The 11.5 micrometer THIR radiances and the 0.36 micrometer and 0.38 micrometer TOMS ultraviolet (UV) reflectivities, together with concurrent Air Force surface temperature measurements, are the primary data sources for the cloud-detection algorithms. For sunlight overpasses, two independent estimates of total cloud amount are produced from both a UV reflectivity algorithm and an IR algorithm, which are then combined to form a composite estimate. At nighttime, only the IR algorithm is used.
In the IR algorithm, each THIR 11.5 micrometer radiance observation (pixel) is classified depending on its magnitude relative to concurrent radiance thresholds. The final classifications include clear sky and low-, middle- and high-altitude cloud. The threshold technique is based on surface temperature estimates to within one-half hour of a Nimbus-7 overpass obtained from time interpolation of Air Force surface temperature analyses archived every three hours. Adjustments are made for errors due to variations in water-vapor attenuation, known systematic biases in the attenuation adjustment, horizontal gradients in the surface temperature field, and for partially cloudy radiometer fields of view (see Fig. 3 in Stowe et al. 1988). Pixels with radiative temperature differences that exceed the threshold criteria are labeled cloudy.
The definitions of low, middle and high clouds differ from the ISCCP definitions, so direct comparisons of the data sets are difficult. In the Nimbus-7 CMATRIX data, an altitude of 2 km separates low- from middle-level clouds. The middle- to high-cloud altitude separation is 7 km equatorward of 30o latitude, but it varies with latitude farther poleward. These definitions are consistent with those in the International Cloud Atlas (WMO 1956).
A primary weakness of the IR algorithm is its inability to detect low-altitude clouds, which have very little thermal contrast with the surface. The TOMS reflectance measurements are used largely to offset this weakness. UV reflectivities measured by TOMS are very low and highly uniform for both land and ocean scenes which are clear and snow-free, and advantage over visible reflectance which is highly variable over land. The UV algorithm is very simple and is based on a linear relationship between UV reflectivity and cloud amount (Stowe et al. 1988). For middle- and high-altitude clouds, UV reflectivities are highly variable because they are very sensitive to cloud-thickness variations as well as to changes in cloud amount. Due to the large thermal contrast of these types of clouds from the surface, the IR scheme is, therefore, more accurate. However, low cloud is easier for the UV algorithm to detect accurately because of the large contrast in reflectance between the low cloud and the snow-free surface.
Once the clear-sky and low-, middle- and high-cloud classifications have been made, the TOMS reflectivity data, together with the IR scheme, can be used to produce estimates of cirrus and deep convective clouds. Essentially, cirrus clouds are defined by low TOMS reflectivities in the presence of a substantial amount of cold high-altitude cloud detected by the IR threshold technique. A shortcoming is that only cirrus without underlying cloud can be datected. From this same principle, deep convective clouds are identified when the IR scheme detects a large amount of cold cloud and the TOMS reflectivities are high.
Several methods have been used to validate the Nimbus-7 cloud climatology, as well as the auxiliary meteorological data used in the IR and UV detection schemes (Stowe et al. 1988). From cloud amounts derived by an analyst using geostationary images, systematic errors in the Nimbus-7 cloud estimates were estimated to be less than 10%, and random errors ranged from 7% to 16%. These empirical error estimates were consistent with results obtained from a theoretical sensitivity analysis, although it must be rememberd that the analyst also contributed systematic and random error to the comparison. Combining cloud estimates from the UV algorithm with those obtained from the IR algorithm changed the IR estimates by 10% or less; thus, the nighttime estimates should only be slightly less accurate than the daytime estimates outside the geographical regions where low clouds are prevalent. Cloud amounts over humid tropical regious were overestimated even when the UV agorithm was used, and cloud amounts over polar regions were less reliable because of the frequent presence of snow.
Many different fields have been archived for Nimbus-7 CMATRIX over the period April 1979 through March 1985 (72 months). As for ERB, ascending (near local-noon) and descending (near local-midnight) observations are given by "A" and "D". The radiances are in units of 0.125 W m-2 str-1.
Climatologies for each month have been produced from the six years of CMATRIX data, so January through March averages include the years 1980-1985, while the other nine months include 1979-1984. All of the fields have been averaged, except for the SIGMA fields. No attempt has been made to correct for possible biases introduced into the data, so the fields should be used with caution. The mean fields have the same processor names and units as means, but they are in "TYPEc = SAVTAV" format at 4.5o resolution.
References: