Table 1: Sentinel-2 Live Fuel Moisture Index (SLFMI) - Live Fuel Moisture (LFM) Correlation Statistics
Correlations below are between Sentinel-2 Live Fuel Moisture Index (SLFMI; based on index and algorithm developed and analyzed by GIS Ag Maps) values and field live fuel moisture (LFM) measurements (percent moisture by weight in vegetation) by United States Forest Service-Wildland Fire Assessment System (USFS-WFAS; offsite page that opens in new tab, http://www.wfas.net/index.php/national-fuel-moisture-database-moisture-drought-103). The following tables are based on Table 1 near bottom of page; correlations are high for both, though higher when representing a single season (as opposed to two seasons combined).
The following are summaries of Table 1, which is near bottom of page:
(average and median of residuals [LFM prediction error] of Row 8 on full table near bottom, are about 2.0 to 2.5)
| All Correlations |
Correlations for Single Seasons | ||||||
| Row | Average R² | Median R² | Row | Average R² | Median R² | ||
| 1 | 0.940 | 0.934 | 1 | 0.938 | 0.934 | ||
| 2 | 0.909 | 0.908 | 2 | 0.911 | 0.908 | ||
| 3 | 0.924 | 0.953 | 3 | 0.926 | 0.962 | ||
| 4 | 0.890 | 0.948 | 4 | 0.888 | 0.939 | ||
| 5 | 0.821 | 0.848 | 7 | 0.939 | 0.933 | ||
| 6 | 0.787 | 0.823 | 9 | 0.924 | 0.916 | ||
| 7 | 0.939 | 0.933 | 10 | 0.909 | 0.962 | ||
| 9 | 0.924 | 0.916 | 12 | 0.885 | 0.948 | ||
| 10 | 0.909 | 0.962 | |||||
| 12 | 0.885 | 0.948 | |||||
| 13 | 0.800 | 0.811 | |||||
| 15 | 0.780 | 0.770 | |||||
| Average | 0.876 | 0.896 | Average | 0.915 | 0.938 | ||
| Median | 0.900 | 0.924 | Median | 0.918 | 0.937 | ||
Correlations are between wild shrubland Sentinel-2 Live Fuel Moisture Index values (SLFMI; based on index and algorithm developed and analyzed by GIS Ag Maps) and linearly interpolated United States Forest Service-Wildland Fire Assessment System values (USFS-WFAS; publicly available at: http://www.wfas.net/index.php/national-fuel-moisture-database-moisture-drought-103). A small amount of LFM values happened to be taken on the same day as imagery, so interpolation was not necessary; also, a small amount of values were extrapolated, but not more than three days. If there was a very large break in values between LFM measurements, where impactful weather could have occurred and, hence, significantly affected linearly interpolated value accuracy, those values were not used (very uncommon). The same pixels, in sunlight (mainly determined by hillshade rasters), were used throughout the year/s for each site, so change was detected at the same map location.
The index was designed for there to be a focus on accuracy at lower, more dangerous LFM values. Imagery was used for solar elevations < 45° because atmospheric correction becomes too difficult at lower sun angles (45° solar elevation is about the beginning to end of October depending on the location in the state of California). Therefore, correlations were made from May 1st (about halfway though spring) to < 45° and separately from LFM < 120 to < 45°.
Number corresponds to row in table below.
1) R² 2020 from May 1st to < 45° solar elevation; the trendline with the highest correlation is listed (Exp = exponential, Lin = linear, Log = logarithmic, Pwr = power).
2) R² linear (linear correlation is always listed; it may or may not be the highest correlation).
3) R² 2019 from May 1st to < 45° solar elevation (trendline with highest correlation is listed).
4) R² linear.
5) R² 2020-19 combined from May 1st to < 45° image solar elevation (N/A if there is LFM data for both years; trendline with highest correlation is listed).
6) R² linear.
LFM Residuals are listed for correlations below based on the highest trendline correlation. Correlations below are for LFM < 120 (LFM in correlations above usually include values > 120) to solar elevations < 45°. The extra information of residuals are included for LFM<120 because that is when LFM values approach or are in the dangerous zone of <100.
7) R² 2020 from LFM < 120 to < 45° solar elevation (trendline with highest correlation is listed).
8) LFM residual (error) based the highest correlation (previously listed).
9) R² linear.
10) R² 2019 from LFM < 120 to < 45° solar elevation (trendline with highest correlation is listed).
11) LFM residual (error) based the highest correlation (previously listed).
12) R² linear.
13) R² 2020-2019 combined from LFM < 120 to < 45° image solar elevation (N/A if there is not LFM data for both years; trendline with highest correlation is listed).
14) LFM residual (error) based the highest correlation (previously listed).
15) R² linear.
N = Amount of image dates for particular correlation
Trendline Type for Correlation; the highest correlation of the following is listed first: Exp = Exponential, Lin = Linear (always listed), Log = Logarithmic, Pwr = Power
The P-value is two-tailed with N-2 degrees of freedom and is < 0.0001 (extremely statistically significant), unless otherwise specified.
Table 1: Correlation Statistics for the 8 Sites (divided into three parts; correlations for additional area calculated after table was developed are listed in Appendix A below table)
Appendix 1 - Additional Correlations
1) Angeles Forest Highway; Chamise; 32 same pixels per dataset; Row 1 correlation is R² = .963 (Lin; N=20); Row 2 correlation is R² = .963; Row 7 correlation is R² = .962 (Log; N=20); Row 9 correlation is R² = .955; all other correlation rows are N/A.
2) Montana, Cooper Creek (1 & 2) and Butte Lookout Average; Huckleberry, Thinleaf; 338 same pixels per dataset; Row 1 correlation is R² = .933 (Log; N=13); Row 2 correlation is R² = .902; all other correlation rows are N/A.