Prediction and mapping of boreal forest fire fuel loads using high-resolution satellite stereo imagery

Abstract

The aim of this study is to evaluate the suitability of very high-resolution satellite stereo-imagery data for creating forest fire-related fuel load maps in the boreal region. We acquired stereo imagery from the GeoEye-1 (GE-1) satellite, which has a ground sampling distance of 50 cm. The images were acquired in August 2021 and 2023 (hence leaf-on). Our study area was centred around the Hiidenportti national park in central Finland, dominated by natural boreal forests. The ground reference was a field dataset consisting of measurements from 33 forested plots, each of 15 m radius. The dominant height (m), foliage biomass (t ha-1) and canopy base height (m) were predicted using multivariate linear regression models, while the understory presence (categorical; present/absent) was predicted using logistic regression analysis. Prediction models using area-based metrics based on airborne laser scanning (ALS) data had the smallest associated root mean square error (RMSE) (between 2.6% and 23.9%). Meanwhile, similar type of area-based metrics of stereo satellite data combined with an ALS-based digital terrain model (DTM) resulted in RMSEs of 6.6–30.3%. We also formulated models suitable for the case when only satellite data is available (i.e. high-quality DTM is absent), such as in remote locations of the boreal forest region. In this case, the models involved several canopy texture metrics and point cloud height and colour intensity-based metrics as predictors. The associated relative RMSEs were in the range of 11–30%. Dominant height, an important global vegetation metric, was predicted with an RMSE of 2.6 m, which compares well with other model predictions under similar circumstances. Our findings suggests that very high-resolution stereo satellite image data is promising for the generation and updating of wall-to-wall boreal forest fuel load maps, including remote areas lacking high resolution DTM data.