Statistical Analysis of CERES Data with Simultaneously Collected S'COOL Data
The Students' Cloud Observations On-Line (S'COOL) data are invaluable to the CERES project. Students from over 500 schools around the world collect data at close to the same time the CERES instrument is passing overhead. These observations are used as the "ground truth" to compare to the cloud data derived for CERES. The ground truth data are used throughout the validation process to help verify the cloud identification used by CERES. If there is a discrepancy in the two sets of data, the specifics of the data are analyzed to try to explain why there is not a perfect match. The reason for S'COOL is to help better understand all the data that are received from CERES and create a more solid, sound set of data for use in statistical studies.As of the summer of 2000, there were 639 entries in the satellite
database, and over 4000 records of ground data. This included data
processed automatically during spring 2000 for the CERES instrument on TRMM for
the period April through August of 1998. During that time, early in the S'COOL
project, a few schools around the globe were making ground observations.
Comparing the ground and satellite databases, 99 cases were found where the
data for the same date and location were taken within 15 minutes of each other.
The cloud fraction and cloud level information from these 99 matches are
summarized below.
| Summary of 99 Ground/Satellite Matches: Comparison of cloud fraction seen by satellite vs. ground | |||||
|---|---|---|---|---|---|
| Clear | Partly | Mostly | Overcast | ||
| S | Clear | 27 | 2 | 2 | 0 |
| A | Partly | 7 | 10 | 2 | 1 |
| T | Mostly | 5 | 3 | 12 | 7 |
| Overcast | 0 | 1 | 8 | 11 | |
The satellite to ground comparison is very important in validating CERES data. In a previous analysis (Rossow et al., 1993) ground observations were also compared with satellite observations. The results from his experiment are similar to those obtained here, with a large number of cases where the two agree exactly and fewer and fewer cases with 1-class and 2-class errors. This is very similar to the data gathered through S'COOL and CERES, which provides a sense that the data were analyzed in a manner consistent with previous experimentation and with similar levels of agreement. In addition, a statistical analysis called the Chi-Squared shows that the chance of obtaining this level of agreement by chance is very small: 1.4e-19 %!!
The cases where there is disagreement offer the possibility of learning something, so let's go on to Examine the Discrepancies.
You can also read the Complete Comparison Report(pdf).
