Statistical inference for spatial statistics defined in the Fourier domain (Q1750275)

The author studies a class of Fourier based statistics for irregular spaced spatial data. The statistic is an approximation of the \(\int g_{\theta}(\omega)|J_{n}(\omega)|^{2}d\omega\) over a grid \(\{\omega_{\Omega,k}\},\) where \(J_{n}(\omega)\) is the irregular sampled Fourier transform. This statistic \(Q(\omega_{\Omega,k},g_{\theta})\) is a quadratic form of a discrete Fourier-type transform of the spatial data. First, sampling properties of \(J_{n}(\omega_{\Omega,k})\) are studied. Then these properties are used to investigate \(Q(\omega_{\Omega,k},g_{\theta}).\) It is shown that the statistic computations require \(O(nb)\) operations, where \(b\) is the number of Fourier frequencies and \(n\) is the sample size. Then the author defines a Studentized statistic which converges to a \(t\)-distributed random variable. The cases of fixed and increasing domains spatial asymptotics are investigated. New results on moments of quadratic forms of irregularly sampled spatial processes are obtained.