particleFilter.sse

Compute the sum of squared errors for particles


Syntax



Description

sse = obj.sse

Computes the sum of squared errors for each ensemble member in each time step. The sum of squared errors evaluate the similarity of each ensemble member to the observations. Lower SSE values indicate greater similarity between an ensemble members and the observations. SSE values are computed as the sum of uncertainty-weighted innovations for each ensemble member in each time step. This method requires the particle filter object to have observations, estimates, and uncertainties. Does not require a prior.

The SSE values are the basic input to the different weighting schemes for a particle filter, so you can use them to implement new weighting schemes. Another application of SSE values is to rank the members of an ensemble by their similarity to the observations.


Output Arguments

sse

numeric matrix [nMembers x nTime]
The SSE values for a particle filter. Each row holds the weights for a particular ensemble member. Each column holds weights for an assimilation time step. Lower values indicate greater similarity to the proxy observations.