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

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 Helisoma48 Thienemannimyia16 Physa4 Cricotopus2 Ablabesmyia1 Helobdella1 TIER 2 COMMUNITY II – Highly sensitive, specialist Taeniopteryx48 Epeorus13 Hexatoma8 Probezzia8 Isoperla7 Pteronarcys1 Capniidae1 Chloroperlidae1 Glossosoma1 Brachycentrus1 III - Sensitive - ubiquitous, generalist Ephemerella127 Acentrella13 Stenonema8 IV - Intermediate tolerance, opportunistic Polypedilum24 Cheumatopsyche5 V - Tolerant Taxa Hydropsyche8 Vs. Data Source: Maine DEP, Susan P. Davies

8. Tiers Exceptional Good Fair Poor Very poor

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

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

11. The set of long rivers Long rivers are > 2000 km 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 ??? What is the probability that the Columbia is long?

12. Membership functions (mathematical fuzzy set theory) Length, km Degree of membership 0 1 Not Long RiversLong Rivers 2000100040003000

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

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

16. Rule development: Tier 2 Example Highly 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

17. Highly Sensitive Taxa (Attribute II) Degree of membership 0 1 0.1 Att II / max of others 0.20.30.40.500.60.70.8 Tier 3 Tier 2 Tiers 4-6

18. 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%)

19. Tolerant Taxa (Attribute V) 70 Degree of membership 0 1 10 60 % abundance 203040500 Tier 3 Tier 2 Tier 4

20. 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

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

22. Model Performance at Matching Panel High gradient Low gradient Model result 1 Tier higher23 Model tied, higher Tier12 Match4634 Model tied, lower Tier52 Model result 1 Tier lower44 Total5845

23. 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

24. Advantages of quantitative fuzzy-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

25. 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

27. Rule development: Tier 2 Example Whole assemblage Overall taxa richness is as naturally occurs; density is not excessively low Linguistic Rule: Taxa richness is somewhat moderate to high and subsample count is near target Quantitative Rule 1: Total taxa > (10-14) genera Quantitative Rule 2: Total individuals > (40-50% of target)

28. Intermediate Sensitive Taxa (Attribute III) Taxa:Parametriocnemus, Baetis, Acroneuria, Neophylax Richness and abundance of intermediate sensitive taxa is high. Some may have increased due to slightly elevated production Linguistic rule: All sensitive taxa (highly sensitive + intermediate sensitive): comprise nearly half or more of all taxa Quant. rule: Taxa (II + III) > Taxa IV+V+VI (-3 to +2)

29. Intermediate tolerant taxa (Attribute IV; Indifferent taxa) Taxa: Nais, Gammarus, Optioservus, Polypedilum, Ceratopsyche May be slight increases in densities of some macroinvertebrate taxa RULE: None