Stratified aboveground forest biomass estimation by remote sensing data
Abstract
Remote sensing-assisted estimates of aboveground forest biomass are essential for modeling carbon budgets. It has been suggested that estimates can be improved by building species- or strata-specific biomass models. However, few studies have attempted a systematic analysis of the benefits of such stratification, especially in combination with other factors such as sensor type, statistical prediction method and sampling design of the reference inventory data. We addressed this topic by analyzing the impact of stratifying forest data into three classes (broadleaved, coniferous and mixed forest). We compare predictive accuracy (a) between the strata (b) to a case without stratification for a set of pre-selected predictors from airborne LiDAR and hyperspectral data obtained in a managed mixed forest site in southwestern Germany. We used 5 commonly applied algorithms for biomass predictions on bootstrapped subsamples of the data to obtain cross validated RMSE and r² diagnostics. Those values were analyzed in a factorial design by an analysis of variance (ANOVA) to rank the relative importance of each factor. Selected models were used for wall-to-wall mapping of biomass estimates and their associated uncertainty. The results revealed marginal advantages for the strata-specific prediction models over the unstratified ones, which were more obvious on the wall-to-wall mapped area-based predictions. Yet further tests are necessary to establish the generality of these results. Input data type and statistical prediction method are concluded to remain the two most crucial factors for the quality of remote sensing-assisted biomass models.
What the paper shows and why it matters
(AI-generated)
Estimates of aboveground forest biomass from remote sensing feed into carbon budgets. It has been suggested that these estimates improve when separate models are built for forest types, for example broadleaved, coniferous and mixed stands; this is a compromise where predicting individual tree species is unrealistic, as in highly mixed stands. Few studies have assessed this benefit systematically, especially alongside other choices that shape model quality, such as the type of remote-sensing data and the statistical method. The paper tests the assumption that stratifying the plots in this way improves the accuracy of biomass models.
The paper argues that the value of stratification has to be judged against the other modelling choices, and finds it small compared with the type of input data and the prediction method. The study tested this in a managed mixed forest near Karlsruhe, southwest Germany. It used airborne LiDAR (laser scanning) and hyperspectral (many narrow colour bands) predictors in separate models, with five prediction algorithms, and repeated each model fit on 500 bootstrap subsamples (random resamples of the plots) to obtain cross-validated error measures. An analysis of variance then ranked how much each factor changed model quality. Models built separately for each forest type were only marginally better than a single unstratified model fitted on a similar number of plots. Maps of predicted biomass covering the whole study area also differed between the two approaches, but only moderately. Input data type, LiDAR versus hyperspectral, and prediction method ranked as the two most important factors for model quality.
The results suggest that, for biomass models, the choice of input data and prediction method affects accuracy more than stratification does. The paper states that further tests on other datasets are needed to show whether the result holds beyond this site. It proposes the factor ranking as a basis for comparing modelling strategies objectively, a possible step towards a quantitative ranking of such strategies. It also notes a practical cost of stratification: each forest type yields its own map, and these would have to be merged into one product.