Modeling the Carbon Cycle of a Subtropical Chinese Fir Plantation Using a Multi-Source Data Fusion Approach

Abstract

Process-based terrestrial ecosystem models are increasingly being used to predict carbon (C) cycling in forest ecosystems. Given the complexity of ecosystems, these models inevitably have certain deficiencies, and thus the model parameters and simulations can be highly uncertain. Through long-term direct observation of ecosystems, numerous different types of data have accumulated, providing valuable opportunities to determine which sources of data can most effectively reduce the uncertainty of simulation results, and thereby improve simulation accuracy. In this study, based on a long-term series of observations (biometric and flux data) of a subtropical Chinese fir plantation ecosystem, we use a model–data fusion framework to evaluate the effects of different constrained data on the parameter estimation and uncertainty of related variables, and systematically evaluate the uncertainty of parameters. We found that plant C pool observational data contributed to significant reductions in the uncertainty of parameter estimates and simulation, as these data provide information on C pool size. However, none of the data effectively constrained the foliage C pool, indicating that this pool should be a target for future observational activities. The assimilation of soil organic C observations was found to be important for reducing the uncertainty or bias in soil C pools. The key findings of this study are that the assimilation of multiple time scales and types of data stream are critical for model constraint and that the most accurate simulation results are obtained when all available biometric and flux data are used as constraints. Accordingly, our results highlight the importance of using multi-source data when seeking to constrain process-based terrestrial ecosystem models.

BiLSTM-I: A Deep Learning-Based Long Interval Gap-Filling Method for Meteorological Observation Data

“…Temperature is a very important variable for agricultural and ecosystem studies, and it is an essential input in agricultural crop growth simulations, agrometeorological disaster monitoring, and ecosystem simulations [ 1 , 2 ]. As agricultural and ecological simulations have improved, the resolution requirements for temperature data have increased; notably, high-resolution data are needed in wind monitoring in dry and hot areas, agrometeorological hazard assessments, and simulations of carbon emissions from forest block ecosystems [ 3 , 4 ]. Temperature observations are usually obtained from field meteorological stations, and the data observed at small weather stations commonly have gaps due to equipment failure, harsh environmental conditions or operational errors [ 5 ].…”