Forecast Lab

Experimental

This panel fits the Gutenberg–Richter relation log10(N) = a − b·M to the live USGS catalog for the Philippine bounding box, then converts the implied annual rate of M ≥ m into a Poisson probability over a given time window. It describes the recent catalog — it does not predict any specific earthquake.

Experimental research forecastThese probabilities are statistical descriptions of the recent catalog using the Gutenberg–Richter relation. They are not predictions of any specific earthquake. For life-safety warnings, refer only to PHIVOLCS and NDRRMC.
b-value
0.65 ± 0.02
a-value
4.67
R² (fit)
0.93
Catalog
164 events / 29d
P(M ≥ 4) within window
7d99.9%
95% CI: 100.00% – 100.00%
30d99.9%
95% CI: 100.00% – 100.00%
90d99.9%
95% CI: 100.00% – 100.00%
P(M ≥ 5) within window
7d99.8%
95% CI: 99.24% – 99.97%
30d99.9%
95% CI: 100.00% – 100.00%
90d99.9%
95% CI: 100.00% – 100.00%
P(M ≥ 6) within window
7d75.2%
95% CI: 56.13% – 90.51%
30d99.7%
95% CI: 99.03% – 99.95%
90d99.9%
95% CI: 100.00% – 100.00%

Omori–Utsu aftershock decay

Mainshock
M 7.8 · 2026-06-07
p
0.71
c (days)
0.001
K
22.11
R²
0.71

Declustered background rate (ETAS-style)

Total events
164
After declustering
111
Window
29 days
Background rate
3.86 events / day

Bayesian rate posterior (Gamma–Poisson)

M ≥ 5 · obs
51 in 0.08 yr
Posterior mean
50.98 /yr
95% CrI
38.41 – 65.32 /yr
P(M≥5, 30d)
98.5%
— 95% CrI
95.7% – 99.5%
M ≥ 6 · obs
5 in 0.08 yr
Posterior mean
1.38 /yr
P(M≥6, 1yr)
74.8%

Prior: Gamma(α=4, β=1) on M≥5 and Gamma(α=2, β=5) on M≥6, reflecting the long-term Philippine catalog.

ETAS productivity (Felzer–Brodsky)

α (productivity)
0.30
K (at M=Mmain)
10.72
Mainshock M ≥
4.0
Mc
2.5
Bins fit
6
Window
7d · 100km

α near 1 is consistent with self-similar aftershock productivity. Lower α implies large mainshocks are relatively less productive than scaling predicts.

Methodology notes

Magnitude of completeness Mc = 3. b-values near 1.0 are typical of tectonically active regions; lower values indicate relatively more large events. Uncertainty on b is the Aki (1965) maximum-likelihood standard error σb = b / √N.

Confidence intervals on probabilities are 95% intervals derived from a Wald-type approximation on log(λ); they widen rapidly when the catalog is thin. The Omori–Utsu fit linearizes log n(t) = log K − p · log(t + c) with a grid search over c. The "background rate" is produced by a simple magnitude-dependent space-time declustering (a stand-in for full ETAS — Ogata 1988 — until full ETAS is implemented).

Bayesian update uses a Gamma(α₀, β₀) prior on the annual rate λ and Poisson likelihood, yielding Gamma(α₀+n, β₀+T) posterior; the 95% credible interval is computed from the Wilson–Hilferty chi-square approximation. ETAS productivity α is estimated by least squares on log10 Naft(M) across mainshock magnitude bins (Felzer & Brodsky 2006). The declustered background rate is a stand-in for full ETAS MLE (Ogata 1988), which is planned.

Planned: full ETAS MLE inference (Ogata 1988), spatial Hawkes intensity maps, and CSEP-style Brier / log-loss / ROC-AUC validation against retrospective Philippine catalogs.

See cited methodology references →