CADDIS Volume 4: Data Analysis
- Controlling for
- Analyzing Trait Data
- Propensity Score
- How do I make these predictions?
- How do I use these predictions in causal analysis?
- More information
Author: L.L. Yuan
Predicting Environmental Conditions from Biological Observations (PECBO)
Different taxa generally require different environmental conditions to persist. If we know the environmental requirements for different taxa, we can predict environmental conditions at a site based on observations of these taxa. These biologically-based predictions can be useful in cases where environmental data are not available or are difficult to obtain. They can also provide valuable supporting evidence in a causal analysis.
The environmental requirements of different taxa can be represented with taxon-environment relationships. A taxon-environment relationship quantifies the relationship between the probability of observing a particular taxon and the value of one or more environmental variables (Figure 1). These taxon-environment relationships can be estimated from field data and then combined statistically with observations of the presence or absence of taxa at a new site to predict environmental conditions.
You can predict environmental conditions from biological observations at a site of interest if an appropriate set of taxon-environment relationships is already available, or if you have a data set that can be used to compute them.
Taxon-environment relationships can be estimated from field data using logistic regression. Logistic regression estimates relationships between the presence/absence of particular taxa in a sample and the values of different environmental variables. Relatively large data sets (i.e., > 500 samples) of matched environmental and biological observations are required for these models. Taxon-environment relationships that relate the occurrence of different macroinvertebrate taxa and certain environmental variables (such as stream temperature and bedded fine sediment) have been computed for western streams and are available for use in the R library bio.infer and in CADStat. Technical details and programs for calculating and using taxon-environment relationships to predict environmental conditions are available as an appendix to this page.
Predictions of environmental conditions based on biological observations can be incorporated into causal analysis as the verified prediction type of evidence. That is, if the level of a certain stressor is elevated at a site, we would hypothesize that the biologically-based prediction for that stressor would also be elevated. For example, if we hypothesize that elevated temperature is a stressor at a site, then we would predict that temperature inferred from the biota collected at the site would be higher than expected. If the biologically-based prediction is in fact elevated, as predicted, then the evidence would support the case for that stressor as a candidate cause.
Assessing whether a prediction of environmental condition differs from reference expectations may require you to control for natural variation in the biologically-based predictions. For example, stream temperature usually decreases as site elevation increases. Therefore, a biologically-based prediction of stream temperature at the site must be compared with a site-specific reference expectation to determine whether stream temperature is higher or lower than expected.
For more detail, see the example of using PECBO in a causal assessment in Volume 3: Examples & Applications.
Environmental variables often covary (e.g., both % fine sediment and stream temperature tend to increase with decreasing elevation). Thus, when developing taxon-environment relationships, covarying environmental factors must be considered. In many cases, it is useful to model different environmental variables simultaneously.
Technical details and programs for calculating and using taxon-environment relationships to predict environmental conditions are available as an appendix to this page.
Taxon-environment relationships (Figure 1) can be used to identify taxa that are indicative of certain environmental conditions. However, predictions of environmental conditions as described here take into account subtle variations in the composition of the entire assemblage, and provide a more informative assessment of site conditions.