A new study led by the Finnish Meteorological Institute (FMI) has used machine learning to substantially reduce a systematic bias in carbon balance estimates for northern ecosystems when using the eddy covariance technique.
The eddy covariance technique is a key method used to measure carbon balances of forests, agricultural fields and wetlands. In principle, this method measures carbon dioxide exchange between ecosystems and the atmosphere continuously every half an hour.
However, some measurements have to be discarded because of unfavorable atmospheric conditions, and also technical failures that lead to data gaps. Replacing, or gap-filling, the missing data to obtain a full timeseries is an essential step in calculating annual carbon balances.
For the first time, the FMI study investigated the impact of different gap-filling methods on the carbon balance estimates of high-latitude ecosystems.
The study showed that the most common gap-filling method, also used by international flux measurement networks, causes a systematic error in the annual carbon balance estimates of northern ecosystems. Because of this error, emissions of carbon sources have been overestimated and the strength of carbon sinks underestimated. The bias is substantial, at some sites even of a similar magnitude to the annual carbon balance. The error is caused by a very skewed distribution of solar radiation in northern latitudes.
To correct for the error, the study utilized a machine learning method, which proved to be more accurate than the old standard method and did not cause any systematic errors. In addition to a new, unbiased method, the old method was modified to better account for the environmental conditions prevailing in northern latitudes.
Boreal and tundra biomes cover an area of more than 20,000,000km2 . Therefore, the systematic error found in the study can have significant impacts on the carbon balance estimates of northern areas if these are based on eddy covariance measurements. When the improved gap-filling methods are taken widely into use, estimates of carbon balances of northern ecosystems can be calculated more accurately. In future, eddy covariance data can also be utilized to increase the reliability of carbon balance estimates produced in national greenhouse gas inventories.
To view the full study published in Scientific Reports, click here.