Day 4

Detailed paper information

Back to list

Paper title SLSTR pre-processor to co-locate reflectance and infrared observations
  1. Claire Bulgin University of Reading & NCEO Speaker
  2. Niall McCarroll University of Reading & NCEO
  3. Owen Embury University of Reading and National Centre for Earth Observation
  4. Christopher Merchant University of Reading
Form of presentation Poster
  • Open Earth Forum
    • C5.03 Open Source, data science and toolboxes in EO: Current status & evolution
Abstract text Bayesian cloud detection is used operationally for the Sea and Land Surface Temperature Radiometer (SLSTR) in the generation of sea surface temperature (SST) products. Daytime cloud detection uses observations at both infrared and reflectance wavelengths. Infrared data have a spatial resolution of 1 km at nadir, whilst the nominal resolution of the reflectance channel data is 500 m. For some reflectance channels, observations are made by a single sensor (Stripe A), whilst others in the near infrared include a second sensor (Stripe B).

Operationally, data at reflectance and infrared wavelengths are transferred independently onto image rasters using nearest neighbour mapping. The reflectance channel observations are then mapped to the infrared image grid by averaging the 2x2 corresponding pixels. This methodology does not achieve optimal collocation of the infrared and visible pixels as it neglects the actual location of the observations, and neglects orphan and duplicate observations.

A new SLSTR pre-processor has been developed that increases the field-of-view correspondence between the infrared and reflectance channel observations. This is beneficial for any application using reflectance and infrared wavelengths together, including for cloud detection.

The pre-processor establishes a neighbourhood map of reflectance channel observations for each infrared pixel. It takes into account orphan pixels excluded when compiling the image raster and ensures that duplicate pixels are not double-counted. It calculates the mean reflectance for a corresponding infrared pixel, using a configurable value of ‘n’ nearest neighbours. The standard deviation of the ‘n’ nearest observations can be calculated in this step, providing an additional ‘textural’ metric that has proved to be of value in the Bayesian cloud detection calculation. The pre-processor can also include data from the Stripe B sensor on request, where these data are available.

We demonstrate the improved collocation of infrared and reflectance channel observations using coastal zone imagery, where steep gradients in temperature and reflectance make it easier to visualise the improved collocation of the observations. We also demonstrate the positive impact that this new pre-processor has on the Bayesian cloud detection algorithm, demonstrating that cloud feature representation is improved.