In developing countries, irrigated agriculture is key to achieve domestic food security and mitigate poverty. The often complex landscapes dominated by traditional smallholder farming (fields ~< 1 ha) pose a challenge with regard to mapping irrigated agriculture. Irrigation mapping efforts are complicated by the spectral overlap between irrigated crops, rainfed crops and other land cover. Efforts to identify smallholder irrigation are often too coarse, and, generally the area under smallholder irrigation is underestimated. This study presents an object-based approach for mapping irrigated agriculture in complex landscapes for both smallholder farming as well as modern large-scale agriculture. The aim here is twofold: 1) a proof-of-concept of irrigation mapping in a case study in the Central Rift Valley, Ethiopia, using SPOT6 imagery and 2) upscaling of the mapping to the Horn of Africa using Sentinel-2 imagery. Our assumption is that the application of irrigation has a positive effect on the crop throughout the field, following the field's borders. Next, we hypothesise that the use of shape, texture, neighbour and location features in addition to spectral information is beneficial for the classification of irrigated agriculture. The case study area is located in the Awash River basin and contains two irrigation schemes serving both traditional smallholder farming as well as modern large-scale agriculture. Three SPOT6 images spanning the dry season were acquired. Objects were generated at the field level on the basis of similarities in vegetation behavior (NDVI change) assuming that this matched the response of the crop on receiving irrigation water. The resulting map shows whether and where fields received irrigation water in between the three satellite acquisition moments and whether it belongs to traditional smallholder farming or modern large-scale agriculture. The object-based classification had an overall accuracy of 94% and a kappa coefficient of 0.85 and showed an improvement for mapping irrigated agriculture compared to a pixel-based classification (overall accuracy of 88% and a kappa coefficient of 0.69). Producer's and user's accuracies of individual classes were generally higher for the object-based approach compared to the pixel-based classification. Type of agriculture (i.e. smallholder or modern large-scale) was mapped with an accuracy of 95% and a kappa coefficient of 0.88. The addition of shape, texture, neighbour and location information in the object-based approach proved essential for the identification of cropland plots, irrigation period, and type of agricultural system. For the upscaling we use Sentinel-2 imagery as they are available with the highest spatio-temporal resolution among publicly-available imagery. The 10-meter resolution and 5-day revisit time has tremendous potential for mapping smallholder irrigation in complex landscapes. At the moment we are processing the data. A mosaic of best-cloud-free images acquired in the dry season (October to March) is segmented at field level. Monthly NDVI composites, combined with shape, texture and neighbour information are used to map irrigated agriculture. The output will be a map showing the spatio-temporal distribution of irrigated agriculture, for both smallholder farming and modern large-scale agriculture, for the Horn of Africa.