A Data‐Driven Global Soil Heterotrophic Respiration Dataset and the Drivers of Its Inter‐Annual Variability

Abstract

Soil heterotrophic respiration (SHR), the CO 2 flux produced by free-living microbial heterotrophs and soil fauna feeding on soil organic matter (Carbone et al., 2016;Hanson et al., 2000), constitutes a key ecosystem-to-atmosphere carbon flux that affects soil carbon storage and carbon-climate feedbacks. Since the magnitude of SHR is roughly four times of global annual anthropogenic fossil fuel emission (Le Quéré et al., 2018) and SHR can regulate the net ecosystem carbon exchange variability in some regions (Liu, Ballantyne, et al., 2018), even small changes in this flux can cause carbon redistribution between soil and atmosphere, and modify the carbon sink. Enhanced microbial dynamics in soil organic matter decomposition have been detected as the dominant factor in an increasing imbalance between higher CO 2 loss rate and CO 2 uptake by plants . Therefore, a detailed understanding of the SHR spatial and temporal dynamics under changing climate conditions is pivotal to improve projections of the carbon-climate feedback (Ballantyne et al., 2017;Bradford et al., 2019).However, unlike other components of the terrestrial carbon cycle like gross primary productivity (GPP) that can be measured through eddy covariance flux tower at plot scale, SHR observations mainly come from small-scale chambers, combined with intrusive methods (trenching, root exclusion, root extraction), or non-intrusive methods of isotope labeling with uncertainty in 14 C measurements (Bond-Lamberty et al., 2004;Hanson et al., 2000) to partition the heterotrophic and autotrophic soil fluxes. Due to the considerable uncertainty underlying these measurements, SHR is the most poorly constrained ecosystem and global carbon flux (Ciais et al., 2020;Konings et al., 2019).

Soil respiration is correlated with rainfall and soil moisture at multiple temporal scales in a seasonal wet tropical forest

“…However, even for the same level of soil moisture, instantaneous Rs tended to be lower in Mulgunda compared to Hosagadde (Fig 2d). While the reasons underlying these differences are currently unclear, it is likely that other factors including edaphic properties – which have been shown to exert significant controls on soil respiration rates, potentially more so than climatic factors (Haaf et al, 2021), soil nutrient availability, litter production and availability of substrate for microbes also contribute to the observed differences (Davidson et al, 2002; Yao et al, 2021).…”

Encoding diel hysteresis and the Birch effect in dryland soil respiration models through knowledge-guided deep learning

“…To the best of our knowledge, this is the first implementation of the DL models in estimating soil respiration in a sub-hourly scale. Most existing studies only adopted vanilla machine learning models (e.g., artificial neural networks (Zhao et al, 2017) and random forest (Lu et al, 2021;Yao et al, 2021)) in emulating soil respiration dynamics at coarser temporal scales from daily to annually, which are too coarse to capture the diel hysteresis. Thus, they fail to recognize the importance of future environmental states to the current respiration state in a dryland ecosystem such as this study site.…”

Twenty Years of Progress, Challenges, and Opportunities in Measuring and Understanding Soil Respiration

“…A third challenge concerns the power, applicability, and interpretability of ML algorithms (Dramsch, 2020; Reichstein et al., 2019). The large size of many Rs data sets, combined with the large number of environmental variables influencing Rs, has made Random Forests a popular algorithm in Rs research (e.g., H. Lu et al., 2021; Warner et al., 2019; Yao et al., 2021) but other approaches have also been applied. However, it is not clear that ML provides more utility over simple approaches: a study of Rs gap‐filling algorithms found that ANNs exhibited larger errors than nonlinear least squares and other techniques (Zhao et al., 2020).…”

“…This explosion in available Rs data has been paired with equally rapid growth in satellite data products, reanalysis data sets, multi‐model intercomparisons, plant trait databases, and other resources (Reichstein et al., 2003), resulting in many openly‐available gridded, global Rs and Rh data sets (S. Chen et al., 2014; Hashimoto et al., 2015; Stell et al., 2021b; Warner et al., 2019; Yao et al., 2021). Advances in remote sensing of soil moisture variability (Y. Liu & Yang, 2022), an important driver of Rs at the biome and ecosystem scale (Hursh et al., 2017) but not typically used in global upscaling efforts, could lead to a more direct consideration of soil moisture in global Rs estimates.…”