Spatiotemporal lagging of predictors improves machine learning estimates of atmosphere–forest CO 2 exchange

[摘要] Accurate estimates of net ecosystem CO 2 exchange(NEE) would improve the understanding of natural carbon sources and sinks andtheir role in the regulation of global atmospheric carbon. In this work, weuse and compare the random forest (RF) and the gradient boosting (GB)machine learning (ML) methods for predicting year-round 6 h NEE over1996–2018 in a pine-dominated boreal forest in southern Finland and analyze thepredictability of NEE. Additionally, aggregation to weekly NEE values wasapplied to get information about longer term behavior of the method. Themeteorological ERA5 reanalysis variables were used as predictors. Spatialand temporal neighborhood (predictor lagging) was used to provide the modelsmore data to learn from, which was found to improve considerably theaccuracy of both ML approaches compared to using only the nearest grid celland time step. Both ML methods can explain temporal variability of NEE inthe observational site of this study with meteorological predictors, but theGB method was more accurate. Only minor signs of overfitting could bedetected for the GB algorithm when redundant variables were included.The accuracy of the approaches, measured mainly using cross-validated R 2 score between the model result and the observed NEE, was high,reaching a best estimate value of 0.92 for GB and 0.88 for RF. In additionto the standard RF approach, we recommend using GB for modeling the CO 2 fluxes of the ecosystems due to its potential for better performance.

[学科分类] 大气科学