Ignore missing values when calculating trend

[scotthosking]

missing_value = scs_full_cube.data.fill_value
slope = np.zeros(cube.data[0,:,:].shape)

for x in range(0,len(lons)):
   for y in range(0,len(lats)):
      arr =cube.data.data[:,x,y]

      ind = np.where(arr < 366)[0]

      if len(ind) < len(arr)*0.8: # if 20% of points are missing then set to missing value (ignore it in plot)
        slope[x,y] = missing_value
      else:
        slope[x,y], intercept, r_value, p_value, std_err = stats.linregress(years[ind], arr[ind])

        
# create new mask and add to cube
new_mask = np.zeros(slope.shape)
new_mask[ slope == missing_value] = 1.
new_mask = new_mask.astype(bool)  # convert to boolean array (True/False)
mx = ma.masked_array(slope, mask=new_mask)