The headline finding is deceptively simple: vineyard pixels inside the fire perimeter showed a mean differenced Normalized Burn Ratio, or dNBR, of 0.130, compared with 0.337 for wildland pixels, a 61.4 percent lower mean. Medians were 0.127 and 0.284 respectively. dNBR is a standard spectral measure that compares pre-fire and post-fire satellite reflectance in near-infrared and shortwave-infrared bands, with positive values indicating declining vegetation and moisture signal. At first glance, this contrast seems to suggest that vineyards burned less severely than the oak woodland, chaparral, and grassland around them, and the temptation to read vineyards as natural firebreaks is obvious for a wine region that adjoins some of California’s most fire-prone terrain.

But the study’s central contribution is precisely that it refuses to let that visual contrast stand as a causal claim. The team sampled 2,000 vineyard-wildland boundary segments and fitted local-linear models with segment-clustered standard errors. At the primary 100-meter bandwidth, the boundary contrast was a small negative tau of −0.0166, but the estimate changed across bandwidths, reversing to +0.0446 at the narrowest 30-meter scale and attenuating to a null −0.0043 at 300 meters. Slope, a key confounder, was discontinuous across the boundary, averaging 3.22 degrees on the vineyard side versus 4.48 degrees on the wildland side. When the immediate 100 meters on either side of the edge were excluded in a donut specification, the contrast vanished entirely, and an interior placebo boundary produced a large wrong-signed estimate of +0.0624. Together, these diagnostics indicate that vineyard placement and background fire gradients, not any protective property of vines, structure the apparent edge effect.

A complementary spatial model reinforced the same caution. On a 250-meter grid of 5,097 cells inside the perimeter, a penalized spatial generalized additive model related mean dNBR to vineyard fraction, elevation, slope, and growing-season evapotranspiration, with smooth terms for location. The joint-smooth vineyard-fraction coefficient was −0.111, with a nominal interval of −0.136 to −0.086, and the model explained 65.5 percent of deviance. Yet residual Moran’s I, a measure of spatial autocorrelation, remained 0.611 with a permutation p-value of 0.001, meaning substantial unmodeled spatial structure persisted. The authors are explicit that the nominal confidence intervals are therefore optimistic, and that the coefficient is a conditional diagnostic rather than a causal or fully spatially corrected effect. A universal firebreak claim, they conclude, is not supported.

The pre-fire picture adds further interpretive complexity. Across the 4,581 fields, mean pre-fire NDMI, a canopy moisture index, was 0.015, mean NDVI was 0.439, and mean April-through-October evapotranspiration was 470.3 millimeters. Canopy indices tracked water use coherently: NDMI correlated with ET at r = 0.567 and NDVI at r = 0.639. Mapped soil available water capacity, by contrast, explained almost nothing, correlating with NDMI at only r = 0.028 and with ET at r = −0.026. Notably, within the 1,206 vineyards with complete covariates, higher evapotranspiration was associated with higher field-mean dNBR, a standardized coefficient of +0.0384. The authors stress this does not mean irrigation increased fire impact; more developed canopies simply have a larger pre-fire spectral baseline, so the differenced response is mechanically larger even if absolute damage is not.

Seasonal timing compounds the ambiguity. The pre-fire composite window spanned 1 September to 22 October and the post-fire window 10 November to 15 December, a period when vine canopies are senescing and autumn rains can green up annual grasslands. Raw dNBR therefore mixes fire-related change with phenological change, and the two land covers do not follow the same seasonal trajectory. The outside-perimeter vineyard mean dNBR of 0.088 is consistent with a substantial seasonal contribution to the inside-vineyard signal. An independent Landsat 8 check confirmed only that the landscape-average response magnitude was similar between sensors, with 5-kilometer export means of 0.086 versus 0.082; it did not validate a land-cover-specific effect. The authors also note that parcel-specific harvest dates, crop moisture, and cover-crop management were unavailable, so no assumption was made that harvest or irrigation reduced fire spread.

Beyond direct burning, the study documents exposures that a burn-severity map alone would miss. NOAA Hazard Mapping System smoke polygons overlapped the study area on 9 of 19 days, and every one of the 4,581 vineyard fields intersected a smoke polygon on at least one day, with a mean of 7.78 potential smoke-days per field and a cumulative 69,101.6 vineyard hectare-smoke-days. The authors are careful to state that these are potential overhead exposures, not ground-level PM2.5 measurements or evidence of grape smoke taint, which requires chemical or sensory assessment of volatile phenols and their glycosides. The correct operational use of such maps, they argue, is to prioritize monitoring and sampling during agriculturally critical periods, not to classify crop loss.

Road access told a similarly two-sided story. The ignition-date OpenStreetMap drive network contained 4,348 nodes, 10,036 directed edges, and 1,581.9 kilometers of road, of which 321.6 kilometers were major roads. The median vineyard centroid lay 404.4 meters from the nearest network node and 719.5 meters from a major road, indicating broad general access. Yet 34.2 percent of nodes, 1,487 in total, were dead ends, a substantial share of network endpoints that points to potential topological fragility. The authors emphasize that topology alone cannot reveal realized evacuation performance, which depends on departure timing, warnings, route choice, and traffic conditions, but it can identify locations where redundancy deserves examination by transport and emergency managers.

Recovery provided the clearest longer-duration signal. The 2019 pre-fire growing-season NDVI baseline was statistically similar for fields inside and outside the perimeter, 0.4282 versus 0.4327. By 2020, recovery ratios relative to that baseline were 0.8918 inside versus 0.9096 outside, and by 2021 they had fallen to 0.8145 versus 0.8543, widening the inside-minus-outside deficit from −0.0178 to −0.0397. Both groups declined during the regional 2020–2021 drought, so the result is an unadjusted differential trajectory rather than proof that the fire alone caused the full decline. Still, within the perimeter, dNBR and the 2019–2021 NDVI change were associated, with a Spearman rho of −0.52, and the persistent greenness deficit offers a target for agronomic assessment. The authors caution that greenness is not vine survival, yield, fruit quality, or economic recovery.

The study also surfaced a critical data gap: no dedicated 2019 airborne lidar collection exists for Sonoma County. Fourteen airborne point-cloud datasets intersect the area between 2003 and 2024, including a 2013 Sonoma acquisition, and GEDI spaceborne lidar contributed 45 monthly images and 87,601 quality-screened height pixels within the perimeter, but neither supports defensible parcel-level estimates of event-day vineyard canopy or inter-row fuels. The authors frame their contribution as a reproducible urban–rural informatics template that separates descriptive contrasts, conditional associations, exposure indicators, and recovery evidence, each tied to an explicit claim boundary. For planners, the practical payoff is a screening workflow: identify vineyard-wildland edges needing field inspection, direct smoke sampling to fields with repeated plume overlap, flag road-redundancy gaps, and specify where event-day fuel data are absent. What the evidence does not justify, they conclude, is converting vineyards into firebreaks on the strength of a satellite map.

Subject of Research: Remote sensing and geospatial analysis of vineyard wildfire impacts, smoke exposure, road access, and post-fire recovery during the 2019 Kincade Fire in Sonoma County, California

Article Title: Multisource remote sensing and geospatial analysis of vineyard wildfire impacts and exposure: the 2019 Kincade Fire

Article References: Farajpoor, P., Pour, M. A., Ghiasi, M. B., & Narimani, M. (2026). Multisource remote sensing and geospatial analysis of vineyard wildfire impacts and exposure: the 2019 Kincade Fire. Frontiers in Sustainability, 7, 1997548. https://doi.org/10.3389/frsus.2026.1997548

Image Credits: AI Generated

DOI: 10.3389/frsus.2026.1997548

Keywords: Kincade Fire, vineyards, wildfire, remote sensing, dNBR, burn severity, Sentinel-2, OpenET, smoke exposure, post-fire recovery, Sonoma County, spatial analysis