The Use of Satellite Data in an Optimal Interpolation Assimilation Scheme

Three hourly cloud data from the International Satellite Cloud Climatology Project (ISCCP) and from the GOES-VAS CO2 slicing technique datasets are used with hourly surface observations of cloudiness to build a comprehensive eight year cloud Climatology over North America. An optimal interpolation based assimilation scheme is employed to combine heterogeneous satellite and surface data containing known errors in a manner such that the resulting error of the final product is minimized. The assimilation makes use of observation weights, defined during quality control, to compute the conditional mean and variance of the cloud data. Cloud observations from hourly surface reports are used to calculate correlations in cloudiness and to develop spatial and temporal correlation models. The correlation models are applied to both surface and satellite data. The assimilation spreads the influence of the data effectively in both space and time via the correlations, and greatly improves the temporal interpolations necessary to build a one hour database.

The ISCCP pixel level (DX) data, which consists of cloud/no cloud determinations, is used in the assimilation system. During daytime, the cloud/no cloud determinations are made using both the infrared (IR) and visible channels while at night only IR determinations are available. "Super observations" (superobs) of the ISCCP data are created by weighting nine pixels in such a way as to yield the same effective viewing area as that of a surface observer (2400 km2). Quality control is applied to the satellite superobs. High clouds get high weight while low clouds are deweighted. This prevents the assimilation from drawing to erroneous detected low clouds particularly at night during the winter. High clouds are thus given more weight than low clouds while surface observations of low clouds are given greater weight than high clouds. Additionally, surface observations are weighted at night according to the amount of lunar illumination. On nights when the moon is not present the surface observer tends to underestimate clouds therefore they are given less weight.

Results indicate the assimilation system provides a unique blend of both satellite and surface observations. The assimilation accurately depicts the seasonal and diurnal variations of cloudiness present over the North America. During nighttime cases in the winter, when the ISCCP algorithm erroneously indicates low clouds, the assimilation accurately clears out the area based on available surface observations. In other cases, when surface observations indicate clear sky at night, the assimilation will draw to the satellite reports of high thin cirrus clouds. Other examples will be shown at the conference.

This paper will be most suitable under theme 2) Climatology and Long-Term Satellite Data Studies. An electronic poster will be given.