Plant-water sensitivity regulates wildfire vulnerability

Wildfire-burned area has increased rapidly in the western United States over recent decades, threatening human populations and altering ecosystems. Contributing factors include historic fire suppression increasing fuel loads, expansion of the wildland-urban interface (WUI) with frequent human ignitions, and climate-driven increases in atmospheric aridity that reduce fuel moisture. However, the sensitivity of burned area to atmospheric aridity varies geographically, implying a role for local ecophysiological and hydrological controls. Live fuel moisture content (LFMC) generally decreases as atmospheric aridity increases, but its response is mediated by soil water availability, rooting depth, plant hydraulic traits, and species composition, which can drive large variations in LFMC and hence flammability under similar meteorology. To quantify the vegetation and soil control on LFMC’s response to climate, the authors define plant-water sensitivity (PWS) as the integrated sensitivity of LFMC to a climate-derived moisture balance that accounts for both precipitation and VPD. They hypothesize that higher PWS regions experience larger increases in burned area per unit rise in VPD due to greater LFMC declines and higher flammability, and test whether spatial variation in PWS regulates the effect of atmospheric aridity on burned area and human exposure, including interactions with trends in VPD and WUI population growth.

The study builds on extensive literature documenting rising wildfire area in the western US and its links to climate variability and change, fire suppression legacies, and WUI expansion. Prior work shows anthropogenic climate change and natural variability have increased atmospheric aridity (VPD), contributing to declines in fuel moisture and increased burned area, though sensitivities vary by region and ecosystem. Research has established that LFMC is influenced by meteorological drought indices but also by plant and soil hydraulic traits, which can cause substantial variation in LFMC and flammability across species and communities. Global and regional fire models often parameterize vegetation effects using plant functional types, yet hydraulic traits vary within these types, potentially limiting model fidelity. Studies also highlight the role of antecedent precipitation in modulating fuel availability, particularly in arid ecosystems, and the influence of human ignitions in the WUI. This work synthesizes these threads by empirically linking a physiologically grounded metric (PWS) to interannual burned area sensitivity to VPD and evaluating its interaction with spatial VPD trends and WUI population growth.

• Definition and computation of PWS: PWS quantifies the sensitivity of LFMC anomalies to a climate-derived moisture balance represented by 100-hour dead fuel moisture content (DFMC), which integrates precipitation, temperature, humidity, and daylight into an index of atmospheric wetness/drying potential. For each pixel, LFMC anomalies are regressed against lagged DFMC anomalies at lags from 0 to 150 days in 15-day steps, using a constrained multiple linear regression with non-negative coefficients to ensure physically consistent responses. PWS is calculated as the unweighted sum of the non-negative slopes across all lags, capturing effects of both concurrent and antecedent conditions. Anomalies are computed by removing pixel-specific seasonal cycles. LFMC maps are derived from microwave remote sensing (SAR-enhanced products).
• Burned area sensitivity to VPD: For 2001–2020, the slope d(burned area)/d(VPD) is computed between annual burned area and mean annual VPD (April–March). Analyses are conducted for bins of PWS that group spatially disparate but physiologically similar locations by equal vegetated area. Relationships are also examined by land cover type (forests, shrublands, grasslands).
• Circularity and confounder checks: To test potential circularity (since VPD influences DFMC), a modified PWS using VPD directly as predictor of LFMC is computed; its low correlation with the original PWS (R2 = 0.13) suggests minimal circularity. Correlations between PWS and possible confounders (mean/variance of VPD, mean/dry-season NDVI, dry-season length) are evaluated and found to be weak or insignificant. The sensitivity of fuel availability (NDVI) to antecedent precipitation (Dec–May) is assessed across land covers and along mean-precipitation gradients; no positive correlation with PWS is found.
• Drivers of PWS: A random forest model predicts spatial variability in PWS from 14 plant and soil hydraulic traits. Variable importance is quantified by average reduction in node impurity. Traits include saturated soil hydraulic conductivity (Ks), soil water retention curve shape parameter (n), root depth, soil texture fractions, canopy height, xylem capacitance, maximum xylem conductance, isohydricity, ψ50 (xylem water potential at 50% loss of conductance), and stomatal conductance slope parameter (gs), among others. Soil and plant trait datasets at continental/global scales are used.
• VPD trends and double-hazard mapping: Trends in VPD (1980–2020) are computed regionally. Joint distributions of PWS and VPD trends identify areas where both exceed their median values (“double-hazard” regions). Geographic mapping highlights hotspots (e.g., Sierra Nevada, eastern Oregon, Great Basin, Mogollon Rim).
• WUI population analysis: WUI spatial datasets (1990 and 2010) are combined with PWS hazard classes (low: PWS <1; medium: 1–1.5; high: >1.5) to quantify changes in WUI population by hazard class and relate them to increases in burned area per unit VPD rise.
• Data sources and products (as cited): LFMC from microwave remote sensing (SAR-enhanced), DFMC (100-hour) from fire danger calculations using gridded meteorology (e.g., GRIDMET), burned area from satellite products (e.g., MODIS Collection 5), land cover (NLCD), canopy height (spaceborne lidar), soil and plant hydraulic trait databases (global soil hydraulic properties, rooting depth, isohydricity), WUI maps (USDA), and PRISM/GRIDMET climate datasets. Statistical analyses use standard linear regression and scikit-learn random forests.

• Burned area sensitivity to VPD strongly increases with plant-water sensitivity (PWS): across 15 equal-vegetated-area PWS bins, R2 = 0.71, P < 0.0001. The slope d(burned area)/d(VPD) ranges from ~350 to ~700 km² hPa−1 depending on PWS.
• Ecosystem-specific robustness: The PWS–d(burned area)/d(VPD) link is strong in shrublands (R2 = 0.64, P = 0.005) and forests (R2 = 0.69, P = 0.003), but weak/insignificant in grasslands (R2 = 0.17, P = 0.24), where burned area shows low sensitivity to VPD.
• Hazard categorization by PWS: Low hazard for PWS < 1 (roughly constant d(BA)/d(VPD)); medium hazard for PWS 1–1.5 (moderate increase); high hazard for PWS > 1.5 (sharp increase).
• Robustness checks: PWS shows no strong correlation with mean/variance of VPD, mean or dry-season NDVI, or dry-season length. Modified PWS using VPD alone poorly correlates with original PWS (R2 = 0.13), suggesting minimal circularity. No significant link between PWS and NDVI sensitivity to antecedent precipitation across land covers or precipitation gradients.
• Drivers of PWS: A random forest explains 58% of PWS variance using 14 plant/soil hydraulic traits. Top contributors: saturated soil hydraulic conductivity (20% importance), soil water retention curve shape parameter n (13%), and root depth (9%). Soil traits contribute 55% and plant traits 45% of the explained importance.
• Co-occurrence of high PWS and rising aridity: From 1980–2020, VPD increased across 91% of the western US (mean trend 0.05 hPa yr−1). PWS tends to be higher where VPD increased more rapidly; about 28% of the region exhibits both above-median PWS and above-median VPD trends (“double-hazard” zones), concentrated in areas such as the Sierra Nevada (CA), eastern Oregon, Great Basin (NV), and Mogollon Rim (AZ).
• WUI population dynamics: WUI population approximately doubled from 10.0 million (1990) to 20.8 million (2010), a 108% increase overall. Growth was fastest in high-hazard regions (160%), compared to 107% in low-hazard and 95% in medium-hazard regions. High-hazard WUI zones also experienced the largest increases in percentage burned area per unit VPD rise (relative to 2001).
• Overall, ecosystems with high PWS show amplified wildfire vulnerability under increasing VPD, and demographic shifts have disproportionately increased exposure in these vulnerable areas.

Findings support the hypothesis that vegetation water sensitivity regulates how atmospheric aridity translates into burned area. In high-PWS ecosystems—characterized by traits such as relatively open stomata under stress, shallow roots, or fast-draining soils—LFMC declines more for a given moisture balance deficit, increasing flammability and fire spread, and thus burned area per unit VPD rise. The nonlinear PWS–burned-area response likely reflects threshold-like relationships between LFMC and flammability. The strong control of PWS, and its linkage to measurable soil and plant hydraulic traits, argues for integrating ecophysiological controls into fire danger assessments and models, rather than relying solely on meteorology or coarse plant functional types. The geographic co-occurrence of high PWS with large VPD increases suggests vegetation distributions have amplified climate-change impacts on wildfire hazard. Moreover, WUI growth has been disproportionate in high-hazard regions, compounding human risk by aligning increased exposure with increased hazard. Grasslands appear to be governed more by fuel availability, ignitions, wind, and phenology than by VPD-driven LFMC variability at the temporal resolution analyzed, explaining the weak PWS influence there. Incorporating PWS into wildfire danger rating systems and global fire models could improve forecasts, situational awareness, and planning. The results likely generalize beyond the western US, given global variation in hydraulic traits and rising atmospheric aridity.

This study introduces and operationalizes plant-water sensitivity (PWS) as a scalable metric capturing how vegetation and soil hydraulics modulate LFMC responses to climate, and demonstrates that PWS regulates the sensitivity of burned area to atmospheric aridity across the western US. PWS, driven by both soil and plant hydraulic traits, explains much of the observed spatial variability in wildfire vulnerability to VPD, with strong effects in forests and shrublands. Hazard is amplified where high PWS co-occurs with rapid VPD increases, and human risk has been further elevated by disproportionate WUI population growth in high-hazard areas. Incorporating PWS and underlying ecophysiological controls into wildfire danger systems and fire models can improve prediction and risk management. Future research should refine PWS estimates at higher temporal resolution, better represent dead fuel moisture and fire behavior interactions, assess socio-economic dimensions of WUI expansion into high-PWS zones, and evaluate how vegetation acclimation and post-fire community shifts may alter PWS patterns under continued climate change.

• Grasslands: Burned area shows low sensitivity to VPD, and the 15-day LFMC temporal resolution may miss rapid moisture dynamics, limiting detection of PWS effects in grasslands.
• Unexplained variance: The overall PWS–d(burned area)/d(VPD) relationship retains ~29% unexplained variance, potentially due to dead fuel moisture responses, demographic shifts, ignitions, winds, phenology, and fire behavior not captured here.
• Potential circularity: Although DFMC incorporates VPD, a modified PWS using VPD alone correlates weakly (R2 = 0.13) with the original PWS, suggesting limited circularity; nonetheless, some dependence remains conceptually.
• Trait data resolution: Plant and soil trait datasets are coarser than PWS maps, potentially dampening detectable plant-trait influences relative to soil traits due to scale mismatches.
• Confounders and causality: While several confounders (mean/variance VPD, NDVI, dry-season length, antecedent precipitation effects) show weak or no correlation with PWS, other untested factors may contribute locally.
• Temporal acclimation: Drought acclimation and post-fire shifts in community composition could alter PWS over time, potentially changing hazard patterns beyond the study period.
• Publication data limitations: Exact timing and regional generalizability depend on available datasets (e.g., MODIS burned area, GRIDMET/PRISM climate), which have their own uncertainties.