Associating Sparse Network with GaMMA

[ev-anderson]

Hello,

I was curious if GaMMA is able to associate phase picks obtained from a very sparse network composed of 3 instruments?

I have good phase picks acquired from PhaseNet, however when I try to associate the events GaMMA yields the dreaded -1 event index for phase arrivals and generates no catalogs. I've played around with the parameters extensively with all tests resulting the same.

Is GaMMA limited by the number of stations and/or the spatial footprint of the network?

Thank you!

[SquirrelKnight]

Hi, I am working with a similarly small network (5 stations). Here are my parameters, naturally the x, y, and z in the config are going to be different for you. I would recommend still playing around with the min_picks_per_eq parameter as well. Mine is currently set to 5, but you may need a lower threshold for only three stations.

Are you getting phases for all three of your stations? You may also need to increase dbscan_eps if the distance between the stations is fairly large, or turn off dbscan entirely. Good luck!

Gamma

config = {}
config["dims"] = ['x(km)', 'y(km)', 'z(km)']
config["use_dbscan"] = True
config["use_amplitude"] = False
config["x(km)"] = (400, 570)
config["y(km)"] = (5190, 5430)
config["z(km)"] = (0, 30)
config["vel"] = {"p": 4.5, "s": 4.5 / 1.73}
config["method"] = "BGMM"
if config["method"] == "BGMM":
config["oversample_factor"] = 10
if config["method"] == "GMM":
config["oversample_factor"] = 10

DBSCAN

config["bfgs_bounds"] = (
(config["x(km)"][0] - 1, config["x(km)"][1] + 1), # x
(config["y(km)"][0] - 1, config["y(km)"][1] + 1), # y
(0, config["z(km)"][1] + 1), # x
(None, None), # t
)
config["dbscan_eps"] = 25.0 # seconds
config["dbscan_min_samples"] = 3

Filtering

config["min_picks_per_eq"] = 5 # 4 has many more events, but 5 seems to have a lot fewer mislocations
config["max_sigma11"] = 2.0
config["max_sigma22"] = 10.0
config["max_sigma12"] = 1.0

[zhuwq0]

@SquirrelKnight Thank you for explaining the issue! That is correct. GaMMA is basically a clustering method following traveltime/ground-motion relationships. It is not limited by the number of stations, while clustering is a bit sensitive to the hyper parameters.