Land cover mapping of Victoria, 2016

Land cover mapping data is an annual component of the Victorian Land Use Information System (VLUIS). The land cover information has been created specifically for the VLUIS using time series analysis of the MOD13Q1 or MYD13Q1 products produced by NASA using data collected by the MODIS sensor and freely available on the Reverb | ECHO website. Ground data is collected annually across Victoria using a stratified random sampling approach for calibration of classification rule sets based on annual seasonal curves derived from MODIS imagery, and validation of the classification output. The ground data is split into three groups with 50% used to develop classification rules, 25% used to produce interim validation results that feed back into the rule development process with the remaining 25% used to independently validate the final classification. Error matrices for each land cover dataset from 2009-2017 have been produced from this final validation. The TIMESAT program is used to calculate annual seasonal parameters from smoothed annual time series for the Normalised Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI) and the Red and Near Infrared (NIR) MOD13Q1 or MYD13Q1 bands. A time series of 23 images is used to calculate 11 seasonal parameters that 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. Each year, ground data and the seasonal parameters are used to generate a rule-based classification, using the See5 data mining program. Each rule used to assign a class to a pixel is given a confidence rating from 0 to 1, with 1 being the most confident. These ratings are generated by the See5 program. Where a rule has a confidence of less than 0.35, the pixel is assigned an unknown classification. This reflects the uncertainty in land cover classification for that pixel in a particular year. Additional input data, such as hydrological features and public land management (available from data.vic.gov.au) and inter-annual smoothing of the classification are used to minimise misclassification. Due to the spatial resolution of the MODIS imagery, land cover classifications are not calculated for urban areas where the size of the land parcels are smaller than 12.5ha. These areas have been attributed as urban areas using a mask. The dominant land cover classification includes the following classes: Water, Native Woody Cover, Deciduous Woody Horticulture, Evergreen Woody Horticulture, Hardwood Plantation, Softwood Plantation, Pasture and grassland, Brassicas, Cereals, Legumes, Non-woody Horticulture, and Bare and non-photosynthetic material. The methodology for creating the Land Cover datasets is fully described in Sheffield, K et al. (2015): Mapping dominant annual land cover from 2009 to 2013 across Victoria, Australia using satellite imagery. Scientific Data, 2, 150069, doi:10.1038/sdata.2015.69 A metadata statement, for the VLUIS product, and ESRI symbology files for the data can be freely downloaded from the VLUIS project page on the Victorian Resources Online website: http://vro.agriculture.vic.gov.au/dpi/vro/vrosite.nsf/pages/vluis