1 / 26

Rule-based models for uniform assessments on the Biological Condition Gradient

This study describes the development of rule-based models for automated assessments on the Biological Condition Gradient (BCG) for New Jersey streams. The study includes the definition and description of BCG, the development of an automated assessment model, and the integration of BCG into uses and standards (TALU). The study also presents the taxonomic attributes and tiers used in the model, as well as the decision rules and membership functions applied. The advantages of quantitative fuzzy-set models for BCG assessments are discussed.

eileeni
Download Presentation

Rule-based models for uniform assessments on the Biological Condition Gradient

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Rule-based models for uniform assessments on the Biological Condition Gradient Jeroen Gerritsen, Kevin Berry, Dean Bryson, Jonathan G. Kennen, James Kurtenbach, Erik W. Leppo, Victor Poretti

  2. New Jersey TALU development • Define and describe BCG for New Jersey streams • Develop model for automated assessment • (Next) Integrate BCG into uses and standards (TALU)

  3. 1 Native or natural condition Minimal loss of species; some density changes may occur 2 Natural Some sensitive species maintained but notable replacement by more tolerant taxa; altered distributions; functions largely maintained Some replacement of sensitive-rare species; functions fully maintained 3 4 Biological Condition Tolerant species show increasing dominance; sensitive species are rare; functions altered 5 Degraded Severe alteration of structure and function 6 Low Stressor Gradient High

  4. Second order stream in a minimally disturbed, forested watershed Source: Maine DEP, Susan P. Davies

  5. Third order stream draining a shopping mall Source: Maine DEP, Susan P. Davies

  6. Taxa Attributes • I: Historic, endemic, long-lived • II: Highly sensitive • III: Intermediate sensitive • IV: Intermediate tolerance (indifferent) • V: Tolerant taxa • VI: Non-native taxa

  7. TIER 6 COMMUNITY II - Highly sensitive, specialist none III - Sensitive - ubiquitous, generalist none IV - Intermediate tolerance, opportunistic none V - Tolerant Taxa Helisoma 48 Thienemannimyia 16 Physa 4 Cricotopus 2 Ablabesmyia 1 Helobdella 1 TIER 2 COMMUNITY • II – Highly sensitive, specialist • Taeniopteryx 48 • Epeorus 13 • Hexatoma 8 • Probezzia 8 • Isoperla 7 • Pteronarcys 1 • Capniidae 1 • Chloroperlidae 1 • Glossosoma 1 • Brachycentrus 1 • III - Sensitive - ubiquitous, generalist • Ephemerella 127 • Acentrella 13 • Stenonema 8 • IV - Intermediate tolerance, opportunistic • Polypedilum 24 • Cheumatopsyche 5 • V - Tolerant Taxa • Hydropsyche 8 Vs. Data Source: Maine DEP, Susan P. Davies

  8. Tiers Exceptional Good Fair Poor Very poor

  9. Assessment Exceptional Good Uncertain zone x Fair Impaired or not? Poor Very poor

  10. Challenge • What are the boundaries of categories? • Nature doesn’t have hard boundaries between categories • Experts use strength of evidence

  11. The Nile (6688 km) is long The Hudson (490 km) is long The Mississippi (3756 km) is long The Rhine (1312 km) is long The Columbia (2044 km) is long TRUE FALSE TRUE FALSE ??? The set of long riversLong rivers are > 2000 km What is the probability that the Columbia is long?

  12. Membership functions(mathematical fuzzy set theory) Degree of membership Not Long Rivers Long Rivers 1 0 1000 2000 3000 4000 Length, km

  13. BCG Development (NJ) • Taxa assignment to attribute groups I-VI • Site assignments to tiers of BCG • Conceptual tier descriptions • Determine decision rules • Develop membership functions • Apply rules in model

  14. Decision Rules • Expert panel assigned sites to Tiers • Nominal Tier: majority choice • Elicit rules • Quantify rules • Model choice follows rules

  15. Rule development: Tier 2 ExampleHighly Sensitive Taxa (Attribute II) • Taxa: Tvetenia, Epeorus, Leuctra, Glossosoma, Helicopsyche • Richness is moderate compared to other attributes • Linguistic Rule: Richness is half or more of any other single attribute group • Quant. Rule: Taxa (II) > 40-60% of any other attribute

  16. 0.4 0.3 0.2 0 0.5 0.6 0.7 0.8 0.1 Highly Sensitive Taxa (Attribute II) Degree of membership Tier 3 1 Tiers 4-6 Tier 2 0 Att II / max of others

  17. Tolerant taxa (Attribute V) • Taxa: Limnodrilus, Cricotopus, Tanytarsus,Physella • Occurrence and densities of Tolerant taxa are as naturally occur. Typically present but a moderate fraction of organisms. • Linguistic Rule: Tolerant individuals comprise a low fraction or less of all organisms • Quant. Rule: Individuals (V ) < (15-25%)

  18. 1 Tier 4 Tier 3 Tier 2 0 60 70 10 20 30 40 50 0 % abundance Tolerant Taxa (Attribute V) Degree of membership

  19. Rule Application • Tier 2: All rules apply (total taxa, Attrib II, Attrib III, Attrib. V) • Tier 3: Any Tier 2 failure, plus Tier 3 rules • And so on, a downward cascade to Tier 6

  20. And the winners are… • Nominal Tier: Tier with the highest membership • Ties are possible • “Runner-up” tiers exist

  21. Model Performance at Matching Panel

  22. Performance (2): non-matches • Ties (10): expert group nominal and minority choices = tied tiers • Flip-flops (10): model and group nominal, minority choices flip-flopped • Complete mismatch (1 Tier): 3 • 78% exact; 97% close

  23. Advantages of quantitativefuzzy-set models (after Klir 2004) • Capture “irreducible measurement uncertainty” • Capture vagueness of linguistic terms • Cost effective for complex, nonlinear models • Enhance the ability to model human reasoning and decision-making

  24. Quantitative Models for BCG • Quantitative rules replicate expert judgment • Quantitative rules are transparent and can be followed by anyone – no black boxes • Rules and metrics can be nonlinear, nonmonotonic • Robust to missing information

More Related