Introduction to DGVM3D

This package renders the vegetation structure of individual and cohort based dynamic (global) vegetation models (DGVM) with gap dynamic. Here I present how to derive the required data with the DGVM LPJ-GUESS [@smith_representation_2001; @smith_implications_2014]. Initial presentation at the EGU in Vienna, Austria, 11. April, 2018.

Code to be part of an external DGVM

So far LPJ-GUESS is the only model included.

C++ code to include in LPJ-GUESS

LPJ-GUESS is hirarchically organized in stands, which contain a certain number of patches, where the vegetation grows. The default output is averaged over those patches, to represent different successional states of vegetation per coarse gridpoint. This library here needs the output of each patch, therefore an additional output module is needed.

To create the required output table in LPJ-GUESS, which is written in C++ a new output module needs to be registered. In the source code the following header and code files need to be placed in the 'modules' subdirectory of the LPJ-GUESS source code. And they have to be added in the CMakeList.txt file at the appropriate position. The files are also located in the R site-library, which can be accessed via system.file(paste0("lpj/modules/vegstructoutput.", c("h", "cpp")), package="DGVM3D").

/// \file vegstructoutput.h
/// \brief Output module for patch based vegetation structure
/// \author Joerg Steinkamp
/// $Date: Thu Nov  3 11:15:37 CET 2016 $
#include "outputmodule.h"
#include "outputchannel.h"
#include "gutil.h"
namespace GuessOutput {
  class VegstructOutput : public OutputModule {
    // implemented functions inherited from OutputModule
    // (see documentation in OutputModule)
    void init();
    void outannual(Gridcell& gridcell);
    void outdaily(Gridcell& gridcell);
    xtring file_vegstruct;
    FILE *out_vegstruct;
/// \file vegstructoutput.cpp
/// \brief Output module for patch based vegetation structure
/// \author Joerg Steinkamp
/// $Date: Thu Nov  3 11:15:37 CET 2016 $
#include "config.h"
#include "vegstructoutput.h"
#include "parameters.h"
#include "guess.h"
namespace GuessOutput {
    REGISTER_OUTPUT_MODULE("vegstruct", VegstructOutput)
    VegstructOutput::VegstructOutput() {
    declare_parameter("file_vegstruct", &file_vegstruct, 300, "Detailed vegetation structure");
  VegstructOutput::~VegstructOutput() {
  void VegstructOutput::init() {
    if (file_vegstruct != "") {
      std::string full_path = (char*) file_vegstruct;
      out_vegstruct = fopen(full_path.c_str(), "w");
      if (out_vegstruct == NULL) {
        fail("Could not open %s for output\n"                         \
             "Close the file if it is open in another application",
      } else {
        fprintf(out_vegstruct, "Lon Lat Year SID PID VID Pft Lifeform LeafType PhenType Pathway Age LAI ShadeType N DBH Height Crownarea\n");
  void VegstructOutput::outdaily(Gridcell& gridcell) {
  void VegstructOutput::outannual(Gridcell& gridcell) {
    if (file_vegstruct == "")
    if (date.year >= nyear_spinup-50) {
      double lon = gridcell.get_lon();
      double lat = gridcell.get_lat();
      Gridcell::iterator gc_itr = gridcell.begin();
      while (gc_itr != gridcell.end()) {
        Stand& stand = *gc_itr;
        while (stand.isobj) {
          Patch& patch = stand.getobj();
          Vegetation& vegetation = patch.vegetation;
          while (vegetation.isobj) {
            Individual& indiv=vegetation.getobj();
            // guess2008 - alive check added
            if ( != -1 && indiv.alive) {
              fprintf(out_vegstruct, "%7.2f %6.2f %4i ", lon, lat, date.get_calendar_year() );
              fprintf(out_vegstruct, " %i ",;
              fprintf(out_vegstruct, " %i ",;
              fprintf(out_vegstruct, " %i ",;
              fprintf(out_vegstruct, " %10s ",  (char*);
              fprintf(out_vegstruct, " %i ",    indiv.pft.lifeform);
              fprintf(out_vegstruct, " %i ",    indiv.pft.leafphysiognomy);
              fprintf(out_vegstruct, " %i ",    indiv.pft.phenology);
              fprintf(out_vegstruct, " %i ",    indiv.pft.pathway);
              fprintf(out_vegstruct, " %4.0f ", indiv.age);
              fprintf(out_vegstruct, " %6.2f ", indiv.lai);
              if (indiv.pft.lifeform == TREE) {
                fprintf(out_vegstruct, " %4.1f ", indiv.pft.alphar);
                fprintf(out_vegstruct, " %4.0f ", indiv.densindiv * patcharea);
                fprintf(out_vegstruct, " %7.2f ", pow(indiv.height/indiv.pft.k_allom2,1.0/indiv.pft.k_allom3));
                fprintf(out_vegstruct, " %8.2f ", indiv.height);
                fprintf(out_vegstruct, " %8.2f ", indiv.crownarea);
              } else if (indiv.pft.lifeform == GRASS) {
                fprintf(out_vegstruct, " %4.1f ", -1.0);
                fprintf(out_vegstruct, " %i ",     1);
                fprintf(out_vegstruct, " %i ",    -1);
                fprintf(out_vegstruct, " %i ",    -1);
                fprintf(out_vegstruct, " %i ",    -1);
              fprintf(out_vegstruct, "\n");
  } // END of void VegStructOutput::outannual
} // END of namespace VegStructOutput

The code writes the following output columns including a header line, if the parameter file_vestruct is defined in the instruction file.

| Column name | Type | Description | |:------------|:---------:|:-------------------------------------------------| | Lon | numeric | longitude | | Lat | numeric | latitude | | Year | integer | year | | SID | integer | Stand ID | | PID | integer | Patch ID | | VID | integer | Vegetation ID; the ID of a cohort or individuum | | Pft | character | Name of the PFT/Species as defined in the guess instruction file | | Lifeform | integer | value of Pft lifeform (enum {NOLIFEFORM, TREE, GRASS}) | | LeafType | integer | value of Pft leafphysiognomy (enum {NOLEAFTYPE, NEEDLELEAF, BROADLEAF}) | | PhenType | integer | value of Pft phenology (enum {NOPHENOLOGY, EVERGREEN, RAINGREEN, SUMMERGREEN, CROPGREEN, ANY}) | | Pathway | integer | value of Pft pathway (enum {NOPATHWAY, C3, C4}) | | Age | integer | value of cohort age | | LAI | numeric | value of cohort leaf area index | | ShadeType | numeric | value of cohort alphar as a measure for shade tolerance class. Is ranked in the package. | | N | integer | number of individual per cohort. In individual mode this is 1. | | DBH | numeric | individual tree diameter (equal for all members of the same cohort) | | Height | numeric | individual height | | Crownarea | numeric | individual crown area |

An additional column 'BoleHeight' (branch free stem) can be given, if not present BoleHeight for broad-leaved trees is 1/3 of 'Height' and 1/4 for needle-leaved trees.

So far not all of those values are used in the code (used ones are maked bold). However, they seemed to be useful in the future. Do not run LPJ-GUESS with too many gridpoints with the above output module. This will create hundreds of GB or even TB of output! An example dataset ('dgvm3d.succession') is included in this package, which contains a named list of 3 data.frames at 3 locations with 12 patches each for the period 1861 to 2006:

DGVM3D workflow

Global options

When the library is loaded a few global default options are set to reduce the number of function parameters passed through several cascading function calls. These options can be modified by dgvm3d.options.

| Name | Value | Function | |------|-------|-----------------------------------------------------------| | patch.area | 1000 | The area of one patch in m^2 | | samples | c(10, 10) | [1] Number of samples to determin next tree position; [2] max. repetition if no suitable position is found | | overlap | 0.5 | fraction of crownradii to overlap with nearest neighbour during establishment | | sort.column | c("Crownarea", "descending") | Determine the position for the largest trees first, then fill the gaps | | establish.method | "random" | How to choose from the above sample for the next tree position ('random', 'row' or 'sunflower' distance) | | color.column | "ShadeType" | which data column to use for the canopy color | | verbose | TRUE | print some messages |

Importing the data

For very large output files I recommend writing an extra input function using data.table, which is way faster than the default data.frames.


The function read.LPJ reads the above output if the output and extend cohorts to N average individuals, which will be rendered in varying shapes and colors based on the other attributes. If LPJ-GUESS was run at several locations, it is recommended to read the gridlist first, and create a list of data.frames. In the example below every fifth year after 1859 is selected for the given locations.

dgvm3d.locations = read.table("gridlist.txt",
                              col.names=c("Lon", "Lat", "Name"), sep="\t",
for (i in 1:nrow(dgvm3d.locations)) {
  dgvm3d.succession[[dgvm3d.locations$Name[i]]] =
  dgvm3d.succession[[i]] = dgvm3d.succession[[i]][!(dgvm3d.succession[[i]]$Year %% 5) &
                                                  dgvm3d.succession[[i]]$Year > 1859, ]

Other models

Please write and contibute your own input function.


For visualization each patch in a stand is represented by a hexagon with an individual number of soil layers as well as variable height. However, the patch size is equal for all patches. This here is a WebGL element, if your browser supports it, otherwise I guess you see nothing.

stand = initStand()

Establish the individual trees

The individual trees are by default randomly distributed within the inner radius of each patch hexagon with establishTrees. The returned data.frame should is the put into the respective vegetation slot of a patch of a stand. There are three algorithms implemented, how the individual trees are distributed:

The later two algorithms can be complemented by setting the option 'jitter' to TRUE and optionally giving 'jitter' parameters for fine tuning. Alternatively additional the columns 'x' and 'y' in the vegetation data.frame can be set with custom coordinates with (0/0) in the center of the hexagon.


stand3D renders the soil hexagons in a rgl window and plant3D renders the trees planted with establishTrees.

veg = data.frame(DBH=rep(0.05, 250))
veg$Height    = veg$DBH * 35
veg$Crownarea = veg$DBH * 5
veg$LeafType  = sample(1:2, nrow(veg), replace=TRUE)
veg$ShadeType = sample(1:2, nrow(veg), replace=TRUE)
veg$LAI = rep(2, nrow(veg))
veg = rbind(veg, data.frame(DBH=-1, Height=-1, Crownarea=-1, LeafType=-1, ShadeType=3, LAI=0.5))
stand = initStand(npatch=3)
stand3D(stand, 1)
dgvm3d.options(establish.method = "random")
stand@patches[[1]]@vegetation = establishTrees(veg, stand@hexagon@supp[['inner.radius']])
stand = plant3D(stand)

stand3D(stand, 2)
dgvm3d.options(establish.method = "sunflower")
stand@patches[[2]]@vegetation = establishTrees(veg, stand@hexagon@supp[['inner.radius']])
stand = plant3D(stand, 2)

stand3D(stand, 3)
dgvm3d.options(establish.method = "row")
stand@patches[[3]]@vegetation = establishTrees(veg, stand@hexagon@supp[['inner.radius']])
stand = plant3D(stand, 3)

rot.z = rotationMatrix(pi/3, 0, 0, 1)
rot.y = rotationMatrix(-pi/8, 1, 0, 0)
rgl.viewpoint(userMatrix = rot.y %*% rot.z, fov=1)

One stand with 3 patches

Overview Applications

With the function snapshot the trees present at the given year and for the given patches will be rendered. The function succession keeps the trees with the maximum distance to the nearest neighbour for each cohort at their position and those trees growing too close together are removed, when the number of individuums decreases. New trees are randomly distributed with the desired method. The removal of killed individuals is done in the function updateStand, which calls establishTrees with the next years vegetation data.frame to distribute new tree saplings.

Temporal snapshots and their application

The succession is often visualized with the Leaf area index (LAI). Starting with grasses in the beginning, then early successional trees (shade intolerant) and later on the late successional trees (shade tolerant). Due to the random component in establishment, mortality and patch distroying disturbance the indivudual patches look quiet chaotic with a low number of repeated patches. However, if averaged over all patches, which is normally done, the expected successional pattern emerges. Allthough still not very smooth here, since we have only 12 patches. Normally a simulation should have at least 50 or better 100 patches (see demo(dgvm3d.ggsuccession)).

location <- 'Canada - boreal forest'
for (y in c(1865, 1915, 2005)) {
  open3d(windowRect=c(0, 0, 800, 600), scale=c(1, 1, 1), FOV=0)
  stand = snapshot(dgvm3d.succession[[location]], year=y)
  rgl.light( theta = -25, phi = 30, specular = "black", diffuse = "#FFFFFF")
  axis3d("z", pos=c(-stand@hexagon@supp$outer.radius, 5*stand@hexagon@supp$inner.radius, NA))
  rot.z = rotationMatrix(pi/6, 0, 0, 1)
  rot.y = rotationMatrix(-pi/3, 1, 0, 0)
  rgl.viewpoint(userMatrix = rot.y %*% rot.z, fov=1)
  rgl.snapshot(paste0("snapshot_", y, ".png"))

In the beginning the patches look very homogen. Location 1 5 years after disturbance However, already after 40 years they grow higher and start to differentiate. Location 1 50 years after disturbance And after 140 years, the patches look very heterogen, they are dominated by different tree types and exhibit an individual age structure (height). Location 1 140 years after disturbance

Succession and fire

If we look at another location (Sahel), we get a grass dominated landscape, where the trees can't really establish beyond the sapling size and with several years of high fire probability.

location <- "Africa - sahel"
dummy = open3d(windowRect=c(0, 0, 900, 400), scale=c(1, 1, 1), FOV=1)
stand = succession(dgvm3d.succession[[location]], = c(2, 8, 10), init.year = 1860, years = seq(1905, 2005, 10))
stand = plant3D(stand)
fire3D(stand, limit=0.2)
rgl.light( theta = -25, phi = 30, specular = "black", diffuse = "#FFFFFF")
rot.z = rotationMatrix(-pi/2, 0, 0, 1)
rot.y = rotationMatrix(-pi/3, 1, 0, 0)
rgl.viewpoint(userMatrix = rot.y %*% rot.z, fov=1.5, zoom = 0.5)

Time series of 4 patches in the Sahel region from 1905 to 2005, every 10 years.

Rotating Animation

The demo('dgvm3d.animation') contains an example with a snapshot of all 12 patches, starting in 1860 rotating around the central z-axis and every 50 frames the year is incremented. Be carefull this takes several minutes. It also includes commented out code to save all pictures, label them and convert them to MP4.


Try the DGVM3D package in your browser

Any scripts or data that you put into this service are public.

DGVM3D documentation built on May 2, 2019, 3:47 p.m.