Automated Recognition of Oceanic Cloud Patterns. Part II: Detection of Air Temperature and Humidity Anomalies above the Ocean Surface from Satellite Imagery

Abstract

The usefulness of cloud classification for detecting and quantifying air temperature and humidity anomalies above the ocean surface is examined. Cloud fields are classified in 20 classes following the automated method of Garand (1988), here applied over the northwestern Atlantic during the winter season. From collocation of the classified cloud fields (scale of ≈130 km) with ship or buoy observations of air temperature and humidity, significant anomalies are found for specific cloud classes while for other classes no anomaly is found. All results are verified from independent data taken in early 1984 and 1986. The results confirm that for the mesoscale cellular convective patterns (MCC), i.e., cloud ?streets?, rolls, and open cells, the air and dew point temperatures are colder than climatology by several degrees, implying large latent and sensible heat fluxes. A latitudinal dependency of the anomaly is also observed. The removal of this bias provides estimates of surface air temperature with an accuracy of 2.8 K for these cloud types. Cirrus cloud classes and low stratus are associated with surface relative humidities above 80% while MCC patterns are associated with relatively dry surface humidity, below 70%. For those classes, the dew point depression can be inferred with an accuracy of 2 K; the corresponding relative humidity is determined with an accuracy of 10%. The implications for numerical weather prediction are discussed by comparing the error statistics of the satellite estimates with those of the trial fields (6-h forecasts) used in the analysis cycle of the Canadian Meteorological Center. The humidity estimates are expected to have a greater influence than the temperature estimates because the temperature field is already well analyzed by conventional means whereas the humidity analyses are often deficient.