地震を予測できるか? 科学 vs 迷信

The Difference: Prediction vs Forecasting

In everyday language, "prediction" and "forecasting" are often used interchangeably, but in earthquake science they mean fundamentally different things. An earthquake prediction予知は将来の地震の正確な時刻・場所・規模を特定することを指すが、これは現在のところ不可能である。予測は、ある期間における地震発生の可能性を確率的に見積もるものである。 specifies the location, magnitude, and time of a future earthquake with sufficient precision and reliability to be useful for evacuations or other concrete actions. An earthquake forecast, in contrast, specifies the probability of an earthquake exceeding a given magnitude in a given area over a given time period — a statistical statement, not a precise prediction. The distinction matters enormously for public policy. Governments cannot evacuate a city based on a 10 percent probability of a Mw 7.0 earthquake in the next decade. But they can — and do — design buildings, update land-use plans, and prepare emergency services based on probabilistic forecasts. As of the current state of science, reliable deterministic earthquake prediction in the operational sense does not exist. Probabilistic forecasting, however, is highly developed and forms the scientific basis for modern seismic hazard assessment.

Failed Prediction Attempts in History

The history of earthquake prediction is littered with claims that initially attracted attention but did not survive scientific scrutiny. The VAN method, developed by Greek researchers Varotsos, Alexopoulos, and Nomikos in the 1980s, claimed to predict earthquakes from anomalous electrical signals in the ground (seismic electric signals or SES). Despite initial enthusiasm, rigorous statistical evaluation showed no predictive skill beyond chance. Radon gas anomalies, groundwater level changes, unusual animal behavior, and electromagnetic anomalies have all been proposed as earthquake precursors at various times, but none has demonstrated consistent, reliable predictive power in controlled scientific tests. The one genuinely successful prediction in earthquake history — the 1975 Haicheng earthquake in China, where unusual animal behavior and 前震同じ地域で本震に先立って発生する地震。前震は事後にしか特定できず、事前に通常の地震と区別する確実な方法は存在しない。 activity led to a successful evacuation — was followed one year later by the 1976 Tangshan earthquake (which killed an estimated 242,000 people) with no warning. The Haicheng success now appears to have been partly luck.

The Parkfield Experiment

The Parkfield segment of the San Andreas Fault in California was considered a prime candidate for earthquake prediction experiments in the 1980s, because it had apparently produced characteristic Mw 6 earthquakes at roughly 22-year intervals: in 1857, 1881, 1901, 1922, 1934, and 1966. Based on this pattern, a Mw 6 earthquake was predicted for approximately 1988, with a 95 percent confidence window extending to 1992. The earthquake finally occurred in 2004 — 12 years late. While the 2004 event was well-recorded by the dense monitoring network deployed in anticipation, its lateness demonstrated that even apparent periodicity in 地震クラスタリング地震がランダムに発生するのではなく、本震・余震系列や群発地震のようにクラスター（集団）として発生する傾向。地震が独立してランダムに発生するという一般的な仮定に反する。 cannot serve as the basis for operational prediction.

Why Earthquakes Are Fundamentally Unpredictable

Modern seismology suggests that deterministic earthquake prediction may be inherently impossible, not merely technically difficult. Earthquakes are the result of stick-slip friction on faults — a process with sensitive dependence on initial conditions. The state of stress on a fault is heterogeneous at all scales; small 前震同じ地域で本震に先立って発生する地震。前震は事後にしか特定できず、事前に通常の地震と区別する確実な方法は存在しない。s or slow-slip events that might seem to signal an impending rupture may alternatively arrest without generating a large event. The transition from stable sliding to dynamic rupture is a nonlinear, potentially chaotic process: tiny perturbations in fault stress — perhaps from small distant earthquakes, changes in groundwater pressure, or even ocean tidal loading — can either trigger or prevent a rupture. This sensitivity means that even with perfect knowledge of current fault stress, predicting the exact time of the next rupture may be physically impossible.

The Role of Foreshocks

前震同じ地域で本震に先立って発生する地震。前震は事後にしか特定できず、事前に通常の地震と区別する確実な方法は存在しない。s — earthquakes that precede the mainshock on the same fault — occur before approximately 40–70 percent of large earthquakes, but they are only recognizable as foreshocks in hindsight. At the time they occur, there is no way to distinguish a foreshock from any other small earthquake. The probability that a small earthquake will be followed by a larger earthquake on the same fault is computable using statistical models (such as ETAS — the Epidemic-Type Aftershock Sequence model), and these models do produce short-term probability increases that can be communicated to emergency managers. But the absolute probability remains low enough that routine evacuations based on foreshock activity alone would cause enormous social disruption for very few genuine precursors.

Earthquake Forecasting: Probabilistic Approaches

While deterministic prediction has failed, probabilistic earthquake forecasting has matured into a rigorous, quantitative science. Probabilistic forecasting combines geological fault data (slip rates, recurrence intervals from 古地震学断層トレンチ、隆起した段丘、津波堆積物といった地質学的証拠を通じて、先史時代の地震を研究する学問。地震の記録を数千年前まで遡らせる。), seismological observations (historical catalogs, b値グーテンベルク・リヒターの頻度・マグニチュード関係の傾きを表す値。1.0前後が一般的で、値が高いほど大地震に対して小地震の割合が多いことを示す。値の変化は応力状態の変化を示唆することがある。 analysis, グーテンベルク・リヒター則地震の頻度とマグニチュードの関係を示す統計法則で、マグニチュードが1単位増えるごとに、地震の発生頻度はおよそ10分の1になる。 statistics), and geodetic data (strain rates from GPS) to estimate the probability of an earthquake exceeding a given magnitude in a given area over a given time period. The Working Group on California Earthquake Probabilities, for example, estimates that there is roughly a 60 percent probability of a Mw 6.7+ earthquake striking the San Francisco Bay Area in the next 30 years. These forecasts are updated as new data become available and are presented in probabilistic seismic hazard maps that inform building codes, insurance rates, and public policy.

Operational Earthquake Forecasting

A newer and more applied form of probabilistic forecasting is operational earthquake forecasting (OEF) — the real-time updating of earthquake probabilities following significant events, especially for aftershock sequences. After a major earthquake, the probability of damaging aftershocks is substantially elevated, following 大森公式（オモリ則）時間経過に伴う余震発生頻度の減衰を示す経験則で、余震の発生率は本震からの経過時間にほぼ反比例して減少する。 in its time decay. Agencies including the アメリカ地質調査所（USGS）地震の監視、National Earthquake Information Centerの運用、世界の地震データのリアルタイム公開を担う、アメリカ政府の主要機関。, the Istituto Nazionale di Geofisica e Vulcanologia (INGV) in Italy, and GNS Science in New Zealand now issue regularly updated OEF products during significant aftershock sequences, providing emergency managers with probabilistic guidance on the elevated risk of additional damaging events. This represents a practical, scientifically defensible application of 地震クラスタリング地震がランダムに発生するのではなく、本震・余震系列や群発地震のようにクラスター（集団）として発生する傾向。地震が独立してランダムに発生するという一般的な仮定に反する。 statistics to real-world emergency management.

PSHA: The Current Best Practice

確率論的地震ハザード解析（PSHA）起こりうるすべての地震の発生源・マグニチュード・地震動レベルを考慮し、特定の揺れの水準を超える確率として結果を表現する、地震ハザードの定量評価手法。 — Probabilistic Seismic Hazard Analysis — is the gold standard for quantifying earthquake hazard for engineering and policy purposes. PSHA integrates over all possible earthquake sources (fault sources and area sources), all possible magnitudes on each source (グーテンベルク・リヒター則地震の頻度とマグニチュードの関係を示す統計法則で、マグニチュードが1単位増えるごとに、地震の発生頻度はおよそ10分の1になる。 magnitude-frequency relationships), all possible distances from source to site, and ground motion prediction equations (empirical models relating magnitude and distance to peak ground acceleration, 最大地動加速度（PGA）地震時における地面の最大加速度で、重力加速度（g）の単位で表される。耐震工学における構造物設計の重要なパラメータ。, and spectral accelerations) to compute the probability that ground shaking will exceed any given level at a site in a given time period. The output is a hazard curve: for every shaking level, the annual probability of exceedance. PSHA results are expressed as the peak ground acceleration with a given probability of being exceeded in 50 years — for example, the 2 percent in 50 years (approximately 2,500-year return period) hazard level used in US building codes. This framework is fundamentally probabilistic, acknowledging the irreducible uncertainty in earthquake occurrence while still providing quantitative, decision-relevant hazard estimates.

The Future of Earthquake Science

The frontier of earthquake science is not deterministic prediction but rather the progressive reduction of uncertainty in probabilistic forecasts. Denser 地震観測網地震活動を継続的に監視する、連携した地震観測点の集合。世界地震観測網（GSN）は150以上の観測点から構成され、世界規模の観測範囲を提供する。s and GPS arrays are improving our knowledge of fault behavior. Machine learning is uncovering subtle seismicity patterns that precede some large earthquakes, potentially providing modest short-term probability increases. New 古地震学断層トレンチ、隆起した段丘、津波堆積物といった地質学的証拠を通じて、先史時代の地震を研究する学問。地震の記録を数千年前まで遡らせる。 studies are lengthening the earthquake record on poorly understood faults. Laboratory experiments on rock friction are revealing the micro-mechanical processes that control the transition from stable creep to dynamic rupture. None of this is likely to produce the kind of hour-specific, location-specific prediction that the public imagines when they ask "Can we predict earthquakes?" — but it is steadily improving the probabilistic tools that save lives through better-designed buildings, more targeted land-use planning, and smarter emergency preparedness.