Presentation at the US EPA Global Change Workshop

November 3, 2005 Arlington, VA

Linking Pollution to Water Linking Pollution to Water Body IntegrityBody Integrity

-- Second Year of ResearchSecond Year of Research

Vladimir NovotnyVladimir NovotnyElias ManolakosElias Manolakos

Northeastern UniversityNortheastern University

STAR WATERSHED PROJECT FUNDED BY STAR WATERSHED PROJECT FUNDED BY USEPA 2003USEPA 2003--20072007

Development of Risk Propagation Model for Development of Risk Propagation Model for Estimating Ecological Response of Streams to Estimating Ecological Response of Streams to

Anthropogenic Stresses and Stream ModificationAnthropogenic Stresses and Stream ModificationProject TeamProject Team•• PI PI -- Vladimir Novotny, NEU Center for Urban Vladimir Novotny, NEU Center for Urban

Environmental StudiesEnvironmental Studies•• CoCo--PIPI’’ss NEU CUERNEU CUER

Elias ManolakosElias ManolakosRamanitharan KandiahRamanitharan Kandiah

•• CoCo--PIPI Univ. of Wisconsin MilwaukeeUniv. of Wisconsin MilwaukeeTimothy Timothy EhlingerEhlinger

•• CoCo--PI PI Illinois State Water SurveyIllinois State Water SurveyAlenaAlena BartosovaBartosova

•• 5 graduate students5 graduate students

Project objectivesProject objectivesA model that will link stresses to biotic end points A model that will link stresses to biotic end points

Pollutant inputsPollutant inputsWatershed and water body modificationWatershed and water body modification

•• Land use changesLand use changes•• ChanelizationChanelization and impoundmentsand impoundments•• Riparian corridor modificationsRiparian corridor modifications•• Development of a quantitative layered risk Development of a quantitative layered risk

propagation from basic landscape and watershed propagation from basic landscape and watershed stressors to the biotic IBI endpointsstressors to the biotic IBI endpoints

Study the possibility of mitigating the stresses that Study the possibility of mitigating the stresses that would have the most beneficial impact on the biotic would have the most beneficial impact on the biotic endpoints endpoints

Apply the model to another geographic regionApply the model to another geographic region

NUMERIC INDICES OF BIOTIC NUMERIC INDICES OF BIOTIC INTEGRITY (ENDPOINTS)INTEGRITY (ENDPOINTS)

FishFishBenthic macroinvertebratesBenthic macroinvertebratesPhysical Physical -- HabitatHabitat

INPUTS INTO THE MODELSINPUTS INTO THE MODELS

About 50 habitat, chemical, land About 50 habitat, chemical, land use and riparian quality parametersuse and riparian quality parameters

(Ohio and Maryland(Ohio and Maryland)

Stream morphology(slope, width, depth,order)

Fish IBI and itsmetrics

Macroinvertebrate IBI and its metrics

LAYER 1

HABITAT WATERCHEMICALS

SEDIMENTCONTAMINATION

FRAGMENTATION

STRESSOR 1 STRESSOR 2 STRESSOR 3 STRESSOR 4 STRESSOR 5 STRESSOR 6

LAYER 2

LAYER 3

Landscape morpho-logical/ riparianfactors and stresses

ECOREGION

Land use changefactors and stresses,imperviousness

Pollutant loads fromland, point sourcesand atmosphere

Hydrologic/hydraulicstresses

LAYER 4

Periphyton and its metrics

BIOTICENDPOINTS

RISKS

IN-STREAM STRESSES

LANDSCAPE/ATMOSPHERIC STRESSES

LINKING IBIs AND STRESSORS -STRUCTURAL AND FUNCTIONAL MAZE

RESEARCH ACCOMPLISHMENTSRESEARCH ACCOMPLISHMENTSFirst and Second YearFirst and Second Year

First yearFirst year•• Teams were formedTeams were formed•• Data base shell development (USGS vs. our own)Data base shell development (USGS vs. our own)•• Technical review reportsTechnical review reports•• Several publicationsSeveral publications

Second YearSecond Year•• Data base assembled (Maryland, Ohio, Minnesota)Data base assembled (Maryland, Ohio, Minnesota)•• Five Technical Reports completedFive Technical Reports completed•• ANN Unsupervised Model combined with Canonical ANN Unsupervised Model combined with Canonical

Correspondence Analysis model completed and ready for Correspondence Analysis model completed and ready for ((ββ –– testing)testing)

•• Three publications submitted and four presentations were Three publications submitted and four presentations were made (IWA Kyoto, Great Lakes Research Association made (IWA Kyoto, Great Lakes Research Association --Ann Ann Arbor, AWRA Conference Seattle) Arbor, AWRA Conference Seattle)

DATA BASE SHELL DATA BASE SHELL STAR Environmental Data Base STAREDSTAR Environmental Data Base STAREDAlenaAlena BartosovaBartosova, Ramanitharan Kandiah and , Ramanitharan Kandiah and DweightDweight OsmondOsmond

Developed from Developed from FoxDBFoxDBSignificant modifications were made. The data Significant modifications were made. The data base includesbase includes•• environmental data, including water quality, environmental data, including water quality, •• sediment chemistry, sediment chemistry, •• biological indices, biological indices, •• stream hydrology, stream hydrology, •• and habitat and habitat

Master database is stored in Microsoft SQL Server Master database is stored in Microsoft SQL Server 2000 format on a server connected to a network 2000 format on a server connected to a network through the Northeastern University system through the Northeastern University system Data base currently includes complete data from Data base currently includes complete data from Ohio, Maryland and Ohio, Maryland and parrtialparrtial data from Minnesota, data from Minnesota, Wisconsin, Illinois and Massachusetts Wisconsin, Illinois and Massachusetts

STAR

Environmental

Database

Structure

Technical Report # 5

Taxa Related Station Related

Sample Related

Project Related

Parameter Related

Results Related

TBLSample

PK UnqID

FK2 IDLocSample_Code

FK4 Project_CodeFK6 Station_ID

Start_DateMissing_TimeEnd_Date

FK3 MediumFK5 Sample_TypeFK1 Composite_statistic_code

Sample_Depth_LengthReplicateComment

TBLNHD_Reach

PK NHD_100K_River_Reach

Waterbody_NameHydrologic_Unit_Code_8Length_metersElevation_UpstreamElevation_DownstreamSlope_ReachStreamOrder_Strahler

TBLProject_Grade

PK,FK1 Project_CodePK,FK2 QAPPCode

QAPP_GradeCU_Grade

TBLBiol_Taxa

PK Hybrid_ITIS_CodePK,FK3 UnqID

FK1 IDLocSample_CodeResult_Value

FK2 Remark_CodeGradeCommentsHybrid

TBLStation_Type

PK Station_Type

Primary_typeSecondary_typeNatural_indicator_type

TBLProjects_Programs

PK Project_Code

Program_ProjectFK1 Organization_ID

Project_Study_AreaProject_PurposeProject_Start_DateProject_End_DateContact_NameContact_Phone

TBLReporting_Units

PK Reporting_Units

Reporting_Units_Description

TBLWatershed_Info

PK Watershed_ID

FK1 Station_IDDrainage_area_sqmiContributing_area_sqmiSlope_avg

TBLOrganization

PK Organization_ID

Organization_ShortOrganization_NameOrganization_TypeContact_NameContact_PhoneAddress

FK1 ZipWeb_Site

TBLCFS_Daily

PK,FK1 Station_IDPK Date

DischargeSource_Number

TBLZip

PK Zip

CityState

TBLSample_Type

PK Sample_Type

Sample_Type_Description

TBLQAPP_Group_Codes

PK QAPPCode

FK1 Media_GroupFK2 Parameter_Group

TBLKingdom

PK Kingdom_ID

Kingdom

TBLStation_Information

PK,FK1 Station_ID

LatitudeLongitude

FK2 Lat_Long_Accuracy_CodeEPA_Station_CodeUSGS_Station_CodestationCode1stationcode2Place_Name_DescriptionTempRiver_Stream_LakeTotal_AreaHydrologic_Unit_Code

FK4 Station_TypeWater_Body_NameRF3_River_Reach

FK3 NHD_100K_River_ReachAmbnt

TBLComposite_Statistic

PK Composite_Statistic_Code

Composite_Statistic_NameComposite_Statistic_Description

TBLResults_Remarks

PK Remark_Code

Remark_Description

TBLQAPPGroups

PK,FK2 QAPPCodePK,FK1 Parameter_Code

TBLParameter_Codes

PK Parameter_Code

FK1 EPAGroup_CodeFK2 Reporting_Units

Decimal_PointShort_NameFull_Name

TblLandUsage

PK Station_ID

Land_UseLand_Use_Code

TBLResults_Val_NonNumeric

PK,FK2 Parameter_CodePK,FK4 UnqID

Sample_CodeResult_Value

FK3 Remark_CodeGrade

FK1 IDLoc

TBLRank_Information

PK Rank_IDPK,FK1 Kingdom_ID

Rank_DescriptionParent_Rank

TBLEPA_Group_Code

PK EPAGroup_Code

EPAGroup_Description

TBLLat_Long_Accuracy

PK Lat_Long_Accuracy_Code

Accuracy_Description

TBLFeeding_Group_Code

PK Feeding_Group_Code

Feeding_Group

TBLParameter_Group

PK Parameter_Group

First_Order_Parameter_GroupSecond_Order_Parameter_Group

TBLMedia_Group

PK Media_Group

Media_Group_Description

TblIDLocations

PK IDLoc

ID_Description

TBLMedium

PK Medium

Medium_Description

TBLIndices_Group_Code

PK Indices_Group_Code

Indices_Group_Description

TBLIndices_Group

PK ITIS_Code

Indices_Group_Code

TBLAssessment_Code

PK Assessment_Code

Assessment_Code_Description

TBLTolerance_Code

PK Tolerance_Code

Tolerance_Code_Description

TBLTolerance_Categories

PK ITIS_CodePK Tol_Region_Code

Tolerance_Code

TBLTol_Region_Code

PK Tol_Region_Code

Tol_Region_Description

TBLITIS_Code

PK ITIS_Code

NativeFK3 Rank_IDFK3 Kingdom_IDFK1 Feeding_Group_Code

Taxa_Name_1Taxa_Name_2Taxa_Name_3

FK2 Assessment_Group_CodeCommon_NameParent_ITIS_CodeNotes

TBLNative_Code

PK Native_Code

Native_Code_Description

TBLTolerance_Values

PK ITISCodePK Tol_Region_Code

Tolerance_Value

TBLResults

PK,FK3 Parameter_CodePK,FK2 UnqID

Sample_CodeResult_Value

FK4 Remark_CodeGrade

FK1 IDLocComments

TBLReplicates

PK Sample_CodePK,I1 Parameter_CodePK IDLoc

Result_ValueRemark_codeGrade

I1 ALTSample_codeI1 AltIDLOC

UnqIDAltUnqID

Parameter relatedProject related

Station relatedTaxa Related

Results relatedSample related

GIS GIS –– based Water Quality Model based Water Quality Model for Nitrogen by PCAfor Nitrogen by PCA

David Beach and V. NovotnyDavid Beach and V. Novotny

SPARROW and similar models SPARROW and similar models provide mean annual or seasonal provide mean annual or seasonal concentrationsconcentrationsThe model developed by the team The model developed by the team predicts both means and predicts both means and logarithmic standard deviation. logarithmic standard deviation. Extreme (e.g., 99 percentile) Extreme (e.g., 99 percentile) concentrations can be estimated concentrations can be estimated

The model was developed for the Fox River (Illinois) and the Miami River (Ohio)

Figure 2b: Great & Little Miami River Watershed PCA -Standard Deviation of TN

Calibration

0.00

1.00

2.00

3.00

4.00

0.00 1.00 2.00 3.00 4.00

Observed Stdev TN (mg/L)

Cal

cula

ted

Stde

v TN

(m

g/L)

R2 = 0.94

Figure 2a: Great & Little Miami River Watershed PCA -

Coefficient of Variation of TN Calibration

0.00

0.20

0.40

0.60

0.80

0.00 0.20 0.40 0.60 0.80

Observed CV TN (mg/L)

Cal

cula

ted

CV

TN

(mg/

L)

R2 = 0.94

UnivariateUnivariate Functions Linking Functions Linking Stressors to Impairment of Metrics Stressors to Impairment of Metrics

(Species Abundance)(Species Abundance)Jessica Brooks and V. NovotnyJessica Brooks and V. Novotny

0

10

20

30

40

50

60

70

0 5 10 15 20 25 30

Total Pb ug/l

Taxa

Cou

ntMaximum Specie Richness (MSR) functions were developed from field data (Ohio) and US EPA toxicity bioassay data.

MSR can be related to risk of species disappearance

Self Organizing Feature Maps Self Organizing Feature Maps combined withcombined with

Ecological Ordination Techniques Ecological Ordination Techniques for for

Effective Watershed ManagementEffective Watershed Management

Hardik Virani Elias S. ManolakosHardik Virani Elias S. Manolakos

Data base for Ohio provided by Ed Rankin Data base for Ohio provided by Ed Rankin Midwest Biodiversity Institute, Columbus, OH Midwest Biodiversity Institute, Columbus, OH

Modeling Approaches Modeling Approaches

Artificial Neural NetworksArtificial Neural Networks•• LearningLearning

Supervised: Output trackingSupervised: Output trackingUnsupervised: Pattern detectionUnsupervised: Pattern detection

•• Neural network applications to ecological modeling or, more Neural network applications to ecological modeling or, more general, to ecology are quite recent (early 90s)general, to ecology are quite recent (early 90s)

•• Initial applications revolved around back propagation Initial applications revolved around back propagation algorithm to predict the outcome from a set of environmental algorithm to predict the outcome from a set of environmental variables, with emphasis on comparison with regression variables, with emphasis on comparison with regression techniques.techniques.

•• References: predicting brown trout density (References: predicting brown trout density (LekLek et al., 1996), et al., 1996), predicting rainfall runoff (predicting rainfall runoff (AnctilAnctil and Tape, 2004), modeling and Tape, 2004), modeling trihalomethanetrihalomethane residuals in water treatment (residuals in water treatment (LewinLewin et al., et al., 2004)2004)

Self Organizing Feature Map (SOM)Self Organizing Feature Map (SOM)

Unsupervised artificial neural network (ANN) Unsupervised artificial neural network (ANN) based on competitive and cooperative learning.based on competitive and cooperative learning.Mapping highMapping high--dimensional data into a twodimensional data into a two--dimensional representation space.dimensional representation space.SOM is an approximation to the probability SOM is an approximation to the probability density function of the input datadensity function of the input dataUsed for data visualization, clustering (or Used for data visualization, clustering (or classification), estimation etc.classification), estimation etc.

Ohio EPAOhio EPA1848 sampling 1848 sampling sites between sites between 1995 and 2000.1995 and 2000.

Sampling period:Sampling period:•• Summer (JulySummer (July--

September)September)Fish, Physical Fish, Physical Habitat, and inHabitat, and in--situ Water situ Water Chemistry.Chemistry.

Metrics based on Metrics based on the type of sites.the type of sites.

Native species

Wading sitesHeadwater sites

Boat sites

Percentage Tolerant fish

Percentage of omnivores

Percent Round bodied Suckers

Number of Darters

Percent lithophilic spawners

Sunfish species

Headwaterspecies

Sucker species

Minnow species

Sensitive species

Intolerant species

Percentage Insectivores

Percentage Carnivores

Pioneering species

Number of individuals

Percent DELT

Number ofSimpleLithophils

SOM U Matrix SOM U Matrix -- OHOHUU--matrix gives the distances matrix gives the distances between neighboring map between neighboring map units. units. Element for each neuron is Element for each neuron is either the min, max, median either the min, max, median (default) or mean of the (default) or mean of the surrounding values. surrounding values. Dark color = large distance Dark color = large distance and thus a gap between the and thus a gap between the codebook values in the input codebook values in the input space. space. Good for visualization.Good for visualization.Light color = codebook Light color = codebook vectors are close to each vectors are close to each other in the input space. other in the input space. Light areas can be thought Light areas can be thought as clusters and dark areas as as clusters and dark areas as cluster separators.cluster separators.

Spatial representation of the Spatial representation of the clustered data in Ohioclustered data in Ohio3 clusters (good in 3 clusters (good in

Cluster 1, Cluster 1, intermediate in intermediate in Cluster 2, and Cluster 2, and marginal in Cluster marginal in Cluster 3) 3)

Clusters scattered Clusters scattered all over the Ohio all over the Ohio state.state.The northwestern The northwestern and Cleveland and Cleveland /Akron regions are /Akron regions are dominated bydominated byCluster 3.Cluster 3.

Fish IBI Fish IBI -- OHOHIBI is patterned on IBI is patterned on the SOMthe SOM

3 distinct clusters in 3 distinct clusters in the fish IBIthe fish IBIDistinct Habitat use Distinct Habitat use ranges by Ohio EPAranges by Ohio EPA•• EWH(> 48): EWH(> 48):

Cluster 1Cluster 1•• WWH (32WWH (32--44): 44):

Clusters 1 & 2Clusters 1 & 2•• MWH(22MWH(22--30): 30):

Cluster 3Cluster 3

Ohio ranking of wadeablewaters

Canonical Correspondence AnalysisCanonical Correspondence AnalysisFirst 2 axes explain First 2 axes explain 28% and 19%28% and 19%Direction of arrow Direction of arrow indicates interindicates inter--correlations correlations between variables.between variables.Neuron Projections Neuron Projections on arrow indicates on arrow indicates relative relative associations. associations. PerPer--cluster median cluster median projection to assign projection to assign cluster labels to cluster labels to environmental environmental variables.variables.Effective Effective segregation of the segregation of the environmental environmental gradients in Cluster gradients in Cluster 1 and 3.1 and 3.

Relative Rankings of environmental variablesRelative Rankings of environmental variablesLength of the Length of the environmental environmental variables in variables in CCA indicates CCA indicates their their importance in importance in explaining the explaining the variation in variation in species species distributiondistribution

Habitat Habitat parameters parameters dominate the dominate the top 10top 10..

Supervised learning by ANNSupervised learning by ANNMultiple stressors relationship to IBIMultiple stressors relationship to IBI

Jessica Brooks, V. Novotny, H. Virani, E. ManolakosJessica Brooks, V. Novotny, H. Virani, E. Manolakos

ANN were used in supervised learning.ANN were used in supervised learning.Data from Ohio data base were usedData from Ohio data base were usedFirst attempts using intuitively selected First attempts using intuitively selected chemical parameters, habitat index, and chemical parameters, habitat index, and macroinvertebrates index as inputs and macroinvertebrates index as inputs and IBIs as output were not successfulIBIs as output were not successfulSOM and Canonical Correspondence SOM and Canonical Correspondence Analysis identified the most important Analysis identified the most important variables. Using top 12 stressor variables variables. Using top 12 stressor variables the ANN performance improved the ANN performance improved dramaticallydramatically

Training Results (Calibration Training Results (Calibration 75% of data)75% of data)

TEMPERATURE DISSOLVED OXYGEN

PH TOTAL SUSPENDED SOLIDS TOTAL AMMONIA BOD, 5-DAY HARDNESS CHLORIDE SULFATE TOTAL CADMIUM TOTAL LEAD TOTAL ZINC INVERTEBRATE COMMUNITY INDEX QUALITATIVE HABITAT EVALUATION INDEX

RIFFLE POOL EMBEDEDDNESS CHANNEL SUBSTRATE RIPARIAN PERCENT FORRESTED HARDNESS TOTAL IRON BOD, 5-DAY, DISSOLVED OXYGEN INVERTEBRATE COMMUNITY INDEX

INTUITION INPUTS TOP 12 INPUTS AFTER SOM AND CCA

DiscussionDiscussionThe results indicate the efficiency of the SOM in visualizing thThe results indicate the efficiency of the SOM in visualizing the e state of streams in Maryland and Ohio and aid the watershed state of streams in Maryland and Ohio and aid the watershed manager in making and implementing decisions which will manager in making and implementing decisions which will ultimately provide knowledge necessary for restoration of the ultimately provide knowledge necessary for restoration of the degraded watersheds. degraded watersheds. Habitat parameters surpassed chemical parameters as important Habitat parameters surpassed chemical parameters as important variables related to species compositions, which in turn decidesvariables related to species compositions, which in turn decidesthe fish IBI. the fish IBI. Embeddedness was found to be an important variable in both the Embeddedness was found to be an important variable in both the datasets; hence work on improving the watershed should revolve datasets; hence work on improving the watershed should revolve around analyzing the effects of embeddedness on the species around analyzing the effects of embeddedness on the species composition. composition. River mouths were the most degraded regions in both the states, River mouths were the most degraded regions in both the states, which supports the general hypothesis about regions with poor which supports the general hypothesis about regions with poor fish IBI. In the case of Maryland, we found that the Coastal arefish IBI. In the case of Maryland, we found that the Coastal areas as around the Chesapeake Bay are the most degraded regions. The around the Chesapeake Bay are the most degraded regions. The Lake Erie region and the northwestern region in Ohio were found Lake Erie region and the northwestern region in Ohio were found to have poor fish IBI compared to the rest of the state. to have poor fish IBI compared to the rest of the state.

SummarySummarySOM analysis provides an ordered set of sites across the state SOM analysis provides an ordered set of sites across the state

with similar metric characteristics, ultimately reflecting in thwith similar metric characteristics, ultimately reflecting in the fish e fish IBI. IBI. Clustering analysis allowed us to divide the examination of the Clustering analysis allowed us to divide the examination of the entire state into 3 clusters, which helps us to compare the clusentire state into 3 clusters, which helps us to compare the clusters ters and define their key ecological characteristics.and define their key ecological characteristics.CCA help in drawing conclusions about specific fish species and CCA help in drawing conclusions about specific fish species and the the role of the environmental variables in maintaining the perfect role of the environmental variables in maintaining the perfect habitat and water quality for fishes. habitat and water quality for fishes. The results provide a comprehensive stateThe results provide a comprehensive state--based study for future based study for future monitoring that can address shortmonitoring that can address short--term and longterm and long--term trends. term trends. Different visualization approaches, based on a detailed analysisDifferent visualization approaches, based on a detailed analysisgive the watershed manager ample scope for working out new give the watershed manager ample scope for working out new strategies, focused on improving the overall heath of the streamstrategies, focused on improving the overall heath of the streams s today. today.

Future Work (This and Future Future Work (This and Future Projects)Projects)

Prediction networks (supervised ANNs) to predict the fish Prediction networks (supervised ANNs) to predict the fish (an macroinvertebrate) IBI.(an macroinvertebrate) IBI.Development of hydrologically and morphologically based Development of hydrologically and morphologically based multidimensional habitat indexmultidimensional habitat indexFinalize effects of impoundments and other fragmentationFinalize effects of impoundments and other fragmentationPossible to integrate ANNs with other modeling approaches Possible to integrate ANNs with other modeling approaches (genetic algorithms), or tools (GIS), or mechanistic models (genetic algorithms), or tools (GIS), or mechanistic models (Hybrid models) (Hybrid models) Specialized networks dealing with individual clusters to Specialized networks dealing with individual clusters to obtain relationships associated with similar metric traits.obtain relationships associated with similar metric traits.Apply the models to Northeast (the Charles River) Apply the models to Northeast (the Charles River) Integrated tool with guidance manual to cater to a wide Integrated tool with guidance manual to cater to a wide section of the audience section of the audience snfsnf s workshop for users. s workshop for users. Consider sediment contamination as a stressor.Consider sediment contamination as a stressor.Work still to be done to realize the final hierarchical risk Work still to be done to realize the final hierarchical risk based modelbased model

Thank you!!Thank you!!