posted 10-21-2005 10:35 AM            

Edit: Well, it does say log term means ... and I initially took (mis-took) that to be a trailing average of some number of days ... but it could be that they're talking about long term average for the actual date, in which case the mismatch would be very curious.

posted 10-21-2005 12:12 PM            

quote:

---

Originally posted by Jeff Norman:
In a graph that purports to report the annual "Long-Term Means" for lake level, wouldn't you expect the December 31 value to be very close to the January 1 level? Closer than the linked graph?

---

Why?

posted 10-21-2005 12:56 PM            

quote:

---

Originally posted by Jeff Norman:
... wouldn't you expect the December 31 value to be very close to the January 1 level? Closer than the linked graph?

---

No, I'd expect someone to just enter the data as Jan through Dec. Then, when the data points are connected with lines, there'd be no end point for Dec. 31. So, the graphing program would just extend the Nov. slope throughout Dec. instead of decreasing it to match the Jan. 1 point.

posted 10-21-2005 01:51 PM            

Assuming the same several years were used for each day and the "several" years was more than ten (say), then I would have thought that the long-term mean for December 31st would be similar to January 1st as December 30th is similar to December 31st.

Is this wrong?

posted 10-21-2005 02:10 PM            

quote:

---

Originally posted by Jeff Norman:
I had assumed that they recorded the water levels over several years and then averaged daily values to create a long-term mean.

---

Then you would have assumed wrong. In clicking around their site it appears they record water levels at 6 minute intervals with multiple samples for each 6 minute result. Further, the data--at least what is available from the site--starts in January of 2001. In looking at the 01/01/01 data the data is very low relative to the same day for other years. Hence it very well could be pulling down the average since there are so few 01/01 observations.

So the problem appears to be:

1. Small sample.
   
2. An "outlier" in that sample.
   

Given more data, you'll probably find that the data will come more in line with your expectations.

Edits: UBB and clarity.

------------------
Steve
"In a nutshell, he [Steve] is 100% unadulterated evil. I do not believe in a "Satan", but this man is as close to "the real McCoy" as they come."
--Jamey Lee West
steve_v@steveverdon.com
Deinonychus antirrhopus

posted 10-21-2005 02:50 PM            

I would have thought that "long-term mean" meant the same thing as in this graph.

How would one get "long-term" from four years?

posted 10-21-2005 04:51 PM            

quote:

---

Originally posted by Jeff Norman:
Where did you find that?

I would have thought that "long-term mean" meant the same thing as in this graph.

How would one get "long-term" from four years?

---

I found a link to NOAA data sites. It isn't on the site you found the graph. Link (hope that works).

As for the longer period, there is a problem trying to reconcile data that is collected via different methodologies. For example in looking at the verified data it looks like they have data that goes back further (to 1996), but not at the six minute level. So it is also possible that the change in methodology is causing something bizzare to be happening. For example, if older data is only daily with one sample for that day then you could get some end up with some unusual data, especially if there are only a few sites where data is being collected.