Skip to main content

A Flexible, Data-driven Model Framework to Predict Soil Responses to Land-use and Climate Change

Periodic Reporting for period 1 - FlexMod (A Flexible, Data-driven Model Framework to Predict Soil Responses to Land-use and Climate Change)

Reporting period: 2020-05-01 to 2022-04-30

Soil organic matter is the largest land carbon (C) pool, vulnerable to land-use change and climate change. Given initiatives to increase land C storage such as ‘4 per mil’, the Bonn Challenge, and UN REDD, there is now a critical need for robust soil C stock change predictions to evaluate the effectiveness of soil management decisions, in the presence of climate change affecting decomposition processes. Yet, soil C models have large disparities and uncertainties in their projections due to an inadequate model structure to relate to existing measurements.

In this project, the first objective is to develop a soil carbon modeling framework that incorporates measurements and their uncertainties from long-term warming and management experiments. The aim is to use standard statistical and Bayesian methods to calibrate the Millennial model at the site scale as well as using soil C pools that are changing decadally in response to management. The second objective is to incorporate the calibrated Millennial model into the Earth System global land surface model (ORCHIDEE) developed and used at LSCE, the host institution, for building an ensemble of future projections on decadal scales.

The main delivered products are a state-of-the-art soil C cycling model representing measurable pools which is publicly available and constrained by data at global scales, as well as advances in understanding about the capacity of soils to store C and their sensitivity to climate and land management.
The Fellow developed scripts to run the Millennial model which are publicly available on Github under an MIT license. In order to find the best function to represent the relationship between mineral surface area and soil organic matter, the Fellow conducted a synthesis study using 400+ sorption experiments upscaled to a global estimate of sorption capacity using relationships with environmental variables (Abramoff et al. 2021). The Fellow calibrated the temperature response using the ForHot Icelandic experiment (in prep) and presented these results at the ForHot Consortium Annual Meeting in March 2021. In collaboration with the group that runs the BBSFA experiment, the Fellow used a similar model to Millennial with representation of mineral surface area and microbes coupled to an Earth System Model similar to ORCHIDEE to simulate the depth profile and temperature sensitivity of SOC stock at BBSFA. This work is in press in a book chapter describing multi-modeling tools for soil C modeling. To better understand data needs for constraining model parameters, a simulated data experiment was conducted with a coupled C and N microbial physiology model using a Bayesian statistical framework to identify which types of data best constrained the parameters in a general microbial soil decomposition model (Saifuddin et al. 2021).

The Fellow conducted a global-scale evaluation of Millennial V2 using measurements of soil C fractions at 1300+ sites (Abramoff et al. 2022). The Fellow fit Millennial V2 with and without croplands to determine the effect of agricultural management on model parameters. The Fellow developed a novel method for estimating total litter input using a data product of aboveground litterfall combined with a separate data product estimating the partitioning between above and belowground plant parts. The global evaluation compared Millennial V2 model output to many of the datasets initially identified in Task 2.2 as well as more detailed datasets gathered to evaluate partitioning of C stock into different fractions and the mean residence time. One of the global datasets of soil fraction measurements used is a new synthesis which is currently under consideration at Nature Communications.
The Fellow considered using ORCHIDEE model output as inputs to the model. However, the Fellow identified data products that are more closely related to measurements than is ORCHIDEE model output. Nevertheless, the Fellow is developing an interface system to couple Millennial V2 and ORCHIDEE that goes beyond what was promised in WP2 because rather than running the model offline, the interface will allow data to be passed between Millennial V2 and ORCHIDEE at each timestep and include coupled C and nitrogen cycling. This work will be completed by postdoctoral fellow E. Bruni who will be co-advised by the Fellow and B.Guenet under the H2020 Grant No. 101000289.

CMIP5 simulations were not yet conducted due to the project ending 4 months early, as the Fellow secured a permanent position as an Associate Scientist at ORNL, USA. However, projections of the coupled ORCHIDEE-Millennial in response to changing climate and land use is planned in collaboration with E. Bruni as mentioned above. Relevant to Tasks 2.2 and 2.3 multiple models were fitted including Millennial V2 to decadal-scale soil C stock changes measured at 17 agricultural sites to determine how much additional C is needed to meet the goal of increasing soil C concentration by 4‰ per year, in prep for Global Change Biology (Bruni et al.). Over the time period of the Fellowship, the Fellow has published 2 first author papers, and 1 corresponding author paper. This represents a total of 3 published papers, 1 book chapter in press, 1 book chapter in revision, 3 papers in review, and 4 papers in advanced prep.
The delivered results of the project include a new understanding of C sequestration potentials as well as explicit quantification of the global potential for organic matter sorption to minerals. The Fellow developed and made public the first soil C model capable of simulating measurable pools, evaluated at the global scale using multiple global-scale data products. Though not completed due to the early end of the project, the fully integrated ORCHIDEE-Millennial will be among the first ESMs with a realistic microbial representation of decomposition and modeled C pools that can be matched with observed pools.

The Fellow disseminated her research via publication, her Twitter account, and invited talks at conferences (e.g. American Geophysical Union) and research institutions.

The Millennial model is already being used within multi-model ensemble tools to inform decision makers and improve model fitting processes, as well as provide uncertainty bounds to projections based on differences in model structure. In addition to the example described above, the Millennial model code has been forked 9 times from the original Github repository by other research groups and is being developed into model ensemble tools in Europe and the US. The ability to update projections when new data become available is a fundamental advance that allows modelers to accelerate the pace with which they improve prediction accuracy.