Monitoring soil carbon flux with in-situ measurements and satellite observations in a forested region

Carbon dioxide (CO2) flux plays a key role in the energy balance of ecosystems and effective monitor CO2 flux from soil at both large scale and high temporal resolution remains a critical scientific problem of soil science. To address this scientific problem, this study tested a novel model to estimate CO2 flux from soil with highly dense vegetation cover. In contrast to previous approaches, the developed model monitored daily CO2 flux at both high spatial and temporal resolutions. This model is applicable for different forest types by fusing the daily TERRA Moderate Resolution Imaging Spectroradiometer (MODIS), Land Remote Sensing Satellite (Landsat) measurements, and monthly in-situ CO2 flux measurements collected by non-permanent chambers in the Great Dismal Swamp (GDS), the U.S. Then, daily vegetation index and surface soil temperature were estimated at high spatial resolution, and the CO2 flux was retrieved through a semi-empirical model. Two-year ground observations in the GDS were used to train and validate the model. Over the GDS area, the validation results of the retrieved CO2 flux had a coefficient of determination (R-2) of 0.47, and a root mean square error (RMSE) of 1.25e + 05 mu g CO2/m(2)/h. Further analysis showed that the performance of the proposed method varied slightly with forest types. The difference between R-2 was less than 0.07 and the difference between RMSE was less than 0.08e + 05 mu g CO2/m(2)/h. The GDS has three main forest types: maple gum (Ater rubrum and Nyssa sylvatica), Atlantic white cedar (Chamaecyparis thyoides), and pine pocosin (Pines serotina). For the maple-gum, the R-2 was 0.51, and the RMSE was 1.14e + 05 mu g CO2/m(2)/h. For Atlantic white cedar, the R-2 was 0.62, and the RMSE was 1.15e + 05 mu g CO2/m(2)/h. For pine pocosin, the R-2 was 0.44, and the RMSE was 1.07e + 05 mu g CO2/m(2)/h. The proposed method enables estimation of CO2 flux within forested areas, and forest type affects the monitoring results.