Competition 2020
Our 2020 AETA Competition Overview & Results
Overview of 2020 Competition
Materials supplied for competition:
Using past data to train the team’s algorithm: Low frequency means value of geo-acoustic and electromagnetic disturbance from Oct 2016 to Jan 2020.
Real-time data for prediction: Low frequency mean value of geo-acoustic and electromagnetic disturbance, provided every Sunday to predict the following week.
List of earthquakes from Oct 2016 to Jan 2020, including epicenter, magnitude, and time.
Competition criteria:
Three major elements teams must test for: Epicenter, magnitude, and accuracy of time.
Prediction of magnitude ≥3.5.
16-week competition, one forecast on every Sunday of the following week, for a total of 16 predictions.
80% of marks on the accuracy of the epicenter, magnitude, and time. And 20% of marks on defending the algorithm’s integrity.
The geographical range for competition: 22~34 north latitude, 98~107 East longitude (Sichuan/Yunnan area where most of the existing sensors are deployed).
Outcome of 2020 Competition
Champion team (szpianpian): 12 correctness in 16 yes/no predictions (75% correctness), and among the 12 correct predictions, 8 are quite close to expected results, with an average mean deviation of 0.4885 on the magnitude, and 103.98 kilometers on the epicenter.
On yes/no forecast: All 10 winners achieved 71% correctness or above, while the current correctness ratio reported in earthquakes is below 20%.
The team from the University of Tsukuba achieved 100% correctness in all 16 yes/no predictions, but unfortunately, the accuracy on epicenter and magnitude was not as good compared to the champion team.
Outcome of 2020 Competition
From the 1st of January 2021, we have started a new competition, inviting teams from all over the world to participate in developing an algorithm designed to forecast earthquakes.