Florian Hartig

← Publications

r3PG – An R package for simulating forest growth using the 3-PG process-based model

Volodymyr Trotsiuk, Florian Hartig & David I. Forrester

Methods in Ecology and Evolution, 11(11), 1470–1475 (2020)
Cite this
@article{trotsiuk2020r3pg,
  author = {Volodymyr Trotsiuk and Florian Hartig and David I. Forrester},
  title = {r3PG – An R package for simulating forest growth using the 3-PG process-based model},
  journal = {Methods in Ecology and Evolution},
  volume = {11},
  number = {11},
  pages = {1470--1475},
  year = {2020},
  doi = {10.1111/2041-210X.13474},
}

DOI: 10.1111/2041-210X.13474
Citation statistics: 59 (Google Scholar) · 48 (OpenAlex), as of 29 October 2026

Google Scholar citations per year, as of 20 October 2026 (yearly counts are fetched separately from the total)

Funding

This study was supported through funding by WSL internal project 'Forecasting forest growth and carbon sequestration: current and novel species mixtures, climatic conditions and management regimes'; SwissForestLab QUPFiS project; Bavarian Climate Research Network bayklif, research project BLIZ.

View article (DOI) Software: r3PG

Abstract

Process‐based forest models (PBMs) are important tools for quantifying forest growth and vulnerability, particularly under climate change. The 3‐PG model (Physiological Processes Predicting Growth) is one of the most widely used forest growth simulators for this purpose worldwide. Here, we present r3PG, a new Fortran implementation of 3‐PG, wrapped into an r package. r3PG can simulate monospecific as well as mixtures of evergreen and deciduous tree species in even‐aged or uneven‐aged stands. The combination of Fortran functions with an r interface makes the model extremely fast. This facilitates the use of r3PG for extensive computer experiments and sensitivity analysis. We demonstrate this in a case study including (a) single model runs; (b) a sensitivity analysis and a full Bayesian calibration of the model and (c) spatial simulations of forest growth across Switzerland. r3PG is faster and easier to use than previous implementations of 3‐PG in visual basic. We believe that this will make 3‐PG even more useful and popular for ecologists and climate change scientists.

What the paper shows and why it matters

(AI-generated)

Forest growth models are used to project how forests develop, particularly under climate change. The paper takes up 3-PG (Physiological Processes Predicting Growth), a process-based model of stand development that the paper calls one of the most popular worldwide. The available implementations, however, did not fit the R-centred workflow of data-oriented ecologists. The original Visual Basic versions, including the extension to deciduous, uneven-aged and mixed-species stands, did not easily interface with R and were relatively slow, which made calibration and sensitivity analysis challenging. Fortran re-implementations existed, but they were not designed for end users, were not on CRAN (the central R package archive) and lacked those extensions. The problem the paper takes on is therefore a tool gap: a fast version of the model, with its extensions, that can be run from within R.

The study presents r3PG, an R package that runs a Fortran re-implementation of the 3-PG forest growth model, and argues that this makes the model, including its mixed-species extensions, fast and convenient enough for computationally heavy work from within R. The package re-implements all major published sub-modules and functions of the model and the major available extensions, and unit tests show results numerically identical to the Visual Basic implementations. The paper states that r3PG is faster and easier to use than those versions; no timing comparison is reported. Three case studies demonstrate the functionality. The first simulates a mixed beech and Scots pine stand. The second applies a Morris screening, a method that ranks parameters by their influence on model fit, to a Norway spruce site, then calibrates the most influential parameters against observed stand data in a Bayesian framework. The third simulates spruce stem biomass on a grid across Switzerland, with uncertainty.

The paper concludes that the package makes sensitivity analysis, Bayesian calibration and parametrisation for new tree species faster and easier, including for mixed and uneven-aged stands. Running the model directly from R, it argues, also supports transparent reporting of the computational protocol and reproducible simulations. As an example of use by others, Nölte et al. (2023) ran the model through r3PG to estimate the productivity of tropical reforestation under climate scenarios.

References

  • Nölte et al. (2023). Sharp decline in future productivity of tropical reforestation above 29°C mean annual temperature. Science Advances 9, eadg9175. doi:10.1126/sciadv.adg9175