Towards Global Mapping of Irrigated Agriculture

We developed an irrigation mapping methodology that relies on remotely sensed inputs from MODerate Resolution Imaging Spectroradiometer (MODIS) instrument, globally extensive ancillary sources of gridded climate and agricultural data and on an advanced image classification algorithm. In the first step, we used climate-based indices of surface moisture status and a map of cultivated lands to provide potential, first-cut at global irrigation. To detect actual irrigation, in the second step, we used spatiotemporal and spectral signatures from MODIS remotely sensed data. In particular, we explored three types of irrigation-related indices: i) Annual – where we exploited the difference in annual greenness variability between irrigated and non-irrigated crops and related this difference to precipitation availability; ii) Spectral – where we exploited a vegetation index (Green Ratio Index) that is sensitive to chlorophyll content; and iii) Inter-annual – where we explored the differences in inter-annual changes in vegetation greenness associated with precipitation between irrigated and non-irrigated crops. In the third step, we combined our potential irrigation dataset, remotely sensed indices, and training examples within a supervised classification tool based on a non-parametric decision-tree algorithm to make a binary (i.e. irrigated vs. non-irrigated) irrigated agriculture map. A test of irrigation mapping procedure in a pilot study over the continental US produced a high spatial resolution (1 km) map of irrigated areas with better than 80 percent map accuracy and expected spatial patterns such as a strong east-west divide with most irrigated areas concentrated on the arid west along dry lowland valleys. Future improvements of the method will include estimation of sub-pixel presence of irrigation using remotely-sensed skin temperature measurements. * Corresponding author.