From discretely located to spatially interpolated forest meteorological data - reconstruction of missing values by approximate estimation of forest meteorological data

In the German state Rheinland-Pfalz the Forstliche Versuchsanstalt Rheinlan d-Pfalz acquires forest relevant data at 31 weather stations. Despite sophi sticated measuring techniques data gaps often occur, so that it becomes ind ispensable to develop new approximation methods to close those gaps. Such m ethods require a spatial distribution of meteorological data, changing at s hort temporal intervals, from measurements taken at discrete locations, inc luding a proper error estimation. For this purpose, two geomathematically f unded deterministic approaches are presented in this article. Realistic app rox imations of missing data as well as smoothing of error-affected data ca n be achieved by a multivariate spline interpolation and smoothing method, taking into account the spherical curvature of the earth as well as the rea l topography. This multivariate interpolation method also enables us to pro duce maps of climatological data. Following a different approach, neural ne tworks determine missing data by utilising measurements acquired in the pas t, thereby neglecting topographical factors. This article presents error an alyses for the estimation of missing data of daily mean air temperature by means of various error types (e.g. mean absolute error). These analyses and comparisons with other studies show that both approaches are suitable to s olve the problem of closing data gaps.