Using spatial interpolation to construct a comprehensive archive of Australian climate data

https://doi.org/10.1016/S1364-8152(01)00008-1Get rights and content

Abstract

A comprehensive archive of Australian rainfall and climate data has been constructed from ground-based observational data. Continuous, daily time step records have been constructed using spatial interpolation algorithms to estimate missing data. Datasets have been constructed for daily rainfall, maximum and minimum temperatures, evaporation, solar radiation and vapour pressure. Datasets are available for approximately 4600 locations across Australia, commencing in 1890 for rainfall and 1957 for climate variables. The datasets can be accessed on the Internet at http://www.dnr.qld.gov.au/silo. Interpolated surfaces have been computed on a regular 0.05° grid extending from latitude 10°S to 44°S and longitude 112°E to 154°E. A thin plate smoothing spline was used to interpolate daily climate variables, and ordinary kriging was used to interpolate daily and monthly rainfall. Independent cross validation has been used to analyse the temporal and spatial error of the interpolated data. An Internet based facility has been developed which allows database clients to interrogate the gridded surfaces at any desired location.

Introduction

A complete and accurate source of rainfall and climate data is a prerequisite for the efficient modelling of a wide variety of environmental processes. While the nature of the individual model may vary, most have the fundamental requirement of a dataset that is complete on a temporal and/or spatial basis. To date, this problem has restricted the research efforts of many workers due to the fact that observational records are typically incomplete, making it difficult to construct a continuous climate record. In particular, such data may: (1) be recorded for discrete periods, not spanning the entire time period of interest; (2) contain short intermittent periods where data have not been recorded; and (3) contain either systematic or random errors (Peck, 1997). Furthermore, erroneous data must be removed once detected, which is a difficult problem in itself. While these points focus on the incomplete temporal aspects of observational data, another inherent problem is the spatial distribution of recording stations. The density of stations in observational networks is of particular interest to those who use models which require point data. In many applications, the success or at least accuracy of point simulations can be critically dependent upon the availability of observational data within an acceptable distance of the location under investigation. Ideally the nearest recording station would be situated such that its climatology was identical to that of the location of interest. However, due to the sparsity of observational networks, the distance to the nearest station can be of the order of several hundred kilometres. As a result, the only available data may not be representative of the climatology at the desired location.

The fact that observational data are both spatially and temporally incomplete has led to the widespread interest in remotely sensed data. While in principle such data have the potential to overcome the deficiencies of observational data, its implementation has been severely limited by a number of factors. In particular, satellite data are unavailable for dates prior to the 1970s and, more importantly, remotely sensed information is in most cases an indirect measure of climatic elements. In order to derive quantitative information from remotely sensed data, one must first construct acceptable calibration models to transform the measured signal. Given this already difficult task, one may also be required to extract the desired component from a number of background effects. While calibration models are rapidly improving, and remotely sensed data are becoming widely available, ground-based observational data remain the preferred source of meteorological data.

The reconstruction of serially incomplete data records has been the subject of a number of review articles (see for example Lam, 1983, Bennett et al., 1984). An associated problem which has received less attention is that of record length. The duration of available data records may impact upon the quality and interpretation of modelled results in two ways. First, unbiased statistical analyses may require all datasets to be of equal length, and furthermore, some analyses may impose the additional constraint that all datasets span identical time periods. Second, certain applications can be critically dependent upon the use of long term climate records. In these situations, short term datasets can only be used if they are correctly interpreted in a historical context. As climate data typically consist of a large number of stations with limited observation periods, the issue of record length must be considered carefully. If reliable algorithms are not used to enable reconstruction of long term records from short term datasets, a large proportion of the observed data may be rendered useless.

The majority of observational data is collected and processed by government agencies. These agencies are typically responsible for the installation and operation of the various data recording methods, and also the quality control of the observed data. Many users of such data accept it as being accurate and are not aware of the inconsistencies and biases which can occur in such data. While the existence of erroneous data may corrupt the output from a numerical model, these problems can be minimised by rejecting input data that is considered to be statistically incorrect. The existence of serially incomplete records is a more serious problem as it can result in the rejection of entire datasets. While numerous techniques for estimating missing data values have been implemented (Creutin and Obled, 1982), it is undesirable that individual researchers should have to expend considerable resources to develop their own databases. These problems could be overcome through the development of a single unified archive of quality climate data, which is publicly accessible. This would relieve individual workers of the problems associated with generating continuous climate records from sets of intermittent data. While extensive archives of observed and gridded datasets are available, these are usually restricted to mean datasets and/or monthly time steps. In many modelling applications one requires continuous daily time step records which are not widely available. Furthermore, spatial modelling usually requires access to high resolution gridded surfaces which are rarely available on a continental scale.

The purpose of this paper is to describe the construction of a climate database which was developed to specifically address the aforementioned issues regarding spatially and temporally incomplete datasets. In particular we describe database construction in 2 Database construction, 5 Conclusion, 3.3 Rainfall, followed by an analysis of the interpolation error in Section 3. Discussion and concluding remarks are presented in Sections 4 and 5, respectively.

Section snippets

Database construction

A climate database has been constructed using observational data collected by the Australian Bureau of Meteorology. The database consists of continuous daily climate records at point locations, and sets of interpolated daily surfaces. The gridded surfaces were constructed to facilitate spatial modelling and the compilation of point datasets. Point records were constructed for stations which had long term observational datasets. At each location, the available data were used as a base to

Estimation of interpolation error

An accurate estimate of interpolation error may be critical for three reasons. First, the output from numerical models must be correctly interpreted in the context of the model's sensitivity to input data. Obviously the outputs of any numerical model must be critically appraised if the model is sensitive to perturbations in the input data that is below the expected magnitude of interpolation error. Second, if management decisions are going to be based upon interpolated data, or the results of

Discussion

In the previous sections we outlined the procedures and algorithms used to construct a database of observed and interpolated data. The datasets described are in widespread use and have assisted in the development of many research, educational and managerial applications which were previously restricted by the difficulty in obtaining complete and continuous datasets. The AussieGRASS project (Carter et al., 2000) is one such example. Given a comprehensive array of agrometeorological data,

Conclusion

A comprehensive archive of Australian rainfall and climate data has been described. The archive was constructed to facilitate research and managerial tasks requiring hydrometeorological data. Prior to the availability of such a database, individuals requiring data had to expend considerable resources to develop their own datasets from observational data. This task is both time consuming and in many cases made difficult by the fact that climate data may never have been recorded at a site within

Acknowledgements

The authors gratefully acknowledge the Australian Bureau of Meteorology for the provision of climate data. This work was supported by the Queensland Department of Natural Resources, the Land and Water Resources Research and Development Corporation and the Rural Industries Research and Development Corporation. Many helpful discussions with Dr Mike Hutchinson have been of great benefit, for which we are most grateful. The authors would like to thank Ken Day and Dr Graeme Hammer for assistance in

References (28)

  • R.J. Bennett et al.

    Review article: The problem of missing data on spatial surfaces

    Annals of the Association of American Geographers

    (1984)
  • C.F. Bohren et al.

    Atmospheric Thermodynamics

    (1998)
  • J. Carter et al.

    Development of data rasters for model inputs

  • J.O. Carter et al.

    Aussie GRASS: Australian grassland and rangeland assessment by spatial simulation

  • J.R. Colquhoun

    A limited test of methods of pressure reduction to mean sea level

    Australian Meteorological Magazine

    (1965)
  • J.D. Creutin et al.

    Objective analyses and mapping techniques for rainfall fields: an objective comparison

    Water Resources Research

    (1982)
  • G. Dedieu et al.

    Satellite estimation of solar irradiance on the surface of the Earth and of surface albedo using a physical model applied to Meteostat data

    Journal of Climate and Applied Meteorology

    (1987)
  • C.R. Dietrich et al.

    Estimation of covariance parameters in kriging via restricted maximumlikelihood

    Mathematical Geology

    (1991)
  • Hounam, C.E., 1961. Evaporation in Australia. A critical survey of the network and methods of observation together with...
  • M.F. Hutchinson

    On thin plate splines and kriging

  • M.F. Hutchinson

    Interpolating mean rainfall using thin plate smoothing splines

    International Journal of Geographical Information Systems

    (1995)
  • M.F. Hutchinson et al.

    Normalisation of rainfall across different time steps

  • K-L. Hsu et al.

    Precipitation estimation from remotely sensed information using artificial neural networks

    Journal of Applied Meteorology

    (1997)
  • K-L. Hsu et al.

    Estimation of physical variables from multichannel remotely sensed imagery using a neural network: Application to rainfall estimation

    Water Resources Research

    (1999)
  • Cited by (0)

    View full text