The effect of digital preprocessing and modelling method on estimation of aboveground carbon stock of Zagros forests using Landsat 8 imagery

The aim of this study, was to evaluate the effectiveness of different preprocessing methods and modeling techniques on the accuracy of aboveground carbon stock estimates in two forest stands with different degradation levels (Gahvareh forest and SarfiruzAbad), in Zagros forests in Kurdistan province. Comparison of different digital pre-processing methods on Landsat 8 images was carried out in different scenarios of radiometric, atmospheric, topographic and their combination. In each scenario, we used four modeling methods included linear regression, generalized additive model, random forest, and support vector machine. In most cases, radiometric correction with improved correction coefficient was 0.71 (R2adj =0.71) and the root means square error of 30% (RMSe% =0.30) was outperformed. Comparison of four modeling methods indicates the lower accuracy of estimates in the SarfiruzAbad area with more degradation severity (R2adj =0.58) compared to the less damaged Gahvareh area (RMSe% =0.74). The random forest method for Gahvareh area and linear regression and a generalized additive model for SarfiruzAbad provides better results, respectively. However, our findings showed that selection of suitable preprocessing and modeling method have a noticeable effect on the accuracies of characteristics estimates in forest ecosystems by Landsat imagery.