Upscaling Net Ecosystem Exchange Over Heterogeneous Landscapes With Machine Learning
Graf, A.
Schmidt, M.
Ketzler, G.
Leuchner, M.
DOI: https://doi.org/10.23689/fidgeo-4388
Schmidt, M.; 2 Agrosphere Institute, Forschungszentrum Jülich Jülich Germany
Ketzler, G.; 1 Department of Geography, RWTH Aachen University Aachen Germany
Leuchner, M.; 1 Department of Geography, RWTH Aachen University Aachen Germany
Abstract
This paper discusses different feature selection methods and CO2 flux data sets with a varying quality‐quantity balance for the application of a Random Forest model to predict daily CO2 fluxes at 250 m spatial resolution for the Rur catchment area in western Germany between 2010 and 2018. Measurements from eddy covariance stations of different ecosystem types, remotely sensed vegetation data from MODIS, and COSMO‐REA6 reanalysis data were used to train the model and predictions were validated by a spatial and temporal validation scheme. Results show the capabilities of a backwards feature elimination to remove irrelevant variables and an importance of high‐quality‐low‐quantity flux data set to improve predictions. However, results also show that spatial prediction is more difficult than temporal prediction by reflecting the mean value accurately though underestimating the variance of CO2 fluxes. Vegetated parts of the catchment acted as a CO2 sink during the investigation period, net capturing about 237 g C m−2 y−1. Croplands, coniferous forests, deciduous forests and grasslands were all sinks on average. The highest uptake was predicted to occur in late spring and early summer, while the catchment was a CO2 source in fall and winter. In conclusion, the Random Forest model predicted a narrower distribution of CO2 fluxes, though our methodological improvements look promising in order to achieve high‐resolution net ecosystem exchange data sets at the regional scale.
Plain Language Summary: Whether ecosystems absorb or emit CO2 plays a major role in the global carbon cycle and impacts climate change. This exchange is already measured at scattered stations, but creating spatially resolved data sets remains a challenge. In this paper, we used satellite images of vegetation and meteorological data to predict the CO2 exchange of the Rur catchment area near the German‐Dutch‐Belgian border for every day from 2010 to 2018. In order to assess the prediction quality, we compared actual measurements from several stations within the catchment with the predictions at the locations of these stations. Results show that our method could increase prediction quality compared to previous process‐based models, though the error remains rather high. Vegetated parts of the catchment including coniferous forests, deciduous forests, grasslands, and croplands were all CO2 sinks on average. In late spring and early summer, they were the strongest sink, but in fall and winter a CO2 source.
Key Points:
CO2 flux upscaling with Random Forest can be improved with a backward feature elimination and strict quality control of flux data.
Vegetated parts of the Rur catchment were predicted to be a CO2 sink on average, with the highest uptake in late spring and early summer.
The Enhanced Vegetation Index and potential evapotranspiration are useful predictors for the regionalization of CO2 flux measurements.
Subjects
carbon fluxeddy covariance
feature selection
net ecosystem exchange
Random Forest
spatial prediction