Comments from GeoCaching Listserv:


On Wed, 28 Mar 2001, Nowhere Man wrote:
>I recently revisited a cache, and my GPS was saying that I was still 0.03 miles away when I was right on top of it. The first time I had visited the cache, the reading was pretty accurate, 0.01 to 0.00 miles. What was different the second time? It was later in the day, cloudy, and colder, and the batteries were pretty low.

>So, here is what I want to learn. What effects do any of the following have on GPS accuracy? I have a Magellan 315 BTW.
Practically speaking, accuracy is most affected by how many satellites are visible, and how well dispersed the visible satellites are in the sky (but away from the horizon). Generally, the more satellites your GPS receiver can see, and the more widely dispersed those visible satellites are in the sky (but substantially above the horizon), the better. Consumer GPS receivers need to be able to see three satellites to provide a 2-d position and four satellites to provide a 3-d position; more is better. You can see the effect of satellite geometry on this U.S. Navy page (Trimble used to have a page that provided a more graphically intuitive display, but it's apparently no longer available):

Note that the system accuracy, in best-case visibility conditions, when consumer grade GPS receivers are used, is specified as a 95% probability that the displayed position will be within some ellipse having an approximate radius of about 15m ~~ 49 ft ~~ 0.0093 miles around the actual location. This means that there's a pretty good chance you will occasionally make an observation when the system error is worse, and therefore, the displayed position is outside that ellipse. If satellite geometry and visibility are unfavorable, accuracy can be much worse.

This topic has been discussed extensively in the sci.geo.satellite-nav Usenet newsgroup; if you're a masochist, you can try to wade through the Google (formerly Dejanews) archives.

>* Cloud cover
Water vapor attenuates microwave signals, so in theory, cloud cover could render an otherwise usable satellite unusable. It might also result in differential delay of signals from various satellites. These factors could affect accuracy. Practically, I haven't noticed significant attenuation despite heavy cloud cover, rain, etc.
>* Tree cover (leaves vs. bare twigs vs. no trees)
Heavy tree cover with dense leaves or needles can cause enough attenuation that a GPS receiver will not be able to see sufficient satellites to calculate a position. Even if you're in a clearing, adjacent tree cover, mountains, buildings, etc., can limit how much of the sky is visible to your GPS receiver, thereby degrading accuracy. These physical obstructions can also increase the receiver's susceptibility to multipath (basically, reflected signals), which also degrades accuracy.
>* Time of day
Accuracy depends on satellite geometry, which varies with time. Try the Navy website mentioned above to see how estimated position error varies as you change the time of day.
>* Temperature
>* Battery charge
In theory, either of these could affect the performance of the GPS receiver hardware, so that in marginal situations, satellites that might otherwise be usable could be rendered unusable. Also, these factors can impair the ability of the GPS receiver to measure time, which could also affect accuracy. In practice, I haven't noticed any difference in performance of my GPS receivers due to these factors (but I haven't carefully collected and analyzed accuracy data either).
>* Speed
Dunno. Some GPS receivers appear to use earlier position and velocity measurements to calculate current position. For example, if I walk around with one hand-held GPS (Brand "G") and abruptly stop, the displayed position continues to change for several seconds thereafter. Also, non-linear deviations from a linear course seem to get filtered out until they exceed some threshold. Your mileage may vary.
>* Idle periods to let it average
There was extensive debate of this some months ago on the sci.geo.satellite-nav newsgroup. I believe the thrust of discussion supposed that you would begin averaging for a short initial period of a few minutes, and considered whether additional averaging was beneficial. One camp claimed that additional averaging was beneficial; another camp claimed that the errors correctable by averaging were non-random and related to variables having long time constants, so that after the initial period, no benefit was achieved unless averaging (or at least sample collection) occurred over many hours. I thought the discussion became extremely pedantic, even for sci.geo.satellite-nav, and gave up. I forgot who won.