Global irrigation mapping - the role of spatial resolution of current and future earth observation missions

Kurzfassung

Irrigated agricultural area is of high importance for global food production. Approximately 20 % of global agricultural area is irrigated, but 40 % of the world's harvested food is produced on these irrigated area. Global agriculture is the largest consumer of freshwater (69 %) and due to a changing climate, irrigated area is expected to increase. Increased water consumption by agriculture would lead to conflicts of interest between sectors such as energy, industry, and households. Therefore, a more efficient use of water by agriculture becomes necessary. Current irrigation techniques mostly consist of surface or sprinkler irrigation. Both techniques use the resource water only very inefficiently, since a high proportion of the water evaporates from the surface into the atmosphere. In order to maintain the agricultural production of these areas in the case of a scarcity of the resource water due to a changing climate and the accompanying landscape changes, such as the melting of the glaciers, an increase in the water use efficiency becomes necessary. In order to model water flows and to analyze future changes for recommendations for policy decisions, information on irrigated area and irrigation techniques, at a high spatial resolution, is needed. Existing data set differ in the extent of global irrigated area. Reasons are the different definition ("irrigated" or "equipped for irrigation"), the investigated time period, and the different methodologies and input data. In the present work, a new methodology is developed that combines different input data and distinguishes irrigated areas from non-irrigated areas. National and sub-national statistics reported by countries to the Food and Agriculture Organization (FAO) are one input. This information is spatially distributed on agricultural area at a spatial resolution of 0.008333 degrees (approximately 1 km at the equator), the only available dataset with a spatial resolution of 1 km. In addition to the statistics, satellite data were used to examine plant growth globally and compare it to agricultural suitability derived from climate and soil data. If the vegetation development on agricultural area does not correspond to the plant growth expected from climate and soil data, it is assumed, that agricultural management, like irrigation, leads to the observed plant growth. The methodology detects 18 % more irrigated areas than officially reported to FAO, revealing a knowledge gap in current research and showing that recommendations for action based on officially reported data are limited. Based on these issues, this thesis delves into the question of uncertainties and possible sources of error in the dataset. The influence of the spatial resolution of the sensor is analyzed systematically for three different regions. The developed data set is based on the SPOT-VGT sensor with a spatial resolution of about 1 km. To systematically quantify the influence of spatial resolution, high-resolution satellite data from Sentinel-2 were scaled from 10 m to 1 km stepwise and irrigation was detected. To save computational time, three regions were selected for the conduction of the experiment. It was shown that in two of the regions (USA and Sudan), decreasing spatial resolution leads to decreasing of detected irrigated area. In China the detected area remains constantly. An analysis of the spatial distribution of the irrigated area shows that the mapping result depends on the spatial arrangement and distribution of the irrigated area. Dense, contiguous fields are detected, loosely distributed fields are neglected at a coarser spatial resolution. The study demonstrated, that the negative areal change can be explained by landscape metrics. The application of landscape metrics showed a regionally independent relationship between the loss of irrigated areas at coarser resolution and the "Landscape Shape Index" (LSI). The index is an aggregation index and calculates how complex one class (in this case "irrigated area") is compared to another class in a landscape (in this case "non-irrigated area"). This finding can be used to identify regions that are prone to underestimating irrigated area, for further analysis using high resolution satellite data. The spatial resolution of a sensor is always a trade-off between technical and financial feasibility, handling, scope of application, and research questions. Existing multi- and hyperspectral satellite missions focused on environment and agriculture have sensor resolutions ranging from 10 m (Sentinel-2) to 30 m (EnMAP, LANDSAT). The suitability of these resolutions for the analysis of agricultural areas has not yet been systematically investigated and is part of this thesis. Field boundaries of the German states Bavaria and Lower Saxony and the Netherlands were converted into a Sentinel-2 geometry and rescaled to the resolutions of 5 m, 10 m, 20 m, 30 m, and 50 m. The fields are analyzed regarding at which resolution the fields can be recorded completely by a satellite and are thus suitable for the analysis of agricultural questions. In addition, it was analyzed how many fields are suitable for precision farming applications in order to establish an in-field management monitored by satellites. Therefore, a minimum of 50 pixels per field was assumed, which are necessary to use precision farming applications. The analysis shows that at a Sentinel-2 resolution of 10 m, 2-4 % of the fields cannot be covered and 20-50 % are not available for precision farming applications. Field crop types were also included in the analysis for a better understanding which crop types will be available for a satellite-based monitoring. This thesis provides a basis for decision making for future satellite missions and helps to assess the feasibility of applications with current satellite missions. Overall, this thesis highlights the high importance of global irrigation information and the high complexity of methods detecting irrigated area. With the development of a new dataset, the spatial resolution could be improved and it was shown that in many regions the irrigated area is significantly underestimated. Furthermore, the influence of spatial resolution was analyzed and it could be shown how the spatial resolution of current and future satellite missions affects the possibility of agricultural monitoring in Europe and the possibility of in-field management and precision farming applications.