Microsoft Research Terrestrial Carbon Model

Inspiration | Features | Download | Case studies | People | Acknowledgements

Predict terrestrial carbon potential, anywhere

The Microsoft Research Terrestrial Carbon Model package contains a model designed to make probabilistic predictions of terrestrial carbon in the year 2000 for anywhere on the global land surface (it is currently awaiting a much better name!). It also includes the multi-component model engineering and refinement framework that was used to infer the parameters to the model and the training data used to infer the parameters to the model. It also show cases how several of our tools can be integrated for academic research purposes.


Currently one of the largest single sources of uncertainty in global climate models is the future of carbon on the global land surface in vegetation and soils. In recent years vegetation (mostly forests) has soaked up 30% of human CO2 emissions, but models disagree drastically about the future of this ‘carbon sink’: will vegetation soak up ever larger amounts of carbon, acting as an ever-stronger brake on climate change? Or will vegetation become a significant carbon source, accelerating climate change which in turn would demand much more radical policy action to reduce CO2 emissions now? These alternative scenarios have been predicted by contemporary global vegetation models but unfortunately technical obstacles make it practically impossible to compare existing models and identify what the most likely future scenario is. The Microsoft Research Terrestrial Carbon Model is our first step towards solving this problem, and the wider problem of developing predictive models of complex natural systems to allow robust measures of confidence in their predictions.


  • Source code and data used in our Biogeosciences research paper
  • Fully-working research example of integrating FetchClimate, Dmitrov and Filzbach
  • Training and evaluation data on terrestrial carbon stocks and flows at equilibrium
  • Output files featuring spatial predictions and parameter distributions
  • User Guide


The Microsoft Research Terrestrial Carbon Model, containing the model engineering framework, terrestrial carbon model, and derivative data necessary to reproduce the results of this paper, as well as the user’s guide is available from here.  The data resulting from conducting the analyses described in this study is available from here. The User Guide is avalable from here. See a Layerscape tour of the study here (requires WorldWide Telescope - free)

Case studies

  • The Computational Ecology and Environmental Science (CEES) group used the Microsoft Research Terrestrial Carbon Model to characterise the climate dependence of the terrestrial carbon cycle.
  • The Microsoft Research Terrestrial Carbon Model was used to model terrestrial vegetation, in particular the biomass of evergreen and decidiuous leaves, in the Madingley Model.
  • Matthew Smith has used the Microsoft Research Terrestrial Carbon Model to estimate the potential carbon credits that could be generated from land use changes and prevention of deforestation for the global land surface (article in review)
  • Isabel Rosa and Sadia Ahmed have used the predictions of the Microsoft Research Terrestrial Carbon Model to estimate various consequences of past and future land use change decisions in the Amazon rainforest (articles in preparation).


The Microsoft Research Terrestrial Carbon Model was developed by the CEES group at Microsoft Research, Cambridge.


We thank the Machine Learning and Perception group at Microsoft Research Cambridge for technical assistance. We thank Florent Mouillot, Ian Wright, Peter Reich, H. Gibbs, Elaine Matthews, Rob Jackson, Nathan Stephenson, Philip van Mantgem, Takeshi Ise, the Joint Research Centre of the European Commission, Springer, John Wiley & Sons, the American Geophysical Union, Oak Ridge National Laboratory Distributed Active Archive Centre (ORNL-DAAC) and the Climate Research Unit and the University of East Anglia, for their kind permission in allowing us to make derivatives of their data available with MSRTCM, and the same authors as well as Mac Post, Robert Hijmans, the Centre for Sustainability and the Global Environment (SAGE) and The IPCC Data Distribution Centre for 15 making their data available for us to use in the associated research project research. We thank G. Mace, A. Friend, P. van Bodegom, Neil Crout, Pete Smith, Jessica Bellarby, Tim Lenton, Ben Adams, Kirsten Thonicke, Stephen Beckett and the GREENCYLCES2 members for providing advice for our study. We thank Pablo Tapia, Carole Boelitz, Chuck Needham, Rachel Free and Rob Knies at Microsoft Research for technical assistance.


  • User Guide [pdf]


Contact Us Terms of Use Trademarks Privacy Statement Microsoft Corporation. All rights reserved.