Nonparametric inference in small data sets of spatially indexed curves with application to ionospheric trend determination

This paper is concerned with estimation and testing in data sets consisting of a small number (about 20–30) of curves observed at unevenly distributed spatial locations. Such data structures may be referred to as spatially indexed functional data. Motivated by an important space physics problem, we model such data as a mean function plus spatially dependent error functions. Given a small number of spatial locations, the parametric methods for the estimation of the spatial covariance structure of the error functions are not satisfactory. We propose a fully nonparametric estimator for the mean function. We also derive a test to determine the significance of the regression coefficients if the mean function is a linear combination of known covariate functions. In particular, we develop methodology for the estimation a trend in spatially indexed functional data, and for assessing its statistical significance. We apply the new tools to global ionosonde records to test the hypothesis of ionospheric cooling. Nonparametric modeling of the space–time covariances is surprisingly simple, much faster than those previously proposed, and less sensitive to computational errors. In simulated data, the new estimator and test uniformly dominate those based on parametric modeling.