| Summary: | Effective agricultural planning requires up-to-date and spatial crop phenology and land use (LU) data for time-critical and location-specific extension, inputs or emergency aid. Usually, this information is most sparse where most needed, (sub)tropical smallholder-dominated landscapes. Our study enhances planning methods for food-security and climate change adaptation in rain-fed smallholder agriculture of South Central Ethiopia. In a case study covering 10,500 km2 and four agroecological zones, we developed a phenology-oriented approach for dynamic classification of cropping systems using Google Earth Engine. Complementary datasets of MODIS Enhanced Vegetation Index were merged, increasing time resolution from 16 to 8 days. Derived pixel-and crop-specific seasonal time series at 250 m resolution reflected vegetation phenology. Random Forest was applied to classify agricultural LU types and cropping systems. Vegetation period onset, validated against farmers’ sowing dates, was used to detect crop rotations and map drought risk for the years 2003–2018. Identified LU types were: single cropping (maize/haricot bean, wheat/barley), and double cropping (maize/other, maize/wheat). Overall accuracy for agricultural LU was 76–94%.We use open source data and online data processing. High temporal resolution phenology data allow identifying needs for agricultural inputs and emergency support “in real time”, assess drought risk and monitor LU dynamics.
|
|---|