These are scatterplots of GOES-derived precipitable water values vs. adjacent points, and I have done the scatterplot for different maxima in separation values between the adjacent points. I have considered data for one 24-hour period in late October 1999; I'm not sure if I can think of a reason why they would vary substantially day-to-day.

Scatterplot of Adjacent Points

Spacing less than 25 km

Spacing less than 30 km

Spacing less than 35 km

Spacing less than 40 km

What do these plots tell you? For close spacing there isn't a lot of difference in precipitable water from one point to another. Not too surprisingly, as the spacing increases, so too does the spread.

I wondered as I looked at these plots: is the difference a function of spacing? To visualize that, I plotted the percentage difference between adjacent points vs. the distance. Those scatterplots are below and they will show you that as the distance between the two points, the percentage difference can also increase.

Scatterplot of Difference vs. Distance

Spacing less than 25 km

Spacing less than 30 km

Spacing less than 35 km

Spacing less than 40 km

I sent this url to some satellite folks who work on the floors below me at the Space Science and Engineering Center, and here is part of the e-mail that I received:

Scott, 

I think the biggest differences are due to comparisons between retrievals over
land vs ocean. Unfortunately we don't (yet) store a land/sea flag in the
MDfile. 

We have seen how our retrievals more favorably compared to raobs at smaller
match distances. In fact, at large distances (say 1 degree), the guess is
better than the retrieval (because it's so smooth). But, by say 0.5 km the
retrievals are better. 

So I recomputed the scatterplot, restricting the comparisons to over land. I also used only GOES-8 or GOES-10 data -- I won't be comparing data from two different birds, as that could introduce variability as well. Here is a map of the locations included in the comparisons. A table with all the scatterplots follows. Recall that GOES 8 is over the equator at 75 W and GOES 10 is over the equator at 135 W. I also include here in the table, for simplicity, the scatterplot that contains all the data, i.e., those at the top of this webpage.

Scatterplot of Adjacent Points

All Data Points	Spacing less than 25 km	Spacing less than 30 km	Spacing less than 35 km	Spacing less than 40 km
Land-locked GOES 8 Points only	Spacing less than 25 km	Spacing less than 30 km	Spacing less than 35 km	Spacing less than 40 km
Land-locked GOES 10 Points only	Spacing less than 25 km	Spacing less than 30 km	Spacing less than 35 km	Spacing less than 40 km

A couple of things I notice about these scatterplots. The most obvious is that there is apparently more scatter in the GOES 10 instrument. This could be a function of many things and I mention it only as something that deserves further study. Note that the precipitable waters in both GOES 8 and GOES 10 fields of view are roughly equivalent, even though there are synoptic patterns where the deep south is significantly moister than the Rocky Mountains. These data were collected when an intense storm was off the east coast of the US and the eastern half of the country was under strong northwesterly flow, so the Gulf of Mexico and its moisture was shut off. (The date is 1 Nov 1999 -- the date the Egypt Air plane went down off Nantucket). (If you notice something interesting, please e-mail me! I also redid the plots showing the difference as a function of distance. They are below.

Scatterplot of Difference as a function of Distance

All Data Points	Spacing less than 25 km	Spacing less than 30 km	Spacing less than 35 km	Spacing less than 40 km
Land-locked GOES 8 Points only	Spacing less than 25 km	Spacing less than 30 km	Spacing less than 35 km	Spacing less than 40 km
Land-locked GOES 10 Points only	Spacing less than 25 km	Spacing less than 30 km	Spacing less than 35 km	Spacing less than 40 km

The first thing I noticed about this is that there is more scatter in the GOES-8 vs. GOES-10 data. So I sent this url to the folks below me who know more about GOES satellites than I do, and I got the following e-mail in return:

Scott, 

When one looks at the instrument noise, GOES-10 is cleaner than GOES-8. 

If GOES-10 has more scatter, I think it's due to the variability of the
terrain. Sfc pressure is very important for TPW determinations. To test this,
do your scatter plots sorting on the sfc pressure differences. 

(Note above: TPW is total precipitable water, which we are trying to measure). Now this made me think to redo the difference scatterplot with surface pressure as the y-axis; furthermore, I first did this in the region where the GOES-8 and GOES-10 satellites overlap -- the high plains of the USA. It is the intersection of the two areas shown in this figure -- between 35 and 41 North and between 100 and 107 West.

Difference as a function of surface pressure

Spacing less than 40 km [GOES 8]

Spacing less than 40 km [GOES 10]

Note above how as the pressure increases the spread difference between adjacent points increases. Can we conclude from this that the GPS/GOES comparison will be better over flat land? I think so. One might argue that the difference between adjacent points as the surface pressure varies is an error -- one could also argue, however, that the difference is real. As the surface pressure drops, the amount of atmosphere holding water vapor is reduced, so the precipitable water will of course decrease. At any rate, this suggests that if there is a big difference between a GPS and a GOES data point, it could be related to pressure differences in the data -- and this effect is much more likely over regions of variable topography (mountains).

Finally, I made a using all data (GOES 8 and GOES 10) and looked at the percentage difference in precipitable water at adjacent points as a function of pressure difference between the adjacent points. There is an obvious relationship between the two.

Difference as a function of surface pressure difference

Spacing less than 40 km [GOES 8 and GOES 10]