Michael Marshall; Kees de Bie; Michael Schlund
Spatio-temporal Analysis of Remote Sensing Data for Food and Water Security / Advanced Image Analysis
Statistics, agronomy and programming background/experience highly recommended
The location and extent of arable fields is important for monitoring the productivity of crops and informing food security specialists and other decision makers. Remote sensing is commonly used to delineate these boundaries, because the workflow can be automated and it can provide consistent estimates over large areas through time. In developing countries, such as Ethiopia, remote sensing-based techniques are difficult to implement, because farm sizes are too small to be resolved with moderate spatial resolution imagery. The mapping approach defined above was successfully tested in an important agricultural region in Ethiopia. This study intends to fill knowledge gaps identified in that study and estimate arable field fractions for the entire Ethiopia.
The purpose of this study is to map wall-to-wall arable field probabilities/fractions at 20-30m spatial resolution for the entire Ethiopia. The maps will be derived from surface reflectance/emissivity acquired by a number of remote sensing platforms, including: Proba-V, Sentinel-1, Sentinel-2 and Landsat-8. Ancillary geospatial data, such as topography will be considered for model-building as well. Any number of data-driven (support vector machine, Random Forest, boosted regression) modelling methods can be employed to translate such information into arable field fractions. Critical to this analysis will be a high-quality reference dataset of arable field fractions. Such a dataset will be compiled from a number of ground-based and high spatial resolution remote sensing sources, including: global dataset of crowdsourced land cover and land use reference data; Radiant Earth LandCover Net; USGS Global Land Cover Validation Reference Dataset; Mohammed et al., 2020; and EENSAT. Undoubtedly, the spatial coverage of these products is incomplete, so the student will fill gaps via visual interpretation of Google Earth or PlanetScope high spatial resolution imagery and a field campaign.
Mohammed, I., Marshall, M., de Bie, K., Estes, L., and Nelson, A. 2020. A blended census and multiscale remote sensing approach to probabilistic cropland mapping in complex landscapes. ISPRS Journal of Photogrammetry and Remote Sensing, 161: 233-245.