LOWESS Smoothing

LOWESS, locally weighted scatterplot smoothing, is a non-parametric regression that fits the trend in noisy data by running a weighted local polynomial across the points, weighting each neighbor by distance. It needs no assumed global function, so it follows curvature that a single straight line or low-order fit would miss. In hydrography and tidal analysis it smooths sounding profiles, tide-residual series, and sensor drift while preserving real structure, and a robust variant downweights outliers across iterations. The span, the fraction of points in each local window, controls the trade-off: a wide span oversmooths real features, a narrow span tracks noise.