How do I get MSL out of National Elevation Dataset?

I received some helpful e-mail that pointed out that the NED
elevations *are* orthometric (NAVD88). That's good to know and
certainly simplifies my life, but it doesn't explain some of the big
(over 400') differences I have observed. Back to the data...

First I eliminated facilities in Hawaii and Alaska. That got rid of
some examples with large differences (like MKK's 321 foot difference).
I then eliminated heliports. Those could be on top of buildings or
have high buildings immediately surrounding them. (CN43 had a 111
foot difference.)

Here's the resulting test data.
http://aviationtoolbox.org/munge/NED_test.out
The format is airport identifier, longitude, latitude, elevation (feet)
according to the FAA, elevation (feet) according to NED, elevation
difference (feet).

The FAA database is here
http://aviationtoolbox.org/old/ATA-1...txt?format=del
and it's described here.
http://aviationtoolbox.org/old/ATA-1...ats/Apt_rf.txt

The summary:
count_total=3959
count_missing_coverage=97 [ouch!]
count_missing_information=1
differences min=-391.917218
differences max=238.982513
differences mean=9.975323
differences standard deviation=18.448923

Only airports with surveyed elevations are examined here. The
elevation of the airport is "MEASURED AT THE HIGHEST POINT ON THE
CENTERLINE OF THE USABLE LANDING SURFACE" so it's not necessarily at
the reference point of the airport. I can easily imagine a 10 foot
difference due to that, so the mean difference of 10 feet seems within
reason but I expected a mean closer to zero.

The huge differences still surprise me. There are some, like 1U1
http://airnav.com/airport/1U1
with its 239 foot difference that I can imagine are victims of
inaccurate position data. Several airports have differences of over
100 feet.
43CA -115.733330833 34.8819330556 2611.0 2436.69845671 174.301543289
CN37 -118.230910278 35.3813477778 4071.0 4297.3856353 -226.3856353
1U1 -114.927333333 46.1207508333 2454.0 2215.0174873 238.982512704
88LL -89.8381755556 42.1916880556 917.0 789.834221466 127.165778534
MT07 -110.159080556 46.0416094444 5785.0 6176.91721816 -391.917218163
MT09 -106.608927778 48.8458569444 2948.0 3055.93195701 -107.931957014
69NC -81.8512119444 35.4340127778 1078.0 972.81228361 105.18771639
37ND -102.5090725 47.2319561111 2360.0 2257.08108279 102.918917211
8OA9 -81.81569 41.0444988889 1045.0 1217.27440486 -172.274404859
OR27 -123.167888611 45.6331680556 670.0 526.229929285 143.770070715
OR89 -120.052520556 44.9998552778 3966.0 3823.289852 142.710148004
WA79 -122.417875833 45.9331672222 451.0 685.433816825 -234.433816825
Maybe I should just assume that backwoods turf strip surveys aren't
accurate?

Others simply seem wrong. Why is it that all of the airports in Denver
are off by almost the same distance?
DEN -104.6731775 39.8616563889 5431.0 5382.27232302 48.7276769813
APA -104.849294167 39.5701283333 5883.0 5835.05466177 47.9453382333
BJC -105.117215833 39.9087866667 5670.0 5622.14489287 47.8551071296
Perhaps they were surveyed using NGVD29? I see that NAVD88 became the
official vertical reference datum in 2002.
http://www2.faa.gov/arp/aso/airporto...5/planning.htm

Plugging the Denver numbers into VERTCON
http://www.ngs.noaa.gov/cgi-bin/VERTCON/vert_con.prl
I only see a difference of under three feet.

I'm willing to believe this is just an accuracy problem, but I'll
welcome other insight.

Thank you.

--kyler