1. Spatial disaggregation in CAPRI
Adrian Leip1, Renate Koeble1
European Commission, Joint Research Centre (JRC)

2. Outline CAPRI modelling framework Disaggregation
Generating a priori Land use maps
Homogeneous Spatial Units
Spatial data
FADN spatial allocation
CAPRI-disaggregation

3. The CAPRI modeling framework
Data Base Tools
Projection
GLOBAL
CAPTRD
EuroStat FaoStat ...
Scenario
CoCo
CAPMOD
CAPREG
Supply
Market
Disaggreg.
Impact Assessment
Environment
Welfare
FADN spatial allocation
CAPRI-disaggregation

4. Integrated economic and agri-environment modelling: why Disaggregation?
1. For several indicators, environmental impact is non-linear or asymmetric
2. Data on the distribution of the environmental impact in space more relevant than the average value (hot or cold spots, shares, thresholds, …)
3. Interpretation of environmental indicators at administrative boundaries if often meaningless for policy decisions (protected areas, areas delimited by natural borders (e.g. watersheds), areas with natural constraints, …)
FADN spatial allocation
CAPRI-disaggregation

5. Integrated economic and agri-environment modelling: why CAPRI?
Provides input data to bio-physical models or to indicators (e.g crop shares, stocking densities, yields, input use (N,P,K; energy; feed ...) which are
based on EU official data sources (FSS, FADN …)
consistent across scales: farm type / region / MS / EU / trade blocks
used in economic modules and passed to environmental modules
CAPRI
is in line with DG-AGRI projections,
provides benchmark for counterfactual scenarios
Therefore consistent modelling chain for economic (farm income, trade flows …) and environmental assessment is ensured
FADN spatial allocation
CAPRI-disaggregation

6. Spatial data: CAPRI as spatial impact assessment tool
Agri-environmental indicator at the country and province scale
Aggregation rules
Agri-environmental indicator at the pixel scale
FADN spatial allocation
CAPRI-disaggregation

7. Sequential disaggregation
Altitude Slope
T-sum
Veg-period
Rain
Fodder
Regression
HPD
Distances
Sequential disaggregation
A priori land shares
NUTS3 crop areas
HPD
1. Crop shares
3. Livestock density
5. Indicator calculation
Irrigat. shares
Pot.+ rainfed yields
Irrigated areas
HPD
2. Yield/ irrigation
4. N inputs
Ndep
MITERRA coeff
Soil data
Rain classes
Distances
Sequential estimation of N input (manure, crop residues, mineral fert.)
FADN spatial allocation
CAPRI-disaggregation

8. [1] http://ec.europa.eu/agriculture/rurdev/eval/guidance/note_f_en.pdf
CMEF
OECD
Eurostat
Greenhouse gas emissions
Baseline 26
IV. Environ. Impacts of agriculture-4
AEI 19 - Greenhouse gas emissions (pressure)
Nutrient surplus
Baseline 20
III. Use of farm inputs and natural resources-1.
AEI 15 - Gross nitrogen balance (pressure)
Risk of soil erosion by water
Baseline 22
IV. Environ. Impacts of agriculture-1
AEI 21 - Soil erosion (pressure)
Agricultural landscape - 
IV. Environ. Impacts of agriculture-1 
Proxy for: AEI 28 - Landscape – state and diversity (state)
Biodiversity friendly farming practices
Baseline 18: Biodiversity: HNV farmland and forestry’
IV. Environ. Impacts of agriculture- 6 Wildlife Habitats
AEI 23 - High nature value farmland (state)
Environmental compensation zones -
[1]
[2]
[3]
FADN spatial allocation
CAPRI-disaggregation

9. Locally Weighted Max. Likelihood
+ + +
LUCAS
SWHE
30% Sand
680mm Rain
…..
GRAS
25% Sand
700mm Rain
SWHE
30% Sand
680mm Rain
…..
GRAS
25% Sand
700mm Rain
Corine Class 1
+ + + di
Corine Class 2
Weighting Function:
wi=f{di} @ di < d + d
example: LUCAS points L, model for soft wheat
x: Environmental attributes at the LUCAS points (e.g. annual rainfall, vegetation period, sand content, …)
BETA: coefficient measuring the impact of the change of parameter x by one unit
LAMBDA: dummy variable expressing the probability that wheat is grown at the LUCAS point.
Logit avoid the unboundness of a linear probability model
Locally Weighted Logit :
Di=exp{b’xi} [1+exp{b’xi}]-1
(= share of land use )
max: ln{L}=Siwi[yi ln{Di} + (1-yi)ln{1-Di}]
FADN spatial allocation
CAPRI-disaggregation

10. Estimating a priori land use shares
Yest
Ycon
Y(h,c)
pdf{Y(h,c)}
60ha m asL
650mm Rain
…..
40 ha m asL
700mm Rain
HSMU
26% SWHE -> 10.4 ha
72% GRAS -> 28.8 ha
70% SWHE -> 42 ha
35% GRAS -> 21 ha
Yest(h,c)=f{bc, xh}
ESTIMATION
25% SWHE -> 10 ha
75% GRAS -> 30 ha
76% SWHE -> 45 ha
25% GRAS -> 15 ha
55 ha SWHE
45 ha GRAS
Nuts II
Data consistent shares Ycon(h, c) for all crops c and HSMU h
max: SiSj[Ycon(hicj)]
s.t. SiYcon(hicj) Ai=Aj
SjYcon(hicj) =1
FADN spatial allocation
CAPRI-disaggregation

11. Quality check of disaggregation on FSS 2000 data
Fig. A1. Percentage of misclassified areas in validated NUTS 2 regions after dis-aggregation. The pies show the contribution of different
crop groups to the total error in the region (Cereals: soft wheat, durum wheat, barley, rye, oats, maize, other cereal; Fallow: fallow land; Rice
and Oil Seeds: rice, sunflower, soya, texture crops, pulses, other crops; Root Crops: potatoes, sugar beet, root crops, rape, nurseries; Per-manent/Industrial Crops: tobacco, other industrial, vegetables, flowers, citrus trees, fruit trees, olive trees, vineyards; Grassland: grassland,
fodder production). Note that the size of the pie is related to the area of the NUTS 2 region for visualization purposes only.
Leip et al., 2008
FADN spatial allocation
CAPRI-disaggregation

12. Update: Lamboni et al. (in review)
Methodological changes
Multinomial logit
Updated spatial layer (HSU2)
Updated explanatory variables
Current challenges
Problems with non frequent crops
Introduction of physical constraints