Bayesian Cloud Detection for Remote Sensing of Land Surface Temperature
Bulgin, Claire1; Old, Christopher1; Merchant, Christopher2; Sembhi, Harjinder3; Ghent, Darren3
1University of Edinburgh, UNITED KINGDOM; 2University of Edinburgh/University of Reading, UNITED KINGDOM; 3University of Leicester, UNITED KINGDOM
Accurate cloud detection is fundamental to remote sensing of surface temperature. Cloud detection error was identified as a major limitation for remotely sensed land surface temperature products in a community paper of the EarthTemp Network (www.earthtemp.net). We present here a Bayesian approach to cloud detection, arguing that this approach can outperform threshold based methods. Bayesian cloud detection is a probabilistic approach comparing observations with clear-sky radiative transfer, cloudy PDFs and associated errors to give a probability of clear sky conditions. It can be applied to a variety of Earth surface types and atmospheric conditions and generalised more easily than thresholds based on a subset of available data. The appropriate probability threshold for cloud clearing can also be determined at application level. We present results from several projects looking at the effectiveness of the approach. In an ESA project, SEN4LST, we demonstrated that Bayesian cloud detection over land using only 11 and 12 micron channels (no visible channels) outperformed an operational cloud mask for AATSR. Joint visible-thermal Bayesian detection has been further developed within the context of the projects for the UK National Centre for Earth Observation (AATSR) and the US NOAA (geostationary imagery). Results for Bayesian cloud detection including visible imagery around 0.66 and 1.6 micrometres together with thermal channels will be presented.