Improve estimate_noise binning routine

[lsetiawan]

Overview

This issue follows up on Issue #1028. Improvements in 2D binning routines have been implemented for both compute_MVBS and compute_NASC, however it may seem that estimate_noise is still left as mentioned in the issue.

echopype/echopype/clean/noise_est.py

Lines 53 to 81 in f98e825

	def estimate_noise(self, noise_max=None):
	"""Estimate noise from a collected of pings
	Parameters
	----------
	noise_max : Union[int, float]
	the upper limit for background noise expected under the operating conditions
	"""
	power_cal_binned_avg = 10 * np.log10( # binned averages of calibrated power
	self.power_cal.coarsen(
	ping_time=self.ping_num,
	range_sample=self.range_sample_num,
	boundary="pad",
	).mean()
	)
	noise = power_cal_binned_avg.min(dim="range_sample", skipna=True)
	# align ping_time to first of each ping collection
	noise["ping_time"] = self.power_cal["ping_time"][:: self.ping_num]
	if noise_max is not None:
	noise = noise.where(noise < noise_max, noise_max) # limit max noise level
	self.Sv_noise = (
	noise.reindex(
	{"ping_time": self.power_cal["ping_time"]}, method="ffill"
	) # forward fill empty index
	+ self.spreading_loss
	+ self.absorption_loss
	)

An investigation and improvements needs to happen for this method to enhance efficiency, memory usage, and robustness against different types of data.

Original description by @leewujung can be found at #1028 (comment).

[MohamedNasser8]

Hello @lsetiawan,
As I understand it, ping_time can refer to actual timestamps or simply be a sequence of indices. We need to account for both scenarios: when ping_time represents real-time data and when it represents indices. I am making sure to handle both cases appropriately to ensure accurate results.

[ctuguinay]

Closing this as coarsen itself is very fast and probably does not need the Flox algorithms to speed up this process.