Fig 1.
Agro-ecological regions of Zimbabwe.
Table 1.
RMSE of cropland area comparing different land cover products to national1 FAO statistics [42].
Fig 2.
Comparison of different LULC classifications and class definitions: (a) IGBP, 1993, (b) GlobCover, 2009(c) MODIS13Q1, 2014 (Band1).
Fig 3.
MODIS and Landsat tiles used in this study.
Landsat tile used for the creation of agro-ecological stratified endmembers is highlighted. The rectangular NDVI dataset represents the extent of the NDVI dataset.
Fig 4.
Flowchart of the processing chain for LULC classification.
Fig 5.
Exemplary visualization of an original and a filtered NDVI-curve, as well as 11 Seasonal Parameters.
Adapted from [59].
Fig 6.
Temporal profiles and spatial location of 750 pure grassland MODIS pixels for phenological season 2004/2005.
Color codes represent the maximum value of the NDVI profile. Maximum values follow a South-North increment, related to agro-ecological conditions.
Fig 7.
Spatial visualization of endmember construction.
Comparison of a pure 'rainfed' endmember (yellow) in Landsat false color image (a), binary classification result for rainfed/other (b), ESRI-basemap (c) and purity grid on MODIS resolution (d).
Fig 8.
Visualization of heterogeneity of land use classes.
Profiles are colored according to longitude. For the selection and verification process of endmember pixels, we made use of two external datasets.
Fig 9.
Examples of pure pixels (250x250m) and mean NDVI profiles (2004/2005) including standard deviations for each LULC class.
Table 2.
Accuracy report for NDVI+SP, 2004/2005.
Table 3.
Accuracy report for NDVI+SP,+NIR+RED, 2004/2005.
Fig 10.
Spatial results of Random Forest classification.
(a) classification result (2004/2005) for former freehold tenure area in AER II, (b) recent ESRI-basemap data for comparison, (c) numbers of seasons classified as rainfed or other land use. Details are overlayed with farm boundaries.