Estimating aboveground biomass using Pléiades satellite image in a karst watershed of Guizhou Province, Southwestern China

Abstract

Biomass in karst terrain has rarely been measured because the steep mountainous limestone terrain has limited the ability to sample woody plants. Satellite observation, especially at high spatial resolution, is an important surrogate for the quantification of the biomass of karst forests and shrublands. In this study, an artificial neural network (ANN) model was built using Pleiades satellite imagery and field biomass measurements to estimate the aboveground biomass (AGB) in the Houzhai River Watershed, which is a typical plateau karst basin in Central Guizhou Province, Southwestern China. A back-propagation ANN model was also developed. Seven vegetation indices, two spectral bands of Pleiades imagery, one geomorphological parameter, and land use/land cover were selected as model inputs. AGB was chosen as an output. The AGB estimated by the allometric functions in 78 quadrats was utilized as training data (54 quadrats, 70%), validation data (12 quadrats, 15%), and testing data (12 quadrats, 15%). Data-model comparison showed that the ANN model performed well with an absolute root mean square error of 11.85 t/ha, which was 9.88% of the average AGB. Based on the newly developed ANN model, an AGB map of the Houzhai River Watershed was produced. The average predicted AGB of the secondary evergreen and deciduous broadleaved mixed forest, which is the dominant forest type in the watershed, was 120.57 t/ha. The average AGBs of the large distributed shrubland, tussock, and farmland were 38.27, 9.76, and 11.69 t/ha, respectively. The spatial distribution pattern of the AGB estimated by the new ANN model in the karst basin was consistent with that of the field investigation. The model can be used to estimate the regional AGB of karst landscapes that are distributed widely over the Yun-Gui Plateau.