Mapping of cropland areas over Africa combining various land cover/use datasets Food Security (FOODSEC) Action Monitoring Agricultural ResourceS (MARS) Unit Institute for Environment and Sustainability (IES) Joint Research Centre (JRC) – European Commission Christelle Vancutsem, Francois Kayitakire, Jean-Francois Pekel, Eduardo Marinho
Pasture & Crop Masks RS time series Agriculture monitoring and early warning NDVI anomalies Vegetation Index profile extraction Context
Objective Map cropland areas at 250m « STATIC » Expert-based combination of existing datasets At the global scale (17) with emphasizing on Africa (10) « DYNAMIC » Every year Sub-saharien african countries Identify potential cropland areas Analyse the inter-annual variability From MODIS time series … … multi-annual mask
10 sources Landsat-based: - SADC (CSIR) - CUI 1988 (USGS) - LULC 2000 (USGS) - Woody Biomass 2002 (World Bank) - Africover 2000 (FAO) - LC Senegal 2005 (GLCN, FAO) - LC Mozambique 2008 (DNTF) - MODIS-derived Crop mask 2009 (JRC, MARS) Low/medium resolution: - Globcover (ESA) - RDC LC 2000 (UCL) Crop mask
10 sources Data preparation Selection of cropland classes Combination of datasets Regularly updated Static crop mask
Validation JRC contributes to the improvement of the tool: -as beta-tester (7 experts) -providing SPOT VGT NDVI profiles
Validation With agriculture.geo-wiki.orgagriculture.geo-wiki.org Two validation datasets: 1.For all Africa, one point every degree (IIASA) 2942pts 2. For Niger-Nigeria (Foodsec), one point every 40km 649pts (200 pts overlap) total 3591 pts without overlap Comparison with two existing crop masks: Fritz et al. (2011) and Pittman et al. (2010)
Validation Agreement (%) between experts for each category of crops taking into account the category concerned only (% of agreement 1cl) and the neighbouring classes (% of agreement 3cl) 130 points Niger- Nigeria
Validation Comparison between the 3 crop masks and two validation datasets MARS IIASA Pittman et al.MARS IIASA Pittman et al. >50% 65.15% 30.26% 21.3% 69.6% 49.8% 17.3% Africa Niger-Nigeria window JRC IIASA
Combination of the best existing datasets available (static mask) - half of the African countries covered by high and medium resolution-derived products - validation shows that the product better agrees with the validation dataset than other existing crop masks - need of up-to-date information and feedback from users ! - in continuous improvement (global) Training and validation datasets with agriculture.geo-wiki.org - Reliable and user-friendly collaborative tool - Allows sharing data and expertise between experts in a win/win approach - As powerful as the number of user is growing - Allows a high productivity of the interpreter Conclusion
Thank you Global Cropland Map (JRC-MARS, 2011)
10 sources Data preparation From feature to Raster Reprojection Resampling at 250m Translation in the LCCS legend (5cl) -Cultivated and managed areas -Post-flooding or irrigated croplands -Rainfed croplands -Mosaic cropland (50-70%)/vgt -Mosaic vgt / cropland (20-50%) Static crop mask
10 sources Data preparation Selection of cropland classes –By default, crops >50% –IF crops <50% Selection by experts based on comparison with HR imagery (GE) - Globcover 20-50% in equatorial countries - CUI 30-50% Static crop mask
10 sources Data preparation Selection of cropland classes Data combination When different sources: 1) Comparison with high resolution imagery (GE) & Analysis by experts 2) Rules: 1 st priority to the highest resolution 2 nd priority to the most recent Static crop mask
10 sources Data preparation Selection of cropland classes Combination of datasets Possible issues Out-dated Global LC data (Globcover) Spatial inconsistencies Spatial resolution 250m not “real” Static crop mask