Smooth maps from clumpy data

We study an estimator for smoothing irregularly sampled data into a smooth map. The estimator has been widely used in astronomy, owing to its low level of noise; it involves a weight function -- or smoothing kernel -- w(\theta). We show that this estimator is not unbiased, in the sense that the expectation value of the smoothed map is not the underlying process convolved with $w$, but a convolution with a modified kernel w_eff(\theta). We show how to calculate w_eff for a given kernel w and investigate its properties. In particular, it is found that (1) w_eff is normalized, (2) has a shape `similar' to the original kernel w, (3) converges to w in the limit of high number density of data points, and (4) reduces to a top-hat filter in the limit of very small number density of data points. Hence, although the estimator is biased, the bias is well understood analytically, and since w_eff has all the desired properties of a smoothing kernel, the estimator is in fact very useful. We present explicit examples for several filter functions which are commonly used, and provide a series expression valid in the limit of large density of data points.