Detection and mapping of irrigated farmland in Canterbury, New Zealand

Identification of irrigated farm land is critical to a variety of environmental models. For this work, it is needed to model the impacts of irrigation on groundwater quality and flow rates. As the ground cover appearance for many land uses is quite dynamic, we take a multi-temporal remote sensing approach and classify each location as likely irrigated if it appears irrigated at any stage. Our model also uses a landform analysis to determine where irrigation is feasible. Either the presence of visible infrastructure or NDVI values higher than expected from rainfall data are then used as evidence for irrigated land. Paddocks identified by fieldwork as irrigated were 75.6% correctly classified, rising to 88.9% for a pasture subset. Cropping fields have proved more difficult as their NDVI also depends on their growth stage and the specific crop type.