Limitations to the ReVA Approach

A fundamental limitation to the ReVA approach is lack of data to describe all relevant features of a region's condition. Therefore, there is an implicit assumption that the available data are sufficient. There is the risk that a watershed will appear to be in good condition when, in fact, an unmeasured resource is seriously impacted.

A second consideration is spatial scale. In the ReVA approach, a single value for each variable is applied to a watershed. This permits one to see the overall pattern of quality across the region. But the watershed value cannot be applied to every point within the watershed. It is quite possible that variability within a watershed is as great as variability between watersheds. Therefore, the ReVA data sets cannot automatically be scaled down to finer resolution.

On the other hand, aggregating finer scaled data up to coarser scales has some advantages. If measurements are unbiased, their average over the watershed has less uncertainty than the individual measurements. Interpretation errors of satellite imagery are smaller at larger scales. In general, spatial models, such as air models, provide a more accurate estimate of large scale averages than of individual points in space. The only real problem occurs when point data are interpolated. This introduces errors that are difficult or impossible to estimate.

Regional data sets have some peculiarities that may affect statistical analysis. Analyses often assume continuous variables but regional data may be discrete or even binary. For example, an individual endangered species may be represented as either absent (0) or present (1) on a watershed. Such variables can usually be aggregated so that the regional variable becomes the total number of endangered species rather than the presence or absence of each individual species. In other cases, the variable can be expressed as the percentage of potential habitat actually occupied by the endangered species.

Most statistical analyses assume that the data are normally distributed. Experience with the pilot study in the Mid-Atlantic region indicates that regional variables will often be strongly skewed. This problem will probably prove to be common in regional data. For example, agriculture on steep slopes is absent on most watersheds but reaches high values in some areas. These skewed distributions may affect some statistical analyses, but there is little that can be done to skirt the problem. Logarithmic or exponential transformation of only a few variables would make the problem as bad or worse. The best approach is to test each method for its sensitivity and use a suite of methods to integrate the data, choosing methods with different sensitivity to data distributions.

Another problem with regional data is imbalance. There may be a single variable characterizing aquatic systems and a dozen describing forest condition. If every variable is given equal weight, then an integrated index may underestimate the value of the aquatic system. If this problem arises, the variables should be grouped and the averages, rather than the original variables, used to characterize overall environmental condition.

At present, ReVA is limited to providing relative assessments. It is possible to assess which watersheds are in better or worse environmental condition, but it is not possible to make the statement in any objective sense of good or bad. To make an objective statement requires some criterion, such as a threshold, that allows one to determine whether a variable value is acceptable. Unfortunately, such criteria do not exist for variables such as persons over 65 living in poverty. As a result, ReVA is limited to calculating spatial patterns of relative environmental condition.

ReVA Home

Research & Development | National Exposure Research Laboratory | Environmental Sciences

EPA Home | Privacy and Security Notice | Contact Us