Approaching the potential of model-data comparisons of global land carbon storage

Impact Assessment of the Ms7.0 Earthquake on Jiuzhaigou Valley from the Perspective of Vegetation Net Primary Productivity

In order to assess the impact of the M_s7.0 Jiuzhaigou earthquake that occurred on 8 August 2017 on vegetation, the Carnegie-Ames-Stanford Approach (CASA) model was adopted to estimate the vegetation net primary productivity (NPP) of Jiuzhaigou Valley, one of the World Heritage Sites, in July, August and September from 2015 to 2019. Then the characteristics of the impact of different earthquake-induced geohazards on vegetation were discussed, and a vulnerability-resilience assessment system concerning the seismic intensity was proposed. The results show that the NPP_max and NPP_mean values in Jiuzhaigou Valley first decreased and then increased and were 151.5-261.9 gC/m² and 54.6-116.3 gC/m², respectively. The NPP value of more than 70% area was 90-150 gC/m² in July. In August, the NPP_mean values decreased, and the areas with lower values became larger; the NPP_mean values of most areas affected by geohazards were 60-150 gC/m². During the earthquake, the NPP_mean values of areas hit by geohazards sharply declined by 27.2% (landslide), 22.4% (debris flow) and 15.7% (collapse) compared with those in the same month in 2016. Vegetation in debris flow zones showed a stronger recovery, with a maximum NPP value increase of about 23.0% in September 2017. The vegetation gradually recovered after the earthquake, as indicated by the uptrend of the NPP values in the corresponding period in 2018 and 2019. In general, the reduction magnitude of NPP values decreased year by year in comparison to that in 2015 and 2016, and the decrease slowed down after the earthquake. The vulnerability and resilience index corresponding to the three seismic intensity ranges were 0.470-0.669 and 0.642-0.693, respectively, and those of Jiuzhaigou Valley were 0.473 and 0.671, respectively. The impact coefficient defined to represent the impact of the earthquake on NPP was 0.146-0.213. This paper provides a theoretical reference and guidance for the impact assessment of earthquakes on the ecosystem.

Deep Learning Optimizes Data-Driven Representation of Soil Organic Carbon in Earth System Model Over the Conterminous United States

Soil organic carbon (SOC) is a key component of the global carbon cycle, yet it is not well-represented in Earth system models to accurately predict global carbon dynamics in response to climate change. This novel study integrated deep learning, data assimilation, 25,444 vertical soil profiles, and the Community Land Model version 5 (CLM5) to optimize the model representation of SOC over the conterminous United States. We firstly constrained parameters in CLM5 using observations of vertical profiles of SOC in both a batch mode (using all individual soil layers in one batch) and at individual sites (site-by-site). The estimated parameter values from the site-by-site data assimilation were then either randomly sampled (random-sampling) to generate continentally homogeneous (constant) parameter values or maximally preserved for their spatially heterogeneous distributions (varying parameter values to match the spatial patterns from the site-by-site data assimilation) so as to optimize spatial representation of SOC in CLM5 through a deep learning technique (neural networking) over the conterminous United States. Comparing modeled spatial distributions of SOC by CLM5 to observations yielded increasing predictive accuracy from default CLM5 settings (R ² = 0.32) to randomly sampled (0.36), one-batch estimated (0.43), and deep learning optimized (0.62) parameter values. While CLM5 with parameter values derived from random-sampling and one-batch methods substantially corrected the overestimated SOC storage by that with default model parameters, there were still considerable geographical biases. CLM5 with the spatially heterogeneous parameter values optimized from the neural networking method had the least estimation error and less geographical biases across the conterminous United States. Our study indicated that deep learning in combination with data assimilation can significantly improve the representation of SOC by complex land biogeochemical models.

Constraining modelled global vegetation dynamics and carbon turnover using multiple satellite observations

The response of land ecosystems to future climate change is among the largest unknowns in the global climate-carbon cycle feedback. This uncertainty originates from how dynamic global vegetation models (DGVMs) simulate climate impacts on changes in vegetation distribution, productivity, biomass allocation, and carbon turnover. The present-day availability of a multitude of satellite observations can potentially help to constrain DGVM simulations within model-data integration frameworks. Here, we use satellite-derived datasets of the fraction of absorbed photosynthetic active radiation (FAPAR), sun-induced fluorescence (SIF), above-ground biomass of trees (AGB), land cover, and burned area to constrain parameters for phenology, productivity, and vegetation dynamics in the LPJmL4 DGVM. Both the prior and the optimized model accurately reproduce present-day estimates of the land carbon cycle and of temporal dynamics in FAPAR, SIF and gross primary production. However, the optimized model reproduces better the observed spatial patterns of biomass, tree cover, and regional forest carbon turnover. Using a machine learning approach, we found that remaining errors in simulated forest carbon turnover can be explained with bioclimatic variables. This demonstrates the need to improve model formulations for climate effects on vegetation turnover and mortality despite the apparent successful constraint of simulated vegetation dynamics with multiple satellite observations.