Reports 1-1 of 1 Clear search Modify search
PEM (Tool)
tatsuki.washimi - 16:12 Tuesday 04 February 2025 (32580) Print this report
Development of a New Microseism Forecast 

To update the Microseism Forecast, I'm scraping the weather forecast page "windfinder.com", since Jan. 13.
This page provides us with the forecast of sea wave height, wave direction, wave frequency, wind speed, wind direction, and other weather information for 10 days with a 3-hour time resolution.
(Otenki.com, used in the current system, forecasts sea wave height, wave period, wind speed, wind direction, and other weather information for 2 days with a 3-hour time resolution, and for 8 days with a 24-hour time resolution.)

Fig.1 shows the wave height at 516 coasts of the Prefecture in Honshu island, and also the microseismic band RMS for a KAGRA seismometer.
Fig.2 and Fig.3 show the correlation coefficient between the microseism and each wave height.

Fig.4 and Fig.5 show a heatmap and a cluster map of the correlation coefficient between each wave height.

These data will be used in a machine-learning based microseism forecast.

Images attached to this report
Comments to this report:
tatsuki.washimi - 16:25 Tuesday 04 February 2025 (32581) Print this report

I simply picked up the 12 coasts (used in the current system) from the windfinder.com data and applied them to the current model & parameters determined by the 2020 data.

The forecasting period and the time resolution are improved.
Also somehow the agreement between the forecast and the actual data looks better.

Images attached to this comment
tatsuki.washimi - 10:50 Friday 07 February 2025 (32611) Print this report

I applied some minor changes (the algorithm itself is the same) to the microseism forecast and compared it to the current forecast.

  • Added coasts : mikuni_hukui, nyu_hukui.csv, akabane_aichi
  • Removed coasts : hukui_hukui, turuga_hukui, simizu_sizuoka
  • Number of sea groups: 3 -> 4
  • Used data : 1 year in 2020 -> recent 10 days
Images attached to this comment
tatsuki.washimi - 18:14 Monday 10 February 2025 (32626) Print this report

I updated the new forecast algorithm:

  • Input 516 coast wave height [Fig. 2]
  • Choose the coasts that have enough correlation (> 0.5) with the seismometer data [Fig.3]
  • Take 12 clusters from each correlation [Fig. 4]
  • Take the first principal components for each cluster [Fig. 5]
  • Perform fitting [Fig.6, 7]
Images attached to this comment
Search Help
×

Warning

×