|
Article Excerpt
Introduction
The reasons for measuring potential ultraviolet (UV) exposure correspond to concerns about health and environment. Prolonged exposure to UV radiation has specific effects on human health, crops, terrestrial ecosystems, and aquatic ecosystems (Tevini 1993; Caldwell and Flint 1994). In humans, UV exposure is critical for understanding melanoma etiology (Elwood 1989; Armstrong and Kricker 1993; Elwood and Koh 1994; Elwood and Jopson 1997; De Fabo et al. 2004) and other common adverse effects, such as conjunctivitis and sunburn (De Gruijl 1997). Measurement of UV exposure, however, has proved challenging. Often, the assessment of the health effects resulting from UV exposure fails to provide useful results due to incomplete exposure measurements.
Measurement of UV exposure at specific body sites has been accomplished using personal passive UV dosimeters (Diffey 1989) which quantify the distribution and the fraction of the ambient UV radiant exposure on a horizontal plane. However, this method, while useful for measuring current individual exposures cannot be used to measure the effects of historical UV exposure (Holman and Armstrong 1984; Autier and Dore 1998). The impact of historical exposure is typically assessed through self-report. The drawback in this subjective approach to UV exposure assessment is that results may be subject to recall bias (Cockburn et al. 2001). In an attempt to better account for historical exposure, mean annual hours of bright sunshine received at location of residence has been used as the metric of individual UV exposure (Armstrong 1988). However, this metric is not the best proxy for UV exposure because of the inconsistent relationship between incoming solar radiation and sunshine hours (see Linacre (1992) and Liou (2002) for factors that influence trends in incoming solar UV).
In recent years, the assessment of the role of historical UV exposure in melanoma etiology has increasingly relied on the interpolation of UV exposure at places of residence based on the available solar radiation networks. Fears et al. (2002) was one of the first to objectively measure UV exposure at location of residence based on the actual UV measurements recorded by approximately 30 radiation stations across the continental U.S.A. However, a principal limitation was that the accuracy of individual UV exposure estimates was unknown. The analysis relied on interpolation of data from a small imperfect measurement network, and it may have fallen short of accurately representing spatio-temporal heterogeneity of incoming UV energy at the local level (e.g. 1[km.sup.2] grid cell). Yet, no proper attention was given to the issue of spatial variability and uncertainty in UV estimates from which measures of individual exposure were derived.
Specifically, the fact that solar radiation varies not only across different latitudes but also across different elevation gradients and various features of the receiving terrain (Wilson and Gallant 2000) has generally been overlooked in recent attempts to quantify UV exposure. Furthermore, the quality of any analysis that relies on interpolation of values at unsampled locations based on the imperfect measurement networks such as that for solar radiation is always subject to a degree of uncertainty, because the derived values are only estimates of what the real values should be at a particular location (Chiles and Delfiner 1999). Different interpolation methods can therefore generate different predictions at the same locations (see Lain (1983), Burrough (1986), Meyers (1994), Burrough and McDonnell (2000), and Cressie (2003) for reviews of different interpolation techniques).
This raises fundamental questions regarding how well potential UV exposure can be modeled, given not only the measurement network and methods available to us at this time but also the resolution of the spatial unit (e.g., place of residence, county, census tract) relevant in the analysis of individual UV exposure. Significant progress has been made in the development of solar radiation models (e.g., Solar Flux model (Dubayah and Rich 1995); SRAD model (Wilson and Gallant 2000), and Solar Analyst model (Fu and Rich 2000)) which are all suitable for fine-scale resolutions. Is it possible to develop models of UV exposure that would effectively cope with the potentially large degree of local variation at, for example, the county level? Studies are needed that will assess the performance of different source data and spatial interpolation techniques for estimating potential UV exposure and the capture of spatial variability and uncertainty in exposure estimates.
A comparison of performance accuracy of different interpolation methods available in standard geographic information systems (GIS) packages was conducted previously (Siska and Hung 2001) using a local, low-vacillating (i.e., relatively uniform) elevation data set. The results suggest that the Thiessen polygon, kriging, and Triangular Irregular Network (TIN) procedures performed almost at the same level, producing the least error in comparison to the Inverted Distance Weighting (IDW) and trend surface analysis interpolation methods. An as yet unanswered question is how would some of the procedures that were identified in this instance as more accurate perform on a non-uniform, elevation-dependent surface such as UV radiation.
In response to the aforementioned needs, the present research attempts to:
* Evaluate the performance of the Thiessen polygon and kriging interpolation procedures in a standard (GIS) package in estimating potential UV exposure across the continental U.S.; and
* Compare these results with those obtained with the ANUSPLIN (Spline) routine (Hutchinson 2003) that runs outside typical GIS through a series of C++ and FORTRAN.
The Thiessen polygon, kriging, and ANUSPLIN procedures were purposely chosen for this type of analysis because of their very different statistical properties, and also computational complexity, and their ability to incorporate additional variables, all of which may differentially affect the predictions. The impetus for stepping outside the typical GIS toolbox and using ANUSPLIN was stimulated by the success of this procedure in predicting precipitation, temperature, and other climate variables while incorporating their spatial dependence on elevation (Hutchinson 1991b; 1993; 1995; 1998; Corbett and Carter 1996; Custer et al. 1996; Stillman 1996). As a part of this effort, our research also includes:
* An examination of the impact of using this alternative methodology on the resulting patterns and estimates of potential UV exposure at the local and county levels; and
* A model for which the accuracy of measurements is known, and which can be used to facilitate objective assessment of historical and cumulative lifetime UV exposure for determining melanoma risk.
In the following section of this paper we describe (1) the measurement network and data used for input in each interpolation procedure; (2) the characteristics of the three interpolation techniques and how they were utilized to generate estimates of potential UV exposure at unsampled locations across the continental U.S.; (3) how model performance was evaluated based on the magnitude and distribution of errors; and (4) the procedures used to examine where the results diverge at the local...
|
