Research
Advancing the frontiers of wildfire science, our lab pioneers physics-based models, data-driven tools, and AI-powered solutions to transform wildfire modeling, risk assessment, and resilience—bridging breakthrough research with real-world challenges.
Ongoing Projects
Transport and deposition patterns of wildfire-produced ash, firebrand, and smoke and their ecological significance
The objective of this work is to develop a mechanistic and unifying framework that captures the full lifecycle of airborne pyrogenic particles generated during wildfires, including their generation, lofting, atmospheric transport, and eventual deposition. By integrating physical modeling with collected field data, we aim to bridge the gap between theoretical predictions and real-world outcomes. Specifically, we compare model outputs with the mass and chemical composition of particles collected in the field at various distances from the 2021 Caldor Fire in California to assess the accuracy and applicability of the proposed framework. Furthermore, we examine the ecological implications of the deposition footprint, highlighting how these particles may influence ecosystem recovery in post-fire landscapes.
Team Members: Facundo Scordo, Majid Bavandpour, Dani Or, Hamed Ebrahimian, Sudeep Chandra
Probabilistic Wildfire Risk Assessment (PWRA): An efficient uncertainty propagation framework
Wildfires are becoming increasingly destructive, turning into a national crisis and highlighting the urgent need for more advanced risk assessment tools to aid in planning and preparedness. This work presents a new framework that redefines wildfire risk assessment by shifting from traditional single-point loss estimates to a probabilistic, uncertainty-aware approach. Unlike conventional “red-zone” maps, which offer limited insight, our method quantifies and propagates uncertainties throughout the modeling process. This enables decision-makers and communities to evaluate outcomes based on specific quantities of interest. By incorporating advanced analytical techniques such as the unscented transform, the framework also improves computational efficiency—reducing computational costs without compromising accuracy, making it highly scalable for large and complex regions.
Team Members: Majid Bavandpour, Hamed Ebrahimian, Dani Or
Data-Driven Fragility Models for Probabilistic Damage Assessment of WUI Structures
Increasing WUI fire conflagrations in recent wildfire incidents highlights the need for tools and methods for pre-fire structural-level damage prediction and risk assessment. We introduce an interpretable machine-learning-based fragility model as part of a modular probabilistic wildfire risk assessment framework to predict WUI structures damage probabilities. The machine-based model is built from multi-source geospatial data, integrating over 50,000 CAL FIRE Damage Inspection (DINS) records with weather, building footprints, NAIP imagery, and canopy-height products, and includes physics-based features to quantify direct flame-contact potentials, radiative heating from surroundings, and ember exposures. The model predicts the probability of structural damage at a structure-level. The predicted damage probabilities are designed to construct fragility functions, which are then used as the damage assessment module within a probabilistic wildfire risk assessment framework.
Team Members: Facundo Scordo, Majid Bavandpour, Dani Or, Hamed Ebrahimian, Sudeep Chandra, Chenzhi Ma
Deep learning for near real-time wildfire monitoring
The objective of this work is to enhance a previously-developed deep learning approach for wildfire monitoring by incorporating additional preprocessing steps aimed at improving data quality. These improvements focus on refining the input satellite data to reduce noise, correct artifacts, and better align spatial and temporal features between different satellite sources such as GOES-17 and VIIRS. By improving the quality of the input data, the deep learning models can more accurately estimate active fire locations and brightness temperatures. The goal is to increase both the spatial resolution and reliability of the outputs, ultimately leading to more accurate and timely wildfire detection, monitoring, and response capabilities using satellite data and deep learning methods.
Team Members: Ryota Yagi, Majid Bavandpour, Hamed Ebrahimian, Dani Or
Firebrand modeling
We are working to advance a physics‑based modeling framework for firebrand generation, transport, and ignition, to be implemented in coupled fire atmosphere platforms. Tree morphology is explicitly constructed to represent vegetation structure and is coupled with a mechanical failure model and combustion-driven fragmentation to simulate firebrand production. Firebrands that achieve sufficient buoyancy within flame‑scale plumes are lofted and become airborne particles, whose transport is resolved using an atmospheric flow model. Upon deposition, individual firebrands are incorporated into an ignition sub‑model to predict spot fire initiation. By explicitly linking vegetation structure, firebrand dynamics, and ignition processes, the proposed framework provides an integrated and physically consistent approach for simulating firebrand‑driven fire spread.
Team Members: Kasra Shamsaei, Majid Bavandpour, Hamed Ebrahimian, Dani Or, Fangjiao Ma
Recent Completed Projects
Review of wildfire modeling considering effects on land surfaces
This work introduces hydrologists, soil scientists, and ecologists to principles and advances in modeling of wildfire dynamics and rates of spread to improve understanding of capabilities and limitations offered by modern wildfire models. It highlights the common omission of wildfire effects on soil processes and emphasizes how wildfire models can help fill this gap by providing boundary conditions necessary for quantifying thermal impacts on soils. By enhancing cross-disciplinary collaboration, the review seeks to support more accurate representation of wildfire impacts on soil and hydrologic systems, as well as the recovery of fire-affected landscapes. It outlines the fundamental characteristics, processes, and metrics central to wildfire science and offers an overview of modeling approaches used for research and prescribed fire planning.
Team Members: Dani Or and Hamed Ebrahimian
Smoke from regional wildfires alters lake ecology
​We analyzed a 2018 event in which wildfire smoke covered Castle Lake, a low-productivity lake in California, for 55 days. Conditions during the event were compared to data from 2014-2017. Smoke reduced incident ultraviolet-B (UV-B) radiation by 31% and photosynthetically active radiation (PAR) by 11%, altering the light regime. Underwater, UV-B and PAR decreased by 65% and 44%, respectively, while the lake's heat content dropped by 7%. Despite this, primary production in shallow waters increased by 109%, likely due to a release from photoinhibition. In contrast, deep-water production decreased, and the chlorophyll-a peak in deep waters failed to develop, likely from reduced PAR. While zooplankton biomass and migration patterns showed little change, trout were absent from littoral-benthic habitats during the smoke period. These findings highlight the varied effects of smoke on lake ecosystems.
Team Members: Facundo Scordo and Sudeep Chandra
Wildfire smoke effects on lake-habitat specific metabolism: Toward a conceptual understanding
​This study evaluated how wildfire smoke alters lake metabolism by comparing ecological parameters in years with and without smoke exposure. We examined offshore and nearshore habitats of Castle Lake, a clear, low productivity lake in California, assessing the light regime, gross primary production (GPP), ecosystem respiration (ER), and net ecosystem production (NEP). During smoke cover, incident ultraviolet-B (UV-B) radiation and photosynthetically active radiation (PAR) decreased by 53% and 28%, respectively while water column extinction coefficients for UV-B and PAR increased by 20% and 18%, respectively. Offshore productivity increased during smoke periods, likely due to reduced solar radiation with sufficient PAR remaining. However, nearshore productivity remained unchanged, possibly due to the adaptation of these habitats to high-intensity UV-B radiation. The study highlights how wildfire smoke affects the energy balance of specific lake habitats.
Team Members: Facundo Scordo and Sudeep Chandra
Wildfire smoke reduces lake ecosystem metabolic rates unequally across a trophic gradient
This study quantified the impact of wildfire smoke on lakes, focusing on trends in smoke coverage and its effects on ecosystem functions such as gross primary production (GPP) and ecosystem respiration. Ten California lakes with varying water clarity and nutrient concentrations were accessed. During heavy smoke years (2018, 2020, 2021), lakes experienced 23 to 45 days of medium to high-density smoke, reducing shortwave radiation by 20% and increasing particulate matter five-fold. The results indicated that ecosystem respiration generally declined during periods of cover of smoke, with the most significant declines observed in low-nutrient, cold lakes. The response of GPP to smoke was more variable, depending on the specific attributes of each lake. The study concluded that both lake characteristics and the timing of wildfires are crucial in mediating the effects of smoke on freshwater ecosystems.
Team Members: Facundo Scordo and Sudeep Chandra
Deep learning to Improve spatial resolution of GOES-17 wildfire boundaries
Wildfires in the U.S. are becoming more frequent and severe, highlighting the need for better emergency response systems with high temporal and spatial monitoring. Satellites like GOES-17 offer high temporal (1–5 min) but low spatial (≥2 km) resolution, while VIIRS provides the opposite: low temporal (~12 h) but high spatial (375 m) resolution. This study uses deep learning (DL), specifically an Autoencoder model, to enhance GOES-17’s spatial resolution using VIIRS data. Multiple DL architectures and loss functions were tested on wildfires from 2019–2021 in the western U.S. Results show DL can generate GOES-17 images with VIIRS-like resolution, enabling high-resolution, near-real-time wildfire monitoring and semi-continuous fire progression tracking.
Team Members: Mukul Badhan, Kasra Shamsaei, Hamed Ebrahimian
Data-driven firebrand generation model
This study presents a data-driven firebrand generation model designed for integration into wildland fire simulation platforms to enable fire spotting prediction. The model was developed by compiling and synthesizing experimental data from the literature and consists of three components that estimate firebrand yield, mass distribution, and the relationship between mass and projected area. Model inputs include fuel type, consumed fuel mass, fuel moisture content, and wind speed; outputs consist of firebrand realizations with assigned mass and projected area. Validation against independent experimental datasets shows good agreement between predicted and observed firebrand mass and projected area distributions. The model is computationally efficient, requires only readily available input parameters from existing wildland fire models, and delivers reliable predictions, making it well suited for incorporation into current simulation platforms.
Team Members: Kasra Shamsaei, Hamed Ebrahimian, Dani Or
Numerical simulation of historic wildfires
Through a series of studies, we evaluate the performance of Weather Research and Forecast-Fire (WRF-Fire), a coupled fire-atmosphere wildland fire simulation platform, in simulating a large historic fire, including the 2018 Camp Fire. The simulation results are compared to high-temporal-resolution fire perimeters derived from NEXRAD radar observations and show non-negligible discrepancies between simulated fire boundary and the observations with regard to rate of spread (ROS) and spread direction. Next, the sensitivity of the model to a series of modeling parameters and assumptions governing the simulated wind field are investigated. Finally in an attempt to improve the heat flux presentation in the model, a canopy fuel model is added to the modeling platform and the heat release scheme into the atmosphere is corrected and improved.
Team Members: Kasra Shamsaei, Hamed Ebrahimian