Disaggregation of CAPRI results Renate Köble Adrian Leip Outline 1. Introduction 2. Existing approaches 3. Tentative solution of the problems 4. Overview on available data sets → → Corine Landcover 2000 LUCAS survey 7. First example WP8 - CAPRI-GIS link “Link of CAPRI to GIS covering soil, climatic and land use maps” Distribution of agricultural activities on the land → Correspondence between geo-coded data and geo-referenced data from land cover maps Distribution of secondary agricultural parameters → → → → livestock densities, feed composition animal wastes (grazing / stabling / application)? mineral fertilizer other parameters? Re-mapping for the calculation of indicators → grid size / thematic maps Allocation of land use changes in the reference year → → net / gross changes ? use of additional information Studies to built on ... (1) Trees species map (Renate) Calculation of Agricultural Nitrogen Quantity for EU River basins (JM Terres, JRC, 2000) → based on a methodology developed in a regional study (Loire and Elbe catchment) 2000 → used by Declan to disaggregate NUTS 2 NewCronos data to NUTS 3 level for the DNDC model Use of a correspondence table Corine / FSS CORINE NUTS 2 allocation matrix NUTS 3 (or gridcell) VARESE PROVINCE Land to be redistributed (FSS 2000 - arable land (D except rice)) → → Lombardia Varese 641 640 ha arable land 6 440 ha arable land Land available (Corine Land Cover 90) → → → class 211: non-irrigated arable land class 242: complex cultivation pattern class 243: land principally occupied by agriculture Lombardia Varese 211 242 Corine Land Cover Area [1000 ha] 856 36 13 0 243 94 12 x distribution factor for potentially available land: Lombardia Varese • non-irrigated arable land 0.95 • complex cultivation pattern 0.80 • land principally occupied by agriculture 0.60 Land available for FSS class arable land 813 22 56 12 0 7 2.2% of land potentially available in Lombardia is located in the province of Varese 891 19 Disaggregation FSS2000 data Varese Como Lecco Sondrio Milano Bergamo Brescia Pavia Lodi Cremona Mantova 2.2% 1.5% 1.0% 1.5% 11.0% 7.6% 17.3% 13.3% 7.1% 16.4% 21.2% 13805 9745 6566 9902 70337 48449 110789 85456 45827 104922 135842 Varese Como Lecco Sondrio Milano Bergamo Brescia Pavia Lodi Cremona Mantova 1.0% 1.1% 0.5% 0.2% 8.9% 6.1% 18.3% 14.1% 7.2% 18.9% 23.8% 6440 7020 3220 1470 56870 38830 117190 90630 46510 120980 152480 Lombardia 100% 641640 Lombardia 100% 641640 → → → Total deviation: 45 000 ha (7% of total arable land distributed) “Transfer” of arable land from the south to the north total potential area: 891 kha LOMBARDIA Non-irrigated land Permanently irrigated land Complex cultivation pattern Land principally occupied by agriculture with sign. areas of natural veg. Natural grassland Pastures Agro-forstry Fruit trees and berry plantations Rice Fields Vineyards Forests Water Urban Problems? → Assumption of equal land use of land classes throughout Europe → → → Non-matching between statistical and land cover data Interpretation errors and interpretation differences Time lag between Satellite images and Statistical census • Corine 1990 has is based on images between 1985-1993 • FSS 2000 represents the situation in 2000 Studies to built on ... (2) Spatial redistribution of statistical data from the Farm Structure Survey (GIM report) → → → → → Match absolute values of land in both data sets regrouping of CLC and FSS to aggregate classes Matrix of land cover classes potentially available Classes are filled-up successively Coefficients are optimized Aj i c ij CLC i Aj: area of Farm Structure category j CLCi: area of Corine class i cij: coefficient for redistribution of Corine class i to FSS category j GIM approach SOIL MAP, DEM NUTS 2 allocation algorithm CORINE NUTS 3 (or gridcell) Ranking of differences FSS-CLC → for all classes except rough grazing successive determination of coefficients • Determination of the fraction of temporary pasture to be distributed to permanent pasture: • when CLC overestimates arable land together with an underestimation of pasture, the possibility to allocate part of (211+212) to (231) up to a max. of 25% is evaluated • Distribution of complex classes Redistribution of complex classes recalculation of FSS and CLC by using coefficients y GROUP or CLASS FSS original CLC class additional CLC classes coeff. CLC coeff. CLC coeff CLC 1 arable land without rice fields D + G05 - I05A - I05B - D07 - aD18 b1 211 + 212 c1 242 d1 243 2 rice fields D07 b2 213 c2 242 d2 243 3 vineyards G04 b3 221 c3 242 d3 243 G01 + G02 b4 222 c4 242 d4 243 5 olive groves G03 b5 223 c5 242 d5 243 6 permanent pasture F01 + Ad18 b6 231 c6 242 d6 243 7 rough grazing F02 d7 243 4 fruit trees and berry plantations b7 or f7 231 or 321 8 combined crops; permanent and annual I05B + E b8 241 c8 242 d8 243 9 combined crops; annual and forestry I05A b9 244 c9 242 d9 243 Interaction of permanent pasture and rough grazing if FSS (pasture) is bigger than CLC (pasture) → the coefficient “a” for allocating part of temporary pasture (D18) to permanent pasture (F01) will be evaluated • • • if F01 < class 2.3.1. up to a maximum value of “a” that FSSarable CLCarable up to a maximum value of “a” that FSSarable CLCarable recalculation of FSS and CLC by using coefficients y GROUP or CLASS FSS 6 permanent pasture F01 + aD18 7 rough grazing F02 original CLC class coeff. b6 additional CLC classes CLC 231 b7 or f7 231 or 321 c6(242) + d6(243) + e6(211+212) + f6(321) + g6(322) + h6(323) + i6(333) d7(243) + g7(322) + h7(323) + i7(333) Parameters used for tuning → → class 2.4.2. (complex cultivation pattern) • a minimum of 25% must be used for “pasture” class 2.4.3. (Land principally occupied by agr. with sign. areas of natural vegetation) → • a maximum of 75% can be used to match agricultural land. • the share of natural vegetation can be used to match rough grazing Maximum tolerable error at district-level • the matching of FSS and CLC can be stopped if the error is <3%, when the addition of new classes makes the distribution too uncertain → class 2.1.1. • a maximum of 25% can be used to match pasture land Problems Assuming pure classes → also pure classes have a proportion of other land uses Assumptions on coefficients and sequence arbitrary → methodology will have to be re-checked Geo-referencing with soil map and digital elevation model → probabilities for aggregated classes only Time difference between data sets → “buy” synchrony with statistical consistency “Solution” Use of updated data → → “Corrected” Corine 1990 landcover map Corine 2000 land cover map as it becomes available Use of LUCAS to determine probabilities of a certain land use in a land cover class → → → clustering of LUCAS points co-location problems derive matching function between Corine and LUCAS land cover classes Use of physical parameters → determine relationships between the occurrence of • • disaggregated arable classes and soil (texture, chemistry), climate (temperature, precipitation, vegetation period), elevation (absolute elevation, slope) Co-location problem Gallego, 2003 Proposed approach NUTS 2 LUCAS SOIL MAP, DEM, climate allocation map+algorithm CORINE GRID What is the target grid / map ? Nested approach for DNDC? → → → 5 or 10 km grid landscape assessment: complexity of landscape 2.5 or 1 km grid Other maps according to the results of WP9 →Catchments ? →homogeneous units ? 1.3·106 km2 x 1 min = 2.5 years CAPRI Definition of parameters to be disaggregated WP8 Allocation of land use into the land cover map WP9 CAPRI-indicators CAPRI-indicators disaggregated (grid) disaggregated (map) WP10 Definition of target scale and target map WP10 WP9 Simulation of nutrient cycling Landscape assessment indicators