AI-based prediction of Alaskan natural fires and its application into NCAR community land model 5 - biochemistry

Wildfires in boreal forests release large quantities of carbon from ecosystem to atmosphere. However, land surface models are still limited in representing the burned area, thus simulating fire impacts on the land surface processes. Our research objectives are to develop the machine learning tool of predicting natural wildfires and to apply it into the land surface model to understand the fire impacts on carbon fluxes. In this study, we developed the new method of predicting burned area with Long Short-Term Memory (LSTM) and Artificial Neural Network (ANN). For this study, fire database (i.e., location of wildfire, start date, end date and total burned area) of Alaska Interagency Coordination Center (AICC) from 2016 to 2020, lightning datasets, fire government policy datasets and daily scaled climate and vegetation datasets from ERA5 were collected. Using them, we tried to find out the land surface conditions of fire ignition and the duration of fire with LSTM. Then, the total burned area is estimated with ANN using the fire duration and climate datasets. The performance of capturing the large fires (>10000ha) improved, comparing the burned area from NCAR community land model 5 - biogeochemistry (CLM5-BGC). (Correlation: 0.79). This method is useful to capture the date of burning as well as total burned area. We then applied the daily-scaled burned area from machine learning (LSTM+ANN) into NCAR CLM5-BGC. The total net ecosystem exchange (NEE) value with machine learning increased by more than 2 times of NEE value of only CLM5-BGC, which could largely affect the co2 level in the atmosphere. We suggested that our fire system could improve the future fire carbon emission and carbon fluxes.

Acknowledgement

This work was supported by the Basic Science Research Program through the National Research Foundation of Korea, which was funded by the Ministry of Science, ICT & Future Planning (grant no. 2020R1A2C2007670) and the Korea Polar Research Institute (KOPRI, PE22900) funded by the Ministry of Oceans and Fisheries.