Towards accurate cropland area estimation with Earth observations and machine learning in Sub-Saharan Africa

Agricultural land area distributions in many African countries are changing rapidly as a result of rural transformations supporting small to medium holder farmers. These dynamics may be underreported systematically due to limitations of existing enumeration methods that are largely based on field or household surveys which are very expensive.

Earth observation(EO) data, machine learning (ML) methods and various geospatial techniques offer avenues to address these limitations and have proven to be effective in continuous monitoring and reporting of agricultural activities in such data-scarce regions. Beyond cropland mapping, an accurate quantitative estimate of cultivated area is required for supporting production estimates, land management, risk assessments and development policy activities. Moreover, information on annual cropland change trajectory is a key indicator for.regional and national climate change reporting.

We estimate annual cropland area using existing global crop datasets for multiple African countries, including Tanzania, Kenya, and Ethiopia, following good practices for area estimation and land change mapping [1]. Our approach is poised to be a practical abstraction of cropland area estimation and annual area cropland conversion, fully integrated with the current state-of-the-art cropland mapping implementations.

[1] Olofsson, P., Foody, G. M., Herold, M., Stehman, S. V., Woodcock, C. E., & Wulder, M. A. (2014). Good practices for estimating area and assessing accuracy of land change. Remote Sensing of Environment, 148, 42-57.