Mapping the forest fire risk zones using artificial intelligence with risk factors data

Spatial prediction of forest fires in India: a machine learning approach for improved risk assessment and early warning systems

Forest fires pose a significant ecological and environmental threat globally, and India has seen a marked increase in both the frequency and severity of these events in recent years. This has led to extensive damage to natural resources, including forests and wildlife habitats. Effective management strategies are essential to mitigate these impacts, and this study aims to contribute to that effort through spatial prediction of forest fires in India using machine learning techniques. The research begins by analyzing spatial patterns and trends of forest fires and identifying key factors contributing to their occurrence. It then explores the relationships between forest fires and these factors. Using data from 2001 to 2020, the study develops a probability map of forest fire occurrences in India, leveraging data from the Moderate Resolution Imaging Spectroradiometer (MODIS) on fire points and burned areas and the European Space Agency Climate Change Initiative (ESA-CCI) land use land cover. The analysis incorporates various fire occurrence conditioning factors, including climate variables (precipitation, air temperature, specific humidity, near-surface wind speed, net longwave radiation, and land surface temperature for both daytime and nighttime), biophysical factors (aspect, slope, elevation, soil moisture, and Normalized Difference Vegetation Index), and other factors such as proximity to railways, roads, and waterways. The maximum entropy model (MaxEnt) was employed to identify these factors and generate the probability map. The results indicate that forest fires are predominantly concentrated in three major regions: Northeast India, Uttarakhand-Himachal, and the Deccan Plateau. The study also highlights the link between forest fires and Sustainable Development Goals (SDGs 1, 2, 3, 5, 6, 9, 12, 13, and 15). The findings can inform the identification of vulnerable areas, enabling the implementation of preventative measures. This research is crucial as it not only pinpoints regions at risk of forest fires but also elucidates their impacts on the environment and local livelihoods, thereby enhancing risk assessment and early warning systems.

Determination of forest fire risk with respect to Marchalina hellenica potential distribution to protect pine honey production sites in Turkey

Turkey is the leading producer of pine honey worldwide, accounting for 90% of global production, largely due to the presence of Marchalina hellenica populations. However, in recent years, devastating forest fires have caused substantial damage to Pinus brutia forests and M. hellenica populations, leading to a dramatic decline in pine honey production areas. The urgency for early intervention procedures against forest fires and relocation of M. hellenica populations to other P. brutia forests has become apparent. A comprehensive assessment of 25 criteria was conducted to investigate the thresholds and behaviors of each criterion, which play a vital role in the distribution of M. hellenica, using the maximum entropy model (MaxEnt). To evaluate the impact of forest fires on the distribution of M. hellenica, the potential locations of pine honey production areas were determined for 2022. Furthermore, the susceptibility of forest fires was modeled for all pine honey production months. The findings revealed that forest fires have destroyed 384.9 km2 (12.8% of the total pine honey production area), predominantly in August and September, with the most severe damage in Marmaris (156 km2) and significant impacts in Ula, Köyceğiz, and Milas. The analysis facilitates the estimation of new areas suitable for M. hellenica and pine honey production while providing valuable insights into strategies for mitigating forest fires and formulating proactive protection protocols.

Deep learning-based burned forest areas mapping via Sentinel-2 imagery: a comparative study

In order to evaluate the effects of forest fires on the dynamics of the function and structure of ecosystems, it is necessary to determine burned forest areas with high accuracy, effectively, economically, and practically using satellite images. Extraction of burned forest areas utilizing high-resolution satellite images and image classification algorithms and assessing the successfulness of varied classification algorithms has become a prominent research field. This study aims to indicate on the capability of the deep learning-based Stacked Autoencoders method for the burned forest areas mapping from Sentinel-2 satellite images. The Stacked Autoencoders, used in this study as an unsupervised learning method, were compared qualitatively and quantitatively with frequently used supervised learning algorithms (k-Nearest Neighbors (k-NN), Subspaced k-NN, Support Vector Machines, Random Forest, Bagged Decision Tree, Naive Bayes, Linear Discriminant Analysis) on two distinct burnt forest zones. By selecting burned forest zones with contrasting structural characteristics from one another, an objective assessment was achieved. Manually digitized burned areas from Sentinel-2 satellite images were utilized for accuracy assessment. For comparison, different classification performance and quality metrics (Overall Accuracy, Mean Squared Error, Correlation Coefficient, Structural Similarity Index Measure, Peak Signal-to-Noise Ratio, Universal Image Quality Index, and KAPPA metrics) were used. In addition, whether the Stacked Autoencoders method produces consistent results was examined through boxplots. In terms of both quantitative and qualitative analysis, the Stacked Autoencoders method showed the highest accuracy values.