A large number of errors can be made in obtaining data. These include errors in the methods used to collect, transport, handle, and analyze field data, and errors introduced by laboratory or other methods used to determine the property of interest.
Many, if not most, of the data used in vulnerability assessments display significant spatial and temporal variability. Thus, large sampling errors can occur because different estimates of attributes or model parameters will be obtained from different samples at different locations or time periods. Bias may also be present because the sites selected for data collection may not truly represent the area or volume they are intended to characterize, or the timing of data collection may not account for seasonal effects on measured values. In fact, the concept of a representative value for processes that vary continuously in time or space is difficult to justify in theory. For practical purposes, however, it is necessary to assume representative values, particularly when data are limited.
Consequently, at any given spatial and temporal scale of interest, the model parameters should represent effective values at the appropriate scale. Any method to obtain averages will not eliminate uncertainty due to variability occurring at smaller scales. Large scale average values have reduced variance compared to values obtained over the same spatial and temporal domain but at a smaller scale, thus increasing the uncertainty of localized behavior. Appropriate averaging schemes and the magnitude of the associated uncertainty are model and problem specific and depend on the spatial and temporal structure of the variability. For example, areas of missing information in the mapped region should be found and data obtained as needed to create as complete a database as possible. Where empirical data are not available, estimates can be made by conditional simulations, interpolation, or extrapolation. One simple method is to assign the nearest data series available to a nearby map cell and assume that the value is constant over each cell. Alternatively, collected data can be used to develop estimates for the data at locations where data are missing. This process may involve a geostatistical technique such as kriging in which a model of the spatial correlation is developed from existing data and used to estimate an unknown value and the estimation variance. Statistical time series methods may be used for data exhibiting temporal variability. In any event, these or any other methods of interpolation or extrapolation will introduce additional sources of error and/or uncertainty.