Integrated Statistical and Optimization Analysis for Computationally Expensive Models of Complex Environmental Systems

研究人员将开发一种综合的、计算效率高的通用方法，用于利用现场和实验室数据，并结合环境模型，为将参数和输入值的不确定性纳入与环境分析、保护和修复有关的科学调查和决策提供特定地点的基础。这项研究将使用最近开发的基于响应面的全局优化方法，该方法为模型参数的校准提供了一种有效的方法，对于复杂的环境模型来说，这是典型的，模型运行成本很高。在贝叶斯框架内，通过使用反映数据中的非正态分布、异方差和相关性的数据转换和加权来解决不同的不确定性来源，还将提高效率(计算和统计)。环境系统的未来状态将通过模拟根据参数值的后验分布选择的多组参数的基础环境模型和使用感兴趣的每个输出变量的响应面近似来预测。将进行灵敏度和不确定度分析，并对其准确性进行调查。这项研究将应用于纽约坎农斯维尔水库和用于饮用水的地下水中有毒氯代乙烯的生物修复。

The investigators will develop an integrated and computationally efficient general method for utilizing field and laboratory data coupled with environmental models to provide a site-specific basis for incorporating uncertainty in parameter and input values into scientific investigations and decisions related to environmental analysis, protection, and remediation. The research will use recently developed methods of global optimization based on response surfaces that provide an efficient method for calibration of model parameters when, as is typical for complex environmental models, the model is expensive to run. Efficiency (computational and statistical) will also be improved by using, within a Bayesian framework, data transformations and weightings that reflect nonnormal distributions, heteroscedasticity, and correlation in the data, to address different sources of uncertainty.Future states of the environmental system will be forecast by simulating the underlying environmental model for a number of sets of parameters selected according to the posterior distribution of parameter values and using a response surface approximation of each output variable of interest. Sensitivity and uncertainty analyses will be done and their accuracies will be investigated. The research will be applied to the Cannonsville reservoir in New York and the bioremediation of toxic chlorinated ethenes in groundwater used for drinking water.