The evaluation of rainfall influence on CSO characteristics: the Berlin case study S. Sandoval*, A. Torres*, E. Pawlowsky-Reusing **, M. Riechel*** and.

The evaluation of rainfall influence on CSO characteristics: the Berlin case study

S. Sandoval*, A. Torres*,

E. Pawlowsky-Reusing **,

M. Riechel*** and N. Caradot*** * Pontificia Universidad Javeriana, Bogotá, Colombia** Berliner Wasserbetriebe, Berlin, Germany*** Kompetenzzentrum Wasser Berlin, Berlin, Germany

Combined sewer system

Separate sewer system

CSO monitoring station N

CSO monitoring in BerlinSub catchment:

• 126000 inhabitants

• 800 ha impervious area

10 km

Combined sewer system

Separate sewer system Rain gauges

CSO monitoring station N

CSO monitoring in BerlinSub catchment:

• 126000 inhabitants

• 800 ha impervious area

Rainfall

• Annual rain 570 mm/a

• > 10 mm: 13/a

10 km

CSO monitoring

• Average contribution of wastewater to

– CSO volume = 11%

– CSO COD load = 16%

• 84% contribution from other sources !

rain runoff wash-off

resuspension of sewer sediments

• Very strong variability of volume and concentrations

What is the influence of rainfall on CSO characteristics ?

Is it possible to predict CSO characteristics from rainfall ?

Canonical Correlation Analysis CCA

• Linear relationship between two multidimensional data sets:– X (input rainfall characteristics) and Y (output CSO characteristics) – Row: events / Columns: characteristics

• A couple of vectors a and b is found by maximizing correlation (a.X , b.Y)

a1 X1 + a2 X2 + … + an Xn ~ b1 Y1 + b2 Y2 + … + bnYn

• Evaluation of correlation with canonical loadings: – linear correlations between each characteristic and CV

Canonical loading Xi = corr (Xi, CVx)

Canonical loading Yi = corr (Yi, CVy)

Canonical variate xCVx

Canonical variate yCVy

Canonical loadings

Rainfall X

CV5

Duration 0.24

Max intensity -0.43

Depth 0.07

Mean intensity -0.52

DW duration -0.21

Canonical loadings

CSO Y  CV5

Duration 0.00

Max. Flow -0.54

Volume -0.40

Mean Flow -0.64

M_COD -0.63

M_TSS -0.51

M_CODd -0.72

mean_TSS -0.24

mean_COD -0.41

mean_CODf -0.47

mean_EC -0.23

Waste ratio (V) -0.29

Waste ratio (M) 0.11

Canonical loadings

Rainfall X

CV6

Duration 0.58

Max intensity 0.35

Depth 0.84

Mean intensity -0.17

DW duration 0.22

Canonical loadings

CSO Y  CV6

Duration 0.64

Max. Flow 0.39

Volume 0.72

Mean Flow 0.09

M_COD 0.12

M_TSS 0.23

M_CODd 0.13

mean_TSS -0.55

mean_COD -0.57

mean_CODf -0.62

mean_EC -0.71

Waste ratio (V) -0.71

Waste ratio (M) -0.60

Max intensityMean intensity

Max flowMean flow

Pollutant loads

DurationDepth

DurationVolume

Mean concentrations

Canonical Variate 1 Canonical Variate 2

Partial Least Square regression PLS

• Linear relationship between a multidimensional input variable X (rainfall characteristics) and individual output Y (CSO characteristic)

– Row: events

– Columns: characteristics

• The PLS method projects original data onto a more compact space of latent variables

• A set of coefficients ai is found by maximizing the covariance between X and Y

Y = a1 X1 + a2 X2 + … + an Xc

Identification of most important rain characteristics (high coefficients)

• For each CSO variable (e.g. max. flow)

• Generation of 1000 sets of random rainfall and CSO values within their uncertainty interval 1000 PLS models

Quality of prediction : coefficient of determination R2 Identification of most important X variables

Max intensity DW duration Rain duration

Identification of most relevant explenatory variables

Probability of being the most important

rainfall variable

Duration

Max intensity

DW duration

Max intensityMean intensityDuration Depth

Max flowMean flowDuration Volume

DW durationMax intensity

Pollutant loads

DurationDepth

Mean concentrations

Conclusion

• PLS and CCA highlight the influence of rainfall on CSO characteristics

• For PLS, low determination coefficients were obtained (< 0.6)

• not suitable for prediction purposes,

• useful for exploring the qualitative influence of rainfall on CSO

• Future researches

• Test of other analysis methods (e.g. Artificial Neural Networks)

• Relation between rainfall, CSO and resulting river impacts

Thank you for your attention !

More information : [email protected]

30 km

3 km

Combined sewer system

Separate sewer system

Area of water bodies

a

b cd

a River monitoring station

CSO monitoring station

N

Integrated monitoring stations