Network-not-connected error in step 1

[falkamelung]

I am frequently getting a Network-not-connected error in the consistency check. My question is why I don't get this error in this case when the network is not connected.

I first ran a dataset using num_nearest_neighbours: 30. As the network is not connected,

I get a Network-not-connected error as expected:

2025-06-20 23:25:04,208 - INFO - ITERATION: 0
2025-06-20 23:25:08,879 - ERROR - Singular normal matrix. Network is no longer connected!

Then I increased the number of nearest neighbors to num_nearest_neighbours: 50. It runs to the end and results are reasonable. However, the plot shows that the network is still not connected.:

Why no network-not-connected error is raised? Th network plot looks exactly as for 30 nearest neighbors. Is this plot not indicative for network connectivity? Any insights are appreciated. I am also confused why you suggest in this response #50 (comment) to lower num_nearest_neighbours to get a connected network. I would expect an increase can yield to a connected network.

These are the parameters:

    consistency_check: {
        coherence_p1: 0.7,
        grid_size: 0,
        mask_p1_file: "",
        mask_p1_file: "maskfiles/narrow2_maskPoly.h5",
        num_nearest_neighbours: 50,
        max_arc_length: null,
        velocity_bound: 0.1,
        dem_error_bound: 250.0,
        num_optimization_samples: 100,
        arc_unwrapping_coherence: 0.5,
        min_num_arc: 3,
    },

The complete sarvey output including config for 50 nearest neighbors is here
and for 30 nearest neighbours here

[mahmud1]

Hi @falkamelung.

Is this plot not indicative for network connectivity?

The plot only shows arcs with temporal coherence < 0.7. This means there is a possibility that arcs with temporal coherence > 0.7, which are not displayed in the figure, are actually connecting the network. The figure does not really indicate network connectivity, rather, it shows which points in the network have lower quality and whether the noisy point removal process was effective (comparing the before/after removal figures).

@Andreas-Piter I think we need to implement an additional plot that is indicative of network connectivity.

I am also confused why you suggest in this response #50 (comment) to lower num_nearest_neighbours to get a connected network. I would expect an increase can yield to a connected network.

In this step, a point is considered noisy, among other factors, if the average coherence value of all its arcs is below consistency_check:arc_unwrapping_coherence. Setting a large number of consistency_check:num_nearest_neighbours increases the chance of longer arcs between P1 points that are far apart -> Longer arcs tend to have lower coherence (atmosphere, etc.) -> Average arc coherence is lower -> The point and all its arcs are removed -> network is disconnected Therefore, when the network is not connected, reducing the number of nearest neighbors may help preserve the points.

[falkamelung]

Thank you for the explanations, @mahmud1 . Yes, something showing network connectivity would be helpful. And some more guidance on the viable ranges for the different thresholds and parameters. It is easy to automatically rerun after a failure with a different parameters, but you want to get into the range where results are not reliable.