1.
Capsis progress, current projects, foreign partnerships
(Francois de Coligny, INRA AMAP, Montpellier) (pdf)
Reminder of the Capsis project objectives : to build a software
platform to integrate many growth and yield or dyamics models for
forestry modellers, forestry managers and education. In Capsis, it is
possible to choose a model, choose an initial situation and then build
scenarios with various iterventions (thinning...). The results can be
checked in integrated control tools (viewers, curves, histograms,
tables...) and all the simulated data can be easily exported to other
software for further analysis.
The Capsis software architecture was built to be stable arround a
strong kernel and evolutive thanks to its extension framework. The
developpers and modellers are part of the Capsis codevelopment
community
inside which they share all the source codes, to comply with the Capsis
charter. The charter explains the few rules needed to work efficiently
together and relies on a free licence (LGPL licence) for all the shared
code. The models themselves are the property of their authors and their
code is also shared with people in the codevelopment community, they
may
be distributed outside the community under a licence, free or not, when
the authors decide.
The main points in Capsis : models can have very different data
descriptions, allowing stand level models as well as diameter class
models or even individual based spatialized models or mixt models. It
is possible to integrate models for plantations as well as for
heterogeneous forests. It was even possible to integrate fish dynamics
models as an experience for thematics enlargement. Simulations can be
run in interactive or script modes (for long simulations), the user
interface is bilingual french/english, A grouping system is proposed
for models handling individuals and it is possible to connect with
other simulators in various technical ways when needed. Capsis has now
about 400 extensions integrated.
There were several training sessions in 2006, one major with 11
trainees in february and 2 others, for people of the National Forest
Inventory and for people of the Chinese Academy of Forestry.
The most active projects this past year have been :
- the ArcGIS - Capsis-Eucalypt connection by Véronique Cucchi
and Laurent Saint-André,
- the Cytisus model studying a Scotch Broom / Grass / Sheep ecosystem
by Estelle Chambon-Dubreuil,
- the joining of the French National Inventory with Antoine Colin,
Marie-Dominique Van Damme and Jean-Luc Cousin,
- the mixing of the Fagacées and Sylvestris models to build a
model for heterogeneous situations of Oak and Pine by Grégory
Decélière and Thomas Perot,
- the Dynaclim model for mixing ecophysiology and dendrometry level
models on the basis of the Ventoux model, by Hendrik Davi,
- the Transpoprege model dedicated to transitionnal populations genetic
study over one generation by Sylvie Oddou,
- the integration of a Wood Quality Workshop (WQW) in Capsis, started
in last november in New-Zealand with Dave Pont and Andrew Gordon
(ENSIS) for Radiata pine and continued with Frédéric
Mothe for the Fagacées model (Oak) and Céline Meredieu
and Thierry Labbé for the PP3 model (Maritime Pine),
- the integration of ISGM, the Integrated Stand Growth Model (Tang
ShouZheng) by Hong LingXia of the Chinese Academy of Forestry.
Recent communications, interventions in networks and recent papers
related to work in Capsis were mentionned. A list of the models
integrated or under integration in Capsis was given, by categories of
models : Individual based models (IBM) spatialized or not, Diameter
class models, spatialized or not, Stand level or Agroforestry models,
Fish dynamics models.
Perspectives: a common 2D / 3D representation system in Capsis at the
stand level with Lollypops, do better what we already do, accept new
projects, try to become more international, try to have a second
developper for the Capsis project to help handle all the partnerships
and do needed background work to make it possible to grow more and
better.
2.
Ecological factors shaping the genetic quality of seeds and
seedlings in forest trees: a simulation study coupled with sensitivity
analyses (Sylvie Oddou, INRA URFM, Avignon) (ppt)
This study deals with the reproduction cycle of trees. The evolution in
space matters a lot. Here, interest in demographic and genetic
composition: study the processes that change the demography and genetic
compositions. Experimental calibration of input factors. First project
for the "Bureau de la Ressource Génétique" (BRG),
focusing on regeneration for 6 species: from seed to sappling
demography and genetic evolution in natural regeneration. Creation of a
module in Capsis named Transoprege for example to spread adults at the
beginning of the simulation and simulate regeneration.
Question: how processes affect sapplings genetic quality, how the way
each process is modelled affects the result ? To answer, run
sensitivity analyses with the Morris method to screen the factors that
mostly affect the variance of output variables. It is an economic
method in simulation and computation (Method presentation).
Transpoprege is interested in the dispersal functions and wether they
are light or fat tailed.
The script mode is running correctly, some troubles are under
inspections, related to memory management problems or fine bugs in the
implementation.
Discussion: the sensitivity
analyses do not need a special extension but they need the script mode.
The analysis of the spatial and genetic structure is done outside
Capsis in an existing C program. A connection with the R software could
be interesting. Jacques Labonne and Sylvie Oddou have needs that
Guillaume Cornu and Sylvie Gourlet-Fleury already had in their Selva
module, possible synergy. The studied factors are supposed to be not
linked. The studied parameters are not stochastic but the runs include
stochasticity.
3.
Ongoing computing activities for the use of national forest
inventory data and models within Capsis for predictive assessments of
forest resources and timber harvests at the regional level (Antoine
Colin, IFN, Nogent sur Vernisson) (ppt)
Projection of potential forest resources and wood removals at the
regional level. The National Forest Inventory method changed two ears
ago.
Context: modernization of NFI tools. Models: Age classes (regular
even-aged stands) and Diameter classes (all stands structures). Develop
a new functionnal and user friendly tool to discuss with local forestry
managers. Improve project management and traceability. Integrate the
NFI into the growth modellers community. Ease the use of the NFI data:
270 000 inventory plots, 3 million trees.
Objectives: Forecast short (5 years) and mid-term (30-50 years)
potential timber harvest. Scale: department, regional or national
levels. Decision making tool to enhance timber harvest, keep adequacy
between forest resource and industry, follow up for national policy and
afforestation programs.
A Study: a 3 steps process:
- analyse the current forest resource,
- define appropriate management practices,
- calculate the potential forest resource and wood removals in the
future.
Important: define forest management scenarios with several methods.
NFI would like to have various hypotheses on growth, on CO2, on
rainfalls, etc and be able to use them in the simulator.
Building sets of Forest Management Units ("Domaines d'étude"):
compromise between homogeneity, detail, statistic accuracy. Purpose:
describe the regional forest resource for forestry management purpose.
Assessment of forest resource (from raw NFI data):
Variables: growing stock, net increment, forest area, number of stems,
basal area...
Criteria: region, ownership, stand structure, tree species, age
classes, dbh classes (small, medium, large), slope, accessibility
classes, section (timber, industrial, fuel wood), others (to be defined
by local manager)...
Preparation of input data for growth models: smooth the raw data:
- to prevent improbable data,
- to obtain one age - one data.
Definition of forest management practices:
- business as usual,
- intensified management practices,
- others (discussed with the managers)
Models:
- iterative model for age class distribution, with area, volume, yield
for each class for even aged stands: simple and robust aproach.
- iterative model for diameter class distribution for every kind of
stand structure: assess forest resource, calculate potential wood
removals, growth and thinning.
Edition of automatic pdf reports + additionnal calculations: economic
balance, fuel wood resource, carbon balance, others.
Discussion: NFI will first use
its own models but is interested in the models of the community. To
have the same management practice for all the Forest Management Units
is realistic: generally management for one species with the same rules.
It is possible to use the site index as a decision tool. The
interactive (gui) approach will be used to search best management
practices, then use of the batch mode for large simulations. Within a
project, availability of NFI data can be discussed.
4.
Integrated Stand Growth Model (ISGM) (Hong LingXia, Chinese
Academy of Forestry, Beijing) (ppt1, ppt2)
The Integrated Stand Growth Model was created by Tang ShouZheng in
1994. It is a group of correlated equations including (1) basal area
growth model, (2) density index definition, (3) basal area equation,
(4) self-thinning model, (5) dominant tree growth model (either
Schumacher or Richards model), (6) average tree growth model, (7) form
height model, (8) stand volume model.
The statistical algorithm of nonlinear error-in-variable simultaneous
equations is used to estimate the parameters of the correlated
equations to ensure that the parameter estimation is unbiased and the
equations are compatible.
ISGM was previously integrated into CAF's ForStat software (in chinese)
in two parts: (1) from measurements, calculate the parameters and
(2) simulate the normal stand growth, thinning growth and
density control graph. This second part was integrated into Capsis (in
english).
Measurements data: several plots for each site type with different ages
and densities and some maximum density plots. For each plot, serial
measurement for different ages. If one plot was thinned, before and
after thinning are considered as two plots.
In Capsis: normal growth simulation from plantation or from current
situation. The thinning procedure is integrated. The density control
graph (originaly from Japan) can be opened when needed afterwards.
Outputs:
- Stand growth table,
- Dbh growth curve,
- Density index curve,
- Number of trees curve,
- Total basal area growth curve,
- Tree height growth curve,
- Volume growth curve,
- Volume growth rate (continuous and yearly),
- Density control graph.
Discussion: Many references are
in chinese but also some in english. Capsis is interesting because the
software can be run in english. Tendancy is to focus on individual
level with correlation with stand level.
5.
Dynamic connection between Eucalypt-Dendro and ArcGIS for the
development of a forest management tool: where do we stand now ?
(Laurent Saint-André, Cirad, Montpellier) (pdf)
Context: a 42000 ha plantation of Eucalyptus in Congo (Pointe Noire),
with about 150 clones in more than 1000 stands. The owner (EFC) wants a
dynamic tool for management support and research (models validation).
Eucalypt is a single tree distance independent model based on
allometrical relationships. It is a chain of models:
(1) growth module: inventory (t) -> site index -> dominant height
growth -> basal area growth -> individual tree basal area growth
-> individual tree height -> inventory (t+1). Sum of the
individual heights = stand height, same for basal area. No mortality,
no climate, no fertilisation yet.
(2) wood properties: volumes, ring patterns, biomass of the trees, by
compartments or by ring.
(3) biogeochemical module: nutrients in rings...
The decision making tool is under construction arround the ArcGIS
geographical information system, the Access database and
Capsis-Eucalypt. The GIS is used (not only) as a Graphical User
Interface, Capsis
is run in background in script mode, launched from the GIS with a file
communication. The simulation results are writen by Capsis in the
database.
Why ArcGIS: it is a standard, it was the one used in Congo.
Database: three parts, was recreated for the tool and the plantation
people are
happy of this.
- Kernel DB: stand characteristics, name, clone, density, management
practices...
- Real DB: observations, trees measuremants, environment.
- Simulated DB: results of the simulations under Capsis.
How to?
- From the GIS, the user selects stands and decides to simulate to a
target date with management.
- Capsis is launched from the GIS user interface, retrieves the user
selection in a file.
- It checks if data is available in the database, then generates for
each stand one inventory file.
- Simulation is run with one Capsis project per stand, results are
stored in the simulated DB.
- Capsis writes a finishing report.
- The results can be seen under the GIS.
Other functionnalities: queries to visualise environment information
(ex: fires in the last 10 years), resulting in DB management and forms
creation.
Conclusion: no important
technical problems but a lot of work under Capsis. Positive
interactions with forestry managers, happy with the new DB, will use it.
Future: finalize management practices procedure, make new GIS+DB
applications (new road = how many m3 ?...). The growth model runs only
two clones, the 150 clones must be grouped, 7 groups must be done.
Growth model validation and calibration has begun with good
results. Need for feedback from biogeochemical models and exchanges /
interactions between stands.
User will be able to see the real data and the simulated data with
layers in the GIS, all simulated data have a simulation number. If same
simulation is run again on same stands, results will be replaced. If
simulation parameters change, results will be added.
The owner works by zones (policies, ex: harvest a zone). Overall
silviculture for zones, not stands.
The geodatabase is complex but the technical link was quite easy to do.
All the ArcINFO users are potential users of the system, able to
produce the DB. The system could be reused in Africa by Cirad. Cirad
has other plots in the same area and the same framework could be
reusable.
Local managers do not use Capsis but they use GISs, maybe this way
Capsis could be accepted in an easier way ? New Caledonia users are
interested. The shape of the DB could be reused: it was designed from
two DBs: INRA and Natural Forests teams in Cirad. A good basis for a
generic forest DB.
Transfer: training courses for users about models, how they are fed, in
one week, they are aware of what is a model. Objective: one target
person in the owner's team will exchange with the modeller: an end-user
in the Capsis charter will be an interface to the modeller.
6.
Coupling deer and forest models in Capsis: why and how ? (Tanguy
Daufresne, INRA CEFS, Toulouse) (ppt)
The Toulouse INRA laboratory is interested in the behaviour and ecology
of widlife. Deer closely interact with the ecosystem and forest
populations, deer density is low (hunting). Recently, in some places,
populations raised again due to many factors. How to manage deer
populations in forests?
In France, Roe deer (20-30kg, generalist, selective browser,
territorial) and Red deer (80-300kg (Cévennes), generalist,
selective browser/grazer (graminates), social). Overall populations
have dramatically increased over the past three decades due to local
density increase and recolonisation.
Deer effect on forest dynamics (Rooney et al., 2003):
- direct effect: selective browsing, understorey and tree regeneration:
diversity and density,
- indirect effect: mediation of plant-herbivore interaction or
plant-plant interaction after nutrient recycling.
Forest effect on deer dynamics: habitat quality:
- transfer of soil fertility to foliage quality,
- understorey density,
- hedge effect.
Illustration: impact of Black tailed deer on forest understorey in
Western British Columbia. Islands: colonized by deer 15, 20 years ago
and without deer: interesting comparison. Decrease is measurable for
species richness and biomass density.
Feedback effect of forest structure on deer population dynamics.
- Reduction of understorey biomass due to change in fire regime
entailed decline of deer in Oregon,
- soil fertility influences foliage quality hence body size in roe deer.
Impact of Red and Roe deer on forest successions in Europe. Gap model
with a negative deer effect on recruitment (seedling -> small
sappling (with deer - effect) -> large sappling). Predictions are
model and site specific.
Caveats:
- models do not detail enough the early stages of the life cycle of
trees,
- indirect effects of deer on trees are ignored,
- feedback of forest structure on deer population dynamics are ignored.
Perspectives:
- theoritical "toy models": mediation tree-understorey with deer
population + understorey + one species of canopy tree ; stoichiometric
model of N and P: soil nutrients + deer population + understorey,
- introducing "deer effect" including indirect effect in forest models,
- creating a deer population in Capsis to be coupled with forest models
in Capsis.
Discussion: Need to consider
life cycle, vertical structure of stands and multispecific stands?
Trials to couple animals and forests stands already done few years ago,
see Benoit Courbaud (Cemagref). To couple forest and animal models,
need a seedling and sappling regeneration phase. Considering seedlings
and sapplings, measurements can be done in many sites (ONF).
Challenges: 1. seedling and sapplings and 2. scale up to Landscape =
good forest management scale? See AMAP Landscape platform under
development. Entomologists of INRA URFM in Avignon couple insects with
forests (Jean-Noel Candeau). Seedlings and sapplings already exist in
some Capsis models (Mountain, Samsara, Selva...).
7.
Simulating mixed virtual stands with the "spatial" library:
application to Samsara module (Francois Goreaud, Cemagref LISC,
Clermont Ferrand) (ppt)
The spatial library: to manage more complex stands, need for new
models. Tree individual based models (IBM) with local neighbourhood.
Spatial structure possibly for locations or local neighbourhood.
structure characterises neighbourhood: aggregated, random, regular
structures. Knowing the local conditions, we know they influence the
natural processes and anthropic actions which alter in return the
spatial structure.
IBMs need detailed initial states to model the evolution of each
individual. From aggregate data (N, G, V...), simulate initial
structures (a virtual stand) to feed the IBM.
Crucial issues: (1) How to describe the structure of a stand? (2) How
to simulate it? (3) How does it influence the dynamics? (4) How to take
it into account in the models? (1) and (2) are processed by the spatial
library in Capsis: to characterise and to simulate initial states.
Pure and regular stands: One population only: a set of points:
Characterised by Clark and Evans index (distance to nearest neighbour)
or Ripley L(r) function (number of neighbours at distance < r):
Examples:
- clumped: CE = 0.57, L(r) positive,
- random: CE = 1.05, L(r) near zero,
- regular: CE = 1.44, L(r) negative.
Simulated by Poisson, Neyman Scott or Gibbs.
Mixed and unevened stands: different populations = different sets of
points, with high variability of spatial structure and interactions
between populations: repulsion, independance, attraction.
Characterisation: define the populations (species, dbh...),
characterise the structure of each population and the relative
structures between populations. Two additional methods in the spatial
library: Inter population CE12 index and Intertype function L12(r).
Examples:
- repulsion CE12 = 3.36, L12(r) negative,
- independance CE12 = 1.89, L12(r) near zero,
- attraction CE12 = 0.65, L12(r) positive.
Simulation: intertype Gibbs process (ex: from younger to older trees).
1. number of sub populations
Then species code ex: 1 for Spruce, dbh, height, wether location is
dependant from others or not (first is inependant) -> Random, Neyman
Scott or Gibbs ;
Then species code ex: 2 for Fir, dbh, height, related to Spruce (ex:
repulsion).
-> Aggegates of Spruce, Fir far from Spruce, etc.
-> Check with L12(r) function.
Applications to Benoit Courbaud's Samsara module (Corentin Lhuillier).
- Spruce-Fir mixed stands,
- there is a sensibility to initial spatial structure with pure regular
stands,
- what about initial mixed stands?
Using a real plot (example plot from Benoit Courbaud): Saisie, 1ha, 375
spruce, 53 fir ; change locations only of trees: keep spruce locations
and change fir, keep fir location and change spruce, change
independance, repulsion, attraction.
Results on spruce regeneration: sapplings in Samsara, the evolution of
young trees is different according to initial structure due to
availability of light. Changing the location of fir (with repulsion
from spruce) may lead to less openings, less light and less
regeneration.
Results on fir regeneration: fewer fir in the stand and also
variability due to the initial state influence.
Basal area changes: few influence for spruce (lot of trees) but lower
basal area for fir in case of fir attraction because fir experiments
more competition by spruce (higher basal area in case of repulsion).
Next steps: improve spatial lib, add options, improve user interfaces,
possibly develop new tools, how to make it easy to integrate the
spatial tools in the modellers work (interact...) ? Simulate realistic
initial states: PhD of Marie-Ange Ngo-Bieng.
Discussion: M.-A. Ngo-Bieng
showed that spatial structure was highly variable from one stand to
another. There is a need first to characterize this spatial structure
into a typology and then for each type, create a point process to
simulate as realistic as possible virtual stands. The oscillations in
the Samsara graphs may be an artifact resulting of the regeneration
process in the mountain module because seeds are not produced every
year, should be checked. The competition in Benoit Courbaud's model
depends on the amount of light: growth depends on the light the tree
receives, light is explicit: light rays.
8.
A logging module for oak connected to the Fagacees model
(Frédéric Mothe, INRA LERFoB, Nancy) (pdf)
Goal: improve the quality assessment of the modelled trees by
considering a logging strategy and target products. Simulate the
logging process using stem taper for the Fagacées, PP3 and
Eucalypt models (Fagacées for collaboration with the french
CTBA, Emmanuel Bucket). There is also a generic logging module in
progress.
Oak: logging oriented towards a list of products: sliced veneers,
staves (barrels), unedged sawing, sawing, LVL, particle boards, fire
wood, with a priority.
Selection using the data available in Fagacées: growth rings,
sapwood/heartwood, knotty core, bark, crown base.
Each product: several criteria established in connexion with CTBA:
- number of logs (ex: max 1),
- length (min / max values),
- diameter (medium or small end diameter),
- criterium linked with wood quality (ex: width of heartwood free of
nodes).
Each tree is considered from bottom to the top. Outputs are mainly text
files: tree data file, log data file (may be big), Dbh class: result by
product. To be completed with prices.
Experimental generic logging module: same but no product names: p1,
p2... with basic rules for geometry (dimension, length...) depending on
the growth model detected capabilities. May be completed. Possibly
define custom rules: write new java classes. Minimal requirements for
using Geolog: the method named getTreeRadius_cm (GTree t, double
height_m, boolean overBark). If the method getCrownBaseHeight () can be
found, then the knotty core information is made available.
Discussion: No file export
towards WinEpifn. This connection could be hard to do because branches
are important to have knots for quality classes.
DEMO+CHAT: The Wood Quality Workshop in Capsis (F. Mothe)
Fagacées simulation with evolution according to a sylvicultural
scenario from an initial rdi of 0.4 and a final rdi of 0.6 with a final
age of 180 years. Graph showing the evolution of rdi over time.
Opening the Wood Quality Workshop on a chosen step in the scenario (ex:
final step), make a tree selection: the listed trees are the trees
alive on the reference step, with possible addition of the thinned
trees.
Choose Geolog for Oak. Sort the list of products by decreasing
priorities, check options to save in text files and keep logs in
memory. Validation launches the job for the selected trees. Results can
be seen in the Logs tab with an inspector (see all the values of the
logs properties) or a Viewer in 2D.
The Generic Logger uses generic products (+ particle and fire) and
generic criteria.
9.
Interactions between gene flow and adaptive divergence in
populations of guppies (Poecilia reticulata): an individual based
modelling approach (Jacques Labonne, INRA ECOBIOP, Saint Pée sur
Nivelle) (pdf)
Theory of evolution predicts that ecological speciation arises when one
species is distributed in two contrasted environments and when
adaptation occurs. Adaptation is the process that improves the fitness
of resident individuals compared to immigrant individuals.
Guppies populations in the watersheds of Trinidad are a good example to
test this prediction: watersheds present contrasted environment, where
predation occurs strongly downstream, and is usually absent upstream.
These environment thus provides different selection pressures, and are
separated by obstacles such as waterfalls. The distribution of male
phenotypes differs between these environments: males display colorful
patterns upstream in Low Predation sites (LP), whereas they are far
more cryptic downstream in High Predation sites (HP). This is the
result of a trade-off between sexual selection that favors male colour,
and natural selection that counter selects colourful males. Such a
divergent selection thus leads to phenotypic adaptation in the wild,
and transplant experiments have shown that this process could be
reproduced.
In order to produce ecological speciation, this adaptation is expected
to limit gene flow efficiently between HP and LP sites, by building pre
and post zygotic reproductive barriers. Still, a recent study by Crispo
et al. (2006) found no genetic isolation due to difference in predation
between sites. Our purpose is to model the evolutionary process of
adaptation in guppies population, accounting for neutral and
non-neutral genetics, environmental contrasts, and behaviour. Thanks to
the Genetics Library provided by CAPSIS, we are able to model
explicitly the transmission of phenotype. The model utilizes various
sets of sexual and natural selection functions (based on real data),
and evaluates the potential for adaptation in these populations under
different dispersal conditions between environments. We aim at
demonstrating that 1) phenotypic adaptation can occur under various
gene flow conditions, 2) adaptation may also prevent gene flow on the
long term, but very strong diverging selection is required, such
conditions are not likely to be found in the wild on a long term scale.
This model is still in progress, but examplifies a realistic approach
for evolutionary problems, accounting for individual variation,
behaviour and phenotype, genetics, demography, environmental
contrasts.
DEMO+CHAT: The fish models in Capsis (J. Labonne)
Demo of the Guppy and Bidasoa modules in Capsis. Bidasoa has one
end-user in Spain (the Spanish region partner in the project). Demo of
the Dynet module with the river network generator, characterised by
indices.