Inverse distance weighting

Content sourced from Wikipedia, licensed under CC BY-SA 3.0.

Inverse distance weighting (IDW) is a simple, deterministic way to estimate unknown values from a set of known data points. It assumes that points closer to the location you’re estimating have more influence than farther points.

How it works
- You have N known points (xi, ui). To estimate the value at a new location x, you measure the distance from x to each xi.
- Each known value gets a weight that is the inverse of its distance raised to a power p: wi = 1 / di^p, where di is the distance from x to xi.
- The estimated value at x is the weighted average: u(x) = sum(wi * ui) / sum(wi).
- The same simple rule applies whether you’re estimating a single point or many points across a region.

Choosing the power p
- The power p controls how strongly nearby points influence the estimate.
- Larger p makes the closest points matter more; smaller p allows farther points to have more influence.
- Common practice is to use p values between 1 and 3. In higher dimensions, the effect of p changes similarly to the data's dimensionality.
- The choice depends on how smooth you want the surface to be and on how data are distributed.

Tips and variations
- You can limit influence by using only nearby points or a fixed radius, which can speed up computation and reduce the impact of distant points.
- IDW can be made faster with spatial data structures (like kd-trees) when dealing with many points.
- It’s often used to create surfaces from scattered data and to build simple spatial weights for analyses (such as Moran’s I).

What IDW is good for and its limits
- Pros: Easy to understand and implement; fast for moderate data sizes; produces intuitive results where closer data points are more relevant.
- Cons: If data are unevenly distributed, results can be biased toward densely sampled areas; it doesn’t model underlying processes and may not extrapolate well beyond the observed region.
- Use cross-validation to choose p and to decide whether to include all points or only nearby ones.

Overall, IDW provides a straightforward way to turn scattered measurements into a continuous surface by giving more weight to closer observations.