When animals — including humans — breathe, they exchange oxygen for carbon dioxide. Plants absorb carbon dioxide from the environment and produce oxygen. The net ecosystem exchange (NEE) is how much carbon is put into the environment compared to how much is removed. This concept is important because an increase in the concentration of carbon dioxide in our atmosphere amplifies Earth's natural greenhouse effect, resulting in climate change.
The eddy covariance technique is one of the most accurate and direct tools to help us measure this carbon exchange between the surface and the atmosphere. However, some conditions — like equipment malfunctions, power outages and extreme weather conditions — result in missing values for 30% to 65% of the data. Thus, other methods have been developed to fill in the gaps — one of which is the artificial neural networks (ANNs) approach. ANNs are a form of artificial intelligence (or machine learning) and provide an approximate solution to a real problem, rather than an exact solution to an oversimplified one.
In a new study, DWFI postdoctoral researcher Babak Safa, DWFI Faculty Fellow Andy Suyker and colleagues compared two different ANNs models on rain-fed corn fields near Mead, Nebraska — the Multi-layer Perceptron (MLP) network trained by the Back-Propagation (BP) algorithm and the Radial Basis Function (RBF) Network. They found that the RBF network provides the best fit for observed values.
Overall, the results show that ANNs, as a technique, are able to estimate the missing NEE data accurately and efficiently. This helps provide a more accurate estimation of net carbon exchange rates to both scientists and policymakers, which is essential in enacting effective future strategies related to climate change.