each fire.
Table 14.
Average accuracy metrics for burn extent across all imagery layers used in each fire.
Fire | Accuracy | Sensitivity | Specificity |
---|
Four Corners | 92.27% | 95.10% | 90.97% |
Mesa | 85.30% | 93.37% | 77.41% |
McFarland | 89.81% | 79.65% | 96.33% |
Table 15.
Average accuracy metrics for burn severity across all imagery layers used in each fire.
Table 15.
Average accuracy metrics for burn severity across all imagery layers used in each fire.
Input Layer | Accuracy | Sensitivity (Classified White Ash Well) | Specificity (Classified Black Ash Well) |
---|
RGB Bands | 92.12% | 78.18% | 99.17% |
RGB–NIR Four-Band PlanetScope | 94.16% | 89.70% | 98.56% |
All Eight-Band PlanetScope * | 92.26% | 66.28% | 99.96% |
PCA-Transformed Bands ** | 89.80% | 70.90% | 98.91% |
ID3-Informed Bands ** | 87.62% | 67.53% | 97.42% |
Average | 91.20% | 74.52% | 98.74% |
Table 16.
Average accuracy metrics for burn severity across all imagery layers used in each fire, separated by fire.
Table 16.
Average accuracy metrics for burn severity across all imagery layers used in each fire, separated by fire.
Fire | Accuracy | Sensitivity | Specificity |
---|
Mesa | 85.88% | 74.15% | 98.89% |
McFarland | 89.97% | 69.37% | 97.31% |
Four Corners | 96.45% | 91.61% | 100% |