Surface water and flooding dynamics data set based on seasonally continuous Landsat data (1986-2011) in a dryland river basin

Animations of the data are available here: https://doi.org/10.5281/zenodo.2438110

If you are using this data set, please cite the following publication:

Tulbure, M.G. and M. Broich (2018). Spatiotemporal patterns and effects of climate and land use on surface water extent dynamics in a dryland region with three decades of Landsat satellite data. Science of the Total Environment. https://www.sciencedirect.com/science/article/pii/S0048969718347466 

The data represent statistically validated surface water and flooding extent dynamics derived from seasonally continous Landsat TM/ETM+ data and random forest models, and summarised to the maximum extent of surface water per season between 1986-2011 over Australia's Murray-Darling Basin. The overall accuracy was over 99% and producer's accuracy for water 87% +/- 3%. 

The method is described in the following publication: 
Tulbure, M.G., M. Broich, S.V. Stehman, A. Kommareddy. (2016). Surface water extent dynamics from three decades of seasonally continuous Landsat time series at subcontinental scale in a semi-arid region. Remote Sensing of Environment. 178: 142-157

URL: https://www.sciencedirect.com/science/article/pii/S0034425716300621 

Data are provided in GeoTIFF format per season per year. File naming convention is as follows:
yy_inund_freq_season_SamplingMethod. For example, "99_inund_freq_winter_max" will represent inundation frequency for winter 1999 resampled using a maximum resampling method. 

Inundation frequency represents the number of times a pixel has been flagged as flooded out of the times that pixel had valid observations * 100. Valid observation exclude no data values and clouds. The valid range of inundation frequency is 0-100 [%], with 255 indicating no data values. Data type is eight bit unsigned integer (uint8). 

The data were resampled to 120m resolution to reduce file size. The resampling methods used include max (e.g. selects the max value of all non-NODATA contributing 30m pixels) and mean (median and min can be provided upon request). If you are unsure which resampling to use, you may want to start with the mean.