Cloud Detection Using Meteosat Imagery and Numerical Weather Prediction Model Data

Abstract

The cloud detection algorithm of the Royal Netherlands Meteorological Institute (KNMI) Meteosat Cloud Detection and Characterization KNMI (Metclock) scheme is introduced. The algorithm analyzes the Meteosat infrared and visual channel measurements over an area from about 25°W to 25°E and from 35° to 70°N, encompassing Europe and a small part of northern Africa. The scheme utilizes surface temperatures from a numerical weather prediction model. Synoptic observations are used to adjust the model surface temperatures to represent satellite brightness temperatures for cloud-free conditions. The measured reflected sunlight is analyzed using a minimum reflectivity atlas. Comparison of cloud detection results with synoptic observations of cloud cover at about 800 synoptic stations over land and 50 over sea were made on a 3-h basis for 1997. In total, two million synoptic observations were used to evaluate the detection method. Of the reported cloud cover, Metclock detected 89% during daytime and 73% during nighttime over land and 86% during daytime and 80% during nighttime over sea. The fraction of pixels labeled as cloud free in reported cloud-free conditions was 92% for daytime and 90% for nighttime over land and 94% during daytime and 90% during nighttime over sea. The largest contribution to the cloud detection capability is the threshold comparison of the satellite brightness temperatures with the adjusted model surface temperatures. The cloud detection method is used for the initialization of a short-term cloud prediction model and testing of cloud parameterizations of atmospheric models that will be used as an aid to meteorologists in analyzing Meteosat data.