A central focus for governing bodies in Africa is the need to secure the necessary food sources to support their populations. It has been estimated that the current production of crops will need to double by 2050 to meet future needs for food production. Higher level crop-based products that can assist with managing food insecurity, such as cropping watering intensities, crop types, or crop productivity, require as a starting point precise and accurate cropland extent maps indicating where cropland occurs. Current cropland extent maps are either inaccurate, have coarse spatial resolutions, or are not updated regularly. An accurate, high-resolution, and regularly updated cropland area map for the African continent is therefore recognised as a gap in the current crop monitoring services. Key PropertiesGeographic Coverage: Continental Africa - approximately 37° North to 35 SouthTemporal Coverage: 2019Spatial Resolution: 10 x 10 meterUpdate Frequency: TBDNumber of Bands: 3 BandsParent Dataset: Digital Earth Africa's Sentinel-2 Semiannual GeoMADSource Data Coordinate System: WGS 84 / NSIDC EASE-Grid 2.0 Global (EPSG:6933)Service Coordinate System: WGS 84 / NSIDC EASE-Grid 2.0 Global (EPSG:6933)

Digital Earth Africa’s cropland extent maps for Eastern, Western, and Northern Africa show the estimated location of croplands in these countries for the period of January to December 2019:

Eastern: Tanzania, Kenya, Uganda, Ethiopia, Rwanda and BurundiWestern: Nigeria, Benin, Togo, Ghana, Cote d'Ivoire, Liberia, Sierra Leone, Guinea and Guinea-BissauNorthern: Morocco, Algeria, Tunisia, Libya and EgyptSahel: Mauritania, Senegal, Gambia, Mali, Burkina Faso, Niger, Chad, Sudan, South Sudan, Somalia and DjiboutiSouthern: South Africa, Namibia, Botswana, Lesotho and Eswatini

Cropland is defined as:

"a piece of land of minimum 0.01 ha (a single 10m x 10m pixel) that is sowed/planted and harvestable at least once within the 12 months after the sowing/planting date."

This definition will exclude non-planted grazing lands and perennial crops which can be difficult for satellite imagery to differentiate from natural vegetation.

The provisional cropland extent maps have a resolution of 10 metres and were built using Copernicus Sentinel-2 satellite images from 2019. The cropland extent maps were built separately using extensive training data from Eastern, Western, and Northern Africa, coupled with a Random Forest machine learning model. A detailed exploration of the methods used to produce the cropland extent map can be found in the Jupyter Notebooks in DE Africa’s crop-mask GitHub repository.

Independent validation datasets suggest the following accuracies:

The Eastern Africa cropland extent map has an overall accuracy of 90.3 %, and an f-score of 0.85 The Western Africa cropland extent map has an overall accuracy of 83.6 %, and an f-score of 0.75 The Northern Africa cropland extent map has an overall accuracy of 94.0 %, and an f-score of 0.91The Sahel Africa cropland extent map has an overall accuracy of 87.9 %, and an f-score of 0.78The Southern Africa cropland extent map has an overall accuracy of 86.4 %, and an f-score of 0.75

The algorithms for all regions tend to report more omission errors (labelling actual crops as non-crops) than commission errors (labelling non-crops as crops). Where commission errors occur, they tend to be focussed around wetlands and seasonal grasslands which spectrally resemble some kinds of cropping.

Available BandsBand IDDescriptionValue rangeData typeNoData/Fill valuemaskcrop extent (pixel)0 - 1uint80probcrop probability (pixel)0 - 100uint80filteredcrop extent (object-based)0 - 1uint80

mask: This band displays cropped regions as a binary map. Values of 1 indicate the presence of crops, while a value of 0 indicates the absence of cropping. This band is a pixel-based cropland extent map, meaning the map displays the raw output of the pixel-based Random Forest classification.

prob: This band displays the prediction probabilities for the ‘crop’ class. As this service uses a random forest classifier, the prediction probabilities refer to the percentage of trees that voted for the random forest classification. For example, if the model had 200 decision trees in the random forest, and 150 of the trees voted ‘crop’, the prediction probability is 150 / 200 x 100 = 75 %. Thresholding this band at > 50 % will produce a map identical to mask.

filtered: This band displays cropped regions as a binary map. Values of 1 indicate the presence of crops, while a value of 0 indicates the absence of cropping. This band is an object-based cropland extent map where the mask band has been filtered using an image segmentation algorithm (see this paper for details on the algorithm used). During this process, segments smaller than 1 Ha (100 10m x 10m pixels) are merged with neighbouring segments, resulting in a map where the smallest classified region is 1 Ha in size. The filtered dataset is provided as a complement to the mask band; small commission errors are removed by object-based filtering, and the ‘salt and pepper’ effect typical of classifying pixels is diminished.

More details on this dataset can be found here.