Scientists use AI to update plant data maps to improve wildfire predictions

A new technology developed by the National Center for Atmospheric Research (NCAR) uses artificial intelligence to efficiently update maps of vegetation that computer wildfire models rely on to accurately predict the behavior and spread of fires.

In a recent study, scientists demonstrated the use of Colorado’s 2020 East Troubleome Fire, which has burned through land incorrectly described in fuel stocks as healthy forest. In fact, the fire, which grew explosively, has burned a landscape devastated by pine beetles and wind gusts recently, leaving large areas of dead and fallen wood.

The research team compared a simulation of a fire generated by the latest wildfire behavior model developed at NCAR using standard area fuel stocks and one that has been updated with artificial intelligence (AI). Simulations that used the AI-modified fuels did a much better job of predicting the area the fires burned down, which eventually grew to more than 190,000 acres of land on either side of the Continental Divide.

“One of the main challenges in modeling wildfires was getting accurate inputs, including fuel data,” said NCAR scientist and lead author Amy DeCastro. “In this study, we demonstrate that the combined use of machine learning and satellite imagery provides a viable solution.”

The research was funded by the US National Science Foundation, which is NCAR’s sponsor. Modeling simulations were run at the NCAR-Wyoming Supercomputing Center on the Cheyenne System.

Using satellites to calculate pine beetle damage

In order for the model to accurately simulate wildfires, it requires detailed information about current conditions. This includes the local weather and terrain as well as the characteristics of the plant material that provides fuel for the flames – what is actually available to burn and in what condition. Is he alive or dead or not? Is it wet or dry? What type of vegetation is this? How many are there? How deep are the fuel layers on Earth?

The gold standard for fuel data sets is produced by LANDFIRE, a federal program that produces a number of geospatial data sets including wildfire fuel information. The process of creating these wildfire fuel data sets is extensive and includes satellite imagery, landscape simulations, and information collected in person during surveys. However, the amount of resources needed to produce it means, in practice, that it cannot be updated frequently, and that disturbance events in the forest—including wildfires, insect infestations, and development—could drastically alter the fuel available in the meantime.

In the case of the troublesome East Fire, which began in Grand County, Colorado, and burned east in Rocky Mountain National Park, the most recent LANDFIRE fuel dataset was released in 2016. In the intervening four years, pine beetles have caused widespread tree deaths in the area.

To update the fuel data set, the researchers turned to the Sentinel satellites, which are part of the European Space Agency’s Copernicus programme. Sentinel-1 provides information about the texture of the surface, which can be used to determine the type of vegetation. (For example, grass has a very different texture than trees.) Sentinel-2 provides information that can be used to infer a plant’s health from greenery. The scientists fed satellite data into a machine learning model known as “random jungle” that they trained on the US Forest Service’s insect and disease detection survey. The survey is conducted annually by trained personnel who estimate tree mortality from the air.

The result was that the machine learning model was able to accurately update the LANDFIRE fuel data, converting the majority of fuels classified as “timber waste” or “minimal timber” into “cut blast,” the designation used for woodland areas with a heavy tree rate. deaths. “LANDFIRE’s data is extremely valuable and provides a reliable platform to build upon,” DeCastro said. “Artificial intelligence has proven to be an effective tool for updating data in a less resource intensive way.”

In a position to make a positive impact

To test the effect of updated fuel stocks on the wildfire simulation, the scientists used a version of NCAR’s weather research and forecasting model, known as WRF-Fire, which was developed specifically to simulate the behavior of wildfires.

When WRF-Fire was used to simulate the troublesome East Fire using the unadjusted LANDFIRE fuel data set, it did not significantly predict how much area the fire would burn. When the model was run again with the adjusted dataset, it was able to predict the burned area with a greater degree of accuracy, indicating that dead and fallen wood helped fuel the spread of the fire more than if the trees were still alive.

Currently, the machine learning model is built to update an existing fuel map, and it can do the job quickly (within minutes). But the project’s success also shows the promise of using a machine learning system to start producing and regularly updating fuel maps from scratch in large areas at risk of wildfires.

The new NCAR research is part of a larger trend to investigate potential AI applications for wildfires, including efforts to use AI to estimate the perimeter of fires more quickly. NCAR researchers also hope that machine learning will be able to help solve other ongoing challenges to modeling wildfire behaviour. For example, machine learning might be able to improve our ability to predict the properties of embers generated by a fire (how big, how hot, and how intense they are) as well as the likelihood that embers will cause spot fires.

“We have a lot of technology, computing power, and a lot of resources at our fingertips to solve these problems and keep people safe,” said Timothy Giuliano, a scientist at NCAR, a co-author of the study. “We are in a good position to make a positive impact; we just need to keep working on it.”

The search was published in Remote Sensing.