Transcript
Elin Vanlierde05-10-2015
Sediment Measurements @ FHR
Overview of presentation• What do we monitor and where?• Translating measurements into sediment
concentrations• Translating sediment concentrations into fluxes
* ISCO sampler- Automatic sampling
(every seven hours)
* Grab samples/weighted bottle sampling
- Manual sampling (every week)
* EWI sampling- Manual sampling (every
week)
Monitoring physical parameters
• water samples for determining: SSC, LOI, grain size
Localisation of water sampling
High water and low water slack(highest and lowest conductivities)
Half tide ebb(highest flow velocities)
Dock sampling
- Water sampling: SSC, Cl-, Cond
- YSI: T°, pH, Cond
- Water sampling: SSC- YSI: T°, pH, Cond
- Water sampling: SSC, Cl-, Cond
- YSI: T°, pH, Cond
Monitoring physical parameters (continued)
• Monthly sailing campaigns:• Sampling with centrifugal pump• Measurements with YSI
Localisation of sailing campaigns
Half tide ebbSlack tide HW & LW
Monitoring physical parameters (continued)
• Measuring physical parameters with multi-parameter probes
• CTD (Conductivity, Temperature Depth)
• Aanderaa (Flow velocity, Flow direction, Temperature, Conductivity, Pressure, Turbidity)
• YSI (Temperature, Conductivity, pH Pressure, Turbidity, Oxygen content)
Where do we measure?
ProsperpolderBoei 84
Liefkenshoek
Oosterweel
Hemiksem
Driegoten
Melle
GavereZulte
Wondelgem
Appels
EppegemAarschot
Duffelsluis
Grobbendonk
Herenthout
Overview of presentation• What do we monitor and where?• Translating measurements into sediment
concentrations• Translating sediment concentrations into fluxes
Translating measurements into SSC
• Sampling• Grab samples• Automatic ISCO samples• Pump samples
yield SSC (mg/l)
• continuous measurements• Turbidity• Discharge• Conductivity• …
yield various proxies needs translating
into SSC (using rating curves)
Creating rating curves• Approach Aanderaa differs from approach YSI
Location driven • Seascheldt = Aanderaa• influx into Seascheldt = YSI
Aanderaa approach• Aanderaa locations have 1 summer and 1 winter
sampling campaign (multiple hours) per year
Aanderaa approach• All campaigns plotted on one graph• 95% Upper and Lower Confidence intervals are
calculated
Aanderaa approach• R²: high• RMSE: 27 – 58 mg/l• MSPE around 20% or lower
Boei84 Regressiemodel R² RMSE MSPE N Turb_Min Turb_Max SSC_Min SSC_MaxSG5 y = 0.000776x2 + 1.133978x 0.91 45.9 23.1 670 19 515 16 872
SG25 y = 1.232555x 0.95 41.7 20.6 549 22 818 16 1004
Oosterweel Regressiemodel R² RMSE MSPE N Turb_Min Turb_Max SSC_Min SSC_MaxSG5 y = 0.000227x2 + 1.237918x 0.86 51.46 14.96 408 40 510 41 804
SG25 y = 1.397138x 0.87 57.85 13.62 597 53 553 52 841
Driegoten Regressiemodel R² RMSE MSPE N Turb_Min Turb_Max SSC_Min SSC_MaxSG5 y = 0.003491x2 + 1.274708x 0.89 45.3 22.8 522 19 340 15 923
SG25 y = 0.003349x2 + 0.999738x 0.91 27.23 14.87 344 20 259 15 523
YSI approach• Continuous on field measurement data
• YSI: Turbidity, Conductivity, etc. (5min)• Discharge (5 or 15 min Daily or Tidal Mean)
• High-frequent calibrational data:
• automatic sampling SSC (7h)
YSI approach• USGS (Rasmussen et al., 2009)
• Simple linear regression (Turb): when MSPE <20%• Multiple linear regression (Turb. + 1 other variable) when:
• MSPE > 20%• p-value (probability value) of the partial F statistic for extra
predictor< 0.025• Pearsons r < 0.95 (no collinearity issues)
Model R² p-value RMSE MSPE PRESSTu 0,31 <2e-16 144 53,62 0,31Tu-Qtij 0,42 <2e-16 132 49,06 0,42Tu-Cond 0,37 <2e-16 142 53,8 0,36
Regression analysis: Sub-relations
• Sub-relations:• Diff. series
• difference between measurement and predictions
• testing different periods• summer-winter• trimester• year
• using Chow test
Quality of the relationExample Melle• R²: low• RMSE:
• MLR: 121 mg/l• SLR (Turb): 144 mg/l
• MSPE:• MLR: 47%• SLR (Turb)= 55%
Meetlocatie Regressiemodel SSC_Var R² p-value RMSE MSPE N Min_Turb Max_Turb Min_QTij Max_Qtij Min_SSC Max_SSClogSSC=f(log Turb, log QTij) 0.42 <2e-16 121.1 46.99 3236 6.9 549.2 0.174 318.297 35 2419
logSSC=f(log Turb) 0.35 144.362 55.45 3261 6.9 549.2 nvt nvt 35 2419
SSC=f(Qtij) 0.20 <2e-16 205.8 73.2 3507 nvt nvt -29.139 318.297 35 3465
Melle
Quality of the relation• 95% Upper and Lower Confidence Intervals
Discussion points• Hysteresis effects
• The variability observed is due to hysteresis • YSI:
• individual peaks, or even individual rising and descending limbs of SSC peaks give very good relations
• Not feasible within framework of data validation• Aanderaa, less hysteresis observed
Campaign 7/11/14 Oosterweel
150 200 250 300 350 400 450 500 5500
100
200
300
400
500
600
700
800
900
1000
320588outliers 320outliers 588
Turb (FTU)
SSC
(mg/
l)
0
20
40
60
80
100
120
D50D90µm
• Masterthesis student requested for further research
Overview of presentation• What do we monitor and where?• Translating measurements into sediment
concentrations• Translating sediment concentrations into fluxes
Translating SSC’s into SSF’s • Fluvial flux towards the Seascheldt
= ∑ fluxes tributaries• Flux tributary = SSC tributary x discharge tributary
• Historical approach (weekly grabsamples)• Assumptions:
• Grabsample = daily average value• Rating curve Q-SSC to determine daily values for
the other 6 days of the week• Results: relationships with wide 95% confidence
bands
Weekly grabsamples (Melle)
Translating SSC’s into SSF’s • Flux towards the Seascheldt = ∑ fluxes tributaries• Flux tributary = SSC tributary x discharge tributary
• Historical approach (weekly grabsamples)• New proposed approach (high frequency
measurements) problem: SSC’s measured or sampled are all point samples
not representative of the cross-sectional SSC
Fluxes based on grabsamples (+CI’s) and ISCO (Melle)
Translating SSC’s into SSF’s • Equal Width Increment
method sampling• results in a depth and
width integrated SSC sample
• Creating rating curves between EWI and grab samples (historically) and ISCO samples (new approach) to correct the calculated flux
EWI corrections (Eppegem)
EWI (mg/l) ISCO (mg/l) Grabsample (mg/l)10 6 250 43 47
100 89 103700 641 777
EWI corrections (Melle)
ISCO (mg/l) EWI (mg/l)10 -8050 8
100 1181000 2094
Translating SSC’s into SSF’s • Equal Width Increment method sampling
• results in a depth and width integrated SSC sample
• Creating rating curves between EWI and grab samples (historically) and ISCO samples (new approach) to correct the calculated flux
• Promising with enough sampling points
To do’s• Establishing good EWI correction factors• Calculating SSF and confidence intervals for SSC’S
corrected with these factors for• grab samples• ISCO samples• Turbidity + other variables
• Comparing the results and confidence bands and selectnig the least demanding approach.
The end…• but let’s have some discussions
Discussion questions• Is the better quality of the Aanderaa relations
compared to YSI relationships due to:• the location, the equipment, the sampling
frequency, less hysteresis between sediment peak and discharge peak in the Seaascheldt…
• Does the increase in predictability of SSC merit the inclusion of other variables in a Multiple Linear Regression?
• Should fluxes be calculated based on measurements (ISCO) rather than (even) higher frequency YSI data?
top related