Results from Specification Meeting on February 11th, 2016

When to use? What is its purpose?

The Resampling-Op is required whenever within a product there are bands present which do not have the identity-scenetransform, i.e., pixel positions cannot be derived via an affine transform.

→ Shall the ResamplingOp also be used when bands have different imageToModelTransforms?

Every time a user attempts to use a tool on a MS product, but the tool requires a single size product. The ResamplingOp will ask what is the master resolution (model crs) and scale all bands to that resolution. The user will be able to determine how the scaling shall take place.

When exactly shall the UI pop up? How often? What settings will/can be saved? How will the product be handled in SNAP? Shall it appear as a second (or third/fourth) product in the tree? Or shall there be an internal linkage to the resampled product(s)? If so, how is this resampled product presented to the user?

 The resampled product shall appear in the product tree as an usual new product 

What to do if the user wants to re-resample a product?

→ Works as for any other operator.

How shall the actual resampling take place?

Suggestion: The user is presented various resolutions present in the product. He/she can choose any of these as the master resolution (or define his own).

How shall masks be handled?

Masks can be used in almost every Sentinel Toolbox functionality. They do have their own UI (The Mask Manager), but are often used without calling it explicitly. Currently, masks are only supported for those bands which have the same resolution/size or scenetransform as the mask. As consequence, the ResamplingOp would need to be called immediately when a product is opened. This undermines the purpose of the tool. Very similar problems exist for layers and valid pixel expressions.

→ Potential Solution: Keep masks as they are (only applicable to bands with same scenetransform). Allow in MaskManager to create a resampled product, use it if it is available.

Question: Since we have the SceneTransforms, which are the actual instances when the ResamplingOp is needed?

The ResamplingOp is needed in the cases when actual raster data values shall be compared to each other. The SceneTransform can serve to find corresponding pixel positions in rasters, but in some instances the user might want to perform an interpolation to arrive at more refined values. The ResamplingOp is also needed when operations are performed tilewise, as in these instances parts of the product are to be kept in memory. Reading parts from other tiles would counteract this approach.

List of Tools that require the ResamplingOp:

List of Tools that do NOT require the ResamplingOp