Visible Product Descriptions
(click image to enlarge)

The formation of a satellite-based convective cloud mask is the first step in the development of a CI nowcast. IR-based CI interest fields are computed only where convective clouds are present, thereby excluding approximately 70-90% (on average) of a given satellite image from further processing. This step greatly reduces the time required to produce a CI nowcast, which allows for nowcasting in near real-time.

The convective cloud mask is based upon a multi-spectral region growing (clustering) technique for classifying all scene types in a GOES image. GOES pixels are clustered based upon statistical similarity. The similarity threshold for each GOES channel is user specified and is adjusted automatically based upon standard deviation. The thresholds chosen for each satellite channel determine how many clusters (i.e. scene types) are created. Currently, we have chosen thresholds that produce 11 clusters. These clusters include "uncertain" (i.e. pixels that do not fall into any cluster), land, water, snow, stratus/fog, semi-transparent cirrus, and 5 categories of convectively induced clouds. A database of labeled clusters is created from a set of VIS and IR training images over the area(s) of interest. The labeled cluster database forms the basis of the pixel classification technique. Pixels are assigned to a cluster based on the relationship between the pixel's multi-spectral properties and those of the clusters identified in the database.

This classification system is produced at the 1 km GOES VIS resolution and currently identifies 5 types of convectively-induced clouds: 1) small, low-level cumulus (blue), 2) mid-level cumulus (cyan), 3) mature cumulus clouds with depths extending through the entire troposphere (red), 4) thick anvil ice clouds (magenta), and 5) thin anvil and cirrus clouds (pink). For CI nowcasting purposes, analysis is primarily focused on low- to mid-level cumulus pixels, as these clouds likely have not begun to precipitate.

(back to top-->)