Victorian Land Cover Mapping 2015

The TIMESAT GUI is used to create smoothed annual time series for the Normalised Differenced Vegetation Index (NDVI) and the Red and Near Infrared (NIR) MOD13Q1 bands using the Savitsky-Golay algorithm. A suite of 11 seasonal parameters are created that each numerically describe features of the annual seasonal curves for each band. In addition the standard deviation of the annual seasonal curve is calculated for each band and used in conjunction with the seasonal parameters.\r\nPrior to classification, the ground data and the MOD13 data is stratified into regions with similar climate, soils and farming systems using the Primary Production Landscapes developed by DEDJTR. For each stratified region a three-tiered hierarchical classification was developed to assign a dominant land cover class to each pixel. Initially, rules developed using expert knowledge were applied to the seasonal parameters and the annual standard deviation in conjunction with a GIS data-set of water bodies greater than 12.5ha in area to classify each pixel as either Tree, Non-tree or Water based on two data sets from the corporate spatial data library, HY_WATER_AREA_POLY.shp and VM_LITE_HY_WATER_AREA.shp; and are combined to form the water bodies layer. In addition, the primary classes are cross checked using data from preceding and following years to reduce misclassification prior to the secondary classification.\r\nA secondary classification developed using rules based on expert knowledge is applied to split the primary class Tree into the secondary classes Native Woody Cover and Treed Production and the primary class Non-tree into the secondary classes Pasture/ Grassland and Crops.