Increased tree mortality has become apparent in recent decades across the world. These tree mortality events have been attributed to changes in drought and fire regimes driven by climate change. As warming and drying is projected to continue in the coming decades, more frequent or severe droughts and fires pose a major threat to the stability of biomass C stocks. However, there is still so much we do not know – or cannot predict – when it comes to tree mortality. A mechanistic understanding of drought responses, drought legacy effects and drought-driven tree mortality results in gross underestimation of biomass carbon loss by Earth System Models (ESMs). The same holds for the representation of fires in ESMs, which do not fully simulate rapidly changing fire regimes and extreme fires in many regions, including the Arctic and Boreal zones. This knowledge gap results from the fact that satellite-observations at moderate resolution until now could only capture large disturbances.
Leveraging the power of new high- and very high-resolution satellite imagery combined with machine learning approaches, we will identify the factors explaining forest vulnerability to drought, heat stress and fire. Taking as ‘real world experiments’ recent droughts and fires worldwide we will use deep-learning methods to identify immediate and lagged mortality areas and C loss at tree and ecosystem levels. We will then develop machine learning models to upscale biomass losses from tree mortality and fires regionally and assess post-disturbance recovery dynamics, and use explainable machine learning approaches to identify relevant predictors of fires and mortality. In a second step, this information will be used to improve the representation of climate induced carbon losses from tree mortality and subsequent recovery in the participating ESMs in CALIPSO.
Soil microbes control the decomposition and stabilisation of organic matter, through their carbon use efficiency (CUE: ratio of microbial growth to microbial C uptake) and other traits. In current ESMs, CUE is assumed to be constant, whereas in the real world it varies depending on complex microscale interactions between microbes, substrate chemistry, and the environment—especially temperature and soil moisture. Constraining microbe-mediated soil C loss in response to warming is important globally, but especially critical for massive soil C stocks in northern peat and permafrost soils, where microbes adapt rapidly. Thawing permafrost and peat drainage are projected to weaken and even reverse the global C sink, but we do not know to what degree and how microbial adaptation modulates these processes.
To quantify soil C vulnerability, we will: 1) construct new microbial CUE and trait databases and integrate them into microbial models, 2) model CUE adaptation to combined warming and soil water changes using novel eco-evolutionary approaches, and 3) upscale those models for ESMs. For peat and permafrost, we will map soil C heterogeneity and represent the sub-grid diversity of C-rich landforms in ESMs based on measured C pools associated with contrasting decomposition rates. We will also include the nonlinear physical processes that increase the availability of organic C for microbes and trigger abrupt soil C losses.
The sinking of carbon from the ocean surface to depth by organic particles keeps the concentration of atmospheric CO2 about 200 parts per million (ppm) lower than it would be otherwise, with geological evidence suggesting this flux has shut down in the past. So far, ESMs have assumed that changes in this flux are controlled by the effect of ocean physics on phytoplankton production through nutrient and light limitations, but the biologically-mediated feedbacks are largely missing in current ESMs. The key challenges in modelling marine biota lie at the two extremes of the particle size structure: viruses and zooplankton. These organisms are key determinants of the fraction of organic carbon which gets remineralised in the surface ocean (and thus can be outgassed back to the atmosphere), versus that which sinks to the intermediate and deep ocean (and thus contributes to the storage of carbon in the ocean in the long term). Our focus in CALIPSO will thus be to develop new components for representing the diversity and activity of virus and zooplankton communities in ESMs.
To identify and model the controls of C recycling in the ocean surface and C export to depth by viral and zooplankton activity, we will make use of the >200 million images in EcoTaxa, a database of size, morphology, and transparency of all particles in each image, complemented by omics and satellite data. We will introduce virus infections and the virus-host interactions in modelled ecosystems, and assess the influence of warming on microbial processes and recycling of organic carbon. We will also reproduce the effects of diverse zooplankton varying over three orders of magnitude in size (from 5 to 5000 μm), on C export to the deep ocean. We will create new data metrics for the evaluation of marine processes, and introduce ML-emulators to speed up model simulations to enable large ensemble and long ESM simulations.
We will share theoretical advances and methodologies across the three challenges.
In particular, we will establish new hypotheses for marine ecosystems by applying ecological theories from plants and soil microbes, and vice versa
We will organize open scientific events with a larger community of experts in ML methodologies, to share and benefit from the application of advances made across Challenges 1-3.
In parallel, the new processes from challenges 1-3 will be integrated into the IPSL-CM7 Earth System Model.
We will conduct suites of dedicated experiments with the IPSL ESM to quantify the interactions and contributions of CALIPSO model improvements to carbon-concentration, carbon-climate and climate-carbon feedbacks.
We will then assess the risks of climate change-induced C losses from plants, soils and changes in marine ecosystems, and explore the fate of C under different future scenarios of anthropogenic CO2 emissions over the 21st century.
