Monitoring smallholder irrigation using remote sensing and object-based image analysis: Case studies in Eastern Africa

Over the last decades, population growth in low- and middle-income (LMI) countries has been very large thereby putting high demands on land and water resources. The ambition of the Sustainable Development Goals of 2015 is to achieve global food security by 2030, which requires large increases in food production over the coming years. In LMI countries, smallholder irrigated agriculture will play a vital role to enhance food production. Currently, the distribution (when, where, how much) of smallholder irrigation is unknown. As efficient water-resource management becomes increasingly important, field-level knowledge on the use of water resources for irrigation is vital. Remote sensing has emerged as an effective tool for irrigation mapping in different regions and climates. However, in LMI countries it is difficult to map irrigated agriculture at field level, because of data-poor systems and complex agricultural landscapes. Smallholder agriculture has small cultivated plots, intercropping systems and dynamic spatio-temporal patterns of sowing, harvesting and irrigation application. The arrival of Sentinel-2 provides high spatial-resolution images and considerably increased the prospects of field-level mapping of smallholder agriculture. In this thesis, the understanding of spatio-temporal dynamics of smallholder irrigated agriculture and water consumption is improved by exploring the value of object-based image analysis on high-resolution imagery in data-poor regions. This thesis exploits universal characteristics of (irrigated) agricultural fields and local context in classification and quantification approaches. The use of objects decreases spectral intra-class variability. The use of local context and field characteristics is relevant for many types of agriculture, and is not dependent on absolute spectral signatures of specific crops, which differ per region and climate zone. Such object-based approaches are especially valuable for data-poor LMI countries, as ground truth to attain spectral signatures are generally unavailable or difficult to acquire. The methodologies developed in this thesis are conceptualized and realized through case studies in Eastern Africa. This thesis provides the first field-level assessment at a monthly time interval of irrigated agriculture and its water consumption in smallholder-dominated complex landscapes. The new methods use smart neighbourhood relations to identify irrigated agricultural land and quantify irrigation-water consumption. Besides irrigation mapping, this thesis provides a workflow for the characterization of the type-of-agricultural system; smallholder or modern large-scale agriculture. This is beneficial for studying irrigation practices specifically for both types of agriculture when they occur side-by-side in the landscape. In addition, this thesis shows a semi-automated object-based approach for the interpretation of black-and-white aerial imagery. The time span of land-use change analysis is thereby expanded to far before the availability of satellite imagery in 1972. The object-based approaches described in this thesis exploit universal characteristics of (irrigated) agriculture and the methodologies are mainly based on open-source readily-available remote-sensing datasets. This makes the methodologies portable to other regions and climates and applicable for large areas. These methods provide information useful for irrigation-development policy, the monitoring of irrigation efficiency and optimization measures, and the analysis of the irrigation-impact on the water balance