Evaluation of Wildfire Occurrences in Pakistan with Global Gridded Soil Properties Derived from Remotely Sensed Data

• Published in: Remote sensing (Basel, Switzerland) Vol. 14; no. 21; p. 5503
• Main Authors: Rafaqat, Warda, Iqbal, Mansoor, Kanwal, Rida, Weiguo, Song
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.11.2022
• Subjects: Air pollution; Annual cycles; Annual variations; Atmospheric models; Bulk density; Climate change; Correlation; Elevation; fire count; Fire hazards; Forest & brush fires; fuel; Global warming; Learning algorithms; Machine learning; Monitoring; Monitoring methods; Multivariate analysis; Precipitation; Rain; Regression analysis; Regression models; Remote sensing; Satellite observation; satellite remote sensing; Soil density; Soil properties; Soils; Taxonomy; Temporal variations; Terrestrial ecosystems; Topography; Vegetation; Wildfires; Pakistan
• DOI: 10.3390/rs14215503
• ISSN: 2072-4292

Wildfires are predicted to occur more frequently and intensely as a result of global warming, posing a greater threat to human society, terrestrial ecosystems, and the atmosphere. Most existing methods for monitoring wildfire occurrences are based either on static topographical information or weather-based indices. This work explored the advantages of a new machine learning-based ‘soil properties’ attribute in monitoring wildfire occurrence in Pakistan. Specifically, we used satellite observations during 2001–2020 to investigate the correlation at different temporal and spatial scales between wildfire properties (fire count, FC) and soil properties and classes (SoilGrids1km) derived from combination with local covariates using machine learning. The correlations were compared to that obtained with the static topographic index elevation to determine whether soil properties, such as soil bulk density, taxonomy, and texture, provide new independent information about wildfires. Finally, soil properties and the topographical indices were combined to establish multivariate linear regression models to estimate FC. Results show that: (1) the temporal variations of FC are negatively correlated with soil properties using the monthly observations at 1° grid and regional scales; and overall opposite annual cycles and interannual variations between and soil properties are observed in Pakistan; (2) compared to the other static variables such as elevation, soil properties shows stronger correlation with the temperate wildfire count in Northern Pakistan but weaker correlation with the wildfire properties in Southern Pakistan; and it is found that combining both types of indices enhances the explained variance for fire attributes in the two regions; (3) In comparison to linear regression models based solely on elevation, multivariate linear regression models based on soil properties offer superior estimates of FC.