SequentialBatchLeastSquares with a new Propagator

Orekit usage
1

I have a question about the SequentialBatchLeastSquares tutorial (tag release-12.2), and the use of the SequentialBatchLeastSquares class during OD estimation. The scenario I am trying to accomplish is estimating an orbit with Drag & SRP enabled, based on a previously estimated orbit and set of measurements, with a new propagator.

In the tutorial class, the propagator first gets created with the initialOrbit, and is used along with a BatchLSEstimator to generate the estimatedOrbitLS Orbit. Using the same propagator, created with that initialOrbit, aSequentialBatchLSEstimator is created, with the previousEstimation.

What I would like to do, is from the initial estimator, propagate that orbit to the end of the provided measurement list, and then use that new laterEstimatedOrbitLS, along with a priori data and new measurements to estimate an additional state.

I have modified the tutorial to first set the estimatedOrbitLS to be equal to the epoch of the last measurement. I have also updated the yaml file to enable Drag and SRP.

       //final Orbit      estimatedOrbitLS = estimatedLS.getInitialState().getOrbit();
       estimatedLS.resetInitialState(estimatedLS.getInitialState());
       final Orbit      estimatedOrbitLS = estimatedLS.propagate(new AbsoluteDate("2016-02-13T23:55:00.000Z", TimeScalesFactory.getUTC())).getOrbit();

With that change and using the existing initializeSequentialEstimator method, I get a new estimation that seems reasonable.

The problem I am running into is when I try to use a new Propagator for the SequentialBatchLSEstimator, created from the later estimatedOrbitLS, the estimation takes a lot longer, and the drag is not close to the previous answer. I did modify the Cd/Cr values used in the force models to be the values that were estimated from the initial run.

I have tried to perform the estimation using both the Optimum from the initial estimation, as well as trying to use the a prior data, normalized from the physical solution. I have also tried calling the initializePropagator method, with the initialOrbit, and updated Cd/Cr values, but that also generates different results.

Any information as to if I am doing something unexpected would be helpful.

2

Bumping, hopefully this is the correct approach - Hoping to get some input on the expected behavior when using the SequentialBatchLSEstimator with a new instance of a propagator vs a propagator that was used in a BatchLSEstimator.

3

I am not sure to understand the question, and in fact I don’t really know the sequential version of batch least squares.
I think it behaves roughly like a Kalman filter, i.e. it allows force models to change during the propagation (which classical least square does not allow). If this is true, then using a regular single propagator with fixed drag values is probably doomed to fail.