Weighted model calibration with spatial conditional information
Cost functions such as mean square error are often used in environmental model calibration. These treat observations as independent and equally important even though model residuals exhibit spatial dependence and additional observations near existing points do not provide as much information on the system as those elsewhere. To address this issue, we develop a method to derive calibration weights based on spatial conditional information. Using simulation experiments with Gaussian processes and the Tephra2 volcanic tephra dispersion model, we show that the additional accuracy and precision from weighted inference increases with the degree of observation clustering and spatial dependence present. To demonstrate real-world relevance, the methods are applied to tephra load observations from the 2014 eruption of the Kelud volcano in Indonesia.