¥Both
methods assume a priori that covariance is maximized at the observation location - problematic for non-local and time-lagged obs.
¥Both
methods are flow-independent (assume same degree of locality for every situation).
¥Localization
can destroy balance.