Satellite Data Problems

- August 4, 2002 -

There are two kinds of common problems on using meteorological satellite data to predict earthquakes. One of them is general that affects the both meteorological study and earthquake prediction. The other one is special that only affects earthquake prediction. This essay will discuss the both problems, and how they affect earthquake prediction.

The General Problems consist of a low frequency problem, a data loss problem, and a data error problem. First, different homepage of satellite imagery has a different frequency, such as one image every 15 minutes, 30 minutes, an hour, 3 hours, 6 hours, and so on. For earthquake prediction, a frequency lower than an image an hour induces four difficulties.

The first difficulty is hard to catch short-life earthquake clouds. For example, the Northridge earthquake cloud existed for 35 minutes [1], so a database with a frequency an image every three hours takes a risk 1 in 5 (=35/180) to capture it.

The second difficulty is hard to identify an earthquake cloud. A main principle to judge if a cloud is an earthquake cloud is to check whether or not it appears suddenly, so low frequent data often make the principle inefficient. For instance, the linear cloud in Image 20000426 12:00 [2] was the only one in the database, whose frequency is an image every 3 hours and sometimes more. Thus, I was confused if the cloud appeared suddenly until the 6.4 Chile earthquake happened on June 16, 2000.

The third difficulty is hard to guess which direction an earthquake cloud comes from. For example, the long, even, linear cloud in Image 20010430 18:00 [3] was sure as an earthquake cloud. However, the direction of the related epicenter was a puzzle until the 7.9 Peru earthquake happened on June 23, 2001 because the cloud had no partner to compare how it moved.

The fourth difficulty is hard to narrow an area window, which increases the probability of a prediction sometimes. For example, if an earthquake cloud moves with a rate 33 km/h from Southern California, a frequency of an image every 3 hours can produce a distance error about 100 km or one degree. The following comparison shows a big increase of the probability just due to one degree in longitude.

Time Latitude (N) Longitude (W) Magnitude(ML) Probability (%)

30 33~35 117~120 >=4 13.1

30 33~35 116~120 >=4 28.9

Second, data loss is another common problem. One or more days without data are not strange; for example, the homepage of the Ohio State University offered nothing from September 17 to October 14, 1999. Without data, it is impossible for a common person like me to guess what happens.

Third, data error is also a common problem. Here are a few examples. Image 20010619 9:00 [4] is blank. Image 20010613 18:30 [5] is black. Image 20020702 3:30 [6] gives a part of data. Image 20010301 13:00 [7] has no meaning. All above data are useless.

In short, the above three kinds of general problems work neither for earthquake prediction, nor for meteorological study. By contrast, the Special Problems do not work just for earthquake prediction. There are two kinds of the special problems. One of them is blind to the trace of earthquake vapor generally. The other of them is a low precision.

Primary, earthquake vapor exists, demonstrated by the Dehydration Theory [8], so its trace should exist, and link to the vapor source. After analyzing why satellite images were blind to the trace, I foresaw a possibility to catch a few examples of the traces while their surroundings are cold. Under this view, I found two traces called geoeruptions later: one from about 38N, and 122.8W, Central California on August 14, 1999 [9], and the other from about 40S, and 73W, Chile on August 15, 1999 [10]. Obviously, meteorology could not explain those sudden increases of the local heats.

Coincidentally, an earthquake of magnitude 5 happened at 37.91N, 122.67W or the primal point of the California geoeruption, on August 18. It was the only one of magnitude equal to or bigger than 5 in the area of 34~41N and 119~126W within 474 days from November 28, 1998 to March 15, 2000. Similarly, the 6.4 Chile earthquake happened at 40.51S, 74.76W or another primal point, on August 22. It was the only one of magnitude equal to or bigger than 6 either within 25~55S, 30~100W, and 461 days from September 4, 1998 to December 11, 1999, or within 37~43S, 30~100W, and 992 days from April 2, 1998 to December 19, 2000.

A more persuasive example is the Hollister, Central California Geoeruption on March 20, 2001[11] and its related 4.1 Hollister earthquake on July 2 [12] This earthquake had been pinpointed and predicted to the US Geological Survey with a time probability 10.2% correctly. The following is a comparison between the prediction and the earthquake.

Time Location Magnitude(ML)

4/3~7/2/2001 Hollister, California >=4

7/2 17:33 36.69N, 121.33W Hollister,California 4.1

Besides geoeruptions, I found the trace of the 4.3 San Fernando earthquake cloud on January 3, 2001 [13]. I had pinpointed its epicenter and mentioned its magnitude equal to or bigger than 4.3 in our homepage [14]. On January 14, the 4.3 San Fernando earthquake occurred exactly at the pinpointed dot (34.29N, 118.40W), San Fernando, Southern California [15].

In short, all above facts prove that the both earthquake vapor and its trace exist objectively, and the trace begins from an impending epicenter indeed. Under a cold surrounding, the trace appears close to its vapor source in meteorological satellite images. However, under a hot surrounding, the trace does not show near its vapor source in modern satellite images. This problem brings three big disadvantages.

The first disadvantage is no idea to predict a good area. For example, I found a long earthquake cloud near Sri Lanka from Images19990716 6:00 to 15:00 [16], and foreknew an earthquake of magnitude more than 7 in the area from Iran to Italy within about 49 days. However, those images did not show the trace to the vapor source, so I had to wait for the earthquake to know the epicenter until the 7.4 Izmit (40.67N, 29.82E), Turkey earthquake on August 17, 1999.

The second disadvantage is to cause failures of predictions. More than 50% of my failed predictions, including recent Japan earthquake prediction, are due to this problem (Refer to "Past Predictions" [17]).

The third disadvantage is sharply increasing the probability of a prediction. For example, I predicted an earthquake of magnitude equal to or more than 5.5 in Turkey and Eastern Mediterranean (>15E) from May 28 to July 12, 1997. I further predicted the area in 25~35E. The both are correct, but they have different time probabilities. The former is 35.6%, and the latter is 14.7%. Moreover, if satellite image had given good data to narrow the area window to 35~37N, 30~32E or +/-1 degree, the probability would reduce to 1%, and the predicted earthquake would be the only one there within 12 years at least.

The other special problem is that modern satellite images are not precise enough to divide very close erupting vapor sources into individual. Therefore, if vapors erupt from a group of adjacent hypocenters together to form one earthquake cloud, there is no way to detect how many earthquakes will happen, and a mistake occurs. For example, according to the linear cloud in Image 19991224 10:00 [18]. I predicted a M7 earthquake in the area of 25~28S, 60~80E, Southern Indian Ocean from December 27, 1999 to February 10, 2000. However, no M7, but a group of six M5 earthquakes happened on February 10 instead. The following is a comparison between the prediction and those earthquakes.

Item Time Latitude (S) Longitude (E) Magnitude

My prediction 12/27/99~2/10/00 25~28 60~80 >=7 ML

Earthquake 1 2/9/00 18:40 27.62 65.72 5 Mb

Earthquake 2 2/9/00 18:40 27.69 65.71 5 Mb

Earthquake 3 2/10/00 14:18 27.58 65.73 5.4 Mb

Earthquake 4 2/10/00 14:18 27.66 65.68 5.4 Mb

Earthquake 5 2/10/00 23:00 27.58 65.78 5.3 Mb

Earthquake 6 2/10/00 23:00 27.63 65.76 5.4 Mb

The group was the only one in whole Indian Ocean in the World Earthquake Database of the USGS from 1990 to 2001 at least. Because of this problem are 3% of my failed predictions.

The precision problem also increases the probability of a prediction. For example, I predicted a M5 or two M4 earthquakes in the area of 40~52N, and >120W from April 28 to May 18, 2001. Later a 5.1 earthquake fell into the all predicted windows correctly, but the time probability of this prediction increases from 27.5% to 63.8% just because the data do not distinguish a M5 from two M4.

Besides above two kinds of common problems, different homepages sometimes offer conflicting images. For instance, the Ohio State University offers a linear cloud and an arc-shaped cloud over Northern California in its image at 3:00 of September16, 1999 [19];while those clouds do not exist over the same place at the same time in another image of the Utah State University [20].

Moreover, the qualities of images among various homepages are very different. For example, American satellite images usually do not show wave-shaped earthquakes clouds although eyes can see and cameras can photograph them clearly. This problem implies that using American images to predict earthquakes will lose many earthquakes. By contrast, some Japanese images can reveal this kind of clouds over Japan clearly [21].

All in all, meteorological satellite data have two common problems, the General Problems, including a low frequency problem, a data loss problem, and a data error problem, and the Special Problems, containing blind to the trace of earthquake vapor, and a low precision. The both problems make predicting earthquakes impossible sometimes, cause mistakes, and increase probabilities sharply. Furthermore, meteorology can explain neither why this cloud [13] came from a fixed point about (34.3N, 118.4W) of the ground on January 3, 2001 suddenly, nor why this heat current [11] appeared from Hollister on March 20, 2001 suddenly. On the other hand, no other seismology can explain why an earthquake, for example the 4.3 San Fernando earthquake[15] on January 14, 2001, or the 4.1 Hollister earthquake [12] on July 2, 2001, occurred in a special time, at a special place, and with a special magnitude, coincidentally following the cloud[13], and the geoeruption [11] respectively. By contrast, the Dehydration Theory [8] answers the both puzzles, and predicted the both earthquakes reliably. Therefore, it is necessary and urgent to create a new satellite data system to overcome those satellite data problems for earthquake prediction. Otherwise, people will dig or will be dug after a devastating earthquake continuously.

References