add some notes to coherence sorting

aurora/transfer_function/weights/coherence_weights.py

@@ -9,11 +9,11 @@
 and that of Ey and Hx.  This yields one single number.
 
 Then, iteratively remove each observation in turn and count whether that partial
-cohernece is bigger or smaller than the baseline and by how much
+coherence is bigger or smaller than the baseline and by how much
 
-Deflection downward means naughty data and defelction  upward means well-behaved data.
+Deflection downward means naughty data and deflection upward means well-behaved data.
 
-This leads to a coherence score for each obervation.  We can boot out all data that
+This leads to a coherence score for each observation.  We can boot out all data that
  say move the estimate down (worse than average)
 """
 
@@ -45,6 +45,15 @@ def coherence_from_fc_series(c_xy, c_xx, c_yy):
 
 def coherence_weights_v00(x, y, threshold=0.95):  # 975):#0.98
     """
+    Note 1: Extremely high coherence can be due to noise
+    Consider ways to pre-filter those events before this is called.
+    - That may need a min_fraction_to_keep=0.2-0.8, which guards against
+    generally poor data being completely discarded.
+    - Consider variations on coh-sorting threshold_min, and threshold_max
+    near perfect coherency can be expected when coherent noise is present
+    - 95% seems extreme for a threshold, maybe not yielding enogh
+    observations for robust statistics.
+
 
     Parameters
     ----------
@@ -101,9 +110,11 @@ def compute_coherence_weights(X, Y, RR, coh_type="local"):
     -------
 
     """
+    # these should be params in the config
     remote_threshold = 0.8
     local_threshold = 0.95
 
+    # redundant -these should already be dropped
     X = X.dropna(dim="observation")
     Y = Y.dropna(dim="observation")
     if RR is not None: