SlideShare a Scribd company logo
1 of 30
Development of Inferential Sensors for
Real-time Quality Control of Water-
Level Data for the Everglades Depth
Estimation Network
Ruby Daamen, Advanced Data Mining Int’l
Edwin Roehl, Advanced Data Mining Int’l
Paul Conrads, USGS – SC Water Science Center
Matthew Petkewich, USGS – SC Water Science Center
Presentation Outline
• What is an “Inferential Sensor” (IS)?
• Background - Industrial application
• Everglades Depth Estimation Network
(EDEN)
• Automated Data Assurance and
Management (ADAM) - Inferential Sensor
development for EDEN
Development of IS in Industry
• A tough environment to monitor
– Emissions regulations require measurements of
effluent gases
– Smoke stack burns up probes
– Need alternative to “hard” sensors
Hard Sensor vs. Inferential Sensor
• Virtual sensor replaces actual sensor
– Temporary gage smoke stack
– Operate plant to cover range of emissions
– Develop model of emissions based on
operations
– Model becomes the “Inferential Sensor”
Inferential Sensors for Real-Time Data
• Problem – Need to minimize missing and
erroneous data
• Use similar approach taken by industry to predict
real-time data – ie. “Inferential Sensors”
– QA/QC hard sensor
– Provide accurate estimates for hard sensor
– Provide redundant signal
EDEN Problem
• EDEN – integrated network of real-time water-
level (WL) stations, ground-elevation models, and
water-surface models
– Real-time data is used to generate EDEN WL surfaces
used by scientists, engineers and water-resource
managers (253+ stations)
– Data used to guide large-scale field operations,
integrate hydrologic and ecological responses, support
assessments that measure ecosystem responses to the
Comprehensive Everglades Restoration Plan (CERP)
– Correcting errors is often time consuming; gaging
stations may be in remote areas
– Need to identify errors and to provide estimates on a
daily basis
EDEN Water Surface Map
Bad values
creates
erroneous
areas on
maps
EDEN / ADAM Overview
Inferential Sensor
application
Some of the Challenges
• 253+ gaging stations
• Inferential sensor uses data from one or
more correlated gages. At any given
datetime do not know what stations will have
available data
• Stations added/removed over time
Implementation
• Create models “on the fly”
– Use “best” available stations
– Simplifies addition / removal of stations
• Needs to consider
– Assessment of data – prevent use of suspect
data in models
– Model inputs need be decorrelated
Methods
• Two algorithms sequenced to analyze data
– Univariate filtering
• Provides information about the quality and behavior
for each stations WL
• A Statistical Process Control (SPC) which consists of
14 univariate filters – uniquely set for each station
– Estimate parameter value
• Select a “pool” of candidate gaging stations using
matrix of Pearson coefficients
• Principal component analysis (PCA) – calculates
decorrelated inputs
• Multivariate linear regression
Univariate Filtering
UNIVARIATE
FILTER CHECK DESCRIPTION PRECEDENCE
WATER LEVEL
LIMIT (ft.)
LOST_SIGNAL no signal 1 NA
GT_RNG_UL x(t) > signal range Upper Range Limit 2 15.19
LT_RNG_LL x(t) < signal range Upper Range Limit 3 6.99
GT_UCL x(t) > signal Upper Control Limit 4 14.73
LT_LCL x(t) < signal Upper Control Limit 5 8.56
Sn_LT_L flatlined: x'(t) = x(t)=x(t-1); SUM[(|x'(t)|,…,|x'(t-n+1)|] < Limit 6 0.00
D1_GT_L_1 vfast vlarge increase: x(t)-x(t-1) > Limit 7 1.92
D1_LT_L_1 vfast vlarge decrease: x(t)-x(t-1) < Limit 8 -2.34
D1Sn_GT_L_1 fast vlarge increase: x'(t)=x(t)-x(t-1); Sum[x'(t),…x'(t-n+1)] > Limit 9 1.98
D1Sn_LT_L_1 fast vlarge decrease: x'(t)=x(t)-x(t-1); Sum[x'(t),…x'(t-n+1)] < Limit 10 -2.52
D1_GT_L_2 vfast large increase: x(t) - x(t-1) > Limit 11 1.69
D1_LT_L_2 vfast large decrease: x(t) - x(t-1)< Limit 12 -0.25
D1Sn_GT_L_2 fast large increase: x'(t)=x(t)-x(t-1); Sum[x'(t),…x'(t-n+1)] > Limit 13 1.98
D1Sn_LT_L_2 fast large decrease: x'(t)=x(t)-x(t-1); Sum[x'(t),…x'(t-n+1)] < Limit 14 -0.27
• Additional filters
– Dry Protocol – set using offset from ground elevation
• Any filter trips are flagged for review
• Data triggering a filter is not used in any
predictions
Synthesize WL Measurements
• WL at candidate stations are correlated – no
surprise
• First approach examined selected the most
highly correlated station as a “standard”
signal and then attempted decorrelating the
other stations by computing differences from
the standard.
Principal Component Analysis (PCA)
• PCA is a statistical technique used to
“reduce the dimensionality of a data set
consisting of a large number of interrelated
variables, while retaining as much as
possible of the variation present in the data
set. This is achieved by transforming to a
new set of variables, the principal
components (PCs), which are uncorrelated,
and which are ordered so that the first few
retain most of the variation present in all of
the original variables” (Joliffe)
PCA – Main Points
• PCA - the main points
– Principal components are uncorrelated
– Transforms a set of correlated variables into a
smaller number of uncorrelated variables
– The first principal component (PC) accounts for
most of the variability in the data
PCA - Analysis
• PCA – a brief description of the analysis:
– Calculate the eigenvectors and eigenvalues of
the covariance matrix
• Create data set of n inputs with no gaps
• Subtract the mean from each n input
• Calculate the covariance matrix (square nXn matrix)
• Calculate eigenvectors
– Sort by eigenvalues (highest to lowest)
• Largest eigenvector = 1st principal component
• Use eigenvalues to determine how many PCs to
include
PCA – A 2-Dimension Example
Original Data
(Mean Subtracted)
Eigenvectors Principal
Components
Eigenvectors
Plotted on
Data
-4
-3
-2
-1
0
1
2
3
4
-4 -3 -2 -1 0 1 2 3 4
Normalized Data
E1
E2
Original Data
ADAM - Functionality
• Setup
– File paths
– PCA setup – period, number of sites to include
– Add / Edit / Remove sites
– Univariate filters
• Inferential Sensor – Option to analyze daily
(hourly and 15 minute data), quarterly and
annual (hourly) daily files
• Review results
• Output daily median files as required for
EDEN water surface map
ADAM – Control Worksheet
Select Daily, Quarterly or
Annual Run Analysis
Resume, redo OR continue from
last analyzed
Fill Setup
Remove , add, or edit sites
included in ADAM
Set Pathnames for files used
by ADAM
ADAM – Control Worksheet
Loads data from
selected run for review
Creates output files
required to generate
EDEN water surface
map
Dumps a listing sites
tripping any filters
ADAM Review Worksheet
First PCA
analysis
GAPFILL
ADAM Review Worksheet
Another Example
GAPFILL
PCA
First PCA
analysis
ADAM – Univariate Filter Setups
Graphs show
settings vs. data
What’s Next
EVEEvolution and Verification of EDEN Inferential Sensor
Questions?
Archive
PCA – The methodology
• From our favorite source -Wikipedia – PCA is:
– A mathematical procedure that transforms a set of correlated variables into a smaller number of
uncorrelated variables. The first principal component accounts for as much of the variability in
the data as possible and each succeeding component accounts for as much of the remaining
variability as possible.
• How to do it: In the broadest sense it is an eigenvector based multivariate analysis. The
methodology used is:
– Assemble the data
• In EDENIS the data will be water level data from up to 5 gaging stations. For 90 days of hourly data this
equates to up to 2160 vectors (8640 for 15 minute data)
– Remove any vectors that contain any missing data (1)
• Resulting matrix X[n,m] where n = number of fully populated vectors; m = number of gages included
– Subtract the mean from each of the data dimensions (m) – lets call this the normalized matrix B
(2)
• B[n,m] stores mean-subtracted data
– Calculate the Covariance matrix (3)
• Covariance matrix is a square matrix with dimension mXm: C[m,m]
– Calculate the eigenvalues and eigenvectors of the covariance matrix
• This is an iterative process. If you want to look up some more on this I used the Jacobi eigenvalue
algorithm. Some important properties of eigenvectors:
– Can only be found for square matrices. If a square matrix (mXm) does have eigenvectors, there are m of them
– All the eigenvectors of a matrix are orthogonal to each other. This is important: when the data is expressed in terms of
these eigenvectors the resulting principal components are uncorrelated
• For a refresher course on eigenvalues / eigenvectors heres a link
http://en.wikipedia.org/wiki/Eigenvalue,_eigenvector_and_eigenspace
– Sort the results: Largest eigenvalue to smallest eigenvalue. (4)
– Decide how many components to keep and calculate the new data set to be used. This is a
simple matrix multiplication of B X E where E is the mXm eigenvector matrix. (5)
• For the 10-12 sites I looked at when presenting the PCA results I would expect we’ll rarely if ever use more
than 2 principal components out of a possible 5 to make the regression predictions.
PCA – A 2-Dimension Example
1. Original Data 2. Mean
Subtracted
3. Covariance
Matrix
4. Eigenvalues /
Eigenvectors
5. New Data Set
Note high
correlation of
orignal X1, X2 vs.
No correlation of
PC1, PC2
E1 and E2 are the 2
eigenvectors layed on
top of the data. Note E1
and E2 are
perpendicular. Also note
that E1 goes through the
middle of the data – like
a best fit. E2 provides
less information about
the variance in the data.
-4
-3
-2
-1
0
1
2
3
4
-4 -3 -2 -1 0 1 2 3 4
Normalized Data
E1
E2
Challenges
• Develop 253+ models using artificial neural
networks (ANNs) (1 model per station)
– Pros
• authors have prior success modeling complex
processes using ANNs
• ANNs use non-linear curve fitting to capture complex
behaviors
– Cons
• Do not know what stations will have “good” data at
any given datetime
• Stations are removed and added
• Possibly insert Yearly run – min/max r2
using pca,gapfill

More Related Content

What's hot

Low power tool paper
Low power tool paperLow power tool paper
Low power tool paperM Madan Gopal
 
Kalman Filter and its Application
Kalman Filter and its ApplicationKalman Filter and its Application
Kalman Filter and its ApplicationSaptarshi Mazumdar
 
ThesisDefensePresentation_KyleIngersoll
ThesisDefensePresentation_KyleIngersollThesisDefensePresentation_KyleIngersoll
ThesisDefensePresentation_KyleIngersollKyle Ingersoll
 
Report kalman filtering
Report kalman filteringReport kalman filtering
Report kalman filteringIrfan Anjum
 
Size measurement and estimation
Size measurement and estimationSize measurement and estimation
Size measurement and estimationLouis A. Poulin
 
presentation 2019 04_09_rev1
presentation 2019 04_09_rev1presentation 2019 04_09_rev1
presentation 2019 04_09_rev1Hyun Wong Choi
 
Kalman Filter Based GPS Receiver
Kalman Filter Based GPS ReceiverKalman Filter Based GPS Receiver
Kalman Filter Based GPS ReceiverFalak Shah
 
Keep Calm and React with Foresight: Strategies for Low-Latency and Energy-Eff...
Keep Calm and React with Foresight: Strategies for Low-Latency and Energy-Eff...Keep Calm and React with Foresight: Strategies for Low-Latency and Energy-Eff...
Keep Calm and React with Foresight: Strategies for Low-Latency and Energy-Eff...Tiziano De Matteis
 
Distributed approximate spectral clustering for large scale datasets
Distributed approximate spectral clustering for large scale datasetsDistributed approximate spectral clustering for large scale datasets
Distributed approximate spectral clustering for large scale datasetsBita Kazemi
 
Pdcs2010 balman-presentation
Pdcs2010 balman-presentationPdcs2010 balman-presentation
Pdcs2010 balman-presentationbalmanme
 

What's hot (19)

Kalman Equations
Kalman EquationsKalman Equations
Kalman Equations
 
Low power tool paper
Low power tool paperLow power tool paper
Low power tool paper
 
Kalman Filter and its Application
Kalman Filter and its ApplicationKalman Filter and its Application
Kalman Filter and its Application
 
ThesisDefensePresentation_KyleIngersoll
ThesisDefensePresentation_KyleIngersollThesisDefensePresentation_KyleIngersoll
ThesisDefensePresentation_KyleIngersoll
 
Report kalman filtering
Report kalman filteringReport kalman filtering
Report kalman filtering
 
Size measurement and estimation
Size measurement and estimationSize measurement and estimation
Size measurement and estimation
 
11 (l)random vibrations
11 (l)random vibrations11 (l)random vibrations
11 (l)random vibrations
 
Nitish 007
Nitish 007Nitish 007
Nitish 007
 
Forecasting methods
Forecasting methods Forecasting methods
Forecasting methods
 
presentation 2019 04_09_rev1
presentation 2019 04_09_rev1presentation 2019 04_09_rev1
presentation 2019 04_09_rev1
 
11 generalsisation of fluent
11 generalsisation of fluent11 generalsisation of fluent
11 generalsisation of fluent
 
Kalman filter
Kalman filterKalman filter
Kalman filter
 
Derrick alan
Derrick alanDerrick alan
Derrick alan
 
11 spectrum analysis
11 spectrum analysis11 spectrum analysis
11 spectrum analysis
 
Kalman Filter Based GPS Receiver
Kalman Filter Based GPS ReceiverKalman Filter Based GPS Receiver
Kalman Filter Based GPS Receiver
 
Kalmanfilter
KalmanfilterKalmanfilter
Kalmanfilter
 
Keep Calm and React with Foresight: Strategies for Low-Latency and Energy-Eff...
Keep Calm and React with Foresight: Strategies for Low-Latency and Energy-Eff...Keep Calm and React with Foresight: Strategies for Low-Latency and Energy-Eff...
Keep Calm and React with Foresight: Strategies for Low-Latency and Energy-Eff...
 
Distributed approximate spectral clustering for large scale datasets
Distributed approximate spectral clustering for large scale datasetsDistributed approximate spectral clustering for large scale datasets
Distributed approximate spectral clustering for large scale datasets
 
Pdcs2010 balman-presentation
Pdcs2010 balman-presentationPdcs2010 balman-presentation
Pdcs2010 balman-presentation
 

Similar to Daamen r 2010scwr-cpaper

5.1 mining data streams
5.1 mining data streams5.1 mining data streams
5.1 mining data streamsKrish_ver2
 
Test different neural networks models for forecasting of wind,solar and energ...
Test different neural networks models for forecasting of wind,solar and energ...Test different neural networks models for forecasting of wind,solar and energ...
Test different neural networks models for forecasting of wind,solar and energ...Tonmoy Ibne Arif
 
The following ppt is about principal component analysis
The following ppt is about principal component analysisThe following ppt is about principal component analysis
The following ppt is about principal component analysisSushmit8
 
Heuristic design of experiments w meta gradient search
Heuristic design of experiments w meta gradient searchHeuristic design of experiments w meta gradient search
Heuristic design of experiments w meta gradient searchGreg Makowski
 
pcappt-140121072949-phpapp01.pptx
pcappt-140121072949-phpapp01.pptxpcappt-140121072949-phpapp01.pptx
pcappt-140121072949-phpapp01.pptxABINASHPADHY6
 
Noha danms13 talk_final
Noha danms13 talk_finalNoha danms13 talk_final
Noha danms13 talk_finalNoha Elprince
 
A Hybrid Method of CART and Artificial Neural Network for Short Term Load For...
A Hybrid Method of CART and Artificial Neural Network for Short Term Load For...A Hybrid Method of CART and Artificial Neural Network for Short Term Load For...
A Hybrid Method of CART and Artificial Neural Network for Short Term Load For...Salford Systems
 
A Deep Learning use case for water end use detection by Roberto Díaz and José...
A Deep Learning use case for water end use detection by Roberto Díaz and José...A Deep Learning use case for water end use detection by Roberto Díaz and José...
A Deep Learning use case for water end use detection by Roberto Díaz and José...Big Data Spain
 
Dependable Systems - Structure-Based Dependabiilty Modeling (6/16)
Dependable Systems - Structure-Based Dependabiilty Modeling (6/16)Dependable Systems - Structure-Based Dependabiilty Modeling (6/16)
Dependable Systems - Structure-Based Dependabiilty Modeling (6/16)Peter Tröger
 
An Introduction to Deep Learning
An Introduction to Deep LearningAn Introduction to Deep Learning
An Introduction to Deep Learningmilad abbasi
 
Introduction to Deep Learning
Introduction to Deep LearningIntroduction to Deep Learning
Introduction to Deep LearningMehrnaz Faraz
 
Matrix Factorization In Recommender Systems
Matrix Factorization In Recommender SystemsMatrix Factorization In Recommender Systems
Matrix Factorization In Recommender SystemsYONG ZHENG
 
"Quantum clustering - physics inspired clustering algorithm", Sigalit Bechler...
"Quantum clustering - physics inspired clustering algorithm", Sigalit Bechler..."Quantum clustering - physics inspired clustering algorithm", Sigalit Bechler...
"Quantum clustering - physics inspired clustering algorithm", Sigalit Bechler...Dataconomy Media
 
My Postdoctoral Research
My Postdoctoral ResearchMy Postdoctoral Research
My Postdoctoral ResearchPo-Ting Wu
 

Similar to Daamen r 2010scwr-cpaper (20)

GIS Data Quality
GIS Data QualityGIS Data Quality
GIS Data Quality
 
Modeling full scale-data(2)
Modeling full scale-data(2)Modeling full scale-data(2)
Modeling full scale-data(2)
 
5.1 mining data streams
5.1 mining data streams5.1 mining data streams
5.1 mining data streams
 
Test different neural networks models for forecasting of wind,solar and energ...
Test different neural networks models for forecasting of wind,solar and energ...Test different neural networks models for forecasting of wind,solar and energ...
Test different neural networks models for forecasting of wind,solar and energ...
 
Pca ppt
Pca pptPca ppt
Pca ppt
 
pca.ppt
pca.pptpca.ppt
pca.ppt
 
The following ppt is about principal component analysis
The following ppt is about principal component analysisThe following ppt is about principal component analysis
The following ppt is about principal component analysis
 
Spc
SpcSpc
Spc
 
CLIM Program: Remote Sensing Workshop, An Introduction to Systems and Softwar...
CLIM Program: Remote Sensing Workshop, An Introduction to Systems and Softwar...CLIM Program: Remote Sensing Workshop, An Introduction to Systems and Softwar...
CLIM Program: Remote Sensing Workshop, An Introduction to Systems and Softwar...
 
Heuristic design of experiments w meta gradient search
Heuristic design of experiments w meta gradient searchHeuristic design of experiments w meta gradient search
Heuristic design of experiments w meta gradient search
 
pcappt-140121072949-phpapp01.pptx
pcappt-140121072949-phpapp01.pptxpcappt-140121072949-phpapp01.pptx
pcappt-140121072949-phpapp01.pptx
 
Noha danms13 talk_final
Noha danms13 talk_finalNoha danms13 talk_final
Noha danms13 talk_final
 
A Hybrid Method of CART and Artificial Neural Network for Short Term Load For...
A Hybrid Method of CART and Artificial Neural Network for Short Term Load For...A Hybrid Method of CART and Artificial Neural Network for Short Term Load For...
A Hybrid Method of CART and Artificial Neural Network for Short Term Load For...
 
A Deep Learning use case for water end use detection by Roberto Díaz and José...
A Deep Learning use case for water end use detection by Roberto Díaz and José...A Deep Learning use case for water end use detection by Roberto Díaz and José...
A Deep Learning use case for water end use detection by Roberto Díaz and José...
 
Dependable Systems - Structure-Based Dependabiilty Modeling (6/16)
Dependable Systems - Structure-Based Dependabiilty Modeling (6/16)Dependable Systems - Structure-Based Dependabiilty Modeling (6/16)
Dependable Systems - Structure-Based Dependabiilty Modeling (6/16)
 
An Introduction to Deep Learning
An Introduction to Deep LearningAn Introduction to Deep Learning
An Introduction to Deep Learning
 
Introduction to Deep Learning
Introduction to Deep LearningIntroduction to Deep Learning
Introduction to Deep Learning
 
Matrix Factorization In Recommender Systems
Matrix Factorization In Recommender SystemsMatrix Factorization In Recommender Systems
Matrix Factorization In Recommender Systems
 
"Quantum clustering - physics inspired clustering algorithm", Sigalit Bechler...
"Quantum clustering - physics inspired clustering algorithm", Sigalit Bechler..."Quantum clustering - physics inspired clustering algorithm", Sigalit Bechler...
"Quantum clustering - physics inspired clustering algorithm", Sigalit Bechler...
 
My Postdoctoral Research
My Postdoctoral ResearchMy Postdoctoral Research
My Postdoctoral Research
 

More from John B. Cook, PE, CEO

Orange Co. Water District's Solution to Water Crisis
Orange Co. Water District's Solution to Water CrisisOrange Co. Water District's Solution to Water Crisis
Orange Co. Water District's Solution to Water CrisisJohn B. Cook, PE, CEO
 
Scwrc2014 savannah basinresourceoptimization-20141021
Scwrc2014 savannah basinresourceoptimization-20141021Scwrc2014 savannah basinresourceoptimization-20141021
Scwrc2014 savannah basinresourceoptimization-20141021John B. Cook, PE, CEO
 
Wqtc2013 dist syswq-modeling-20131107
Wqtc2013 dist syswq-modeling-20131107Wqtc2013 dist syswq-modeling-20131107
Wqtc2013 dist syswq-modeling-20131107John B. Cook, PE, CEO
 
Wqtc2013 invest ofperformanceprobswitheds-20130910
Wqtc2013 invest ofperformanceprobswitheds-20130910Wqtc2013 invest ofperformanceprobswitheds-20130910
Wqtc2013 invest ofperformanceprobswitheds-20130910John B. Cook, PE, CEO
 
Wrf4285 climate changepresentation-20121008
Wrf4285 climate changepresentation-20121008Wrf4285 climate changepresentation-20121008
Wrf4285 climate changepresentation-20121008John B. Cook, PE, CEO
 
Wqtc2011 causes offalsealarms-20111115-final
Wqtc2011 causes offalsealarms-20111115-finalWqtc2011 causes offalsealarms-20111115-final
Wqtc2011 causes offalsealarms-20111115-finalJohn B. Cook, PE, CEO
 

More from John B. Cook, PE, CEO (18)

Orange Co. Water District's Solution to Water Crisis
Orange Co. Water District's Solution to Water CrisisOrange Co. Water District's Solution to Water Crisis
Orange Co. Water District's Solution to Water Crisis
 
Scwrc2014 savannah basinresourceoptimization-20141021
Scwrc2014 savannah basinresourceoptimization-20141021Scwrc2014 savannah basinresourceoptimization-20141021
Scwrc2014 savannah basinresourceoptimization-20141021
 
Asset management-cda
Asset management-cdaAsset management-cda
Asset management-cda
 
Adm graphics-2003
Adm graphics-2003Adm graphics-2003
Adm graphics-2003
 
Neiwpcc nps 2010
Neiwpcc nps 2010Neiwpcc nps 2010
Neiwpcc nps 2010
 
Hic06 spatial interpolation
Hic06 spatial interpolationHic06 spatial interpolation
Hic06 spatial interpolation
 
Integrated river basin management
Integrated river basin managementIntegrated river basin management
Integrated river basin management
 
Wqtc2013 dist syswq-modeling-20131107
Wqtc2013 dist syswq-modeling-20131107Wqtc2013 dist syswq-modeling-20131107
Wqtc2013 dist syswq-modeling-20131107
 
Caw toronto presentation-20121031
Caw toronto presentation-20121031Caw toronto presentation-20121031
Caw toronto presentation-20121031
 
Wqtc2013 invest ofperformanceprobswitheds-20130910
Wqtc2013 invest ofperformanceprobswitheds-20130910Wqtc2013 invest ofperformanceprobswitheds-20130910
Wqtc2013 invest ofperformanceprobswitheds-20130910
 
Ewri2009 big data_jbc
Ewri2009 big data_jbcEwri2009 big data_jbc
Ewri2009 big data_jbc
 
Ad mi floridan-aquiferwls-for-pps
Ad mi floridan-aquiferwls-for-ppsAd mi floridan-aquiferwls-for-pps
Ad mi floridan-aquiferwls-for-pps
 
Wrf4285 climate changepresentation-20121008
Wrf4285 climate changepresentation-20121008Wrf4285 climate changepresentation-20121008
Wrf4285 climate changepresentation-20121008
 
Wqtc2011 causes offalsealarms-20111115-final
Wqtc2011 causes offalsealarms-20111115-finalWqtc2011 causes offalsealarms-20111115-final
Wqtc2011 causes offalsealarms-20111115-final
 
Neiwpcc2010.ppt
Neiwpcc2010.pptNeiwpcc2010.ppt
Neiwpcc2010.ppt
 
Hic06 spatial interpolation
Hic06 spatial interpolationHic06 spatial interpolation
Hic06 spatial interpolation
 
Ad mi floridan-aquiferwls-for-pps
Ad mi floridan-aquiferwls-for-ppsAd mi floridan-aquiferwls-for-pps
Ad mi floridan-aquiferwls-for-pps
 
Modeling full scale-data(2)
Modeling full scale-data(2)Modeling full scale-data(2)
Modeling full scale-data(2)
 

Recently uploaded

(PARI) Viman Nagar Call Girls Just Call 7001035870 [ Cash on Delivery ] Pune ...
(PARI) Viman Nagar Call Girls Just Call 7001035870 [ Cash on Delivery ] Pune ...(PARI) Viman Nagar Call Girls Just Call 7001035870 [ Cash on Delivery ] Pune ...
(PARI) Viman Nagar Call Girls Just Call 7001035870 [ Cash on Delivery ] Pune ...ranjana rawat
 
Hot Call Girls |Delhi |Preet Vihar ☎ 9711199171 Book Your One night Stand
Hot Call Girls |Delhi |Preet Vihar ☎ 9711199171 Book Your One night StandHot Call Girls |Delhi |Preet Vihar ☎ 9711199171 Book Your One night Stand
Hot Call Girls |Delhi |Preet Vihar ☎ 9711199171 Book Your One night Standkumarajju5765
 
VVIP Pune Call Girls Koregaon Park (7001035870) Pune Escorts Nearby with Comp...
VVIP Pune Call Girls Koregaon Park (7001035870) Pune Escorts Nearby with Comp...VVIP Pune Call Girls Koregaon Park (7001035870) Pune Escorts Nearby with Comp...
VVIP Pune Call Girls Koregaon Park (7001035870) Pune Escorts Nearby with Comp...Call Girls in Nagpur High Profile
 
Call Girls In Okhla DELHI ~9654467111~ Short 1500 Night 6000
Call Girls In Okhla DELHI ~9654467111~ Short 1500 Night 6000Call Girls In Okhla DELHI ~9654467111~ Short 1500 Night 6000
Call Girls In Okhla DELHI ~9654467111~ Short 1500 Night 6000Sapana Sha
 
VIP Call Girls Service Chaitanyapuri Hyderabad Call +91-8250192130
VIP Call Girls Service Chaitanyapuri Hyderabad Call +91-8250192130VIP Call Girls Service Chaitanyapuri Hyderabad Call +91-8250192130
VIP Call Girls Service Chaitanyapuri Hyderabad Call +91-8250192130Suhani Kapoor
 
Low Rate Call Girls Bikaner Anika 8250192130 Independent Escort Service Bikaner
Low Rate Call Girls Bikaner Anika 8250192130 Independent Escort Service BikanerLow Rate Call Girls Bikaner Anika 8250192130 Independent Escort Service Bikaner
Low Rate Call Girls Bikaner Anika 8250192130 Independent Escort Service BikanerSuhani Kapoor
 
Call Girls South Delhi Delhi reach out to us at ☎ 9711199012
Call Girls South Delhi Delhi reach out to us at ☎ 9711199012Call Girls South Delhi Delhi reach out to us at ☎ 9711199012
Call Girls South Delhi Delhi reach out to us at ☎ 9711199012sapnasaifi408
 
VIP Call Girls Moti Ganpur ( Hyderabad ) Phone 8250192130 | ₹5k To 25k With R...
VIP Call Girls Moti Ganpur ( Hyderabad ) Phone 8250192130 | ₹5k To 25k With R...VIP Call Girls Moti Ganpur ( Hyderabad ) Phone 8250192130 | ₹5k To 25k With R...
VIP Call Girls Moti Ganpur ( Hyderabad ) Phone 8250192130 | ₹5k To 25k With R...Suhani Kapoor
 
VIP Call Girls Mahadevpur Colony ( Hyderabad ) Phone 8250192130 | ₹5k To 25k ...
VIP Call Girls Mahadevpur Colony ( Hyderabad ) Phone 8250192130 | ₹5k To 25k ...VIP Call Girls Mahadevpur Colony ( Hyderabad ) Phone 8250192130 | ₹5k To 25k ...
VIP Call Girls Mahadevpur Colony ( Hyderabad ) Phone 8250192130 | ₹5k To 25k ...Suhani Kapoor
 
Russian Call Girls Nashik Anjali 7001305949 Independent Escort Service Nashik
Russian Call Girls Nashik Anjali 7001305949 Independent Escort Service NashikRussian Call Girls Nashik Anjali 7001305949 Independent Escort Service Nashik
Russian Call Girls Nashik Anjali 7001305949 Independent Escort Service Nashikranjana rawat
 
VIP Call Girls Service Bandlaguda Hyderabad Call +91-8250192130
VIP Call Girls Service Bandlaguda Hyderabad Call +91-8250192130VIP Call Girls Service Bandlaguda Hyderabad Call +91-8250192130
VIP Call Girls Service Bandlaguda Hyderabad Call +91-8250192130Suhani Kapoor
 
The Most Attractive Pune Call Girls Shirwal 8250192130 Will You Miss This Cha...
The Most Attractive Pune Call Girls Shirwal 8250192130 Will You Miss This Cha...The Most Attractive Pune Call Girls Shirwal 8250192130 Will You Miss This Cha...
The Most Attractive Pune Call Girls Shirwal 8250192130 Will You Miss This Cha...ranjana rawat
 
(ANAYA) Call Girls Hadapsar ( 7001035870 ) HI-Fi Pune Escorts Service
(ANAYA) Call Girls Hadapsar ( 7001035870 ) HI-Fi Pune Escorts Service(ANAYA) Call Girls Hadapsar ( 7001035870 ) HI-Fi Pune Escorts Service
(ANAYA) Call Girls Hadapsar ( 7001035870 ) HI-Fi Pune Escorts Serviceranjana rawat
 
VIP Call Girls Ramanthapur ( Hyderabad ) Phone 8250192130 | ₹5k To 25k With R...
VIP Call Girls Ramanthapur ( Hyderabad ) Phone 8250192130 | ₹5k To 25k With R...VIP Call Girls Ramanthapur ( Hyderabad ) Phone 8250192130 | ₹5k To 25k With R...
VIP Call Girls Ramanthapur ( Hyderabad ) Phone 8250192130 | ₹5k To 25k With R...Suhani Kapoor
 
(ZARA) Call Girls Talegaon Dabhade ( 7001035870 ) HI-Fi Pune Escorts Service
(ZARA) Call Girls Talegaon Dabhade ( 7001035870 ) HI-Fi Pune Escorts Service(ZARA) Call Girls Talegaon Dabhade ( 7001035870 ) HI-Fi Pune Escorts Service
(ZARA) Call Girls Talegaon Dabhade ( 7001035870 ) HI-Fi Pune Escorts Serviceranjana rawat
 
Call Girls Mumbai Gayatri 8617697112 Independent Escort Service Mumbai
Call Girls Mumbai Gayatri 8617697112 Independent Escort Service MumbaiCall Girls Mumbai Gayatri 8617697112 Independent Escort Service Mumbai
Call Girls Mumbai Gayatri 8617697112 Independent Escort Service MumbaiCall girls in Ahmedabad High profile
 
Mumbai Call Girls, 💞 Prity 9892124323, Navi Mumbai Call girls
Mumbai Call Girls, 💞  Prity 9892124323, Navi Mumbai Call girlsMumbai Call Girls, 💞  Prity 9892124323, Navi Mumbai Call girls
Mumbai Call Girls, 💞 Prity 9892124323, Navi Mumbai Call girlsPooja Nehwal
 
Call Girls In Faridabad(Ballabgarh) Book ☎ 8168257667, @4999
Call Girls In Faridabad(Ballabgarh) Book ☎ 8168257667, @4999Call Girls In Faridabad(Ballabgarh) Book ☎ 8168257667, @4999
Call Girls In Faridabad(Ballabgarh) Book ☎ 8168257667, @4999Tina Ji
 

Recently uploaded (20)

(PARI) Viman Nagar Call Girls Just Call 7001035870 [ Cash on Delivery ] Pune ...
(PARI) Viman Nagar Call Girls Just Call 7001035870 [ Cash on Delivery ] Pune ...(PARI) Viman Nagar Call Girls Just Call 7001035870 [ Cash on Delivery ] Pune ...
(PARI) Viman Nagar Call Girls Just Call 7001035870 [ Cash on Delivery ] Pune ...
 
Escort Service Call Girls In Shakti Nagar, 99530°56974 Delhi NCR
Escort Service Call Girls In Shakti Nagar, 99530°56974 Delhi NCREscort Service Call Girls In Shakti Nagar, 99530°56974 Delhi NCR
Escort Service Call Girls In Shakti Nagar, 99530°56974 Delhi NCR
 
Hot Call Girls |Delhi |Preet Vihar ☎ 9711199171 Book Your One night Stand
Hot Call Girls |Delhi |Preet Vihar ☎ 9711199171 Book Your One night StandHot Call Girls |Delhi |Preet Vihar ☎ 9711199171 Book Your One night Stand
Hot Call Girls |Delhi |Preet Vihar ☎ 9711199171 Book Your One night Stand
 
Call Girls In { Delhi } South Extension Whatsup 9873940964 Enjoy Unlimited Pl...
Call Girls In { Delhi } South Extension Whatsup 9873940964 Enjoy Unlimited Pl...Call Girls In { Delhi } South Extension Whatsup 9873940964 Enjoy Unlimited Pl...
Call Girls In { Delhi } South Extension Whatsup 9873940964 Enjoy Unlimited Pl...
 
VVIP Pune Call Girls Koregaon Park (7001035870) Pune Escorts Nearby with Comp...
VVIP Pune Call Girls Koregaon Park (7001035870) Pune Escorts Nearby with Comp...VVIP Pune Call Girls Koregaon Park (7001035870) Pune Escorts Nearby with Comp...
VVIP Pune Call Girls Koregaon Park (7001035870) Pune Escorts Nearby with Comp...
 
Call Girls In Okhla DELHI ~9654467111~ Short 1500 Night 6000
Call Girls In Okhla DELHI ~9654467111~ Short 1500 Night 6000Call Girls In Okhla DELHI ~9654467111~ Short 1500 Night 6000
Call Girls In Okhla DELHI ~9654467111~ Short 1500 Night 6000
 
VIP Call Girls Service Chaitanyapuri Hyderabad Call +91-8250192130
VIP Call Girls Service Chaitanyapuri Hyderabad Call +91-8250192130VIP Call Girls Service Chaitanyapuri Hyderabad Call +91-8250192130
VIP Call Girls Service Chaitanyapuri Hyderabad Call +91-8250192130
 
Low Rate Call Girls Bikaner Anika 8250192130 Independent Escort Service Bikaner
Low Rate Call Girls Bikaner Anika 8250192130 Independent Escort Service BikanerLow Rate Call Girls Bikaner Anika 8250192130 Independent Escort Service Bikaner
Low Rate Call Girls Bikaner Anika 8250192130 Independent Escort Service Bikaner
 
Call Girls South Delhi Delhi reach out to us at ☎ 9711199012
Call Girls South Delhi Delhi reach out to us at ☎ 9711199012Call Girls South Delhi Delhi reach out to us at ☎ 9711199012
Call Girls South Delhi Delhi reach out to us at ☎ 9711199012
 
VIP Call Girls Moti Ganpur ( Hyderabad ) Phone 8250192130 | ₹5k To 25k With R...
VIP Call Girls Moti Ganpur ( Hyderabad ) Phone 8250192130 | ₹5k To 25k With R...VIP Call Girls Moti Ganpur ( Hyderabad ) Phone 8250192130 | ₹5k To 25k With R...
VIP Call Girls Moti Ganpur ( Hyderabad ) Phone 8250192130 | ₹5k To 25k With R...
 
VIP Call Girls Mahadevpur Colony ( Hyderabad ) Phone 8250192130 | ₹5k To 25k ...
VIP Call Girls Mahadevpur Colony ( Hyderabad ) Phone 8250192130 | ₹5k To 25k ...VIP Call Girls Mahadevpur Colony ( Hyderabad ) Phone 8250192130 | ₹5k To 25k ...
VIP Call Girls Mahadevpur Colony ( Hyderabad ) Phone 8250192130 | ₹5k To 25k ...
 
Russian Call Girls Nashik Anjali 7001305949 Independent Escort Service Nashik
Russian Call Girls Nashik Anjali 7001305949 Independent Escort Service NashikRussian Call Girls Nashik Anjali 7001305949 Independent Escort Service Nashik
Russian Call Girls Nashik Anjali 7001305949 Independent Escort Service Nashik
 
VIP Call Girls Service Bandlaguda Hyderabad Call +91-8250192130
VIP Call Girls Service Bandlaguda Hyderabad Call +91-8250192130VIP Call Girls Service Bandlaguda Hyderabad Call +91-8250192130
VIP Call Girls Service Bandlaguda Hyderabad Call +91-8250192130
 
The Most Attractive Pune Call Girls Shirwal 8250192130 Will You Miss This Cha...
The Most Attractive Pune Call Girls Shirwal 8250192130 Will You Miss This Cha...The Most Attractive Pune Call Girls Shirwal 8250192130 Will You Miss This Cha...
The Most Attractive Pune Call Girls Shirwal 8250192130 Will You Miss This Cha...
 
(ANAYA) Call Girls Hadapsar ( 7001035870 ) HI-Fi Pune Escorts Service
(ANAYA) Call Girls Hadapsar ( 7001035870 ) HI-Fi Pune Escorts Service(ANAYA) Call Girls Hadapsar ( 7001035870 ) HI-Fi Pune Escorts Service
(ANAYA) Call Girls Hadapsar ( 7001035870 ) HI-Fi Pune Escorts Service
 
VIP Call Girls Ramanthapur ( Hyderabad ) Phone 8250192130 | ₹5k To 25k With R...
VIP Call Girls Ramanthapur ( Hyderabad ) Phone 8250192130 | ₹5k To 25k With R...VIP Call Girls Ramanthapur ( Hyderabad ) Phone 8250192130 | ₹5k To 25k With R...
VIP Call Girls Ramanthapur ( Hyderabad ) Phone 8250192130 | ₹5k To 25k With R...
 
(ZARA) Call Girls Talegaon Dabhade ( 7001035870 ) HI-Fi Pune Escorts Service
(ZARA) Call Girls Talegaon Dabhade ( 7001035870 ) HI-Fi Pune Escorts Service(ZARA) Call Girls Talegaon Dabhade ( 7001035870 ) HI-Fi Pune Escorts Service
(ZARA) Call Girls Talegaon Dabhade ( 7001035870 ) HI-Fi Pune Escorts Service
 
Call Girls Mumbai Gayatri 8617697112 Independent Escort Service Mumbai
Call Girls Mumbai Gayatri 8617697112 Independent Escort Service MumbaiCall Girls Mumbai Gayatri 8617697112 Independent Escort Service Mumbai
Call Girls Mumbai Gayatri 8617697112 Independent Escort Service Mumbai
 
Mumbai Call Girls, 💞 Prity 9892124323, Navi Mumbai Call girls
Mumbai Call Girls, 💞  Prity 9892124323, Navi Mumbai Call girlsMumbai Call Girls, 💞  Prity 9892124323, Navi Mumbai Call girls
Mumbai Call Girls, 💞 Prity 9892124323, Navi Mumbai Call girls
 
Call Girls In Faridabad(Ballabgarh) Book ☎ 8168257667, @4999
Call Girls In Faridabad(Ballabgarh) Book ☎ 8168257667, @4999Call Girls In Faridabad(Ballabgarh) Book ☎ 8168257667, @4999
Call Girls In Faridabad(Ballabgarh) Book ☎ 8168257667, @4999
 

Daamen r 2010scwr-cpaper

  • 1. Development of Inferential Sensors for Real-time Quality Control of Water- Level Data for the Everglades Depth Estimation Network Ruby Daamen, Advanced Data Mining Int’l Edwin Roehl, Advanced Data Mining Int’l Paul Conrads, USGS – SC Water Science Center Matthew Petkewich, USGS – SC Water Science Center
  • 2. Presentation Outline • What is an “Inferential Sensor” (IS)? • Background - Industrial application • Everglades Depth Estimation Network (EDEN) • Automated Data Assurance and Management (ADAM) - Inferential Sensor development for EDEN
  • 3. Development of IS in Industry • A tough environment to monitor – Emissions regulations require measurements of effluent gases – Smoke stack burns up probes – Need alternative to “hard” sensors
  • 4. Hard Sensor vs. Inferential Sensor • Virtual sensor replaces actual sensor – Temporary gage smoke stack – Operate plant to cover range of emissions – Develop model of emissions based on operations – Model becomes the “Inferential Sensor”
  • 5. Inferential Sensors for Real-Time Data • Problem – Need to minimize missing and erroneous data • Use similar approach taken by industry to predict real-time data – ie. “Inferential Sensors” – QA/QC hard sensor – Provide accurate estimates for hard sensor – Provide redundant signal
  • 6. EDEN Problem • EDEN – integrated network of real-time water- level (WL) stations, ground-elevation models, and water-surface models – Real-time data is used to generate EDEN WL surfaces used by scientists, engineers and water-resource managers (253+ stations) – Data used to guide large-scale field operations, integrate hydrologic and ecological responses, support assessments that measure ecosystem responses to the Comprehensive Everglades Restoration Plan (CERP) – Correcting errors is often time consuming; gaging stations may be in remote areas – Need to identify errors and to provide estimates on a daily basis
  • 7. EDEN Water Surface Map Bad values creates erroneous areas on maps
  • 8. EDEN / ADAM Overview Inferential Sensor application
  • 9. Some of the Challenges • 253+ gaging stations • Inferential sensor uses data from one or more correlated gages. At any given datetime do not know what stations will have available data • Stations added/removed over time
  • 10. Implementation • Create models “on the fly” – Use “best” available stations – Simplifies addition / removal of stations • Needs to consider – Assessment of data – prevent use of suspect data in models – Model inputs need be decorrelated
  • 11. Methods • Two algorithms sequenced to analyze data – Univariate filtering • Provides information about the quality and behavior for each stations WL • A Statistical Process Control (SPC) which consists of 14 univariate filters – uniquely set for each station – Estimate parameter value • Select a “pool” of candidate gaging stations using matrix of Pearson coefficients • Principal component analysis (PCA) – calculates decorrelated inputs • Multivariate linear regression
  • 12. Univariate Filtering UNIVARIATE FILTER CHECK DESCRIPTION PRECEDENCE WATER LEVEL LIMIT (ft.) LOST_SIGNAL no signal 1 NA GT_RNG_UL x(t) > signal range Upper Range Limit 2 15.19 LT_RNG_LL x(t) < signal range Upper Range Limit 3 6.99 GT_UCL x(t) > signal Upper Control Limit 4 14.73 LT_LCL x(t) < signal Upper Control Limit 5 8.56 Sn_LT_L flatlined: x'(t) = x(t)=x(t-1); SUM[(|x'(t)|,…,|x'(t-n+1)|] < Limit 6 0.00 D1_GT_L_1 vfast vlarge increase: x(t)-x(t-1) > Limit 7 1.92 D1_LT_L_1 vfast vlarge decrease: x(t)-x(t-1) < Limit 8 -2.34 D1Sn_GT_L_1 fast vlarge increase: x'(t)=x(t)-x(t-1); Sum[x'(t),…x'(t-n+1)] > Limit 9 1.98 D1Sn_LT_L_1 fast vlarge decrease: x'(t)=x(t)-x(t-1); Sum[x'(t),…x'(t-n+1)] < Limit 10 -2.52 D1_GT_L_2 vfast large increase: x(t) - x(t-1) > Limit 11 1.69 D1_LT_L_2 vfast large decrease: x(t) - x(t-1)< Limit 12 -0.25 D1Sn_GT_L_2 fast large increase: x'(t)=x(t)-x(t-1); Sum[x'(t),…x'(t-n+1)] > Limit 13 1.98 D1Sn_LT_L_2 fast large decrease: x'(t)=x(t)-x(t-1); Sum[x'(t),…x'(t-n+1)] < Limit 14 -0.27 • Additional filters – Dry Protocol – set using offset from ground elevation • Any filter trips are flagged for review • Data triggering a filter is not used in any predictions
  • 13. Synthesize WL Measurements • WL at candidate stations are correlated – no surprise • First approach examined selected the most highly correlated station as a “standard” signal and then attempted decorrelating the other stations by computing differences from the standard.
  • 14. Principal Component Analysis (PCA) • PCA is a statistical technique used to “reduce the dimensionality of a data set consisting of a large number of interrelated variables, while retaining as much as possible of the variation present in the data set. This is achieved by transforming to a new set of variables, the principal components (PCs), which are uncorrelated, and which are ordered so that the first few retain most of the variation present in all of the original variables” (Joliffe)
  • 15. PCA – Main Points • PCA - the main points – Principal components are uncorrelated – Transforms a set of correlated variables into a smaller number of uncorrelated variables – The first principal component (PC) accounts for most of the variability in the data
  • 16. PCA - Analysis • PCA – a brief description of the analysis: – Calculate the eigenvectors and eigenvalues of the covariance matrix • Create data set of n inputs with no gaps • Subtract the mean from each n input • Calculate the covariance matrix (square nXn matrix) • Calculate eigenvectors – Sort by eigenvalues (highest to lowest) • Largest eigenvector = 1st principal component • Use eigenvalues to determine how many PCs to include
  • 17. PCA – A 2-Dimension Example Original Data (Mean Subtracted) Eigenvectors Principal Components Eigenvectors Plotted on Data -4 -3 -2 -1 0 1 2 3 4 -4 -3 -2 -1 0 1 2 3 4 Normalized Data E1 E2 Original Data
  • 18. ADAM - Functionality • Setup – File paths – PCA setup – period, number of sites to include – Add / Edit / Remove sites – Univariate filters • Inferential Sensor – Option to analyze daily (hourly and 15 minute data), quarterly and annual (hourly) daily files • Review results • Output daily median files as required for EDEN water surface map
  • 19. ADAM – Control Worksheet Select Daily, Quarterly or Annual Run Analysis Resume, redo OR continue from last analyzed Fill Setup Remove , add, or edit sites included in ADAM Set Pathnames for files used by ADAM
  • 20. ADAM – Control Worksheet Loads data from selected run for review Creates output files required to generate EDEN water surface map Dumps a listing sites tripping any filters
  • 21. ADAM Review Worksheet First PCA analysis GAPFILL
  • 24. ADAM – Univariate Filter Setups Graphs show settings vs. data
  • 25. What’s Next EVEEvolution and Verification of EDEN Inferential Sensor Questions?
  • 27. PCA – The methodology • From our favorite source -Wikipedia – PCA is: – A mathematical procedure that transforms a set of correlated variables into a smaller number of uncorrelated variables. The first principal component accounts for as much of the variability in the data as possible and each succeeding component accounts for as much of the remaining variability as possible. • How to do it: In the broadest sense it is an eigenvector based multivariate analysis. The methodology used is: – Assemble the data • In EDENIS the data will be water level data from up to 5 gaging stations. For 90 days of hourly data this equates to up to 2160 vectors (8640 for 15 minute data) – Remove any vectors that contain any missing data (1) • Resulting matrix X[n,m] where n = number of fully populated vectors; m = number of gages included – Subtract the mean from each of the data dimensions (m) – lets call this the normalized matrix B (2) • B[n,m] stores mean-subtracted data – Calculate the Covariance matrix (3) • Covariance matrix is a square matrix with dimension mXm: C[m,m] – Calculate the eigenvalues and eigenvectors of the covariance matrix • This is an iterative process. If you want to look up some more on this I used the Jacobi eigenvalue algorithm. Some important properties of eigenvectors: – Can only be found for square matrices. If a square matrix (mXm) does have eigenvectors, there are m of them – All the eigenvectors of a matrix are orthogonal to each other. This is important: when the data is expressed in terms of these eigenvectors the resulting principal components are uncorrelated • For a refresher course on eigenvalues / eigenvectors heres a link http://en.wikipedia.org/wiki/Eigenvalue,_eigenvector_and_eigenspace – Sort the results: Largest eigenvalue to smallest eigenvalue. (4) – Decide how many components to keep and calculate the new data set to be used. This is a simple matrix multiplication of B X E where E is the mXm eigenvector matrix. (5) • For the 10-12 sites I looked at when presenting the PCA results I would expect we’ll rarely if ever use more than 2 principal components out of a possible 5 to make the regression predictions.
  • 28. PCA – A 2-Dimension Example 1. Original Data 2. Mean Subtracted 3. Covariance Matrix 4. Eigenvalues / Eigenvectors 5. New Data Set Note high correlation of orignal X1, X2 vs. No correlation of PC1, PC2 E1 and E2 are the 2 eigenvectors layed on top of the data. Note E1 and E2 are perpendicular. Also note that E1 goes through the middle of the data – like a best fit. E2 provides less information about the variance in the data. -4 -3 -2 -1 0 1 2 3 4 -4 -3 -2 -1 0 1 2 3 4 Normalized Data E1 E2
  • 29. Challenges • Develop 253+ models using artificial neural networks (ANNs) (1 model per station) – Pros • authors have prior success modeling complex processes using ANNs • ANNs use non-linear curve fitting to capture complex behaviors – Cons • Do not know what stations will have “good” data at any given datetime • Stations are removed and added
  • 30. • Possibly insert Yearly run – min/max r2 using pca,gapfill

Editor's Notes

  1. WL data at real-time gages are related to ungaged ares andusing ground elevation data, water depths throughout the Everglades are computed Data used by scientists, engineers, and water resource managers to support CERP – Comprehensive Everglades Restoration Plan Used to : 1) Guide large scale field operations, 2) integrate hydrologic and ecological responses and 3) support biological and ecological assesssments that measure ecosystem responses to the CERP