Assessing forest fire vulnerability using artificial neural networks in Almora district, Uttarakhand, India

Forests are vital to terrestrial ecosystems and they offer essential services for climate regulation and human welfare. However, the increasing trend in forest fires poses a significant threat to these ecosystems. This study aims to map and assess forest fire vulnerability zones within Almora district, Uttarakhand, India, using geospatial technologies and the Artificial Neural Network (ANN) technique. Twelve environmental indicators related to forest fire vulnerability, including elevation, slope, land use/land cover (LULC), Normalized Difference Vegetation Index (NDVI), Vegetation Health Index (VHI), temperature, precipitation, humidity, wind speed, Land Surface Temperature (LST), and distance from settlements and roads, were considered. The study revealed that a strip running from northern to southern Almora, including Someshwar, Dwarahat, and Ranikhet, is highly vulnerable to forest fires. This region is characterized by moderate to high elevation, a moderate to steep slope, and well-connected roads and settlements, particularly in Dwarahat and Ranikhet tehsils. The central and southern parts of Almora also exhibit good road connectivity, dense human settlements, and receive moderate to low precipitation, all of which contribute to a higher fire risk. In contrast, the eastern and western parts of Almora, comprising northern Sult, northern Bhikiyasain, and Banoli tehsils, are significantly less vulnerable to forest fires. These areas have moderate slopes, low to moderate elevation, higher precipitation in the eastern parts, and lower precipitation in the western parts, making them comparatively less prone to fire incidents. Validation through the Receiver Operating Characteristic (ROC) curve confirmed the accuracy of the model, with an 82% area under the curve.