Remotely Sensed Data Fusion for Spatiotemporal Geostatistical Analysis of Forest Fire Hazard

Improving the prediction of wildfire susceptibility on Hawai'i Island, Hawai'i, using explainable hybrid machine learning models

This study presents a comparative analysis of four Machine Learning (ML) models used to map wildfire susceptibility on Hawai'i Island, Hawai'i. Extreme Gradient Boosting (XGBoost) combined with three meta-heuristic algorithms - Whale Optimization (WOA), Black Widow Optimization (BWO), and Butterfly Optimization (BOA) - were employed to map areas susceptible to wildfire. To generate a wildfire inventory, 1408 wildfire points were identified within the study area from 2004 to 2022. The four ML models (XGBoost, WOA-XGBoost, BWO-XGBoost, and BOA-XGBoost) were run using 14 wildfire-conditioning factors categorized into four main groups: topographical, meteorological, vegetation, and anthropogenic. Six performance metrics - sensitivity, specificity, positive predictive values, negative predictive values, the Area Under the receiver operating characteristic Curve (AUC), and the average precision (AP) of Precision-Recall Curves (PRCs) - were used to compare the predictive performance of the ML models. The SHapley Additive exPlanations (SHAP) framework was also used to interpret the importance values of the 14 influential variables for the modeling of wildfire on Hawai'i Island using the four models. The results of the wildfire modeling indicated that all four models performed well, with the BWO-XGBoost model exhibiting a slightly higher prediction performance (AUC = 0.9269), followed by WOA-XGBoost (AUC = 0.9253), BOA-XGBoost (AUC = 0.9232), and XGBoost (AUC = 0.9164). SHAP analysis revealed that the distance from a road, annual temperature, and elevation were the most influential factors. The wildfire susceptibility maps generated in this study can be used by local authorities for wildfire management and fire suppression activity.

The adoption of climate-smart agriculture to address wildfires in the Maya Golden Landscape of Belize: Smallholder farmers' perceptions

Ecosystems around the globe are enduring wildfires with greater frequency, intensity, and severity and this trend is projected to continue as a result of climate change. Climate-smart agriculture (CSA) has been proposed as a strategy to prevent wildfires and mitigate climate change impacts; however, it remains poorly understood as a strategy to prevent wildfires. Therefore, the authors propose a multimethod approach that combines mapping of wildfire susceptibility and social surveys to identify priority areas, main factors influencing the adoption of CSA practices, barriers to their implementation, and the best CSA practices that can be implemented to mitigate wildfires in Belize's Maya Golden Landscape (MGL). Farmers ranked slash and mulch, crop diversification, and agroforestry as the main CSA practices that can be implemented to address wildfires caused by agriculture in the MGL. In order to reduce wildfire risk, these practices should, be implemented in agricultural areas near wildlands with high wildfire susceptibility and during the fire season (February-May), in the case of slash and mulch. However, socio-demographic and economic characteristics, together with a lack of training and extension services support, inadequate consultation by agencies, and limited financial resources, hinder the broader adoption of CSA practices in the MGL. Our research produced actionable and valuable information that can be used to design policies and programs to mitigate the impacts of climate change and wildfire risk in the MGL. This approach can also be used in other regions where wildfires are caused by agricultural practices to identify priority areas, barriers and suitable CSA practices that can be implemented to mitigate wildfires.

The Effectiveness of Pan-Sharpening Algorithms on Different Land Cover Types in GeoEye-1 Satellite Images

In recent years, the demand for very high geometric resolution satellite images has increased significantly. The pan-sharpening techniques, which are part of the data fusion techniques, enable the increase in the geometric resolution of multispectral images using panchromatic imagery of the same scene. However, it is not trivial to choose a suitable pan-sharpening algorithm: there are several, but none of these is universally recognized as the best for any type of sensor, in addition to the fact that they can provide different results with regard to the investigated scene. This article focuses on the latter aspect: analyzing pan-sharpening algorithms in relation to different land covers. A dataset of GeoEye-1 images is selected from which four study areas (frames) are extracted: one natural, one rural, one urban and one semi-urban. The type of study area is determined considering the quantity of vegetation included in it based on the normalized difference vegetation index (NDVI). Nine pan-sharpening methods are applied to each frame and the resulting pan-sharpened images are compared by means of spectral and spatial quality indicators. Multicriteria analysis permits to define the best performing method related to each specific area as well as the most suitable one, considering the co-presence of different land covers in the analyzed scene. Brovey transformation fast supplies the best results among the methods analyzed in this study.

Statistical analysis of active fire remote sensing data: examples from South Asia

Active fires emit aerosols and greenhouse gases in the atmosphere. In this paper, the behavior of active fires over a period of 1 year in Nepal, Bhutan, and Sri Lanka is studied using spatial statistics. In these countries, these fires are mainly forest and vegetation fires; they wreak havoc to the environment by damaging flora and fauna and emitting toxic gases. This study is based on data acquired through remote sensing of data acquisition platform, NASA's MODIS. Spatial statistics is used here to study the incidence of such fires with respect to geographical location. The behaviors of parameters of various autoregressive models like Spatial Durban Model, Spatial Lag Model, Spatial Error Model, Manski Model, and Kelegian Prucha Model are minutely analyzed. The best model with the highest pseudo R² is selected. The spatial behavior of the fire radiative power (FRP) for the three countries is also predicted using spatial interpolation and kriging. The burning potential of vegetations in unsampled areas is envisaged by thus predicting FRP. This study gives a country-wise perspective to the behavior of fire; this is with reference to South Asia. It holds a great significance for countries of the developing world which lack a strong backbone of good-quality official records. Through the statistical analyses of data collected by such platforms, important information on impact of forest fires can be indirectly assessed.

Wildfire Risk Assessment of Transmission-Line Corridors Based on Naïve Bayes Network and Remote Sensing Data

Considering the complexity of the physical model of wildfire occurrence, this paper develops a method to evaluate the wildfire risk of transmission-line corridors based on Naïve Bayes Network (NBN). First, the data of 14 wildfire-related factors including anthropogenic, physiographic, and meteorologic factors, were collected and analyzed. Then, the relief algorithm is used to rank the importance of factors according to their impacts on wildfire occurrence. After eliminating the least important factors in turn, an optimal wildfire risk assessment model for transmission-line corridors was constructed based on the NBN. Finally, this model was carried out and visualized in Guangxi province in southern China. Then a cost function was proposed to further verify the applicability of the wildfire risk distribution map. The fire events monitored by satellites during the first season in 2020 shows that 81.8% of fires fall in high- and very-high-risk regions.