Statistical estimation of missing data in life cycle inventory: an application to hydroelectric power plants

In evaluating the impacts of products and processes, life cycle assessment has taken a prominent role. Missing data, however, continue to affect the completeness and accuracy of such environmental assessment tools. Data are missing for many reasons, for specific processes, at random in existing inventories or worse, nonexistent. This article proposes a statistical approach to address the lack of data in life cycle inventories and applies it to hydroelectric power plants. Among large power production technologies, hydroelectricity varies considerably in scale and from one site to another. The procedure relies on relationships between the technical properties of a system or process, in this case hydropower plants, and the embodied material and energy flows in construction, operation and eventually dismantling. With highly flexible estimators known as kriging, predicting the value of material and energy flows becomes more accurate. From relatively small sample sizes, kriging allows better estimation without averaging out any of the observed data. Similarly, parameter estimation and model validation can be performed through cross validation which assumes very little on the data itself. Calculations of mean absolute errors for various forms of kriging and regression show that the former estimates the values of material and energy flows more accurately than the latter, more so in cases of incomplete data. Accounting for several technical characteristics as well as joint estimation of covariate material and energy flows provide different ways to further reduce the errors and improve the completeness of inventories where data are missing. The specificity of hydroelectric power makes existing inventory data largely unrepresentative and kriging offers new possibilities to increase the reliability of estimated, representative data. (C) 2012 Elsevier Ltd. All rights reserved.