waidyanatha / quasar Goto Github PK

Provide a map with hover over to review stations and colour code by station type (or location type)

1. Amend the current dataset with other strong motion data - but first understand and describe the attributes in each file
2. Apply SVM to see if the margins can be increased between the clusters
3. Also see if MLP can classify the data to find other classes of interest.
4. Try Naive Bayesian multi classify because the data is parametric or LabelEncoding will parametrize the categorical data

K-means is ideal because it's unsupervised statistical learning but can we come up with a meaningful set of classes that describes the behavior of the SM DB data?
same with Random Forests to allow for using N-samples of features and datasets. It requires labeled data for supervised learning

Apply basic trigger and picking algorithms on GeoNet waveform data.

Read traces from all stations for a given time window. Filter the high and low frequency noises. Convert to ACC data to VEL data. Thereafter, calculate the Pd of the traces. Present the PGV distribution over geographic clusters.

Borrow some of the USGS ground motion processing code, coding style, and methods for GeoNet. Apply during code cleanup and completeness

Determine station cluster topography relative to the fault lines and earthquake detection role and capacity

Essentially plot the nearest neighbour coordinates of the stations and faults with a distance line

The obspy.geodedic.inside_geobounds() function failed with an error saying Attribute: Latitude and Longitude are required; even though the float values are assigned to the attributes.

Assumed that KMeans can 'fit' or 'predict_fit' an array of tuple elements. Need to add an enumeration of the station types that are based on channel codes and so forth. May also consider fault types as an enhancement.

enhance denclue.py be used with 'haversine' metric and lat lon pairs.

Hover over stations plot to view station type and coordinates
plot_station_faultlines notebook.

Classify the GeoNet strong motion database to find any trends. Consider the cross tabs for selected attributes. Void the location data

Current fault station clusters into 8 clusters using K-means clustering. DBSCAN also provides a large cluster with 100s of stations. Need to improve the clustering, using a combination of DBSCAN and NN to provide a sample of smaller cluster groups of stations that are closer to respective fault lines

It is a prerequisite for forming the rings and

Using the LineString.length function to calculate the distance between two decimal lat/lon pairs. However, the outputs seem to be in Meters even though the length is multiplied by earth's radius 6350Km to convert to Kilometers.

Modify the NN model to improve the accuracy and achieve the training long tail. Consider the strengths and weaknesses of each layer and the number of layers.

use the GeoNet Aug 2020 fault line database to plot them along side the stations

Ask GeoNet to review the code for pulling station, fault line, and wave form data.

• how does one rank the station types?
• how does one rank the fault types?

Each feature corresponds to a fault line. each fault line should have a different colour. Also add a legend with the colour codes and fault name to identify the fault line (Fault line names are in the _WGS84.json file)

notebook: 1a_shield_network_cluster. Color code the stations by their type as defined in the station_type_dict[] in stations.py

Build a nearest neighbour map of clusters of all the operational stations within a 30Km radius of each other.

• Clique them by type (i.e. borehole, accelerometer, broadband, and so on).