Deep cognitive imaging systems enable estimation of continental-scale fire incidence from climate data

Area of Science:

• Ecology
• Climate Science
• Artificial Intelligence

Background:

• Wildfires significantly impact terrestrial ecosystems and cause substantial losses.
• Climate change is projected to alter global fire regimes.
• Accurate prediction of fire incidence is crucial for mitigation and management.

Purpose of the Study:

• To develop a deep neural network (DNN) for accurately estimating continental fire incidence.
• To assess the relationship between climate data and wildfire occurrence.
• To evaluate the efficacy of DNNs in predicting fire incidence at a landscape scale.

Main Methods:

• Compared ten artificial neural networks (ANNs) for predicting fire incidence.
• Utilized a deep recurrent Elman neural network as the best-performing model.
• Employed monthly climatic data surfaces for a decennium data experiment.

Main Results:

• The Elman neural network demonstrated superior performance in modeling fire-climate relationships.
• Highly accurate estimations of fire incidence were achieved using monthly climate data.
• The model achieved over 90% global accuracy for Australia with a low false negative rate.

Conclusions:

• Deep neural networks, particularly the Elman network, are effective tools for estimating wildfire incidence.
• This technique can accurately predict the spatial consequences of climate scenarios on wildfire occurrence.
• The developed method is suitable for landscape-scale fire management and climate change impact analysis.