Seasonal Surface Reflectance - Landsat, JRSRP Algorithm, Australia Coverage

Brief description

The dataset consists of composited seasonal surface reflectance images (4 seasons per year) created from the full time series of Landsat TM/ETM+/OLI imagery. The imagery has been composited over a season to produce imagery which is representative of that period, using techniques which will reduce contamination by cloud and other problems. This creates a regular time series of reflectance values which captures the variability at seasonal time scales. The benefits are a regular time series with minimal missing data or contamination from various sources of noise as well as data reduction. Each season has exactly one value (per band) for each pixel (or is null, i.e., missing), and the value for that season is assumed to be the representative of the whole season. The algorithm is based on the medoid (in reflectance space) over the time period (the medoid is a multi-dimensional analogue of the median), which is robust against extreme values.

Notes

Lineage

Data Creation
Image Pre-Processing: All input Landsat TM/ETM+/OLI imagery was downloaded from the USGS EarthExplorer website as level L1T imagery. Images which the EarthExplorer site rated as having greater than 80% cloud cover were not downloaded, on the assumption that they would contribute little, and would add extra noise in the form of undetected cloud, shadow, and mis-registration (which is a greater risk in very cloudy images). The imagery has been corrected for atmospheric effects, and bi-directional and topographic effects, using the methods detailed by Flood et al. (2013). The resulting imagery is expressed as surface reflectance. Cloud, cloud shadow and snow have been masked out using the Fmask automatic cloud mask algorithm. Topographic shadowing has been masked using the Shuttle Radar Topographic Mission DEM at 30 m resolution, and the methods described by Robertson.
Seasonal Compositing: The seasonal composites were calculated using the medoid in the 6-dimensional space of reflectance values from the six Landsat reflective bands. The medoid is a “measure of centre” of a multi-variate set of points, similar in nature to the median of a univariate dataset. The medoid has similar desirable properties to the univariate median. In a general cluster of points, in n-dimensional space, the medoid will lie roughly in the centre of the cluster, making it a good choice as representative of that set of points. Most importantly, it is robust against the presence of outliers in the set, until at least half of the points are to be considered as outliers, after which it breaks down. Two variations of seasonal surface reflectance are available and can be distinguished by the ‘satellite category code’ component of the filename. (Refer to file naming convention information). ‘lz’ indicates that the composite was derived from all possible Landsat data, and ‘l8’ indicates that the composite was derived from Landsat8 OLI data only. For the latter case, the OLI Band 1 is omitted so that the resulting composite has the same bands as the generic Landsat ‘lz’ variant.
Image Restrictions: In estimating a value which is representative of the season, we choose to add a restriction on the number of input points. If a given pixel has less than three observations available for the season, after masking, we define the result as missing, on the principle that we do not have enough data to know how representative our choice might be. Neither the medoid nor the geometric median are robust against a single outlier in the case of less than three observations.

Notes

This product captures surface reflectance at seasonal (ie three-monthly) time scales, forming a consistent time series from 1987 - present. For applications that focus on vegetation changes, the fractional cover and ground cover products may be more suitable.