Research on Southern California wildfires.
-
Layer Interactions: Formalize bidirectional feedback between layers using coupled dynamical systems: $$ \begin{aligned} \frac{\mathrm{d}E_{h}}{\mathrm{d}t}&=f_h(H, W)\ W_r&=g_r(E, H)\ P(x,y,t)&=\exp\left(\beta_0 + \beta_1 H(x,t) + \beta_2 E(x,t) + \phi(x,t)\right) \end{aligned} $$
- Human Activities (H): Migration density, energy consumption (e.g., per capita kWh), land-use policies
- Environmental Factors (E): Temperature, humidity, NDVI (vegetation health), soil moisture
-
Wildfire Dynamics (W): Ignition probability, fire spread rate, burn severity
- Wildfire probability of Spatio-Temporal Bayesian Hierarchical Model where
$\phi(x,t)$ is a spatial random effect (e.g., Gaussian process) - Wildfire probability of pair interpretable model: Cox Proportional Hazards Model with ML (Random Forests) for non-linearities (e.g., humidity thresholds for ignition)
- Wildfire probability of Spatio-Temporal Bayesian Hierarchical Model where
-
Spatio-Temporal Granularity: data over 30m-1km resolution 2D grids (potentially 3D in future) and time to capture microclimates
-
Layer Feedback Loops: system dynamics with stock-flow diagram analysis
- Human Layer
- Environmental Layer
- EU Climate Data Store
- NASA USGS - Soil Moisture Active Passive
- North America CMIP6 climate projections
- U.S. Climate Projections - Mean Projections
- Program for Climate Model Diagnosis & Intercomparison
- Access to next generation climate data
- How well have CMIP3, CMIP5 and CMIP6 future climate projections portrayed the recently observed warming
- Summary for Policymakers
- CMIP6: the next generation of climate models explained
- Wildfire Layer
- Grid Data
-
Wildfire mediated impacts on human activity
- Drinking Water: burn severity correlating with watershed sediment load via HEC-HMS hydrologic models
- Energy Prices: infrastructure damage (e.g., power lines) with input-output networks; correlate with PG&E price volatility data, additional data
- Land Use: post-fire rezoning via cellular automata (e.g., transition from residential to restricted zone)
-
Policy Suggestion
- WUI Building Restrictions: reduce ignition sources (by 40%?)
- Renewable Energy Adoption: simulate decarbonization altering regional temperature trends via WRF-Chem climate simulations
- Rank drivers (e.g., humidity vs. migration) via Sobol Sensitivity Analysis
- Build Environmental Layer using Python's xarray (climate data), Google Earth Engine (NDVI/soil moisture), etc.
- Develop the Bayesian model for probabilistic inference in Stan, PyMC, etc.
- Propagate climate/model uncertainty into risk maps via Monte Carlo Ensemble
- Couple with system dynamics for policy testing in Vensim, AnyLogic, etc.
- Environmental Research Letters (focus on impact of fires on water/energy)
- Spatial Statistics (focus on model methodology)
- Nature Sustainability (focus on policy implications)