Eucalyptus carbon stock estimation in subtropical regions with the modeling strategy of sample plots - airborne LiDAR - Landsat time series data

Updating eucalyptus carbon stock data in a timely manner is essential for better understanding and quantifying its effects on ecological and hydrological processes. At present, there are no suitable methods to accurately estimate the eucalyptus carbon stock in a large area. This research aimed to explore the transferability of the eucalyptus carbon stock estimation model at temporal and spatial scales and assess modeling performance through the strategy of combining sample plots, airborne LiDAR and Landsat time series data in subtropical regions of China. Specifically, eucalyptus carbon stock estimates in typical sites were obtained by applying the developed models with the combination of airborne LiDAR and field measurement data; the eucalyptus plantation ages were estimated using the random localization segmentation approach from Landsat time series data; and regional models were developed by linking LiDAR-derived eucalyptus carbon stock and vegetation age (e.g., months or years). To examine the models' robustness, the developed models at the regional scale were transferred to estimate carbon stocks at the spatial and temporal scales, and the modeling results were evaluated using validation samples accordingly. The results showed that carbon stock can be successfully estimated using the age-based models (both age variables in months and years as predictor variables), but the month-based models produced better estimates with a root mean square error (RMSE) of 6.51 t center dot ha = 1 for Yunxiao County, Fujian Province, and 6.33 t center dot ha = 1 for Gaofeng Forest Farm, Guangxi Zhuang Autonomous Region. Particularly, the month-based models were superior for estimating the carbon stocks of young eucalyptus plantations of less than two years. The model transferability analyses showed that the month-based models had higher transferability than the year-based models at the temporal scale, indicating their possibility for analysis of carbon stock change. However, both the month-based and year-based models expressed relatively poor transferability at a spatial scale. This study provides new insights for cost-effective monitoring of carbon stock change in intensively managed plantation forests.