vsd+ training session, indianapolis 2014 vsd+ props gert jan reinds
TRANSCRIPT
VSD+ training session, Indianapolis 2014
VSD+ PROPS
Gert Jan Reinds
VSD+ tool set
VSD
o dynamic modeling of soil acidification
o soil eutrophication (N availability)
o carbon sequestration
VSD+ tool set
VSD+ (VSD + explicit C and N modeling)
o dynamic modeling of soil acidification
o soil eutrophication (N availability)
o carbon sequestration
VSD+
VSD+ tool set
VSD+abiotic conditions for vegetation
input of fresh organic material
temperature, moisture
MetHyd(hydrology, modifying factors)
vegetation model
(PROPS)
GrowUP (growth, litterfall
and uptake)
VSD+ tool set
VSD+
MetHyd(hydrology, modifying factors)
GrowUP (growth, litterfall
and uptake)
vegetation model
(PROPS)
How to prepare input for VSD+
VSD+ input
• essentialo hydrologyo uptake of N and BC, and input of fresh organics
• optional
• maintain as default
• need calibration
• period
• thick
• bulkdens
• CEC
• pCO2fac
• cRCOO
• deposition
• X_we (non calcareous soils)
• parentCa (calcareous soils, default = -1)
Essential
thick should be depth of rooting zone:
0.5 - 1 m for forestapprox. 0.25 m for grasslands
start before first obs. (> 10 yrs)
if bsat_0 = -1 start at low deposition period
total deposition (as in EMEP),not throughfall (as in measurements)
In VSD+ Help: How to calculate total deposition from throughfall and bulk deposition.
Hydrology
• temperature (TempC)
• average moisture content (theta)
• precipitation surplus (percol)
• modifying factors for mineralisation, nitrification and denitrification (rfmiR, rfnit, rfdenit)
alternative: use MetHyd tool
Uptake and input of organic material
• net uptake of Ca, Mg, K (Ca_upt, Mg_upt, K_upt)
• total uptake of N (N_gupt)
• input of organic C and N (Clf, Nlf)
for forests you can use the GrowUp tool
Optional
• bsat_0 (ECa_0/EMg_0/EK_0)
• Nfix
if not given (default = -1):
bsat_0 in steady state with initial deposition
only necessary for areas with very low N inputs (e.g. north Scandinavia)
Defaults
• kmin_x
• frhu_x
• CN_x
• expAl
• RCOOpars
organic C and N turnover
parameters for protonation of organic acids
(default if ‘RCOOmod’ = Oliver)
exponent for H+ in Al (hydr)oxide equilibrium(default = 3)
exchange constants
means and st.dev. in Mapping Manual (soil types)
Calibrate
■lgKAlBC
■lgKHBC
■lgKAlox
■Cpool_0
■CNrat_0
equilibrium constant for Al (hydr)oxides
mean = 9, st.dev. = 1initial Cpool size and C/N ratio- give values if observation during
large period- calibrate if few observations
Methyd
GrowUp
tool to calculate:
- uptake of N, Ca, Mg and K
- input of C and N from litterfall and root turnover
for forests only
includes management actions (planting, thinning, clear-cut)
two forest types:
- uniform age
- mixed uneven aged (natural rejuvenation)
Demo VSD+ straightforward runs
PROPS; model for computing species occurrence probabilities
Based on a data base with 3400 sites from NL, AT, IR, (UK, DK, ICP Forest) with observed plant species composition and measured abiotic conditions (pH, C/N) etc.
Temperature and precipitation: climate database
From this set we compute optimal values for each abiotic conditions
Use this to assign abiotic conditions to 800000 sites in Europe with observed plant species composition (if possible)
Derive response functions for each species in the large data set
PROPS model versions
Relationship between abiotic conditions and plant species occurrence.
Evapotranspi
ration
Hydrology
Precipitation
Temperature
Npoo
l
cNO
3
pH C
/N
Possible plant species diversity indices
Diversity indices
General indicesCompare to a
reference stateDesired species
Simpson index
Shannon index
Czekanowski (Bray- Curtis) indexBuckland occurrence index
Red List Index
Habitat Suitability index
Habitat Suitability (HS) Index
pj = probability/suitability/possibility of plant jpopt,j = optima (maximum) prob. of plant jn = number of plants
Which species?Suggestion: n = number of desired (typical) species
𝐻𝑆= 1𝑛 ( 𝑝1
𝑝𝑜𝑝𝑡 ,1
+𝑝2
𝑝𝑜𝑝𝑡 ,2
+…+𝑝𝑛
𝑝𝑜𝑝𝑡 ,𝑛)
Probability isolines: single species
0.0010000.0100000.0500000.1000000.200000
Calluna_vulgaris in 1996
NO3 concentration (mg NO3/kg)28272625242322212019181716151413121110987654321
pH
7.2
7
6.8
6.6
6.4
6.2
6
5.8
5.6
5.4
5.2
5
4.8
4.6
4.4
4.2
4
3.8
3.6
3.4
3.2
3
2.8
2.6
2.4
2.2
2
Assigning species to EUNIS classes
E10 - Frisian-Danish coastal heaths on leached dune-sands
Dominant and most frequent species in different layers
Herb layer
Calluna vulgaris, Empetrum nigrum, Genista anglica, Genista pilosa, Carex arenaria, Carex pilulifera, Erica tetralix, Salix
repens subsp. dunensis, Deschampsia flexuosa, Danthonia decumbens, Festuca ovina, Nardus stricta, Molinia caerulea,
Polypodium vulgare, Genista tinctoria, Lotus corniculatus, Orchis morio, Potentilla erecta, Ammophila arenaria
Moss layer (incl. lichens)
Dicranum scoparium, Pleurozium schreberi, Scleropodium purum, Hypnum cupressiforme, Platismatia glauca, Cladina
portentosa, Cladina arbuscula, Cladonia pyxidata, Cetraria aculeata
Diagnostically important species
Calluna vulgaris, Empetrum nigrum, Erica tetralix, Genista anglica, Genista pilosa, Salix repens subsp. dunensis, Carex
arenaria, Pyrola rotundifolia, Pyrola minor, Scleropodium purum, Pleurozium schreberi
Map of the natural vegetation of Europe
Combined probability isolines (British lowland blanket
bogs, 15 species); climate dependency
T=12°CT=3°C
PROPS: results
y = 0.47x + 3.0202R² = 0.496
3
4
5
6
7
8
9
3 4 5 6 7 8 9
Calc
ulat
ed
Measured
pH curves GJpH
1:1
Robustness...
0.0010000.0010000.0100000.0100000.0500000.0500000.1000000.1000000.2000000.2000000.3000000.3000000.5000000.500000
All selected species in 1996
NO3 concentration (mg NO3/kg)5048464442403836343230282624222018161412108642
pH
9
8.5
8
7.5
7
6.5
6
5.5
5
4.5
4
3.5
3
2.5
2
0.0010000.0010000.0010000.0100000.0100000.0500000.0500000.1000000.2000000.3000000.500000
All selected species in 1996
NO3 concentration (mg NO3/kg)5048464442403836343230282624222018161412108642
pH
9
8.5
8
7.5
7
6.5
6
5.5
5
4.5
4
3.5
3
2.5
2
PROPS demo
Bayesian Calibration of the model VSD+
Gert Jan Reinds
Contents
Introduction
Theory
Method
What to calibrate
Examples for VSDplus
Conclusions
Introduction
For application of models at sites we need to calibrate the model because there is an uncertainty and variability in input parameters
In VSD we can calibrate by fitting to the observations:
How to deal with uncertainty in observations and multi signal calibration
Often there is uncertainty in the measurements
We have output parameters that are influenced by more than one input parameter
Pr(A|B) is the posterior probability of A given BPr(A) is the prior probability of A not taking into account information about B. L(B|A) is the standardized likelihood of B given AIn the calibration of VSD, a prior distribution (A) of each VSD input parameter is defined based on available knowledge; for candidate points from normal distributions close to the mean the probability will be large, for points in the ‘tail’ of the distribution the probability will be low.
Then the posterior distribution of input parameters (Pr (A|B)) is computed based on the prior probability in combination with comparison of the model outcome with a set of uncertain measurements giving the likelihood L(B|A): the better the model is able to reproduce the measurements, the higher the likelihood
Bayes Theorem
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0 0.5 1 1.5 2 2.5 3 3.5
parameter value
pro
bab
ilit
y
prior
low prior probility
high prior probility
)Pr()|()|Pr( AABLBA
3
3.2
3.4
3.6
3.8
4
4.2
4.4
4.6
4.8
1985 1990 1995 2000 2005 2010
year
sim
ula
ted
pH
simulation with lowlikelihood
simulation with highlikelihood
observed values
Procedure
Determine for each model parameter suited for calibration its prior distribution (normal, uniform,..)
Run the model with samples from these distributions and compare the results from each run with measurements of output parameters (concentrations in soil solution and their standard deviation)
Accept the run if the goodness of fit is sufficient and store the associated input parameters
The vectors of stored input parameters provide the posterior distribution of the model parameters
How to sample
The method relies on a large number of runs, so we have to take many samples from the input data distributions (104 – 105)
We use a Markov Chain Monte Carlo (MCMC) approach (known as Metropolis-Hastings Random Walk)
Each point is accepted or rejected; accepted points are stored and so is the point with the highest posterior probability (i.e. the point with a combination of high prior probability and good model fit); this is what you see in the VSDp calibration output
Metropolis Hastings Random Walk
What to calibrate
lgKAlox: requires observations of pH and Al
lgKAlBc, lgKHBc; requires observation(s) of base saturation (EBc). Note: we start the calibation assuming EBc to be in equilibrium with deposition (inputs): start the calibration run preferably in pre-industrial time (<=1900)
Cpool_0: requires observation(s) of the Cpool
CNrat_0: requires observation(s) of C/N
DEMO
Standard calibration
Support
Support for you:
For support on VSD+ modeling you can contact CCE
Support for us:
To further develop, test, calibrate and validate VSD+ we like your input!
Forest not in NW-Europe
Non-forest vegetation
Questions?
latest version of• VSD+• GrowUp• MetHydcan be downloaded soonfrom: www.wge-cce.org
we will distribute USB sticks for now