Multi-spectral cloud classification for precipitation estimation

Global observations in the visible, water vapor, and thermal IR bands are currently available at relatively high temporal and spatial resolution from the existing constellation of GEO satellites. As more spectral bands become available from recently launched (e.g, MSG_SEVIRI) and future (e.g, GOES-R_ABI) sensors, more emphasis on the analysis and development of regional/global multi-spectral precipitation retrieval algorithms will be expected.

A technique that integrates two previously developed Precipitation Estimation from Remotely Sensed Information using the cloud classification system (PERSIANN-CCS) and multi-spectral analysis (PERSIANN-MSA) is presented. The technique employs multi-spectral data to support classification of cloud types both in cloud-patch and grid-box scales. The proposed method was tested using GOES12 visible (0.65µm) and infrared (10.8 µm) data over the US land mass east of 115W longitude line. The integration technique derives fine scale (0.04°x0.04° lat/long every 30 min) rain rate for each grid-box through three major steps: a) segmenting clouds into a number of cloud-patches using infrared images, albedo images, or both, b) associating each grid-box with the appropriate corresponding cloud type, obtained from bi/single spectral cloud classification, and c) extending the probability matching technique to establish a multi-spectral-rain rate relationship, for each cloud type.

Radar estimates were used to evaluate the technique at a range of temporal (3 hourly and 6 hourly) and spatial (0.04°, 0.08°, 0.12°, and 0.24° lat/long) scales. The evaluation at 3-hour 0.04° resolution showed 66 % improvement in equitable threat score and 26% in the correlation coefficient. This indicates that using albedo during day time hours results in significant gains over infrared-only techniques.