📄 earth_science

Exponential Distribution for Assessing Earthquake Probabilities (Mw &gt;6.7) within a 100 km Radius in Northeastern Crete

This study utilizes an exponential distribution model on historical and instrumental seismic data to demonstrate that major earthquakes (Mw > 6.7) in northeastern Crete follow a memoryless Poisson process, revealing an 85% probability of a subsequent event occurring within 90 years of the last 1935 mainshock.

Original authors: Nikos Petrakis

Published 2026-07-21

📖 3 min read☕ Coffee break read

CC BY 4.0

Original authors: Nikos Petrakis

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the Earth's crust as a giant, restless rubber band. Sometimes, it stretches quietly for decades, storing up tension, until—snap!—it releases that energy in a violent shudder we call an earthquake. For scientists who study these tremors, the big question is often: "When is the rubber band going to snap next?" To answer this, they often use a concept called a "Poisson process." Think of this like a cosmic slot machine that has no memory. If the machine just paid out a jackpot, it doesn't "remember" that it just did, so the odds of hitting the jackpot again right now are exactly the same as they were a second ago. It's a "memoryless" game. While this might sound scary (like the Earth could strike at any random moment), it actually gives scientists a powerful way to calculate the odds of a big event happening over a long period, turning chaos into a predictable pattern of probability.

This is the playground for Nikos Petrakis, an independent researcher who decided to look at the rubber band in a very specific neighborhood: the northeastern part of Crete, Greece. He wanted to know the chances of a massive earthquake (stronger than magnitude 6.7) hitting within a 100-kilometer radius of that area. To do this, he didn't just guess; he went on a time-traveling detective hunt, digging through historical records and modern instruments to find every major quake that happened there between the years 1500 and 1935. He found exactly 10 big events in that window, with the last one occurring on February 25, 1935.

Petrakis treated these 10 earthquakes like a set of data points to test his "memoryless" theory. He asked: Do the gaps between these earthquakes follow a simple, random pattern, or is there a hidden schedule? Using a statistical tool called an exponential distribution (which is the math behind that "memoryless" idea), he ran the numbers. The results were surprisingly neat. The data fit the random pattern almost perfectly, with a statistical confidence so high that the "randomness" theory was confirmed. He calculated that, on average, a major earthquake of this size happens in this region about 0.021 times per year.

But here is where the story gets urgent. Since the last big "snap" happened in 1935, the region has been sitting in a quiet period. Petrakis used his math to ask: "If we wait 90 years since that last event, what are the odds another one happens?" The answer came out to a striking 85%. In other words, by the time the region crosses the 90-year mark of silence (which it has now passed, entering 2025 and 2026), the statistical likelihood of a major earthquake occurring is very high. The paper doesn't claim to predict the exact date, but it suggests that the region has entered a zone where the probability of a major event is 85%, based on the simple, memoryless rhythm of the past. It's a reminder that while the Earth doesn't keep a calendar, the math of its history can tell us when the odds are stacked against us.

Technical Summary: Exponential Distribution for Assessing Earthquake Probabilities in Northeastern Crete

Problem Statement
This study addresses the challenge of evaluating local seismic hazards and recurrence intervals for major earthquakes (Moment Magnitude, $M_w > 6.7$ ) within a 100 km radius of northeastern Crete, Greece. The primary objective is to determine if the temporal distribution of these events follows a memoryless stochastic process, thereby allowing for the calculation of conditional probabilities for future seismic activity based on historical data.

Data Selection and Catalog Construction
The research utilizes a homogeneous dataset of $N = 10$ major to catastrophic seismic events spanning from 1500 to 1935 AD. To ensure data reliability and mitigate the incompleteness often found in historical archives, the authors restricted the temporal window and applied a strict magnitude threshold ( $M_w > 6.7$ ).

Sources: Historical event details were extracted from the European Archive of Historical Earthquake Data (AHEAD), while instrumental data were cross-referenced with the catalog by Papazachos & Papazachos (2000).
Spatial Domain: A circular domain with a radius of approximately 100 km centered on northeastern Crete.
Completeness: The catalog ends with the well-documented earthquake of February 25, 1935.

Methodology
The authors modeled the temporal distribution of mainshocks using a stochastic renewal process, operating under the standard seismological assumption that mainshocks exhibit memoryless behavior (with aftershocks properly filtered).

Distribution Model: The inter-event times ( $\Delta t$ ) were tested against an exponential distribution, defined by the probability density function $f(\Delta t) = \lambda e^{-\lambda \Delta t}$ , where $\lambda$ is the mean annual occurrence rate.
Parameter Estimation: The rate parameter $\lambda$ was estimated using the Maximum Likelihood Estimation (MLE) method.
Goodness-of-Fit: To validate the fit between empirical data and the theoretical exponential model, two non-parametric tests were conducted:
1. Kolmogorov-Smirnov (K-S) test.
2. Anderson-Darling (A-D) test (utilizing the Brown approximation for small sample sizes).

Key Results

Recurrence Rate: The estimated mean annual occurrence rate ( $\lambda$ ) is 0.02108, with a standard error of 0.00701. The 95% confidence interval for $\lambda$ ranges from 0.007338 to 0.03482.
Statistical Validation: The goodness-of-fit tests yielded high p-values, indicating that the null hypothesis (that the data follows an exponential distribution) cannot be rejected:
- K-S test: $p = 0.959$
- A-D test: $p = 0.802$
  These results confirm that the regional recurrence intervals closely adhere to an exponential distribution, supporting the assumption of a memoryless Poisson process for mainshock generation in this region.
Probability Estimation: Using the Cumulative Distribution Function (CDF), $P(T \le t) = 1 - e^{-\lambda t}$ , the study calculated the conditional probability of a subsequent major earthquake. Given that the last event occurred in 1935, the model estimates an 85% probability ($0.85$) that a major earthquake ( $M_w > 6.7$ ) will occur within a time horizon of 90+ years (i.e., by 2025/2026).

Significance and Claims
The paper claims to provide a "simplified yet robust statistical framework" for regional seismic hazard assessment. By validating that the inter-event times for major earthquakes in northeastern Crete follow an exponential distribution, the study demonstrates that a memoryless Poisson process is an appropriate model for this specific spatial and temporal cluster. The primary contribution is the quantitative estimation of a high conditional probability (85%) for a future major event, derived directly from the elapsed time since the 1935 earthquake and the calculated mean recurrence rate. The authors present these findings as a direct application of statistical renewal processes to historical seismic data, without proposing new physical mechanisms or future experimental protocols.

Technical Summary: Exponential Distribution for Assessing Earthquake Probabilities in Northeastern Crete

More like this