Search for: All records

Eddy Covariance measurements are often subject to missing values, or gaps in the data record. Methods to fill short gaps are well-established, but robustly filling gaps longer than a few weeks remains a challenge. Marginal Distribution Sampling (MDS) is a standard gap-filling method, but its effectiveness for long gaps (> 30 days) is limited. We compared the performance of a machine learning algorithm, eXtreme Gradient Boosting (XGB) against MDS, using various artificial scenarios of gap lengths and locations. We gapfilled half hourly CO2 flux from a temperate deciduous forest, Bartlett Experimental Forest, from 2010 to 2022. Whereas the standard implementation of MDS uses a narrowly-prescribed set of predictor variables, with XGB we were able to include additional variables. The Green Chromatic Coordinate (GCC), derived from PhenoCam imagery, and diffuse photosynthetic photon flux density, emerged as two of the three most important predictor variables. Compared to MDS, the root mean square error (RMSE) of XGB decreased by 9.5 %, and the R2 increased by 2.7 % in a randomized 10-fold cross validation test. XGB outperformed MDS for both day and night times across different seasons. But annual NEE integrals varied across methods, with weaker annual net carbon uptake, by -110 ± 74 g C m-2 y-1 for XGB compared to MDS (214 ± 11 g C m-2 yr-1). In artificial gap experiments, when trained using the 13-year data record, XGB reliably filled gaps, showing little change in RMSE for gaps up to 240 days. In contrast, the performance of MDS steadily decreased as gap lengths increased. MDS was unable to fill gaps longer than 2 months. In summary, XGB demonstrates excellent performance as an alternative method to MDS, providing reliable predictions for temperate deciduous forest carbon fluxes under different gap lengths and location scenarios. Implementation of XGB is facilitated by easy-to-use packages. 

AmeriFlux is a network of hundreds of sites across the contiguous United States providing tower-based ecosystem-scale carbon dioxide flux measurements at 30 min temporal resolution. While geographically wide-ranging, over its existence the network has suffered from multiple issues including towers regularly ceasing operation for extended periods and a lack of standardization of measurements between sites. In this study, we use machine learning algorithms to predict CO2 flux measurements at NEON sites (a subset of Ameriflux sites), creating a model to gap-fill measurements when sites are down or replace measurements when they are incorrect. Machine learning algorithms also have the ability to generalize to new sites, potentially even those without a flux tower. We compared the performance of seven machine learning algorithms using 35 environmental drivers and site-specific variables as predictors. We found that Extreme Gradient Boosting (XGBoost) consistently produced the most accurate predictions (Root Mean Squared Error of 1.81 μmolm−2s−1, R2 of 0.86). The model showed excellent performance testing on sites that are ecologically similar to other sites (the Mid Atlantic, New England, and the Rocky Mountains), but poorer performance at sites with fewer ecological similarities to other sites in the data (Pacific Northwest, Florida, and Puerto Rico). The results show strong potential for machine learning-based models to make more skillful predictions than state-of-the-art process-based models, being able to estimate the multi-year mean carbon balance to within an error ±50 gCm−2y−1 for 29 of our 44 test sites. These results have significant implications for being able to accurately predict the carbon flux or gap-fill an extended outage at any AmeriFlux site, and for being able to quantify carbon flux in support of natural climate solutions.