Tap any colored zone to explore earthquake hazard
✕
Expected Magnitude
6.77.58.08.5+
Dot Size = Magnitude
Zone Statistics
Hazard Curve
Return period (yr) vs expected magnitude
Return Level Estimates
HorizonMw (EVT)Model
Poisson Exceedance Probability
ExposureThresholdProbability
P = 1 − exp(−t / Tr)  •  Poisson model
5 Largest Recorded Earthquakes
— | — earthquakes shown | USGS ComCat · Indonesia · 1926–2026
Earthquake Catalog
Mainshocks after declustering · Mw ≥5.0 · 1926–2026
YearMagnitudeDepthLatLonZoneLinks

Indonesia Seismic Hazard Explorer

Interactive research tool estimating earthquake return periods across Indonesia using Deep Neural Networks (DNN) for zone classification and Extreme Value Theory (EVT) for hazard modeling. Extended with exceedance probability analysis and geological ceiling validation.

How to Use

Select Return Period
Choose 10, 50, 100, or 500 years in the sidebar. Zone colors update to show EVT-estimated earthquake magnitude.
Click a Zone
Click any colored region to open the detail panel with hazard curve, return levels, and exceedance probabilities.
Inspect Earthquakes
Click any dot on the map for earthquake details and links to news, Wikipedia, and BMKG.
Browse Catalog
Use the Catalog tab to browse, sort, and search all mainshock events.

Key Findings

Zone100-yr Return LevelBest ModelNote
PapuaMw 8.40GPHighest hazard zone
SumatraMw 8.75RT-POT2004 Mw 9.1 megathrust
JavaMw 6.72RT-GRLargest underestimation (22.8%)
Nusa Tenggara—RT-POTBanda Arc subduction
Maluku—GPArc-arc collision zone
Sulawesi—GPSmallest underestimation (0.5%)
DNN accuracy: 95.4%  •  Catalog: 1,135 mainshocks  •  Period: 1926–2026

Methodology

#StepDetails
1
Data & Preprocessing
USGS ComCat 1926–2026. Mw conversion via Scordilis (2006). Gardner-Knopoff (1974) declustering for independent mainshocks only.
2
DNN Zone Classification
4-layer fully-connected network (128-64-32-6 neurons). Classifies earthquakes into 6 seismic zones by lat, lon, depth. 95.4% test accuracy.
3
Extreme Value Theory
Four models per zone: GEV, GP, RT-GR, RT-POT. Best model auto-selected by AIC/BIC (lowest score = best fit).
4
Exceedance Probability
Poisson model: P = 1 − exp(−t/Tr). Gives probability of exceeding a return level within any exposure window t.
5
Geological Validation
EVT results compared to PuSGeN 2017 geological ceiling magnitudes. Underestimation error: E = (Mgeo−MEVT)/Mgeo×100%. Bootstrap (500 reps) gives 95% CI.

Understanding Return Periods

A 100-year return period does not mean the earthquake happens every 100 years exactly. Under the Poisson model, earthquakes occur randomly — the probability of at least one event within t years is:

Poisson Exceedance Probability
P(Tr, t) = 1 − exp( −t / Tr )

For a 100-year return period: 9.5% chance in 10 years, 39.3% in 50 years, 63.2% in 100 years.

Study Limitation

The 100-year observation window is shorter than megathrust recurrence cycles (300–500 yr), causing systematic underestimation in subduction zones. The geological underestimation error (Egeo) quantifies but does not fully correct this bias.

References

Ma, G. et al. (2021). Return Period Evaluation of the Largest Possible Earthquake Magnitudes in Mainland China Based on Extreme Value Theory. Sensors, 21(10), 3519.
PuSGeN (2017). Peta Sumber dan Bahaya Gempa Indonesia Tahun 2017. Kementerian PUPR & BNPB.
Irsyam, M. et al. (2020). Development of New National Seismic Hazard Maps of Indonesia. Earthquake Spectra, 36(1_suppl), 112–136.
Gardner, J.K. & Knopoff, L. (1974). Is the sequence of earthquakes in Southern California, with aftershocks removed, Poissonian? BSSA, 64(5), 1363–1367.
USGS Earthquake Hazards Program. ComCat Earthquake Catalog. earthquake.usgs.gov