The earliest applications of thin plate smoothing splines were described by Wahba and Wendelberger (1980), Hutchinson et al. (1984) and Hutchinson and Bischof (1983). The methodology has been further developed into the ANUSPLIN software package at the Australian National University over the last 20 years. ANUSPLIN has become one of the leading methods in the development of spatial climate models and maps from ground-based data and has been applied in North American and many regions around the world (e.g., New et al. 2002; Hijmans et al. 2005; Rehfeldt 2006; Hutchinson et al. 2009; McKenney et al. 2011). The climate surfaces produced by ANUSPLIN are also fundamental components of the ANUCLIM climate application package which can be used to model species distributions in current and projected future climates.
The thin plate smoothing spline method is able to account for spatially varying dependences on elevation, a dominant predictor that is closely aligned with many controlling physical factors (Fig 1).
A key strength of the method, in contrast to local regression methods, is its dependence on all the data (i.e., every data point contributes to the fitted model). This permits robust and stable determination of dependencies on the predictor variables, particularly in data sparse, high elevation regions.
Comparisons with kriging are discussed by Hutchinson (1993) and Hutchinson and Gessler (1994). Comparison with other climate interpolation methods in Price et al. (2000) and Jarvis and Stuart (2001). Extension to additive regression splines by Sharples and Hutchinson (2004). Assessment of scales of vertical and horizontal dependence of precipitation by Hutchinson (1995a, 1998b) and Sharples and Hutchinson (2005).
The method has been applied to the interpolation of Australia-wide monthly mean surfaces describing the principal climate variables that determine biological activity and hydrological processes. These include daily maximum and minimum temperature, precipitation, pan evaporation and solar radiation for the standard period 1976-2005. These surfaces can be used as baseline data for assessing potential impacts of projected climate change. They can also be used to underpin interpolation of climate variables over actual monthly and daily timesteps. An early application to the interpolation of actual monthly climate data has been to develop an index describing the occurrence of drought (Smith et al. 1992).
Key references on methodology of fitting thin plate smoothing spline surfaces to climate data are Hutchinson (1991a, 1995a). Applications to rainfall interpolation in mountainous regions described in Hutchinson (1998ab). Application to the interpolation of evaporation is described by McVicar et al. (2007). A review of techniques for modelling the impact of topography on earth surface processes, including surface climate, given by Hutchinson (2008). See also a spatial analysis of agro-climatic classes across Australia by Hutchinson et al. (2005). A recent review of applications to the interpolation of North American climate is given by McKenney et al. (2011).
1. Further refinement of methods for standardising monthly means of observed data to standard periods, to maximise the information obtained from stations with incomplete records, particularly in areas with a sparse data network.
2. Extension to real time simulation and surface fitting for Australia by Hutchinson (1995b) and Kesteven and Hutchinson (1996). This has been applied to Australia’s National Carbon Accounting System (NCAS) by Kesteven and Landsberg (2004). This has also been used to assess impacts of climate change on Australia’s snow conditions by Hennessy et al. (2008).
3. Application of the techniques to the interpolation of daily temperature and precipitation across Canada is described by McKenney et al. (2006, 2011) and Hutchinson et al. (2009). This supports a wide range of regional, national and international climate modelling related projects by Natural Resources Canada.
4. Further development of the techniques for improved interpolation of actual monthly and daily interpolation of precipitation and temperature across Australia. This is in further support of the National Carbon Accounting System (NCAS) and of the Ecosystem Modelling and Scaling Infrastructure facility (e-MAST) for the Terrestrial Ecosystem Research Network (TERN). This facility enables testing, updating, and improvement of models used for such applications as future climate scenarios and assessment of primary production, at a wide range of temporal and spatial scales.
1. Further development of the methods to underpin assessments of projected climate change.
2. Assessment of climate extremes, their trends over time, and in projected future climates.