Heki, K., Seasonal Modulation of Interseismic Strain Buildup in Northeastern Japan Driven by Snow Loads

Science, 293, 89-92, 2001.

　The baselines on the Japan Sea side have large annual components: they shorten by a few millimeters in winter (Fig. 2B,E), a result consistent with (4). On the other h and, seasonal signatures are smaller on the Pacific side of the arc (Figure 2A,D) and have the opposite sense (shortening in summer). Such a contrast across the backbone range is seen for most pairs of baselines. If the annual signal reflects changes in the convergence ra te or the coupling strength at the Japan Trench, its amplitude would be proportional to th e secular shortening rate irrespective of which side of the arc the baseline lies. Anothe r feature not mentioned in (4) is the annual signals in vertical components (Fig. 2C,F). Most of the sites subside relative to Tsukuba in the winter, and the amplitudes, up to ~2 cm peak- to-peak, are larger along the backbone range than along the coasts. Neither faster subductio n nor stronger coupling predicts such a subsidence pattern for the arc. These feature s make the trench-origin mechanisms untenable.
　Four seasons characterize the environment in Japan. In winter, cold and dry a ir over Siberia becomes humid as it travels across the Japan Sea, leaving heavy snowfall s as it collides with the backbone range of the northeastern Japan. Snow mainly falls o n the western side of the arc but the amounts vary from place to place. The deepest s now, seen along the western flank of the backbone range, starts to accumulate in late autu mn reaching a few meters deep in March; it then disappears in May (Fig. 1) except at the highe st peaks. Loads on Earth’s surface cause subsidence beneath and around the load, along wi th horizontal deformation such that the land shortens beneath the load and extends outside of it. Snow loads distributed along the western half of the arc would cause crustal deformat ion qualitatively consistent with the GPS baseline data (Fig. 2), arc-normal crustal shortening beneath the snow cover (Fig. 2B,E), and smaller extension outside the cover (Fig . 2A,D). It would also cause the subsidence whose maximum lies below the snow load center (F ig. 2C,F).

　Next, I fixed an arbitrary station, and modeled each component of the relative position time series of 88 GPS points in northeastern Japan with the linear, annual and biannu al terms (8). I discarded 13 sites (9) whose root-mean-squares of the post-fit residuals excee ded 4 mm (horizontal) or 15 mm (vertical). Then, I obtained instantaneous values of the periodic components on 15 March (relative to 15 August) for all the sites (Fig. 3) and es timated the load (heights of the water columns above ground) on each of the rectangular area s as large as 0.5 o in latitude and 0.75 o in longitude covering the land area in Fig. 3 that best explained the observed winter-summer site displacements in a least-squares sense (10). I simu ltaneously estimated the translation of the whole network to ensure that the obtained loads were free from the fixed-point problem. To calculate site displacements due to a unit loa d on an elastic half space, I used the formulation by Farrel (11), 30 GPa for rigidity a nd 0.25 for Poisson’s ratio. Typical formal errors of the estimated load were around 20 to 25 cm, and the post-fit residuals of the input data were 1.3 mm, 0.9 mm, and 2.9 mm for the north, east, and up components, respectively.
　In Fig. 3, some stations have anomalously large residuals, a part of which mig ht be explained by localized high snow masses often found over the high mountains. No ntheless, the observations agree with the calculated displacements. To convert the load i nto snow depth, I needed data for average snow density, which can be obtained only by fie ld observations of snow cross sections. Such observations have been carried out at Mount Zao (Fig. 1D), and Yamaya et al. (12) report that the average snow densities there d uring the 1999 to 2000 winter increased with time, i.e. 0.30 to 0.32 g/cm³ from January to February, 0.42 g/cm³ in late March, and 0.50 to 0.53 g/cm³ from the end o f April to May.

　I assumed a snow density of 0.4 g/cm³ in the middle of March to con vert the load into snow depth, and compared the resulting estimates with snow depth data acquired o n 15 March 2000 at AMeDAS (Automated Meteorological Data Acquisition System) stations (13) (Fig. 4). The data sets were fairly consistent (14), which suggested that the s now could explain most of the annual GPS variations. Other hydrological factors (e.g. soi l moisture and groundwater) would be less important, but they would reduce the sharpness of the load decrease by spring thawing. Atmospheric pressure increase in winter (~1 kPa (15 ) in northeastern Japan) would act as the equivalent of additional snow as deep as 25 cm for the entire land area and would enhance the annual signature by ~10 %. Snow covers a re ubiquitous in high- to mid-latitude areas on Earth. Snow loads, together with v arious water loading mechanisms (16), would annually distort the terrestrial reference frame composed of world-wide distributed stations. This would substantially affect the determinat ion of GPS orbits because the current accuracy of the terrestrial reference frame is ~1 cm or better (17). (Here I discussed relative displacements within a local network, which are littl e affected by such errors, but annual variations in global GPS networks are more sensitive to the annual orbital errors and need to be examined carefully.)
　The snow loads modulate interseismic stress build-up in northeastern Japan (Fi g. 2 B and E). Seasonality in earthquake occurrences is a century-long issue (18) in Japan ese seismology, and Murakami & Miyazaki (4) suggested the relevance of the seasonal strain changes to the seismicity. Mogi (19) compiled historical (M>7.9) and modern (M> 7.5) earthquakes, and found that the peak in seismicity exists in March in northeaste rn Japan and in December in central Japan. Ohtake & Nakahara (20) confirmed the statistical significance of the seasonality in central Japan, and evaluated its relevance to the atmospheric pressure increase of ~1 kPa in winter. They showed that this high pressure qual itatively encourages reverse slips at the plate interface but the stress changes are only a few tens of Pa in terms of the Coulomb failure function (CFF) (21). This is not more than the semi-diurnal tidal stress increase of up to ~1 kPa, which is not effective in triggering eart hquakes (21).

　 Recently, Gao et al. (22) suggested that the higher activity of the 1992 Lan ders event aftershocks in summer might be due to the atmospheric pressure change of ~2 kPa, which is an order of magnitude smaller than the minimum static stress changes previously considered to trigger aftershocks. The snow loads inferred in the present study are equiva lent to an increase of up to ~10 kPa in atmospheric load, and a calculation similar to that in (20) shows that they cause positive CFF change at the plate interface dipping westward from the Japan Trench (Fig.1). However, the same calculation also indicates that the large dis tance between the load and the fault reduces the CFF change down to the level of 0.1 kPa. Thi s would be too small to let us conclude that there is a causal relation between earthquakes and the snow loads, even if we took into account that the earthquake-triggering mechanisms ar e not well understood.

REFERENCES

1. S. Miyazaki, K. Heki, J. Geophys. Res. 106, 4305 (2001).
2. H. Tsuji, Y. Hatanaka, T. Sagiya, M. Hashimoto, Geophys. Res. Lett. 22, 1669 (1995).
3. K. Heki, S. Miyazaki, H.Tsuji, Nature 386, 595 (1997).
4. M. Murakami, M. S. Miyazaki, Geophys. Res. Lett. in press.
5. D.S. MacMillan, Geophys. Res. Lett. 22, 1041 (1995).
6. R. Ichikawa et al., Proceedings of the Third Joint Meeting of U.S.-Japan Natu ral Resources Panel on Earthquake Research (2001).
7. S. Miyazaki, T. Saito, M. Sasaki, Y. Hatanaka, Y. Iimura, Bull. Geog. Survey Inst. 43, 23 (1997).
8. The annual variation curve is asymmetric about the peak. For example, the sub sidence in winter is somewhat slower than the uplift in spring (Fig.2C, F), which can be mo deled by combining annually and biannually changing terms. It was also confirmed that th e results are not dependent on the particular station chosen as the fixed point. 9. The discarded stations mostly belong to the first generation GPS points (esta blished in 1994) of GEONET and are not concentrated in a particular area. They had larger post-fit residuals than the newer points.
10. A weak continuity constraint was imposed to stabilize the solution so that t he difference of the loads of the adjacent rectangles is 0 +/- 20 cm (or +/- 50 cm for the dep th of snow as dense as 0.4 g/cm³).
11. W.E. Farrel, Rev. Geophys. Space Phys. 10, 761 (1972).
12. M. Yamaya et al., Tohoku J. Natural Disaster Sci. 36, 171 (2000).
13. Japan Meteorological Agency, Monthly report of Japan Meteorological Agency, vol. 3, No. 5 (2000).
14. Because AMeDAS stations need public electricity and communication lines, the y are generally located in inhabited lowlands. This may have negatively biased observ ed snow depths, and would partly explain the apparent absence of snow on the Pacifi c side in Fig. 4A. 15. National Astronomical Observatory (ed.), Chronological Scientific Tables 74 (Maruzen, Tokyo, 2000).
16. T. van Dam et al., Geophys. Res. Lett. 28, 651 (2001)
17. P. Sillard, Z. Altamimi, C. Boucher, Geophys. Res. Lett. 25, 3223 (1998).
18. F. Omori, Pub. Earthquake Investigation Comm. 8, 1 (1902).
19. K. Mogi, Bull. Earthq. Res. Inst. 47, 419 (1969).
20. M. Ohtake, H. Nakahara, PAGEOPH 155, 689 (1999).
21. R.S. Stein, Nature 402, 605 (1999).
22. S.S. Gao, P.G. Silver, A.T. Linde, I.S. Sacks, Nature 406, 500 (2000).
23. K. Heki et al., J. Geophys. Res. 104, 29147 (1999).
24. I thank Makoto Murakami, who first reminded me of the geophysical importance of the annual signals. I also thank Tetsuya Iwabuchi for providing AMeDAS snow depth d ata, and Shin’ichi Miyazaki for reviewing the earlier version of the manuscript.


26 March 2001; accepted 31 May 2001