From a paper on analemmatics, linked below:
In Kennett Square, Pennsylvania USA, not far from Philadelphia, there is a 1050 acre
horticultural park, Longwood Gardens. The park is on the former country estate of
Pierre S. duPont (1870-1954). The park’s last construction project overseen by duPont
was the design of a 37 by 24 foot analemmatic sundial in what is now a Topiary garden in
the park. The dial was completed in 1939 after more than six years of daily noon-time
observations.
"After about eight months of trying to use [calculated] measurements, Mr.
duPont got disgusted and said we would build the sundial ourselves after
working out our own measurements. Both Roland Taylor and I [Knowles
R. Bowen] began the task of taking actual observations on the sun at 12
noon every day, after checking our time with the Naval Observatory at
Annapolis. If the sun was not out, we could not get that day’s sighting
until the following year, or maybe two years later, which explains why the
project took so long."
At some point during this long process, a visit was made "to France to check out a sundial
there; we copied some details, but the dial was not too accurate...." - presumably this
was the dial at Brou. Some years later, in 1946, duPont commissioned additional work on
the dial, to reposition the hour-markings on the ellipse. One can only wonder if the
purpose behind this change was to attempt to correct the errors which no doubt were
becoming obvious in the dial readings. The errors would have resulted from placing a
standard analemma at the center of an analemmatic sundial, thus perpetuating the design
flaw begun at Brou - a flaw which could not be corrected by a change in the hourmarkings.
In the late 1960’s another analysis was done on the Longwood Gardens dial, enlisting the
assistance of P. Kenneth Seidelman of the U.S. Naval Observatory. Measurements
showed that the hour-points were positioned for standard time readings, but the double
analemmas then at the center of the dial proved to be little more applicable to a proper
design than a single analemma. To address this problem, Seidelman developed a weighted
average approach to defining substitute analemma curves which results in a close
approximation to mean time.
click here