The effect of sampling frequency and strategy on …The effect of sampling frequency and strategy on water quality modelling driven by continuous monitoring data in a boreal catchment
Post on 29-Jun-2020
0 Views
Preview:
Transcript
The effect of sampling frequency and strategy on water quality modelling
driven by continuous monitoring data in a boreal catchment
Paweł Marcinkowski1, Mikołaj Piniewski1, Jari Koskiaho2, Sirkka Tattari2
1. Warsaw University of Life Sciences (email: p.marcinkowski@levis.sggw.pl)
2. Finnish Environment Institute
Water quality monitoring
• Two broad categories of water quality sampling:• Base-flow sampling used for routine monitoring
programmes
• Event-based sampling, used for estimating soil and nutrient erosion rates and constituent load exports
• HELCOM (2015) recommends that the riverine water sampling strategy used by monitoring agencies should be designed to cover the entire flow cycle, i.e. low, mean and high flow.
Figure 1: Riverine monitoring during 2010-2016 in Baltic Sea Region (BSR) countries. Source:HELCOM Pollution Load Compilation
Low-frequency sampling is likely to miss many storm events, and thus fail to disclose the close coupling between the hydrological and chemical dynamics…
Examples of sampling strategies
0
500
1 000
1 500
2 000
2 500
3 000
Tota
l P lo
ad (
kg)
Date
Regular sampling, 5th day of every month (12 samples per year)
PtotOBSkg
Water sample taken
0
500
1 000
1 500
2 000
2 500
3 000
Tota
l P lo
ad (
kg)
Date
Flow-proprtional sampling (12 samples per year)
PtotOBSkg
Water sample taken
High-frequency monitoring
• These issues can be addressed by field-deployable water quality analysers.
• Automatic monitoring devices produce reliable, high-frequency water quality data, provided that quality assurance is taken care of.
Objective
Evaluating the continuous, six-year-long, high-frequency water quality monitoring data set as a source for the calibration and testing of the SWATmodel in a case study carried out in a medium-sized, boreal catchment in southern Finland.
Tasks
• calibration and validation of the model against thehigh-frequency data set of TSS, NO3-N, TN and TP loads.
• Developing a set of 'sub-sampling' scenariosfocused on quantifying the effect of the sampling frequency (12 vs. 26 samples per year) and strategy (regular vs. flow-proportional sampling) on the model performance and output uncertainty
Study area
• Vantaanjoki catchment
• One of the case study catchments of the Bonus RETURN project (www.bonusreturn.com)
• The river flows through Helsinki metropolitan area of ½ million people huge recreational value Figure 2: Case study location
Major point sources, tributaries & lakes
• Catchment area: 1,680 km2
• 75% of the catchmentenclosed by automatic monitoring station
Land use and soil map
Land use (as in 2012)AGRR 17%FRST 56%HAY 7%URBN 9%UIDU 8%WETL 1%WATR 2%
Soil Clay 45%Moraines 12%Rocky 21%Peat 7%Coarse 12%Water 3%
Pitkäkoski automatic monitoring station
• S::can nitro-lyser sensor(www.s-can.at/products/spectrometer-probes#)
• Hourly data from years 2011–2016, virtually without breaks
• Turbidity (FNU)• NO3-N (mg/l)• TOC & DOC (mg/l)• High correlation between
turbidity and water-sampled total P (and sediment) conversions with linear relationships
Assembling the sensor
in October 2010
y = 1,35x + 1,69R² = 0,82
0
20
40
60
80
100
120
140
0 20 40 60 80 100 120 140
Wat
er-
sam
ple
d T
SS c
on
cen
trat
ion
(m
g/l)
Turbidity (FNU), raw sensor data
TSS
y = 2,03x + 45,28R² = 0,73
0
50
100
150
200
250
0 50 100 150 200 250
Wat
er-
sam
ple
d P
tot
con
cen
trat
ion
(µ
g/l)
Turbidity (FNU), raw sensor data
Ptot
y = 0,61xR² = 0,86
0
1
2
3
4
5
6
0 1 2 3 4 5 6
Wat
er-
sam
ple
d N
O3
-N (
mg/
l)
NO3-N (mg/l), raw sensor data
NO3-N
y = 0,83xR² = 0,87
0
1
2
3
4
5
6
0 1 2 3 4 5 6
Wat
er-
sam
ple
d N
tot
(mg/
l)
NO3-N (mg/l), raw sensor data
Ntot
Raw sensor data vs. water samples
conversions to TSS, Ptot and Ntot
concentrations by linear regression
equations
conversion into average daily loads
Sub-sampling scenarios
• 1. Flow-proportional sampling with 26 samples per year (FP-26)
• 2. Flow-proportional sampling with 12 samples per year (FP-12)
• 3. Regular sampling with 12 samples per year (R-12)
• All tested through6 realizations
Work flow
Modelcalibration
against high-frequency data considered as ‘benchmark’
Sub-sampled observational
data set model calibration using
SUFI-2.
54 scenarios
Comparison of calibration
results based on sub-sampleddata with the benchmark
Model performance
KGE = 0.88 (cal) 0.86 (val) KGE = 0.79 (cal) 0.73 (val)
KGE = 0.76 (cal) 0.72 (val) KGE = 0.83 (cal) 0.69 (val)
Model performance in sub-samplingscenarios
PBIAS vs KGE
Model output uncertainty under sub-sampling scenarios
Conclusions
• The model calibrated against low-frequency data performed worse when evaluated against high-frequency data and compared with the benchmark model for each of the studied water qualityparameters.
• Increasing the sampling frequency from 12 to 26 samples per year led to an improvement in the model performance (as quantified by KGE),particularly for TSS and TP loads.
Conclusions
• Several of the lowest KGE values were usually associated with high over- or under-estimation bias
• In contrast to the sampling frequency, no strong evidence for the effect of sampling strategy on the model performance was found.
• More attention should be paid to calibration data as a source of uncertainty in model predictions
• Evidence for a wider adoption of modern sensors providing high-frequency data
• Evidence for increasing the grab sampling frequency to reduce the uncertainty of loading estimates (particularly in rivers with flashy flow regime)
Funding
• BONUS RETURN has received funding from BONUS (Art 185), funded jointly by the EU and Swedish Foundation for Strategic Environmental ResearchFORMAS, Innovation Fund Denmark, Academy of Finland and National Centre for Research and Development in Poland.
top related