ECOCLIMAP-II a climatologic global database of ecosystems and surface parameters Jean-Louis Roujean et al. CNRM/GMME, Météo-France 42, avenue Gaspard Coriolis 31057 Toulouse cedex, France jean-louis.roujean@meteo.fr ECOCLIMAP is a land cover and surface parameters database specifically designed to serve the meteorological community: it aims at initializing and constraining surface models such as Méso-NH and Surfex. The first version was achieved in 2003, a second version is now ready for Europe.
Scientific Context & Objectives Better answering to climate users requirements Soil-Vegetation-Atmosphere Transfer (SVAT) model ISBA (used at Météo-France) adopted a ‘tile’ approach -> 12 vegetation types ECOCLIMAP database is twofold A land cover map of ecosystems (functionally homogeneous) Maps of land surface parameters (used for meteorological applications) establish correspondance between vegetation types and ecosystems
ECOCLIMAP-I => ECOCLIMAP-II Land cover Maps UMD,IGBP,CORINE Surface parameters Soil types Temporal profiles NDVI 1 year NOAA/AVHRR Climate maps FIRS, Koeppe 215 ecosystems (global) 255 ecosystems (Europe) GLC2000, CORINE2000 7 years SPOT / VGT Monthly FAO 10-days
The database includes several sets of surface parameters depending on soil, vegetation, or both: Depending on soil only, are … depending on vegetation only, are … Finally, albedo and emissivity are parameters depending on both soil and vegetation properties. A set of these parameters has to be defined for each new ecosystem. So, how will it be done?
Desaggregation ‘tools’ Each ecosystem is a potential fraction of 12 vegetation types Use of climate zoning LAI (vegetation) To reach this objective, we will follow aggregation rules as in ECOCLIMAP-I. A priori, desaggregation parameters are LAI for vegetation covers and ALBEDO for bare soil. ALBEDO (bare soil)
ECOCLIMAP-I MODIS CYCLOPES LAImin and LAImax fixed for each ecosystem LAI inter-comparison ECOCLIMAP-I MODIS CYCLOPES Evergreen needle-leaf forest (Finland) Evergreen needle-leaf forest (France) This is a 2 years comparison of LAI with other satellite products for 2 forests, representative of high and temperate latitudes. in black, LAI resulting from the NDVI data in ECOCLIMAP-I, in green, LAI from CYCLOPES derived from SPOT/VEGETATION, In red, MODIS LAI. CYCLOPES LAI is too low while MODIS LAI is too noisy. In such situations, ECOCLIMAP LAI appears the more reliable.
GLC-2000 (coordination: Joint Research Center) ~ 1 year (2000) of SPOT/VEGETATION data The first land cover maps used are the different tiles of GLC2000. The GLOBAL LAND COVER 2000 Project is coordinated and implemented by the Global Vegetation Monitoring Unit from the Joint Research Centre in collaboration with a network of partners around the world. GLC 2000 makes use of the VEGA 2000 dataset: a dataset of 14 months of pre-processed daily global data acquired by the VEGETATION instrument on board the SPOT 4 satellite. Here you can see the different tiles available for GLC2000 at the time of the realization of the initial map. We can notice that on Europe and Asia in particular many different tiles are available. A global map also exists. 23 classes : tree cover (10), shrub (5), crop (3), bare area, water body, snow/ice, artificial surfaces, irrigated agriculture
CORINE Land Cover - 2000 44 classes Resolution : 100m CORINE2000 proceeds from the European Environment Agency. Data are derived from visually interpreted satellite images from SPOT and LANDSAT. Here you can see the area covered by the project: some countries do not appear: Swissland, Serbia-Montenegro, Andorre, Norway.
Merged GLC2000+CORINE (as initial LC map) 76 classes some minor CORINE classes Here you can sea the final map obtained after these operations, on Europe. The process was applied on the whole world, we show here the map used for the rest of the study. (rajouter une legende avec unites de paysage dominantes)
Choice of the clusters’ number 1st classification: big numbers of clusters (from a splitting of 76 merged classes GLC2000 +CORINE) Analysis of mean profiles, standard deviation, geographic localisation for each cluster 2, 3 iterations Classification with reduced numbers of clusters As we don’t use any quality criterion to quantify the result of the classification, we decide to proceed by several iterations. First, we launch the program with bigs numbers of clusters for all classes. Then, we observe mean profiles and standard deviation profiles for all clusters, along with geographical localisations associated. We try to pick out the more divergent profiles, and to decide which ones could be grouped together. Next, we restart the classification with new numbers of clusters deduced, lower. We compare new results with the previous, and see if we obtain what we were waiting for. After repeating 2 ou 3 times this process, we finally obtain 201 classes on Europe. Examining the map we find clusters from different classes which can be put together, and we finally get 161 classes. + gathering of ‘parent’ clusters 305 classes over Europe 255 classes over Europe
After classification, the resultinf map comprises 161 classes (but it’s not finished yet, they will potentially be more).
Broadleaf forests 6 clusters GLC2000+CORINE ECOCLIMAP-II Example of spliting for the broadleaf forest class is shown here. A class in Spain clearly sorts out, in orange on the graph, Characterized by a small amplitude, a maximum in spring. It is typical of dry summer and cool winter. Blue profile corresponds to dense forests of France and Italy mainly. Amplitude is also small: high values span from spring to summer, Relative low values in winter may be explained by the permanence of litter under temperate climates. Moving north, NDVI profiles like red, green and purple show larger amplitudes due to sparse forest canopies and the occurrence of snow in winter. Finally, mountain clusters appears in red and yellow. -> 6 classes ECOCLIMAP-II
Crops 11 clusters GLC2000+CORINE ECOCLIMAP-II Example of 11 clusters is presented here for the class: « cultivated and managed areas », which is the largest class of crops in the initial map. ECOCLIMAP-II
To conclude, we can sum up first improvements brought by the second version of ECOCLIMAP: The resolution of satellite data is better, by the use of SPOT/VEGETATION data rather than NOAA/AVHRR. The clusters are more precisely discriminated by the use of the automatic classification The initial land cover maps are more recent too Finally, the interannual variability of some parameters, notably the central parameter LAI, will be introduced. The difficulties encountered comprise the determination of pertinent clusters, knowing that NDVI is a limited parameter to characterize surface coverage. It also deals with the attribution of surface parameters, in particular determination of parameters for the 12 elementary cover types, and the desagregation of the clusters LAI or ALBEDO profiles.
Cross-validation of ECO-II with HR imagery FORMOSAT
Besoins des utilisateurs ECOCLIMAP-SG Besoins des utilisateurs Global 300 m ? (GLOBCOVER, ESA-CCI LC)
ECOCLIMAP-SG Réunion interne CNRM du 18 novembre 2014 Résultats attendus : Occupation des terres plus réalistes (forte urbanisation en 20 ans) Meilleure résolution spatiale Amélioration du temps de calcul Scénario retenu : Utilisation d’ESA-CCI Land Cover (300 m) Abandon des « covers » au profit de cartes de paramètres Etablir une fonction de transfert permettant d’intégrer ESA-CCI LC Utilisation LAI/albédo PROBA-V (à 300 m) Utilisation de nouvelles profondeurs de sol Echéance : fin 201