Next Day Wildfire Spread Dataset

Home » Dataset Download » Next Day Wildfire Spread Dataset

Next Day Wildfire Spread Dataset

Datasets

File

Next Day Wildfire Spread Dataset

Use Case

Next Day Wildfire Spread Dataset

Description

Explore the Next Day Wildfire Spread dataset, featuring nearly a decade of U.S. wildfire data. Utilize 2D fire masks, environmental variables, and meteorological data to develop machine learning models for accurate wildfire prediction with a one-day lead time.

Description:

Accurate prediction of wildfire spread is essential for effective land management and disaster response. To support these efforts, we introduce the Next Day Wildfire Spread dataset, a comprehensive and large-scale multivariate dataset encompassing nearly a decade of historical wildfire data gathered from remote sensing across the contiguous United States. This dataset differs from traditional wildfire datasets, which often rely solely on Earth observation satellite data. It incorporates a blend of 2D fire information with various explanatory features, such as topographical, vegetative, weather-related, and drought-related variables, all aligned spatially over 2D regions. The dataset offers a valuable resource for machine learning researchers, providing a robust benchmark for modeling and predicting wildfire spread with a one-day lead time.

Download Dataset

Data Overview

Spanning the years from 2012 to 2020, this dataset aggregates data from regions across the contiguous United States where wildfires occurred, resulting in a total of 18,445 samples. Each sample captures a 64 km x 64 km area at a 1 km resolution, representing the region’s fire activity at two consecutive time points. The dataset includes a ‘fire mask’, which differentiates between areas with active fire, areas without fire, and regions with uncertain data due to factors like cloud cover.

In addition to the fire masks for time t and time t + 1 day, the dataset is enriched with a variety of environmental and meteorological features crucial to wildfire prediction. These include elevation, wind direction and speed, minimum and maximum temperatures, humidity levels, precipitation, drought index, and vegetation indices (such as NDVI). Additional variables include the energy release component (ERC) and population density, which play significant roles in determining fire behavior and spread. All data is aggregated using Google Earth Engine (GEE), ensuring high spatial and temporal accuracy.

Features

Each sample in the dataset comprises the following:

Fire Mask (Previous Day): Fire activity at time t over the region.
Fire Mask (Next Day): Fire activity at time t + 1 day, showing the spread of the fire.
Topography: Elevation data over the region.
Wind Data: Wind direction and wind speed metrics.
Temperature: Minimum and maximum daily temperatures.
Humidity: Levels of atmospheric moisture in the region.
Precipitation: Daily precipitation levels.
Drought Index: A measurement of drought severity in the region.
NDVI (Normalized Difference Vegetation Index): An indicator of vegetation health.
ERC (Energy Release Component): A measure of the available energy from burning vegetation.
Population Density: The number of people living within the affected area.