The spatial anatomy of urban wildfire vulnerability: a spatially validated GeoAI framework reveals the roles of building density and vegetation moisture in structure loss during the 2025 Palisades Fire

Parastoo Farajpoor; Mohammadreza Narimani

physics.geo-phAug 23, 2026

The spatial anatomy of urban wildfire vulnerability: a spatially validated GeoAI framework reveals the roles of building density and vegetation moisture in structure loss during the 2025 Palisades Fire

Authors: Parastoo Farajpoor, Mohammadreza Narimani

Organizations: Department of Biological and Agricultural Engineering, University of California, Davis, Davis, CA, 95616, USA

Abstract

Urban wildfire resilience depends on interactions among built form, vegetation condition, and extreme fire weather, yet city-scale risk models often overlook whether predictive skill transfers across neighborhoods. We developed a spatially validated GeoAI workflow for the January 2025 Palisades Fire, linking 12,081 CAL FIRE damage inspections to pre-fire Sentinel-2 vegetation indices, Landsat surface temperature, LANDFIRE fuels, terrain, and OpenStreetMap buildings and roads. Among 9,883 inspected residential structures, 5,566 were destroyed. Random cross-validation yielded ROC-AUC 0.92 for the integrated XGBoost model, but 1 km spatial block validation reduced performance to 0.75; logistic regression performed similarly and was better calibrated. Building count within 100 m was the strongest predictor, with destruction odds increasing 4.12-fold per standard deviation. Vegetation moisture and greenness showed opposing conditional associations: NDMI at 100-300 m was protective (OR 0.52), whereas NDVI at 30-100 m was positively associated with destruction after accounting for moisture (OR 1.74). Predictive information was concentrated at the 100-300 m neighborhood scale. A separate post-fire track mapped burn severity and vegetation recovery without leakage. The results support neighborhood-scale susceptibility screening, moisture-aware vegetation management, and spatial block validation as a minimum standard for single-event urban wildfire modeling.

Figures & tables

Figure 1: Analytical workflow. Public data are harmonized to NAD83 / UTM zone 11N (EPSG:26911), quality controlled, and carried into two complementary analytical tracks. Track A uses only pre-fire predictors to model residential destruction, while Track B interprets event impact and community context without using post-fire indicators as predictors.

Figure 2: Study area and observed structure damage. Final Palisades Fire perimeter, 2-km study-area buffer, terrain and road context, and CAL FIRE DINS-inspected residential structures colored by original damage class. The inset locates the fire in California.

Figure 3: Antecedent and event fire-weather context. (A) October-December precipitation totals for 1980-2024 with the historical median and the exceptionally dry 2024 season highlighted; (B) 100-hour dead-fuel moisture and vapor-pressure deficit from April 2024 through January 2025; and (C) 10-m wind speed and relative humidity during the January 2025 event window. These variables summarize the broader climatic and event setting and were not used as structure-level predictors.

Dataset	Provider	Resolution	Temporal window used	Variables derived	Analytical role
CAL FIRE DINS	CAL FIRE / California Open Data	Point (structure)	Post-fire inspections, 2025	Damage class, structure category, year built	Primary outcome
WFIGS Interagency Fire Perimeters	NIFC	Polygon	Final 2025 perimeter	Fire perimeter geometry	Study area definition
Sentinel-2 L2A (via GEE)	Copernicus/ESA	10-20 m	Oct 2024-Jan 2025 (pre); Oct-Dec 2022-2024 (baseline); Jan-F	NDVI, NDMI, NBR/dNBR	Pre-fire vegetation predictors; severity & recovery
Landsat 8/9 C2 L2 (via GEE)	USGS	30 m	Oct 2024-Jan 2025	Land surface temperature	Pre-fire thermal predictor
LANDFIRE LF2024	USGS/USDA-FS	30 m	2024 update (pre-fire vintage)	FBFM40 fuel models, canopy cover	Fuel type/continuity predictors
USGS 3DEP DEM (via GEE)	USGS	10 m	Static	Elevation, slope, aspect, TPI, TRI	Terrain predictors

Table 4

Figure 4: Sentinel-2 visual context before, during, and after the fire. The top row shows true-color composites and the bottom row shows shortwave-infrared false-color composites for pre-fire, active-fire, and post-fire conditions. The active-fire panels retain smoke and shortwave-infrared hotspots.

Figure 5: Pre-fire environmental conditions. (A) Pre-fire NDVI; (B) pre-fire NDMI; (C) LANDFIRE fuel groups; and (D) Landsat land-surface temperature. These variables summarize the pre-fire vegetation and thermal setting used in the predictive track.

Figure 6: Post-fire environmental conditions. (A) Post-fire NDVI; (B) post-fire NDMI; (C) study-derived burn-severity classes; and (D) post-fire land-surface temperature. These are impact indicators used only in the post-fire interpretation track.

Figure 7: Built environment and structure arrangement. (A) Building count within 100 m; (B) assessor year built; and (C-E) distributions comparing nearest-building distance, local building density within 100 m, and construction era for destroyed versus not-destroyed residential structures.

Figure 8: Pre-fire predictor distributions by original DINS damage class. Violin-box-scatter panels summarize pre-fire NDVI, pre-fire NDMI, wildland-fuel fraction, slope, nearest-building distance, local building density within 100 m, year built, and distance to wildland fuels across the five DINS damage classes.

Figure 9: Spatial validation and calibration. (A) One-kilometer spatial cross-validation folds; (B) out-of-fold destruction probability; (C) ROC-AUC by model block and validation scheme for logistic regression and XGBoost; and (D) calibration curves.

Validation	Block	Learner	ROC-AUC	PR-AUC	Balanced accuracy	F1	Brier
Random	M0	Logistic	0.709 (0.01)	0.737 (0.014)	0.662 (0.01)	0.706 (0.011)	0.217 (0.003)
	M0	XGBoost	0.801 (0.004)	0.834 (0.009)	0.714 (0.006)	0.754 (0.006)	0.182 (0.001)
	M1	Logistic	0.718 (0.011)	0.744 (0.017)	0.673 (0.01)	0.72 (0.012)	0.212 (0.004)
	M1	XGBoost	0.891 (0.019)	0.913 (0.018)	0.804 (0.022)	0.827 (0.018)	0.135 (0.012)
	M2	Logistic	0.779 (0.011)	0.829 (0.009)	0.719 (0.01)	0.735 (0.009)	0.192 (0.004)
	M2	XGBoost	0.892 (0.009)	0.917 (0.007)	0.803 (0.007)	0.818 (0.008)	0.134 (0.006)

Table 11

Figure 10: Model interpretation. (A) SHAP summary for the twelve most influential XGBoost predictors; (B) SHAP dependence for building count within 100 m colored by nearest-neighbor distance; (C) standardized logistic-regression odds ratios with 95% confidence intervals; and (D) SHAP dependence for NDMI in the 100-300 m ring colored by NDVI in the 30-100 m ring.

Figure 11: Scale and block-size sensitivity. (A) Spatial cross-validation ROC-AUC by ring scale; (B) summed mean absolute SHAP contribution by ring; and (C) spatial-block-size sensitivity of ROC-AUC.

Figure 12: Fire impact and early vegetation response. (A) Study-derived dNBR; (B) dNBR within the 30-100 m ring by DINS damage class; and (C) monthly NDVI recovery by burn-severity class relative to month-matched 2022-2024 climatology from February 2025 through July 2026.

Figure 13: Community context. (A) Residential destruction fraction by census tract; (B) CDC/ATSDR SVI 2022 overall percentile by tract; and (C) tract-level SVI versus residential destruction. Point size represents the number of inspected structures and point color represents median network distance to the nearest fire station.

Building Damage Assessment

Building damage assessment uses aerial and satellite imagery to locate buildings and estimate disaster-related structural damage for response and recovery.

See all 25 papers

Multisource Remote Sensing and Geospatial Analysis of Vineyard Wildfire Impacts and Exposure: The 2019 Kincade Fire
Sep 14, 2026Parastoo Farajpoor, Mahla Ardebili Pour, Mohammad Bagher Ghiasi +1Disaster Damage Assessment
Department of Biological and Agricultural Engineering, University of California, Davis, Davis, CA, 95616, USA · Department of Civil and Environmental Engineering, University of California, Davis, Davis, CA, 95616, USA · Department of Electrical and Computer Engineering, University of California, Davis, Davis, CA, 95616, USA
WildfireSpreadBench: The Metric Decides the Model in Wildfire Spread Prediction
Aug 29, 2026Arin Gopakumar, Marco PannozzoWildfire Forecasting Forecasting Benchmarks
UC Berkeley · Purdue University
FireGraphNet: Sub-daily wildfire spread prediction with directed graph neural networks
Oct 13, 2026Aiyshwariya Paulvannan Kanmani, Miguel Esparza, Ali MostafaviWildfire Forecasting Spatiotemporal GNNs
Urban Resilience.AI Lab, Zachry Department of Civil and Environmental Engineering, Texas A&M University, College Station, TX 77840, United States of America

Sep 14, 2026J/K move · Enter open · S save

Multisource Remote Sensing and Geospatial Analysis of Vineyard Wildfire Impacts and Exposure: The 2019 Kincade Fire

Parastoo Farajpoor, Mahla Ardebili Pour, Mohammad Bagher Ghiasi, Mohammadreza Narimani

Department of Biological and Agricultural Engineering, University of California, Davis, Davis, CA, 95616, USA · Department of Civil and Environmental Engineering, University of California, Davis, Davis, CA, 95616, USA · Department of Electrical and Computer Engineering, University of California, Davis, Davis, CA, 95616, USA

Disaster Damage Assessment

Open paper AskPDF arXiv

Working agricultural landscapes are often treated as background to wildfire disasters, even though they are managed fuel mosaics, productive assets, and parts of regional infrastructure systems. We examine vineyard wildfire resilience during the electrically initiated 2019 Kincade Fire in Sonoma County, California, using an open, event-anchored geospatial framework spanning 4,581 vineyard fields (8,813.2 ha), wildland vegetation, surveyed structures, roads, overhead smoke, and post-fire greenness. Sentinel-2, OpenET, gridMET, soils, terrain, NOAA smoke polygons, an ignition-date OpenStreetMap network, and three-dimensional data inventories were analyzed at native decision scales. Vineyard pixels showed substantially lower descriptive dNBR than wildland pixels inside the perimeter (means 0.130 and 0.337). This contrast did not identify a universal vineyard firebreak effect: a segment-clustered boundary model gave a small negative contrast at 100 m (tau = -0.0166) but changed across bandwidths, failed slope continuity, disappeared in a 100 m donut, and produced a wrong-signed placebo. A 250 m spatial GAM reversed the unconditional pattern: after conditioning on location, terrain, and water use, vineyard fraction was positively associated with dNBR, while residual Moran's I remained 0.519. Beyond spectral impact, all mapped vineyards intersected overhead smoke on at least one day (mean 7.78 potential smoke-days per field), 34.2% of road-network nodes were dead ends, and inside-perimeter vineyards showed a larger greenness deficit through 2021 (recovery ratios 0.815 inside and 0.854 outside). Lower immediate spectral impact therefore did not imply complete resilience. The study provides a reproducible urban-rural informatics template separating descriptive contrasts, conditional associations, exposure indicators, and recovery evidence for decisions in working landscapes.

Abstract

Building Damage Assessment

Explore similar work

Multisource Remote Sensing and Geospatial Analysis of Vineyard Wildfire Impacts and Exposure: The 2019 Kincade Fire

WildfireSpreadBench: The Metric Decides the Model in Wildfire Spread Prediction

FireGraphNet: Sub-daily wildfire spread prediction with directed graph neural networks