Florian Hartig

← Publications

An R package facilitating sensitivity analysis, calibration and forward simulations with the LPJ-GUESS dynamic vegetation model

Maurizio Bagnara, Ramiro Silveyra Gonzalez, Stefan Reifenberg, Jörg Steinkamp, Thomas Hickler, Christian Werner, Carsten F. Dormann & Florian Hartig

Environmental Modelling & Software, 111, 55–60 (2019)
Cite this
@article{bagnara2019r,
  author = {Maurizio Bagnara and Ramiro Silveyra Gonzalez and Stefan Reifenberg and Jörg Steinkamp and Thomas Hickler and Christian Werner and Carsten F. Dormann and Florian Hartig},
  title = {An R package facilitating sensitivity analysis, calibration and forward simulations with the LPJ-GUESS dynamic vegetation model},
  journal = {Environmental Modelling \& Software},
  volume = {111},
  pages = {55--60},
  year = {2019},
  doi = {10.1016/j.envsoft.2018.09.004},
}

DOI: 10.1016/j.envsoft.2018.09.004
Citation statistics: 19 (Google Scholar) · 17 (OpenAlex), as of 29 October 2026

Google Scholar citations per year, as of 21 October 2026 (yearly counts are fetched separately from the total)

Funding

This study was supported through funding by Deutsche Forschungsgemeinschaft (DO 786/8–1).

View article (DOI) Software: rLPJGUESS

Abstract

Dynamic global vegetation models (DGVMs) are of crucial importance for understanding and predicting vegetation, carbon, nitrogen and water dynamics of ecosystems in response to climate change. Their complexity, however, creates challenges for model analysis and data integration. A solution is to interface DGVMs with established statistical computing environments. Here we introduce rLPJGUESS, an R-package that couples the widely used DGVM LPJ-GUESS with the R environment for statistical computing, making existing R-packages and functions readily available to perform complex analyses with this model. We demonstrate the advantages of this framework by using rLPJGUESS to perform several otherwise laborious tasks: first, a set of single simulations, followed by global and local sensitivity analyses, a Bayesian calibration with a Markov-Chain Monte Carlo (MCMC) algorithm, and a predictive simulation with multiple climate scenarios. Our example highlights the opportunities of interfacing existing models in earth and environmental sciences with state-of-the-art computing environments such as R.

What the paper shows and why it matters

(AI-generated)

Dynamic global vegetation models (DGVMs) are process-based simulators of how vegetation and the carbon, nitrogen and water cycles of ecosystems respond to climate change. They are complex and often slow, so studying them takes very many runs: sensitivity analysis (finding which parameters matter), Bayesian calibration (fitting parameters to data while quantifying uncertainty) and uncertainty propagation need thousands to millions of simulations. The statistical methods for this are well known and available in R, MATLAB and Python, yet they are rarely applied to DGVMs. The paper states that, in the authors’ experience, an important reason is the work of connecting models written in lower-level languages such as C/C++ or Fortran to statistical software.

The paper presents rLPJGUESS, an R package that runs the vegetation model LPJ-GUESS from within the statistical software R, and argues that this lets researchers use existing R tools for demanding model analyses with little coding. Model set-up, simulation and output collection all happen in R, and runs can be spread over many processors on a personal computer, a server or a computing cluster. A case study on one forest site in Germany, using weather data from Erfurt and analysing beech, carries out the whole workflow. Screening 57 parameters with a one-at-a-time and a global (Morris) sensitivity analysis took about 5 and 15 minutes using parallel runs. A Bayesian calibration of the 16 most influential parameters with a Markov chain Monte Carlo sampler returned parameter values close to those used to generate the (model-generated) reference data; it took about 118 hours without parallelisation. Ten climate scenarios were then simulated in about 5 seconds on 10 processor cores. Because the reference data were themselves model output, the calibration shows that the workflow works, not that the model fits real observations.

The paper concludes that coupling LPJ-GUESS to R will make it easier to study the model systematically and will facilitate a better integration of the model with data, and that the approach could be adopted for other vegetation models and even other model classes, such as climate or hydrological models. Breier et al. (2024) cite it when presenting an R toolkit for the vegetation model LPJmL, noting that LPJ-GUESS, unlike LPJmL, already had such an interface.

References

  • Breier et al. (2024). lpjmlkit: A toolkit for operating LPJmL and model-specific data processing. Journal of Open Source Software 9, 5447. doi:10.21105/joss.05447