Ural Federal University-Developed Neural Network Helps Estimate Carbon Dioxide Flows at Ural Carbon Polygon

A neural network helps Ural-Carbon test site specialists to estimate carbon fluxes in the Urals. The network has been trained to link data from space-based Modis multispectral sensors (NASA, Aqua and Terra satellites). It links air temperature and precipitation data with data from the Fluxnet network of ground stations to determine carbon dioxide fluxes in the atmosphere. The model was developed by specialists from the Laboratory of Climate and Environmental Physics of the Ural Federal University and tested at the Carbon Polygon in Kourovka.

“Our student built a neural network, trained it, and now it calculates carbon fluxes, latent heat, and net ecosystem exchange. Overall, it shows good accuracy in reproducing data. We tested it at the carbon test site in Kourovka and compared the results with data from the ground-based Fourier IR spectrometer during atmospheric sounding. The discrepancy was only 15%, which is within the RMS error of the neural network. Currently, we receive new data every eight days,” explains Konstantin Gribanov, Head Specialist at the UrFU Climate and Environmental Physics Laboratory.

The main thing that can be done with the neural network developed by UrFU specialists is to assess the situation throughout Russia. There are dark coniferous forests (an ecosystem similar to the Ural Carbon Polygon).

“At the Kourovka carbon polygon, we made the first assessments of the carbon balance of the forest ecosystem. Based on the data obtained by the Fourier IR spectrometer, we were able to estimate the amount of carbon dioxide absorbed from the atmosphere during the growing season, which lasts from April to September in the Urals. In particular, it was calculated that the amount of carbon dioxide absorbed from the atmosphere by the forest ecosystem is 1.5 tons per hectare. This means that during this period each hectare of forest grows more than 1.5 tons of biomass. As for the neural network, it estimated the amount of carbon dioxide absorbed from the atmosphere during the same period at 1.3 tons per hectare,” explains Vyacheslav Zakharov, Head of the UrFU Climate and Environmental Physics Laboratory.

In other words, one hectare of dark coniferous forest at the Kourovka Carboniferous Site absorbs 1.5 tons of carbon dioxide from the atmosphere during the growing season, which in terms of carbon is about 410 kg of absorbed carbon.

“Until April, it is winter, it is cold, there are no leaves or grass, so the concentration of carbon dioxide in the atmosphere is maximum. As soon as the leaves appear, photosynthesis begins, taking carbon dioxide out of the atmosphere until September. Then the leaves fall off, decompose and return some of the accumulated carbon to the atmosphere, and its concentration in the atmosphere begins to increase,” explains Vyacheslav Zakharov.

The task of experts is to understand how much carbon dioxide the plants and trees of the Ural Carbon Polygon (mixed or dark coniferous forests) absorb from the atmosphere during the growing season. This is not easy, because the annual increase in the level of carbon dioxide in the atmosphere is largely due to its emissions from industry and transportation.

“We have been measuring the total carbon dioxide content in the atmospheric column in Kourovka since 2009. The data show an increasing trend, the atmospheric carbon dioxide content is increasing by about 3 ppm every year, but this is related to global changes, not to our forests. In other words, we measure in the atmosphere where there is an annual increase, and we need to somehow eliminate the global trend and assess the situation at the level of multi-year fluctuations. Then maybe we will see the difference that is absorbed by our forest,” concludes Konstantin Gribanov.

The data obtained by the neural network are still preliminary, explain the laboratory specialists. The model is constantly being improved. This year, for example, they plan to verify the neural network’s estimates with measurements of carbon dioxide fluxes from masts and calculations based on the atmospheric fluid-location model.