Classifying agricultural land uses with time series of satellite images

Use of temporal signatures to classify the complex land use patterns in Canterbury New Zealand is investigated. Two datasets of detailed field histories spanning 5-7 years are used to develop a rule set for classifying from satellite imagery (SPOT/Landsat). While many individual crops proved inseparable, a broader set of land use classes such as `winter forage' or `summer arable crop' were defined that could be separated with accuracies between 71% and 96% (averaging 82% over 8 classifications). These broader land use classes will be used in a second stage of the project to guide and constrain classifications incorporating spectral information from the crop's peak NDVI phase.