In a world affected by climate change, wildfires are becoming a greater menace. Now, scientists at Aalto University have created a neural network model that can precisely forecast when peatland fires start. Using the new model to measure the effects of different fire risk management measures, they came up with a set of steps that will cut the number of fires by 50–76%.
The study concentrated on Indonesia’s Central Kalimantan area of Borneo, which has Southeast Asia’s greatest density of peatland fires. Peatlands are more susceptible to fires as a result of drainage for agricultural or residential growth. Peatland fires not only endanger lives and livelihoods but also emit a large amount of carbon dioxide. But it’s hard to prevent fires when there aren’t clear, measurable links between suggested treatments and fire risk.
The new model forecasts the location of peatland fires based on observations made before each fire season from 2002 through 2019. While the results are generally applicable to different peatlands, various situations would require a fresh investigation. Alexander Horton, who did the work as a postdoctoral researcher, says that our methods can be used in different situations, but that this model would need to be retrained with the new data.
The researchers utilized convolutional neural networks to analyze 31 factors, including the kind of land cover, drought and vegetation pre-fire indices. After it was trained, the network figured out how likely a peatland fire would be at each spot on the map, making an estimate of how many fires would happen that year.
Predictions made by the neural network were accurate 80–95% of the time overall. Although the model typically correctly predicted a fire, it also failed to anticipate numerous fires that actually broke out. The model was only able to forecast around half of the recorded fires, which means that it cannot be used as an early-warning predictive system. While solitary fires were frequently missed by the network, larger groups of fires tended to be reliably forecast. Researchers want to improve the network so that it can be used as an early warning system in the future.
The team used the fact that fire predictions were typically accurate to examine the impact of various land management techniques. The most effective plausible option, according to their simulations of various interventions, would be to turn scrubland and shrubland into swamp forests, which would cut the likelihood of fire by 50%. Fires would be reduced overall by 70% if this was done in addition to sealing all drainage canals aside from the big ones.
Such a course of action, meanwhile, would have glaring economic disadvantages. Horton says that the neighborhood’s economy needs long-term, steady farming more than anything else.
Adding more plantations would be a different course of action, as well-managed plantations greatly lower the risk of fire. Nevertheless, plantations are one of the main causes of forest loss. They are generally owned by larger firms, generally situated outside Borneo, which means the earnings aren’t immediately funneled back into the local economy beyond the provision of labor for the local workers.
According to Professor Matti Kummu, who oversaw the study team, fire protection techniques ultimately need to strike a balance between risks, benefits, and costs. This research gives researchers the knowledge they need to do so. We made an effort to quantify the effectiveness of the various tactics. More emphasis is placed on educating policymakers than on offering specific answers.
