THE AUSTRALIAN NATIONAL UNIVERSITY
CENTRE FOR RESOURCE AND ENVIRONMENTAL STUDIES
CANBERRA
ANUSPLIN VERSION 4.36
USER GUIDE
M.F. Hutchinson
The ANUSPLIN package contains FORTRAN programs for fitting surfaces to noisy
data as functions of one or more independent variables. The package includes programs
for interrogating the fitted surfaces in both point and grid form. Procedures for
calculating standard error surfaces have also been developed.
The programs are normally distributed as binary executables for:
Sun Microsystems Solaris 2.x on SPARC hardware
Silicon Graphics Irix 6.x on MIPS-3 hardware (or later)
Compaq Digital Unix on Alpha hardware
AIX5.x on IBM PowerPC
Linux on Intel or AMD hardware.
Microsoft Windows 95, 98, 2000, NT, ME, XP
Last revision to this document: 17 July 2006
ANUSPLIN Version 4.36 II
The publishing program of the Centre for Resource and Environmental Studies at the
Australian National University is designed to present the results of the Centre’s research
and the proceedings of conferences and workshops. Views expressed in CRES
publications are the views of the authors and are not necessarily those of the Centre or
any associated institution.
Director: Prof Will Steffen
Executive Officer: Sharon McInnes
© CRES 2006
ISBN 086740 512 0
This book is copyright. Apart from any fair dealing for the purposes of study, research,
criticism or review as permitted under the Copyright Act, no part may be reproduced by
any process without permission. Enquiries should be made to the publisher.
All CRES publications are available from:
Publications Section
Centre for Resource and Environmental Studies
Australian National University
Canberra ACT 0200
Tel. +61 2 6125 4277
Fax +61 2 6125 0757
Email: publications@cres.anu.edu.au
URL: cres.anu.edu.au
ANUSPLIN Version 4.36 III
TABLE OF CONTENTS
INTRODUCTION .........................................................................................................1
PROGRAM SUMMARY..............................................................................................3
SPLINA and SPLINB....................................................................................................5
Program Inputs ..........................................................................................................5
Program Outputs........................................................................................................6
Knot Selection for SPLINB .......................................................................................6
Interpretation of Output Statistics..............................................................................7
Calculation of Standard Errors ..................................................................................9
Dependent Variable Transformations......................................................................11
Fitting Climate Surfaces ..........................................................................................12
SPLINA and SPLINB User Directives ........................................................................14
GCVGML....................................................................................................................19
SELNOT......................................................................................................................20
ADDNOT ....................................................................................................................23
DELNOT .....................................................................................................................24
LAPPNT ......................................................................................................................25
LAPGRD .....................................................................................................................27
ANNOTATED EXAMPLES.......................................................................................31
Spline smoothing of uni-variate data.......................................................................32
Partial spline smoothing of monthly mean temperature data ..................................40
Tri-variate spline smoothing of monthly mean precipitation data using knots and
the square root transformation.................................................................................46
Bi-variate and tri-variate spline smoothing of monthly mean solar radiation data
using surface independent variables ........................................................................49
REFERENCES ............................................................................................................52
ANUSPLIN Version 4.3 1
INTRODUCTION
The aim of the ANUSPLIN package is to provide a facility for transparent analysis and
interpolation of noisy multi-variate data using thin plate smoothing splines. The package
supports this process by providing comprehensive statistical analyses, data diagnostics
and spatially distributed standard errors. It also supports flexible data input and surface
interrogation procedures.
The original thin plate (formerly Laplacian) smoothing spline surface fitting technique
was described by Wahba (1979), with modifications for larger data sets due to Bates and
Wahba (1982), Elden (1984), Hutchinson (1984) and Hutchinson and de Hoog (1985).
The package also supports the extension to partial thin plate splines based on Bates et
al. (1987). This allows for the incorporation of parametric linear sub-models (or
covariates), in addition to the independent spline variables. This is a robust way of
allowing for additional dependencies, provided a parametric form for these
dependencies can be determined. In the limiting case of no independent spline variables
(not currently permitted), the procedure would become simple multi-variate linear
regression.
Thin plate smoothing splines can in fact be viewed as a generalisation of standard multi-
variate linear regression, in which the parametric model is replaced by a suitably smooth
non-parametric function. The degree of smoothness, or inversely the degree of
complexity, of the fitted function is usually determined automatically from the data by
minimising a measure of predictive error of the fitted surface given by the generalised
cross validation (GCV). Theoretical justification of the GCV and demonstration of its
performance on simulated data have been given by Craven and Wahba (1979).
An alternative criterion is to minimise the generalised maximum likelihood (GML)
developed by Wahba (1985,1990). It is based on a Bayesian formulation for the thin
plate smoothing spline model and has been found to be superior to GCV in some cases
(Kohn et al. 1991). Both criteria are offered in this version of ANUSPLIN.
A comprehensive introduction to the technique of thin plate smoothing splines, with
various extensions, is given in Wahba (1990). A brief overview of the basic theory and
applications to spatial interpolation of monthly mean climate is given in Hutchinson
(1991a). More comprehensive discussion of the algorithms and associated statistical
analyses, and comparisons with kriging, are given in Hutchinson (1993) and Hutchinson
Gessler (1994). Recent applications to annual, monthly and daily precipitation data have
been described by Hutchinson (1995, 1998ab) and Price et al. (2000). The book by
Schimek (2000) provides a good overview of the subject of smoothing and non-
parametric regression with extensive references.
It is often convenient, particularly when processing climate data, to process several
surfaces simultaneously. If the independent variables and the relative weightings of the
data are the same for each surface then many surfaces can be calculated for little more
computation than one surface. ANUSPLIN allows for arbitrarily many such surfaces
with significant savings in computation. ANUSPLIN also introduces the concept of
"surface independent variables", to accommodate independent variables that change
systematically from surface to surface. ANUSPLIN permits systematic interrogation of
these surfaces, and their standard errors, in both point and grid form.
ANUSPLIN also permits transformations of both independent and dependent variables
and permits processing of data sets with missing data values. When a transformation is
ANUSPLIN Version 4.3 2
applied to the dependent variable ANUSPLIN permits back-transformation of the fitted
surfaces, calculates the corresponding standard errors, and corrects for the small bias
that these transformations induce. This has been found to be particularly convenient
when fitting surfaces to precipitation data and other data that are naturally positive or
non-negative.
A summary of the eight programs that make up the ANUSPLIN package is tabulated in
the following section, accompanied by a flow chart showing the main connections
between the programs. This is followed by detailed documentation for each program in
the package. The User Guide concludes with a comprehensive discussion of example
smoothing spline analyses of uni-variate data and multi-variate climate data. The data
supporting these analyses are supplied with the package. These analyses can be used as
a tutorial on the basic concepts of data smoothing, with particular applications to the
spatial interpolation of climate.
