Artifacts in Simulated TOPO phase

[pmar]

@AndreaMinchella reported that in some cases artifacts in simulated topographic phase can be observed, as marked in the attachment:

Note: This problem exposes rarely, and it is rather exception than rule.

[pmar]

Status: Confirmed

The problem is in how the tile extents are being computed. This will be fixed in v0.7.

Context:

These tile extents are need to be able to complete cover the SAR tile with input DEM tile. Note, that we are mapping (DEM in WGS84 Geometry in DEM Sampling) -> (DEM SAR Geometry in DEM Sampling) -> (DEM SAR Geometry SAR Sampling).

This tile extent is computed dynamically, not to overhead too much processing of tiles that do not need large extent. For this dynamic computation of the approx max height of the tile is performed -- grid of X points distributed, and search for max heights. However, this might not be robust enough for every and any case study. Especially, if there is strong slope at the edges of the tile, or lot of height difference between tile edges.

Fixes ideas:

• Quick&dirty fix 1: Include more sampling points in computation of the approx max height.
• Quick&dirty fix 2: Increase static +extra extent to 30% - 40%. Currently I work with ~25%.
• Quick&dirty fix 3 = Quick solution 1 + Quick solution 2

---

• Optimal solution: Redesign how extra extent is being computed with smarter reverse mapping.

[pmar]

Update:

Quick and dirty fix 3 is implemented, and it is under evaluation.

[pmar]

Resolved with: 1b1f8d63b1294a1e54243ad2107334e289e3afb9. All test pass.

---

For disclosure and completeness, also attached screenshots of 'before' vs 'after' results. Results with artifacts and without artifacts, as a result of the fix. The input for both results is exactly the same data set.

Note histogram values, and smoothness of the signal.

Before fix

---

After fix

---

[pmar]

Reopening because the implemented fix https://github.com/ppolabs/jlinda/commit/1b1f8d63b1294a1e54243ad2107334e289e3afb9 degrades the performance.

Depending, on the area the slow-down of couple of orders is observed.

[pmar]

Solution to be implemented:

• Work with old version of code, that is quite efficient.
• Make old version of code 20% 'conservative' in terms of dynamic scaling of the tile extension.
• Introduce the processing flag in the operator that would allow user that in extreme cases, or in case of observed problem manual set the tile percent for which tiles will be extended.