LAUSR

Evaluating accurate aboveground carbon (AGC) of mangrove forests is a challenging task owing to the complex canopy structure of mangroves and physiographic features. This study thoroughly assessed the potential ability of high-resolution PLEIADES texture metrics at different window sizes for quantifying AGC of mangrove forests in Jiulong River Estuary. Field measurements of AGC were obtained in 31 plots (10 × 10 m), and the data ranged from 78.747 to 143.393 t C ha−1, with an average of 102.233 t C ha−1 per plot. Various possibilities were examined, including spectrals bands, band ratios and various types of texture metrics, and regression modeling was applied in a five-step framework. To ensure good performance of model during the calibration process, we determined coefficients of determination (R2), p value of analysis of variance and root mean square errors (RMSE). Additionally, variable inflation factor was used to avoid the problem of multi-collinearity among the independent variables. Results showed that the spectral-based model only predicted AGC with an uncertainty (RMSE) of 9.14 t C ha−1, and R2 of 0.631. The texture-based model had a much better potential for AGC estimation with a higher R2 value of 0.934, and a lower RMSE of 3.76 t C ha−1. With increasing of window size for the texture calculations, the R2 values increased and the RMSEs decreased. Additionally, we observed negative effects for AGC predictions, when spectral variables were added to texture variables during model development. Based on the calibrations, four texture-based models were selected and validated using another set of field data. Indicators including R2, relative error, Nash–Sutcliffe efficiency (ENS), and RMSE were examined to measure for deviations between estimated and observed data. Model 7 with an R2 value of 0.878 was finally chosen for relatively accurate quantification of AGC. The AGC values at a selected site derived from the model ranged from 1 to 153 t C ha−1.