PREDATOR  predictive model outputs and what they mean

When a user submits properly formatted predictive model files to the Western Center for Monitoring website, the predictive model software generates four output files.  For a basic analysis we are primarily concerned with two files: “Site Test Results” and “O Over E”. 

Site Test Results
The “Site Test Results” file shows if a sample was within the experience of the model.  Samples outside the ranges of reference site environmental variables used to build the model are considered outliers and are highlighted in red.  For example, in the MWCF model the reference sites were sampled between the day numbers of 176-267.  A test site that was sampled earlier or later than these dates would be considered an outlier. The screen shot to the right shows an example from the Walla Walla watershed (WC+CP model ouput).

It is important to note, that the "Site Test Results" test is only for the predictor variables used in constructing the models. Other important variables, such as year or sample abundances are not included. It is the users’ responsibility to ensure their samples are in the correct regions, were sampled in the correct season, follow similar collection and processing protocols, have enough bugs (at least 200), and were sampled within 1998-2004.  Failure to comply with any one of these parameters may lead to inaccurate predictions of O/E.

O Over E
The “O Over E” file, like the screen shot to the right, contains model scores (O/E), as well as the number of observed reference taxa (O), and the number of expected reference taxa (E).  The output includes calculations for two probability of capture (P_c) thresholds: 0 and 0.5.  The PREDATOR models are based on P_c > 0.5.

Example Applications

Two ways in which DEQ plans to use PREDATOR within their monitoring programs are: 1) population assessments and 2) individual site assessments. 

Population assessments:  These types of studies allow us to make statements about the status and trends of stream resources.  In a study where sites are randomly selected, we can estimate the percent of stream miles that are in good biological condition vs. the extent in poor biological condition.  We can get an idea of improvement or degradation of biological integrity over time (trends) in a study area by incorporating repeat surveys.  DEQ has several ongoing status and trends bioassessment programs: the Environmental Monitoring and Assessment Program (EMAP), the Oregon Plan for Salmon and Watershed, and the Grande Ronde National Monitoring Program.

In a population assessment, the distribution of reference O/E scores is used to establish benchmarks for describing biological integrity.  The DEQ biological benchmarks are the 5^th and 25^th percentiles of reference distributions (Table 4).  To interpret O/E results, think of O/E less than 1.0 as the loss of common native taxa.  In this context, the 5^th percentile benchmark for both models means that samples with “poor” PREDATOR scores have lost approximately 25% of their expected native taxa.  “Fair” samples have lost between ~10-25% of their common native taxa.  We consider samples with a loss of ~10% of common native taxa as “good”, or meeting our biological expectations for reference conditions. 

DEQ has established an upper biological benchmark (> 95^th percentile), so that samples with > 25% of the expected common native taxa are considered to be of “Enriched” condition. The uncertainty in biological condition comes from the potential for streams to show an increase in diversity due to small to moderate levels of disturbance.  A high PREDATOR score may be an early warning sign that human activities are altering the biological community, but not yet at a level that has led to community degradation.  Identification of streams in this category may prompt resource managers to proactively implement best management practices. 

Alternatively, a high PREDATOR score may simply indicate that a stream reach has exceptionally high diversity, potentially representing unique communities worthy of special protection or preserve status.  DEQ recommends samples in both the “Fair” and “Enriched” categories be resampled to obtain a more certain estimate of biological condition.

Table 4.  Benchmarks for describing biological integrity for both PREDATOR models.

Example population assessments are shown for all samples assessed by the MWCF model (Figure 5) and the WC+CP model (Figure 6).  In these two assessments, the “Test” populations are not randomly selected; they simply represent all of the samples collected by DEQ in the regions that fit each model’s specifications.  In this type of assessment, we calculate the percent of samples which fall below or above the benchmarks.  In the Coast Range and Willamette Valley ecoregions (MWCF model) 20% of the samples show biological impairment (O/E scores are “Poor”).  In the Klamath Mountains, Cascades, East Cascades, Blue Mountains, and Columbia Plateau ecoregions (WC+CP) 10% of samples are considered impaired.  For both models, 56% of assessed samples show biological integrity equivalent to reference conditions.  However, 23% and 30% of samples in the MWCF and WC+CP ecoregions (respectively) were found to be of uncertain biological integrity (“Fair”).  A small percentage of streams (1% in MWCF and 4% in WC+CP) showed higher than expected biodiversity (“Enriched”).

Figure 5.  Extent of biotic condition classes for samples in the Coast Range and Willamette Valley ecoregions (MWCF model).

Figure 6.  Extent of biotic condition classes for samples in the Cascades, Klamath Mountains, East Cascades, Blue Mountains, and Columbia Plateau ecoregions (WC+CP model).

With the examples given above (Figures 5 and 6), we can begin to prioritize future monitoring activities.  For instance, we may want to go back to the locations of the 23-30% of samples that are “Fair” and the 1-4% of samples that are “Enriched”.  With repeated sampling, we can determine if these locations are significantly different from reference conditions (see “individual site assessments” below). We may also decide to prioritize restoration activities or landuse restrictions based on which areas show a higher amount of streams designated as impaired.   In these examples, it may make more sense to focus on improving conditions in the Coast Range and Willamette Valley ecoregions which had 10% more impaired samples than those assessed in other ecoregions.

Individual site assessments:  Assessing biological integrity at a single site involves direct comparisons to the mean reference condition.  In other words, is the average O/E score at a site significantly different from the reference average?  Unlike population assessments where we utilize percentiles of the reference distribution to determine the percent of resource within a given quality category, here the intent is to determine if the biological integrity at a single location is significantly different from reference. 

If a single sample falls below the 5^th percentile of the reference distribution, the sample is considered to be significantly different from reference conditions.  We feel confident that the results of a single sample are not different simply by chance, but rather a true difference in biological integrity exists.  In this case, a single sample is sufficient to classify the stream reach as biologically impaired.  However, if a sample falls between the 5^th and 25^th percentiles of the reference distribution, there is less confidence that the O/E score is significantly different from reference.  In this case, DEQ would request repeated measures of O/E to determine if a significant difference in biointegrity exists.   

For a sample with higher than expected O/E, distinguishing  between “Fair” and “Excellent” condition would also require further monitoring of the stream and its watershed.  First, a combination of on-site and remote sensing screens of the watershed could be performed to identify potential sources of human disturbances.  Those sites with low levels of human disturbances may be deemed “Excellent”, while those with higher levels of human activities may require further field sampling to determine potential impacts of the existing activities.  In this case, repeated macroinvertebrate sampling, sampling for water quality, and instream and riparian habitat sampling may provide insight into the likelihood of potential sources of impairment.

Home • PREDATOR • Macro taxonomy • Macro input file • Habitat input file • Run the models
Interpretation • Resources • Thanks

Copyright © 2006, The Xerces Society www.xerces.org; please send comments to info@xerces.org