Optimized stratification approach enhances the weight-of-evidence method: Transparently uncovering wildfire probability and drivers-wildfire relationships in the southwest mountains of China
•The optimized weight-of-evidence method enhances prediction of wildfire probabilities.•Improvement in the discrete method boosts spatial interpretation of wildfire drivers.•Variables related to ignition sources largely influence wildfire distribution in Yunnan Province.•The relationship between dri...
Saved in:
Published in | Ecological indicators Vol. 158; p. 111500 |
---|---|
Main Authors | , , , |
Format | Journal Article |
Language | English |
Published |
Elsevier Ltd
01.01.2024
Elsevier |
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | •The optimized weight-of-evidence method enhances prediction of wildfire probabilities.•Improvement in the discrete method boosts spatial interpretation of wildfire drivers.•Variables related to ignition sources largely influence wildfire distribution in Yunnan Province.•The relationship between drivers and wildfires cannot be simply summarized by monotonicity.
The mountainous region of southwest China serves as a significant treasure for resources and ecological security. Due to its topographic heterogeneity and cultural diversity, the wildfire regime in this area exhibits unique features. To uncover the distinct mechanisms that drive wildfire incidents and assess their occurrence probability, an enhanced weight-of-evidence (WofE) method was proposed by improving the stratification of evidence layers. This approach seeks to heighten effectiveness while maintaining procedural transparency and robustness. In our case study focusing on Yunnan Province, we extracted wildfire data from medium-resolution remote sensing imagery from 2006 to 2020. Various wildfire drivers were considered including wildfire environment, fuel conditions, and ignition sources. We utilized the WofE method grounded on Bayesian principles to clearly calculate the spatial association strength between these drivers and wildfires, thereby developing a reliable wildfire occurrence probability map. To enhance the explanatory power of the drivers within the WofE method, a discretization method based on the theory of spatial stratified heterogeneity was incorporated. Our results suggested that the WofE method can be optimized using the spatial stratified heterogeneity measured via GeoDetector, leading to an improved solution. Implementing optimized discrete drivers amplified the spatial explanatory power of wildfires by an average of 7.55%, supporting their inclusion as evidence layers within the WofE method. The optimized discrete WofE method yielded a wildfire occurrence probability map with an Area Under the Curve (AUC) value of 0.91, indicating high predictive accuracy. This map revealed notable spatial clustering and regional variations in wildfire occurrences across Yunnan Province. Additionally, we observed variable spatial correlation between each driver and wildfire occurrence, and the wildfire drivers indicative of ignition sources characteristics were relatively more dominant locally. This research contributes valuable insights towards enhancing the WofE method and provides a helpful reference for local wildfire management practices and resource allocation strategies. |
---|---|
ISSN: | 1470-160X 1872-7034 |
DOI: | 10.1016/j.ecolind.2023.111500 |