In search of earthquake precursors in the water-level data of 16 closely clustered wells at Tono, Japan

Transcription

1 Geophys. J. Int. 2000) 143, 469±477 In search of earthquake precursors in the water-level data of 16 closely clustered wells at Tono, Japan C.-Y. King, 1,2, * S. Azuma, 1 M. Ohno, 2 Y. Asai, 3 P. He, 1 Y. Kitagawa, 1 G. Igarashi 2 and H. Wakita 2 1 Tono Geoscience Center, Japan Nuclear Cycle Development Institute, Toki, Gifu, 509±5102, Japan 2 Laboratory for Earthquake Chemistry, University of Tokyo, Bunkyo-ku, Tokyo, 113±0033, Japan 3 Tono Research Institute of Earthquake Science, Mizunami, Gifu, 509±6132, Japan Accepted 2000 July 11. Received 2000 July 5; in original form 2000 February 17 SUMMARY Water-level data recorded during 1989±1999 at 16 closely clustered wells near the Tono mine in central Japan are searched for possible premonitory changes before earthquakes. The results suggest that some water-level drops at a sensitive well before an M5.8 local earthquake on 1997 March 16 and some large distant earthquakes during 1994±1995 in Japan may have been premonitory in nature. The high sensitivity of this well to earthquake occurrence is attributed to the near-critical hydrological conditions and the relatively high permeability of the tapped aquifer. Key words: earthquake prediction, earthquakes, faulting, geochemistry, hydrology, permeability. INTRODUCTION Short-term earthquake prediction depends on the discovery of various earthquake precursors, but true precursors have been very dif cult to nd. The main dif culties include the following. 1) Anticipated earthquakes have not occurred in intensively studied areas such as Park eld in California and the Tokai area in Japan where many instruments are deployed. 2) Many earthquake-related geophysical and geochemical changes recorded at different sites, some next to each other, are commonly quite different, and the reasons behind this are not well understood. 3) Usually, among many monitoring sites only a very few appear to be sensitive enough to record earthquake-related anomalies. 4) Unlike data from a controlled experiment in the laboratory, eld data for earthquake prediction are simultaneously in uenced by many environmental variables and thus contain various types of background noise; many recorded changes previously thought to be premonitory have turned out under more careful examination to in fact be noise. * Corresponding author. Now at: Earthquake Prediction Research Inc, 381 Hawthorne Avenue, Los Altos, CA 94022, USA. chiyuking@aol.com 5) To nd earthquake precursors requires persistent monitoring over long periods of time, during which instrument failures and personnel/program changes inevitably occur, resulting in data loss or deterioration. 6) There is a lack of genuine understanding of the underlying mechanisms of various earthquake-related anomalies. In a previous paper King et al. 1999) we reported a study of water-level data, in particular many coseismic changes, recorded during 1989±1998 at a cluster of 16 wells in the Tono mine area in central Japan. We found that the response of water wells to earthquakes may be largely controlled by local hydrologic conditions. Thus, different wells may respond differently even if they are situated near to each other. Some wells may be sensitive to earthquake occurrence because the tapped aquifer/ barrier system is in a near-critical condition, such that a slight shaking or stress change may cause a permeability change that allows signi cant groundwater movement. In such a situation, the sensitivity of a sensitive well may change with time as the local stress condition changes. In the present paper, we focus our attention on water-level changes recorded before earthquakes, checking whether they may be premonitory in nature. MONITORING WELLS The Tono mine is located at uE, 35.39uN in Gifu prefecture, about 350 km southwest of Tokyo, within a relatively stable block in central Japan bounded by major faults where # 2000 RAS Knopoff Festschrift 469

2 470 C.-Y. King et al. large earthquakes have occurred historically Fig. 1). The study area Fig. 2) is traversed by a reverse fault, the Tsukiyoshi Fault, in an east±west direction. The fault has a dip angle of about 60±70u to the south, an offset of 30 m and an ordinarily impermeable fault gouge zone about 10±30 cm thick sandwiched between two fracture zones [as in the case of many other active faults see King et al. 1996)], each several metres thick Kiode et al. 1997). On the south side of the fault, there is a 126 m deep underground gallery, from which the in owing groundwaterhas been continuously pumped out at a rate of about 50 t per day for more than 20 yr in order to keep the gallery dry. This pumping activity has decreased the water pressure on the south side while leaving the pressure on the north side almost hydrostatic, thus creating a considerable hydraulic gradient across the fault gouge zone. The waterlevel has been monitored continuously with submersible differential pressure sensors since April 1989 at the SN1 and SN3 wells nearthe upperend of a sloping valley and at the KA1 and KA2 wells, all located on the north side of the fault Fig. 2). The sensors had a resolution of 1 cm, a full-scale range of 10 m and a data sampling interval of 10 min. In April 1996, the pressure sensors at SN1 and SN3 were replaced, and the data have since been telemetered, with an improved resolution of 3 mm, to the Earthquake Chemistry Laboratory of the University of Tokyo. Water pressure has also been monitored since mid-march 1997 at ve different depths with the help of packers) in two 200 m deep wells, TH7 and TH8, located about 140 m and 80 m to the south and north of the fault, respectively. The three deepest wells SN3, TH7 and TH8) were drilled through the sedimentary layers to the basement rock, the Toki Granite Fig. 3). The resulting pressure data were used King et al. 1999) to study why waterwells located neareach otherrespond to earthquakes differently. Barometric pressure and rainfall have also been continuously monitored at the mine. The barometric pressure and theoretical solid earth tide data are routinely used to correct for their effects on the measured water level and pressure. Earlier water-level data from the 131 m deep SN3 well were examined by Igarashi et al. 1996), who observed relatively large tidal variations compared with data recorded at the nearby 91 m deep SN1 well) and water-level changes seemingly related to three large but distant earthquakes M8.1 and 7.5 nearhokkaido, about 1000 km away, and M7.2 at Kobe, about 220 km away) during 1994±1995 the three curves in the middle of Figs 4a and b). These changes are similar to other groundwater-level/ ow/temperature changes observed elsewhere that are believed to be related to earthquakes e.g. Kawabe et al. 1988; Rojstaczer& Wolf 1992; King et al. 1994; Kitagawa & Koizumi 1996), suggesting that SN3 might be a site `sensitive' to tidal and earthquake-related crustal strain changes. King et al. 1999) reported co- and post-seismic water-level changes at SN3 that are correlated with many large distant and moderate local earthquakes, including a rare M5.8 earthquake on 1997 March 16 about 50 km away. They attributed the high sensitivity of SN3 to a relatively high permeability of the aquifer it taps, which is the weathered top layer of the basement rock, the Toki granite see Kiode et al. 1997). MAN-MADE VERSUS EARTHQUAKE- RELATED WATER-LEVEL CHANGES Water-level data recorded at the SN1 and SN3 wells during 1989±1999 are shown in Fig. 5 together with corresponding monthly rainfall data. The water-level curves are broadly similar, showing large changes that are not signi cantly in uenced by rainfall. As discussed below, most such changes are attributable Figure 1. Location of the Tono Mine. Knopoff Festschrift # 2000 RAS, GJI 143, 469±477

3 Earthquake precursors from wells at Tono, Japan 471 Figure 2. Locations of water wells circles; larger circles for deeper wells), the Tsukiyoshi Fault dashed line), the underground gallery, the KNA2 and KNA6 holes, topographic contour lines, and roads and various buildings in the Tono area. to water ow through the fault zone at various locations caused by earthquake or human activity. We rst describe the water-level changes caused by known human activity. In the gallery, which is 126 m deep and on the south side of the fault, an 80 m long hole KNA2) was drilled through the fault zone to the north side Fig. 6). In the uncased lower part of the hole, two interconnected packers were used to isolate a section, from which the groundwater was drained for continuous geochemical monitoring. This isolated section crosses the unconformity between the sedimentary and the basement rocks, and is in the depth range of relatively high permeability tapped by the SN3 well. The packers were normally kept in ated by pressurized nitrogen from a supply tank. As shown in Fig. 7, twice in 1994 the packers failed the pressure dropped to zero), and the failure coincided approximately in time with the preearthquake water-level drops at SN3 mentioned above middle three curves in Fig. 4) reported by Igarashi et al. 1996). At about the same times, increases in water ow from the KNA2 hole were noticed by a worker in the gallery. These observations raised the possibility that the observed water-level drops at SN3 were not earthquake-related, but were caused by increased out ow of water from the KNA2 hole as a result of packer failure. To check whetherpackerfailure at KNA2 could affect water levels at the SN1 and SN3 wells, we conducted a 90 day leak test beginning on 1998 June 19 by releasing the packer pressure at KNA2 to zero. As shown in Fig. 7, the water levels began to drop about 4 days and 9 hr later, respectively, at SN1 and SN3 see also Fig. 9a of King et al. 1999). The rate of water-level drop at SN3 was initially comparable to the rate of waterlevel drop that occurred at the end of September 1994, 10 days before the earthquake on 1994 October 4 second curve in Fig. 4b), but was slightly smaller afterwards compare curves 12 and 4 in Fig. 8b). Packer-failure-related increase in the ow rate at KNA2 can thus account for only part of the water-level drop observed at SN3 in September Notice also that the tank pressure curve in Fig. 7 second panel from the bottom) shows a relatively large slope shortly before, during and shortly after the time when the packer pressure dropped twice to zero, indicating an increased rate of nitrogen leak perhaps through small defects) during that period. Also, as shown in Fig. 8 d) curves 4 and 11), the water-level recovery rate after the 1994 October4 earthquake was signi cantly higherthan that after the leak test, indicating reduced leakage across the fault not only at KNA2, but also elsewhere. These changes the waterlevel drop at SN3, the increased nitrogen leak from the packers # 2000 RAS, GJI 143, 469±477 Knopoff Festschrift

4 472 C.-Y. King et al. Figure 3. Geological columns of the nine deepest wells in the Tertiary sedimentary sequences and the Cretaceous bedrock the Toki granite, the top 30±40 m of which is weathered and has a relatively high permeability). The solid triangles at TH7 and TH8 indicate the different depths at which water pressure was monitored with the help of packers). and the increased water ow at KNA2) were thus all probably the result of tectonic deformation of the fault zone, which affected the packer±rock interfaces allowing faster nitrogen leakage through the defects) and created ssures that are rapidly healed in the fault gouge zone. The large water-level drops at SN1 and SN3 in 1995 Fig. 7) coincided approximately in time with water release at another hole, KNA6, in the gallery Fig. 2). The chemistry of the released water was measured for several months by a different research group. This 100 m long hole is situated about 40 m west of the KNA2 hole and is approximately in the same direction. The water levels began to drop shortly before the water release began, and to rise shortly after the hole was permanently sealed. The effect of water release at KNA6 on water levels at the wells, as measured by the drop and rise rates, was about the same as for KNA2 compare curves 6 and 12 in Figs 8a and b, and curves 8 and 11 in Figs 8c and d). Towards the end of 1998, we noticed that the water levels at SN1 and SN3, as well as at some otherwells, had begun to drop again Fig. 7). After an extensive search, we found the culprit to be the drilling activity of a new 1 km deep well Miu2) some distance away, about 1 km southeast of SN3 on the south side of the Tsukiyoshi Fault, by another research group. The waterlevel drop at SN3 began about 2 days see curves 13 in Figs 8a and b) after the drilling reached a depth of approximately 900 m, where it broke through the fault. Since then, groundwaterhas owed out of the new drillhole, orinto some shallower aquifers when efforts were made to stop the leak at the ground surface. This observation shows that the observed water levels are affected quite rapidly by water leakage through the fault over a wide area of the fault plane, probably via one of the highly permeable fracture zones that sandwich the fault gouge zone. The rate of water-level drop in this case is smaller than the cases mentioned above forkna2 and KNA6, which is understandable in view of the longer ow path in the present case. As mentioned in King et al. 1999), we conducted a pumping test in November1997 at the then newly drilled 167 m deep well 97FT01 to nd out how the waterlevels at the nearby SN1 and SN3 wells were affected Figs 2 and 3). The water levels dropped only slightly, by about 5 cm almost undetectable in Figs 5 and 7), afterthe waterlevel at 97FT01 was pumped down by as much as 48 m forseveral days. This test indicates a low interlayer vertical permeability in the vicinity of these wells away from the fault zone. In our search for pre-earthquake water-level changes at SN3 like the ones examined previously by Igarashi et al. 1996), we found two othercases top and bottom curves in Fig. 4b). The corresponding changes at SN1 are shown in Fig. 4 a). In the rst case, an M6.6 earthquake occurred at a hypocentral distance of 290 km on 1990 September24. However, this event was within a 600 day period when a shaft excavation effect experiment was being conducted. The recorded water-level changes may thus have been caused by the excavation activity. The second case is more interesting. The recorded waterlevel changes at SN3 and SN1 coincided approximately in time with the occurrence of an M6.1 earthquake at a hypocentral distance of 510 km on 1996 September5. They were followed two weeks later by a 6 month period of water-level drops at these wells. During this period, the water ow at the KNA2 hole increased considerably. Several unsuccessful attempts were made to reduce the ow. At the end of this period, an M5.8 earthquake the largest in the vicinity of Tono since 1983) occurred about 50 km away on 1997 March 16 see King et al. 1999). Knopoff Festschrift # 2000 RAS, GJI 143, 469±477

5 Earthquake precursors from wells at Tono, Japan 473 (a) (b) Figure 4. Water-level data recorded at a) SN1 and b) SN3 during periods of 60 days centred at the times of several large distant earthquakes for which pre-earthquake changes were recorded. The earthquake dates are shown on the left side of the gures. Immediately afterthe earthquake, the water ow at KNA2 and the water levels at SN1 and SN3 returned to normal without any human intervention Figs 5 and 7). The rates of both water-level drop at the beginning of this 6 month period 22 September 1996) and water-level recovery after the earthquake of 1997 March 16 are higher than the corresponding rates observed during the leak test mentioned above curves 10 and 12 in Figs 8a and b, and curves 10 and 11 in Figs 8c and d). This result indicates the occurrence of extensive water leakage through the fault zone, not only at KNA2 but also elsewhere during this period. The observations mentioned above the occurrence of the M6.1 earthquake on 1996 September 5, the increased ow at KNA2 and the water-level drops at SN1 and SN3 at the beginning of the 6 month period, and the recovery # 2000 RAS, GJI 143, 469±477 Knopoff Festschrift

6 474 C.-Y. King et al. Figure 5. Long-term water-level data at the SN1 and SN3 wells located on the north side of the Tsukiyoshi Fault together with monthly rainfall data. Arrows indicate the occurrence of earthquakes. of ow at KNA2 and the waterlevels at SN1 and SN3 at the end of this period) may all indicate some broad-scale low-amplitude incremental crustal deformation that created suf cient ssures in the fault gouge zone during this anomalous period. On the otherhand, the water-level changes in the middle of the 6 month period curves 9 in Figs 8a±d) have comparable rates to the leak test. They were probably caused by attempts to prevent the ow increase at KNA2. Although we do not have packer pressure and tank pressure data, the water chemistry represented by the ph data in the bottom panel of Fig. 7) shows that ow changes indeed occurred during this period. As reported previously King et al. 1999), the water-level data at SN1 and SN3 showed a short-term reduction in their diurnal variation beginning about two days before the 1997 March 16 earthquake. However, this feature is attributable to Figure 6. A sketch of the 80 m long KNA2 hole drilled from the gallery on the south side of the Tsukiyoshi Fault through the fault to the north side. The upper 56 m of the hole is cased. Two packers are used in the lower part of the hole, which is not cased, to let water ow out from the unconformity between the sedimentary and basement rocks for continuous geochemical measurement. Knopoff Festschrift # 2000 RAS, GJI 143, 469±477

7 Earthquake precursors from wells at Tono, Japan 475 Figure 7. Water-level data at the SN1 and SN3 wells since 1993 together with the packer pressure, the tank pressure and the groundwater ph data recorded at the KNA2 hole in the gallery. The sawtooth shape of the tank pressure data is due to the periodical change of nitrogen supply tanks, usually before the nitrogen was exhausted. Arrows indicate the occurrence of earthquakes. the combined effect of solid earth tide and barometric pressure changes, and is thus not premonitory in nature. However, another short-term change, a water-level drop, recorded at the 116 m deep KA2 well beginning 2 days before the earthquake Fig. 8 in King et al. 1999) may possibly be premonitory. CONCLUSIONS The water-level drops at the SN3 well before some of the earthquakes in 1994±1995, and especially the 6 month drop before the M5.8 earthquake of 1997 March 16, which was also recorded at SN1, may be premonitory in nature. The SN3 well appears to be sensitive not only to seismic shaking from many large distant and moderate local earthquakes as reported in King et al. 1999), but also to low-amplitude and broad-scale episodic deformation in the crust. The high sensitivity of SN3 is probably due to its tapping a near-critical hydrologic system, namely, a permeable aquifer weathered granite) connected to a permeable fracture zone on the higher-pressure side of the Tsukiyoshi Fault, which has a weak, pliable and normally impermeable fault gouge zone. This near-critical condition allows SN3 to sense water leaks through the ssures, easily created and healed, in the gouge zone over a broad area of the fault plane. A near-critical condition such as this may possibly be a common characteristic of all earthquake-sensitive sites. ACKNOWLEDGMENTS We are deeply indebted to many colleagues at Tono Geoscience Center for the use of the data gathered by their persistent efforts over many years. This work was carried out whilst CYK was on leave from Earthquake Prediction Research Inc, Los Altos, California, serving as an International Fellow at JNC in its Earthquake Frontier Research Project. # 2000 RAS, GJI 143, 469±477 Knopoff Festschrift

8 476 C.-Y. King et al. (a) (b) (c) (d) Figure 8. Comparison of signi cant water-level drops and subsequent rises recorded at the SN1 a, c) and SN3 b, d) wells during 1990±1998. The dates shown on the left side of the gures are the start dates about 10 days before the water-level changes at SN3) of 30 day time windows, which are the same for both wells. The barometric pressure and theoretical earth tide effects have been removed from most of the data except data prior to 1996 at SN1, because their quality was not suf ciently good) for better data comparison. Knopoff Festschrift # 2000 RAS, GJI 143, 469±477

9 Earthquake precursors from wells at Tono, Japan 477 REFERENCES Igarashi, G., Wakita, H. & Umeda, K., Precursory and coseismic changes in well water levels observed for some large earthquakes in Japan, EOS, Trans. Am. geophys. Un., 77, Fall Mtng Suppl., F457. Kawabe, I., Ohno, I. & Nadano, S., Groundwater ow records indicating earthquake occurrence and induced Earth's free oscillation, Geopyhs. Res. Lett., 15, 1235±1238. King, C.-Y., Basler, D., Presser, T.S., Evans, W.C. & White, L.D., In search of earthquake-related hydrologic and chemical changes along Hayward fault, Appl. Geochem., 9, 83±91. King, C.-Y., King, B.-S., Evans, W.C. & Zhang, W., Spatial radon anomalies on active faults in California, Appl. Geochem., 11, 497±510. King, C.-Y., Azuma, S., Igarashi, G., Ohno, M., Saito, H. & Wakita, H., Earthquake-related water-level changes at 16 closely clustered wells in Tono, central Japan, J. geophys. Res., 104, ± Kiode, K., Yamane, M. & Kobayashi, K., Heterogeneity of hydraulic conductivity of a fault in sedimentary sequences at the Tono Mine, central Japan, Power Reactor and Nuclear Fuel Development Corp. Rept, PNC TN 7410, 97±035, 30±39. Kitagawa, Y. & Koizumi, N., Comparison of postseismic groundwater temperature changes with earthquake-induced volumetric strain release: Yudani hot spring, Japan, Geophys. Res. Lett., 23, 3147±3150. Rojstaczer, S. & Wolf, W., Permeability changes associated with large earthquakes: an example from Loma Prieta, California, Geology, 20, 211±214. # 2000 RAS, GJI 143, 469±477 Knopoff Festschrift