[Feat] Heatmap interpolation (continous map with heat defined by points)

[mvorisek]

Target Use Case

Currently it seems there is a no way to render a continous heatmap. Currently the HeatmapLayer can render individual points with optional weight only.

That is great for individual points like paying customers. Each customer has some "point" location with some weigh like spending.

But for "sampled area data" like temperature or population density, the values must be represented using interpolation.

Proposal

1. compute "convex hull" [1]
2. for each point in the hull (render pixel) calculate the heat using interpolation based on the nearest neighborhoods

[1] https://en.wikipedia.org/wiki/Convex_hull

Examples:

Point heatmap (already possible):

Continous heatmap (this feature request):

(in both cases the "heat" must stay between the points no matter the zoom level)

[felixpalmer]

Is this the sort of behavior that you're after? #9753

[mvorisek]

No. The #9753 is about h3 source data support.

The goal and the wanted feature of this feature request is to be able to render "sampled data as continuous".

Example:

(currently supported render)

(wanted supported render)

The expected behaviour/algoritm is described in the feature request description.

As long as the rendered area is in complex hull of the data points, interpolate the nearest values.

Optionally, there should be an option to define maximum physical radius for the hull computation, ie. if two nearest points are more than N meters apart, compute multiple isolated hulls with rendered result like:

And expected result for 3 data points:

[felixpalmer]

Thanks for the explanation. My feeling is that this is too niche to warrant a specific layer in deck.gl, but if you take the time to implement something yourself, feel free to contribute it to the https://github.com/visgl/deck.gl-community repo

[mvorisek]

It is not niche, it is needed for example for rendering temperature, rent prices, anything that has continous meaning.

[felixpalmer]

For the use cases you describe then using some aggregation layer (e.g. H3Layer), or just a raster is the way to go. That is what your "USA" example above is doing. Your other example around LA is just a normal heatmap (like we already have), it has gaps in it.

The reason I say it is niche, is that it is not clear how we would go about implementing this, there are many unanswered questions - for example, how is the value interpolated between 3 or more points. If you would like to create a layer to demonstrate how this could work, we can consider it - but my feeling is that what you want to do is already achievable with the current layers we have