Wildfire Risk Scoring: A Quantitative Wildfire Risk Score Derived from GNSS-R Soil Moisture Data and Thermal Stress

Abstract

Wildfire Risk Scoring is proposing a calibrated and testable quantitative equation for estimating the probability of a wildfire in a designated ground grid cell from two satellite observables: surface soil moisture retrieved by GNSS reflectometry (GNSS-R) and land surface temperature (LST) from a thermal imager. Three published equations are combined: a monthly soil moisture anomaly (adapted from an NDVI anomaly formulation), the Crop Water Stress Index (CWSI), and the Normalized Difference Temperature Index (NDTI); with the monthly soil moisture anomaly utilizing data from the GNSS-R and the CWSI and NDTI utilizing LST from the thermal sensor respectively. They feed into a logistic model that includes soil moisture anomalies at 1 and 5 month lag intervals, so the opposite pre-fire soil moisture signals reported for arid and humid regions are learned from data rather than imposed. A three-part evaluation is proposed: the equation's own skill and calibration, its incremental value over a weather and fuel baseline (ΔAUC), and whether its fitted lags follow the expected climate patterns. On synthetic data, the equation was well calibrated (expected calibration error 0.002 to 0.003), recovered the correct lagged soil moisture effect in both climate regimes, showed a ΔAUC of +0.05 to +0.13 when a true effect was planted, and correctly showed no gain when none was. About 140 fire events in the training set detected a planted ΔAUC of 0.05 in 93% of repeats. Earlier sensitivity tests indicate that thermal accuracy near 1 K is required for the thermal inputs to carry information. Whether soil moisture and thermal stress improve real fire prediction remains a hypothesis that can possibly be answered by testing using existing CYGNSS, MODIS and gridMET records over Arizona. The motivating application is PyroSat, a proposed CubeSat that would supply both inputs.