Do airborne laser scanning biomass prediction models benefit from Landsat time series, hyperspectral data or forest classification in tropical mosaic landscapes?

摘要

Airborne laser scanning (ALS) is considered as the most accurate remote sensing data for the predictive modelling of AGB. However, tropical landscapes experiencing land use changes are typically heterogeneous mosaics of various land cover types with high tree species richness and trees outside forests, making them challenging environments even for ALS. Therefore, combining ALS data with other remote sensing data, or stratification by land cover type could be particularly beneficial in terms of modelling accuracy in such landscapes. Our objective was to test if spectral-temporal metrics from the Landsat time series (LTS), simultaneously acquired hyperspectral (HS) data, or stratification to the forest and non-forest classes improves accuracy of the AGB modelling across an Afromontane landscape in Kenya. The combination of ALS and HS data improved the cross-validated RMSE from 51.5 Mg ha(-1) (42.7%) to 47.7 Mg ha(-1) (39.5%) in comparison to the use of ALS data only. Furthermore, the combination of ALS data with LTS and HS data improved accuracies of the models for the forest and non-forest classes, and the overall best results were achieved when using ALS and HS data with stratification (RMSE 40.0 Mg ha(-1), 33.1%). We conclude that ALS data alone provides robust models for AGB mapping across tropical mosaic landscapes, even without stratification. However, ALS and HS data together, and additional forest classification for stratification, can improve modelling accuracy considerably in similar, tree species rich areas.