Dynamics Thresholding for Assessment of Land Fires Susceptibility, Brantas Hulu

Abstract

Abstract Forest fires are often an environmental problem that has the potential to cause environmental damage. It is predicted that as temperatures increase due to climate change, the risk of land and forest fires will become greater. This research aims to (1) model the vulnerability of land and forests to fire and (2) examine inherent land factors such as the type and physical properties of soil and land-use typology against fire. This research is quantitative, combining remote sensing data and field observations. Machine learning algorithms and dynamic thresholding will be tools for data analysis. Model verification will be carried out using national fire disaster data as well as field observations. The Brantas Hulu watershed has a high probability distribution of vulnerability. The training data Area Under the Curve (AUC) value is 0.923, and the test data The training data Area Under the Curve (AUC) value is 0.923 value is 0.912. The variables used provide different contribution values. Land Cover 2019 has the highest contribution to the model, namely 26.5%, and the lowest contribution to the model, namely Evapotranspiration 2023, Normalized Burn Ratio (NBR) 2023, and Normalized Difference Vegetation Index (NDVI) 2023, namely 0.1%. Forest and land fires can occur due to triggers from human activities. It is necessary to explain to the public not to burn grass and debris. Avoid burning when the weather is windy. Strong winds are the main factor in forest fires becoming more widespread.