Date Published: February 6, 2019
Publisher: Public Library of Science
Author(s): Simon Besnard, Nuno Carvalhais, M. Altaf Arain, Andrew Black, Benjamin Brede, Nina Buchmann, Jiquan Chen, Jan G. P. W Clevers, Loïc P. Dutrieux, Fabian Gans, Martin Herold, Martin Jung, Yoshiko Kosugi, Alexander Knohl, Beverly E. Law, Eugénie Paul-Limoges, Annalea Lohila, Lutz Merbold, Olivier Roupsard, Riccardo Valentini, Sebastian Wolf, Xudong Zhang, Markus Reichstein, Dafeng Hui.
Forests play a crucial role in the global carbon (C) cycle by storing and sequestering a substantial amount of C in the terrestrial biosphere. Due to temporal dynamics in climate and vegetation activity, there are significant regional variations in carbon dioxide (CO2) fluxes between the biosphere and atmosphere in forests that are affecting the global C cycle. Current forest CO2 flux dynamics are controlled by instantaneous climate, soil, and vegetation conditions, which carry legacy effects from disturbances and extreme climate events. Our level of understanding from the legacies of these processes on net CO2 fluxes is still limited due to their complexities and their long-term effects. Here, we combined remote sensing, climate, and eddy-covariance flux data to study net ecosystem CO2 exchange (NEE) at 185 forest sites globally. Instead of commonly used non-dynamic statistical methods, we employed a type of recurrent neural network (RNN), called Long Short-Term Memory network (LSTM) that captures information from the vegetation and climate’s temporal dynamics. The resulting data-driven model integrates interannual and seasonal variations of climate and vegetation by using Landsat and climate data at each site. The presented LSTM algorithm was able to effectively describe the overall seasonal variability (Nash-Sutcliffe efficiency, NSE = 0.66) and across-site (NSE = 0.42) variations in NEE, while it had less success in predicting specific seasonal and interannual anomalies (NSE = 0.07). This analysis demonstrated that an LSTM approach with embedded climate and vegetation memory effects outperformed a non-dynamic statistical model (i.e. Random Forest) for estimating NEE. Additionally, it is shown that the vegetation mean seasonal cycle embeds most of the information content to realistically explain the spatial and seasonal variations in NEE. These findings show the relevance of capturing memory effects from both climate and vegetation in quantifying spatio-temporal variations in forest NEE.
Forests cover about 30% of the terrestrial surface of our planet, accounting for 75% of gross primary production (GPP), and store 45% of all terrestrial carbon (C) [1–3]. This fundamental role highlights the importance of understanding forest C dynamics, which are generally driven by climatic conditions and vegetation dynamics as well as natural and anthropogenic disturbances [4–6]. Changes in climate and disturbance regime can influence the development, structure, and functioning of forest ecosystems [7–12], therefore causing anomalies in the net carbon dioxide (CO2) exchange of terrestrial ecosystems (NEE). As a result, quantifying the effects of climatic variations and forest disturbances on biosphere-atmosphere CO2 fluxes across-scales has considerable importance for understanding the net CO2 balance of forest ecosystems [13–16].
In this study, we present a novel framework for assessing the potential of the memory effects of climate and vegetation on forests’ NEE using the Landsat satellite imagery and in-situ eddy covariance observations. The results presented here for the whole FLUXNET dataset reveal a variable memory effect on NEE across-scales, but that is mainly apparent at the seasonal scale and across-sites. We also find that the effects of memory vary between FLUXNET sites suggesting site-specific memory effects. Although instantaneous observations of the contemporaneous vegetation states may already carry information from the past, current analysis suggests that extracting antecedent observations of vegetation and climate are beneficial for estimating NEE more realistically (the difference between LSTM and LSTMperm, as well as between LSTM and RF). Such effects can emerge from the information contained in the course of the seasonal cycle or from the effects of interannual variation on NEE. However, the close agreement between LSTM and LSTMmsc suggests that either the effect is smeared out by the impact of instantaneous climate on NEE or the interannual variation’s memory effect in NEE is implicitly captured by this approach. The results are contingent on the length of observations and few recently disturbed forests but do emphasize the possibility of dynamic statistical methods that include memory effects to better estimate the contribution of forest ecosystems in the global terrestrial C cycle, hence for further improving statistically-based prediction methods.