Partitioning net carbon dioxide fluxes into photosynthesis and respiration using neural networks

• DOI: 10.1111/gcb.15203
• Publication Date: 2020-06-04T18:29:00Z
• Authors: Gianluca Tramontana; Mirco Migliavacca; Martin Jung; Markus Reichstein; Trevor F. Keenan; Gustau Camps‐Valls; Jerome Ogee; Jochem Verrelst; Dario Papale
• Keywords: Carbon respiration; Carbon sink

Abstract The eddy covariance (EC) technique is used to measure the net ecosystem exchange (NEE) of CO 2 between ecosystems and atmosphere, offering a unique opportunity study responses climate change. NEE difference total release due all respiration processes (RECO), gross carbon uptake by photosynthesis (GPP). These two fluxes are derived from EC measurements applying partitioning methods that rely on physiologically based functional relationships with limited number environmental drivers. However, applied in global FLUXNET network observations do not account for multiple co‐acting factors modulate GPP RECO flux dynamics. To overcome this limitation, we developed hybrid data‐driven approach combined neural networks (NN C‐part ). NN incorporates process knowledge introducing photosynthetic response light‐use efficiency (LUE) concept, uses comprehensive dataset soil micrometeorological variables as We method 36 sites FLUXNET2015 found high consistency results those other standard both ( R > .94) .8). High was also (a) diurnal seasonal patterns (b) responses. performed more realistic than traditional predicting additional fluxes, such as: VPD, direct effects air temperature dynamics, (c) hysteresis diel cycle (d) sensitivity LUE diffuse radiation ratio, (e) post rain pulse after long dry period. In conclusion, valid provide estimates complementary existing methods.