Forecasting fire using satellite imagery, GIS and Machine Learning

Forecasting fire using satellite imagery, GIS and Machine Learning

SHARE

Predicting and forecasting fire, its occurrence and intensity both is crucial not only for safety considerations and preserving loss of property but also in understanding the complex phenomenon of climate change. In this scenario, satellites have emerged as an important resource in monitoring fires and providing accurate information. Using information fetched from satellites, we can create models that can help predict fire more efficiently and help in damage control too.Let’s look at the ways through which fire forecasting is done sucessfully.

Global Fire Weather Database

NASA has developed a Global Fire Weather Database (GFWED) that offers wind, temperature and humidity data that can be used along with high-resolution satellite imagery or any remote sensing software to predict the location of a flare-up. The model provides a unique fire score that pinpoints the area which is more prone to catching fire. While determining the likelihood and predicting the intensity of fire, it is important to factor wind speed as well, as wind fans fire. The NASA model combines different natural factors that contribute to the spread of fire.

GFWED combines meteorological data from several sources. Temperature, relative humidity, and wind speeds come from NASA’s MERRA-2 dataset of the Global Modeling and Assimilation Office (GMAO). Precipitation data come from ground-based rain gauges and from the Integrated Multi-Satellite Retrievals (IMERG), a product of the Global Precipitation Measurement mission. Using the GMAO weather forecasts, GFWED also includes experimental 8–day global forecasts of fire danger.

GFWED was created by Robert Field, a climate scientist at NASA’s Goddard Institute for Space Studies

Field has said that the model has been of immense help in Indonesia, which had an intense fire season during El Niño years.  Weather stations with rain gauges in Indonesia’s fire-prone regions can be sparse, so the satellite data helps fill in the gaps for the region. As a result, the model is able to provide a more accurate picture of potential fire danger, adds Fields.

Emerging technologies in forecasting fire

Machine Learning algorithms also have the capability to predict wildfires using Moderate Resolution Imaging Spectroradiometer (MODIS) data. Though Machine Learning has its own set of challenges and the patterns of land cover and topography are important in making an accurate assessment.

There are also methods that take into account both the human and natural factors in risk prediction. For Example, MODIS data uses methods to evaluate different regions. In the landlocked southern African nation Swaziland GIS-based methods were used to evaluate regional factors and it came to notice that land conditions were responsible for fire. The model had an accuracy of more than 90%.

There are also methods that take into account both the human and natural factors in risk prediction. For Example, MODIS data uses methods to evaluate different regions. In the landlocked southern African nation Swaziland GIS-based methods were used to evaluate regional factors and it came to notice that land conditions were responsible for fire. The model had an accuracy of more than 90%. Satellite data can also be used for correlating with past data and tracking historical patterns.

Landsat Data

Fire prediction using Landsat. Image Courtesy: Science Direct

Fire can also be predicted using Landsat data and a long-term forecast model can be built. With the help of past burning records, fire events, data on slope, aspect and weather conditions can be used to create forecasting models. Observation of historical data trends helps us understand the occurrence of fire in an area and coupled with the latest insights, it becomes a powerful tool.

Landsat is not deployed usually in monitoring short-term fires, but it is vastly useful in mapping prolonged fires in inaccessible terrains. It provides a fair indication of vegetation regrowth. Integrating this field information with the amount of biomass and ground cover with Landsat data helps in creating fuel maps that help in determining risky areas.

There are also other methods that have merged analytics, GIS, remote sensing and insights of local environmentalists for a better grasp of fires. It has been observed that in a large number of cases, human activity is behind fire. The first step lies in understanding the right conditions, relating them to human factors and then improvising for the most accurate results.