Evaluating statistical and combine method to predict stand above-ground biomass using remotely sensed data

• Published in: Arabian journal of geosciences 2022, Vol.15 (9), Article 838
• Main Authors: Bulut, Sinan, Sivrikaya, Fatih, Günlü, Alkan
• Format: Article
• Language: English
• Subjects: Accuracy; Biomass; Carbon content; Coefficients; Correlation coefficient; Correlation coefficients; Earth and Environmental Science; Earth science; Earth Sciences; Environment models; Forest management; Original Paper; Prediction models; Reflectance; Remote sensing; Satellite imagery; Spaceborne remote sensing; Spatial distribution; Statistical analysis; Statistical methods; Stocks; Vegetation; Vegetation index
• ISSN: 1866-7511; 1866-7538
• DOI: 10.1007/s12517-022-10140-3

Above-ground biomass (AGB) is a key parameter as indicator of forest carbon content and ecosystem productivity. Producing highly predictive models and spatial distribution of AGB stocks support the development of forest management plans. In this research, AGB was modeled using multiple linear regression (MLR) analysis and regression kriging (RK) techniques. Reflectance and vegetation indices data derived from the Sentinel-2 satellite images were used as an auxiliary variable. The highest correlation coefficients between the AGB and remote sensing data were obtained in broadleaf stands. These values were 0.743 and 0.869 for reflectance values of B8 band and PSSR vegetation index values, respectively. The AGB contents of coniferous, broadleaf, and mixed stands were modeled separately with MLR method. Then, RK method performed with combining MLR and ordinary kriging methods. The R 2 values of MLR method were lower than 0.80. Ordinary kriging method was applied to MLR residuals, and it was determined that the highest spatial dependence was in coniferous stands (spatial correlation index = 57–60). The highest explained variance proportions for RK method were 0.88, 0.84, and 0.83 in coniferous, broadleaf, and mixed stands, respectively. Overall, the accuracy of MLR method was not sufficient for AGB estimation, but the combination of kriging helped to take advantage in terms of improving the accuracy of AGB predictions. To increase the accuracy of the AGB model, it is thought that it would be beneficial to diversify the remote sensing data and to combine the kriging method with different techniques.