THE EXETER EXPERIMENT

by David F. Webb
Copyright 1973 David F. Webb

Editor's Note: In the March issue of UFO Investigator, NICAP reported on a small network of UFO detectors set up around Exeter, New Hampshire, by NICAP member John Oswald.  The detectors were deployed in homes of people who agreed to monitor the devices and record any times the detectors were activated.  A total of 15 detectors were operated at 13 different sites (two sites used two detectors)  .  The farthest site from Exeter was 7.8 miles (No.12), and the greatest distance between two sites was 13 miles(Nos. 1 and 12) .  Not all detectors were operated for the same length of time; Oswald, who ran site No. 1, maintained a detector for the longest period: 675 days (approximately 22 months).  Other units functioned for shorter periods, ranging from 31 to 650 days.  Only six sites operated on a 24 hour basis. The experiment began in 1970  and ended in September of 1972.
.
All detectors were identical in design.  They used a suspended magnet that triggered an alarm when affected by changes in local magnetic fields.  During operation, the detectors were protected from air currents, vibrations, and other non-magnetic forces that might cause the magnet to activate the alarm.  An effort was made to place the other detectors away from areas where unwanted disturbances might occur, such as effects from passing cars.  This effort was not totally successful.

It was not always possible for people who monitored the detectors to observe the sky when the detectors went off. In some cases, an alarm occurred at an inconvenient time or when weather conditions were poor.  Most of the detectors were operated only during periods when people were at home.

Oswald routinely maintained the system and kept careful records on the status of each unit.  He also collected sighting reports from the Exeter area for the duration of the experiment.  These totaled 46.  Of that number, he classified 22 as good or reasonably good.

The number of alarms recorded during the experiment was 659.  This was too large to attribute to any one cause  In assessing these results, Oswald sought the help of another NICAP member.  David Webb, a physicist in Boston.  Webb prepared a detailed report on his analysis of the detector data, concluding that most of the alarms could be linked to geomagnetic phenomena and other variables not related to UFOs.

Webb recently submitted a copy of his report to NICAP, which we are pleased to publish here in condensed form.  Some of the material we have included was prepared especially for this article.

===
Early in our study of magnetometer data, it became obvious to Oswald and myself that many of the alarms were trigger by geomagnetic activity, i.e, the action of the earth's magnetic field [as it] is related to activity on the sun, such as solar flares.  One conclusion of my analysis is that half of the data [obtained by Oswald]  can be reasonably removed as noise.

To produce the tables [included with this report], the following procedure was used.  Using the basic alarm data, times were converted to Universal Time (UT), so that the times for each alarm could be directly with international geomagnetic observatory data.  I then compared the alarm data with the various indices of geomagnetic activity which would best serve for filtering the data.  I chose the Kp index primarily because it has the best time resolution (3 hours) and is conveniently available in graphs showing many months of activity at a glance.  Every alarm was correlated with its corresponding 3-hour Kp index, and the alarm was placed in one of three categories depending on the values and general geomagnetic storm level.

The Kp index data are contained in NOAA's Solar-Geophysical Data Bulletins (1970-72).  Their derivation and reliability are discussed in detail by Rostoker (1972).  Briefly Kp, as well as most other magnetic indices, is derived from data recorded at individual magnetic observatories on 3-component magnetometers.  The data are recorded in either the local magnetic coordinate system (H,D,Z) or the more familiar geographic coordinate system (X, Y, Z).  It should be noted that Oswald's magnetometer only measures the Earth's local magnetic field in the horizontal (H,D, or X, Y) plane.  The field deviations at each observatory are analyzed in 3-hour intervals and converted to a quasilogarithmic index K.  The values of K range from 0 (low activity) to 9 (high activity).  The Kp index is then simply the mean of the K indices from 13 observatories between magnetic latitudes 47 and 63 degrees.

Rostoker emphasizes several possible pitfalls for anyone using the Kp index.  The index can only give a lower limit of the level of geomagnetic activity for individual events.  The spacing in the observatories used in the index is large enough that local perturbations may not be recorded at all in the index.  Thus the more localized field around, say, New England could be significantly disturbed and not recorded in the world-wide Kp index.  So we cannot say that storm activity was necessarily low everywhere just because Kp was small during the period in question.  However, a large value of Kp is usually a guarantee of general high geomagnetic activity.  This last point served as the basis for my filtering of the alarm data.

Following the above guideline, Rostoker suggests that a good indicator of storm activity [is a Kp of at least 2].  For alarm classification I used three categories based on geomagnetic activity:

        LOW  alarm occurred when [Kp was no more than 2] during the 3-hour interval in question and for the 3-hour immediately proceeding and following (i.e. for a 9-hour period roughly centered on the alarm time.)
        MODERATE  alarm occurred during a 3-hour interval when [Kp was no more than 2] but did no meet  the other criterion for LOW, or when [Kp was more than 2] but not during an obvious solar storm.
        PROBABLE  alarm occurred during a 3-hour interval when [Kp was greater than 2] and a general storm was occurring.

Obviously these ratings are somewhat subjective, but I feel they are descriptive since the first two categories are intentionally conservative.  LOW, in fact, , should be a good indicator of an alarm triggered by a non-geomagnetic event, while PROBABLE should be a good indicator of an alarm set off by a geomagnetic storm.

The accuracy of the PROBABLE category was tested using data from Oswald's own magnetometer (No.1), which I consider the most carefully managed detector, and also the detector with the highest number of alarms (probably not coincidentally).  During nearly every major storm during the study period, Magnetometer No.1 was quite active.

Many of the other sensitive magnetometers, on the other hand, did not consistently have alarms during the same strong magnetic events.  In fact rarely did more than two magnetometers have alarms on the same day.  This is puzzling, but may be due to such factors as poor calibration, poor management (e.g. not resetting the device after an alarm), local magnetic disturbances, or local geomagnetic field variations.  Also the detectors were not always in operation simultaneously.

A word on detector calibration is necessary at this point. If we assume that the relative calibration of the magnetometers is valid (i.e.all the detectors are measured against two standard magnets), we should be able to arrive at a rough Kp-vs.-magnetometer sensitivity threshold for each magnetometer.  The attempt fails with this data probably for the reasons cited in the last paragraph but also because the Kp index is not necessarily a good indicator for the individual magnetic events at one location.  Therefore, the fact that a detector alarm occurred during a period of low geomagnetic activity (LOW) does not necessarily imply a non-geomagnetic cause; the magnetometer may be sensitive to even slight magnetic fluctuations or localized activity.

The converse statement also applies to events in the PROBABLE category.  Thus an alarm occurring during a period of high geomagnetic activity does not necessarily imply a geomagnetic cause.  Other sources such as automobile engines, AC fields, or even UFOs (may have triggered) some alarms. However, I decided to exclude all alarms in the PROBABLE category [for purposes of correlation] simply because magnetic activity could easily have triggered if nothing else had.

The magnetometer data presented in Table A [include] so-called "Reliability Index"which is an indicator of the reliability of a given magnetometer's operation. It is the product of the magnetometer's sensitivities times the total number of it produced divided by the total days of operation, This quantitative scale indicates that Oswald's detector is indeed the best managed detector, and reinforces Oswald's statements about the detectors with the lowest RI.

Once [filtered alarm data - those in the LOW and MODERATE categories - ] had been produced, I could then try to extract UFO "signal" from the filtered data by checking for times correlation with Oswald's list of UFO reports from the Exeter region during the same general time period that the magnetometers were operating.  My expectations were low for several reasons;  1) I doubted that the response time and sensitivity of the detectors were sufficient to record the field of a fast, probably distant sky object;  2) The statistical quality of the reports is low (more than 2/3 are uncertain in either data or location or both;  3) Most of the filtered alarms might very well have been caused by local geomagnetic events, passing autos, or other mundane electromagnetic disturbances.  Therefore, I was surprised to discover that 13 of the UFO reports were correlated in time with 24 alarms,, although only 4 of these reports had alarms occurring within 80 minutes of the [reported] sighting time.

Table B lists the correlated alarm data under two classifications, "Primary" and "Secondary."  The primary correlations are those rated highly because Oswald rates the UFO report as probable or unknown (this is class A or B), and the time difference between the report and the alarm is known and small.  Note that I did not try to include a weighting factor for the proximity   of the detector to the UFO [location] : it is very difficult in most cases to determine the actual location UFO as opposed to the location of the witness.  Even so it is interesting that 3 of the 4 Primary reports have known distances [to the UFO] that are no more than 4 miles.

The secondary correlations have insufficient data (mainly the exact date and/or time are unknown), are rated low by Oswald, or do not correlate closely in time.  A UFO report was included in Table B if an alarm occurred within 12 hours of the report time.  At first glance this conservative figure implies that all secondary correlations are suspect; only [one report] has an alarm occurring within a reasonable time of the sighting, but it is classified as C.  The list may prove useful, however, if all the reports with time uncertainties are further investigated to better determine the date and time of each.

My study compared only the filtered alarm data against the UFO report list.  No attempt was made to study the reverse correlation:  the UFO reports against all of the alarm data.  This processing might uncover some interesting material.  If the quality of the report data were improved, a statistical study of the UFO reports vs. all alarm data would be useful.  But one would have to be quite stringent in the time correlation because of the inclusion of the alarms during high  geomagnetic activity.  If carefully done, the reports could be further correlated with distance, detector sensitivity, and report class.

Observation of the behavior of the two sets of dual magnetometers (sites No. 1 and 5) gives insight for a method of minimizing noise.  Oswald maintained two detectors at his site (No.1), one very sensitive (33 in.) and the other relatively insensitive (24 in.)  The 33 in. gave more alarms than any other detector, while the 24 in. gave only 11 alarms.  All of these latter alarms occurred during intense geomagnetic activity.  The two detectors at location No. 5 were more closely matched in sensitivity, only 4 to 6 inches different in calibration distance.  They recorded most of their alarms simultaneously.  The implication is that all such magnetometers should be used in tandem, with their individual sensitivities chosen such that one will discriminate against geomagnetic activity.  Oswald's 33 in. detector was obviously too sensitive to storm activity, while the 24 in. was probably nearly correct for detecting only the largest storm events and large local disturbances.  It must be remembered that the field intensity falls off as the inverse square of the distance, so the difference in calibration of the two detectors need not be large.  The use of carefully calibrated tandem detectors also will permit the cross-measurement of field strength of a source if it triggers both detectors simultaneously.