Shopping cart
0 Items | US$ 0.00
A.4 Computed trends in driving forces
178
B Maps of the
EF IIs
181
B.1 Hands-off flow reliability indicator HOF
rel
181
B.2 Hands-off flow resilience indicator HOF
res
183
B.3 Pike spawning reliability indicator P IKE
rel
185
B.4 Pike spawning resilience indicator P IKE
res
187
B.5 Floodplain vegetation reliability indicator F V C
rel
189
B.6 Floodplain vegetation resilience indicator F V C
res
191
Acknowledgements
193
vi
List of Figures
1.1
Conceptual schematic overview of this thesis.
5
1.2
Schematic illustration of the scope of this thesis.
7
2.1
Map of the study area.
10
2.2
Map of geological units and glaciation extent in the NRB (source:
Geologic Map of Poland from the Polish Geological Institute).
11
2.3
Topography and hydrographic network of the NRB (hydrographic source:
Digital Map of Hydrological Division of Poland from the Institute of
Meteorology and Water Management).
11
2.4
Generalised soil types in the NRB (after Gielczewski, 2003).
12
2.5
Land cover map of the NRB (source: CORINE Land Cover 2000 from
the Chief Inspectorate of Environment Protection).
13
2.6
Map of the interpolated mean January (A.) and July (B.) temperatures
for 1989-2008.
14
2.7
Map of interpolated mean annual precipitation for 1989-2008.
15
2.8
Observed daily discharge hydrographs at the basin outlet (Zambski
Kocielne station on the Narew) during a wet year (1994) and a dry
year (2003).
16
2.9
Administrative subdivision in the NRB: NUTS 2 (provinces) and NUTS
3 (sub-regions) levels.
17
3.1
Visualisation of HRU delineation from an overlay of land use and soil
maps within sub-basins.
22
3.2
Schematic pathways of water movement in SWAT (Neitsch et al., 2011).
24
3.3
Preliminary conguration of SWAT for the NRB as in Piniewski and
Okruszko (2011).
25
3.4
SWAT input soil map of the NRB and benchmark soil profiles used to
create the map.
27
3.5
New SWAT model set-up for the NRB (cf. Table 3.3 for flow gauges
codes).
28
vii
3.6
Daily outflow from the Siemianówka reservoir measured at the Bond-
ary gauging station on the Narew for the time period 1990-2008.
30
3.7
Spatial calibration and validation approach.
36
3.8
Initial calibration and uncertainty analysis using SUFI-2 in 10 calib-
ration areas.
37
3.9
Basin-averaged intra-annual variability in mean temperature for the
baseline period and under two GCMs for the 2050s.
46
3.10 Spatial and seasonal variability in GCM projections of temperature
change for the 2050s according to IPSL-CM4 (A, C, E, G) and MIROC3.2
(B, D, F, H); used notations: DJF: December - February; MAM: March
- May; JJA: June - August; SON: September - November).
47
3.11 Basin-averaged inter-annual variability in mean precipitation for the
baseline period and under two GCMs for the 2050s.
48
3.12 Spatial and seasonal variability in GCM projections of precipitation
change for the 2050s according to IPSL-CM4 (A, C, E, G) and MIROC3.2
(B, D, F, H); used notations: DJF: December - February; MAM: March
- May; JJA: June - August; SON: September - November).
49
3.13 Schematic representation of the scenario development process in the
NRB Pilot Area (PA; after Gielczewski et al., 2011).
52
3.14 Division of the NRB into sub-regions.
60
3.15 Example fuzzy membership functions for descriptive terms associated
with question 1A.
65
3.16 Groups of wetland vegetation communities in the NRB, according to
the Spatial Information System of Polish Wetlands (http://www.gis-
mokradla.info/). Typical alliances for each community are given in
parenthesis.
75
3.17 Building blocks of the flow regime and dependent elements of the river
ecosystem (after Acreman et al., 2009)
77
3.18 Building blocks of the flow regime and dependent elements of the river
ecosystem the NRB case.
77
3.19 Optimal duration of inundation for NRB reaches (FVC categories from
Tab. 3.15) assigned by ITP (2010).
82
3.20 Bankfull flows estimated at selected gauging stations and interpolated
across river reaches.
85
3.21 An example trapezoidal membership function.
90
4.1
Observed and SUFI-2 simulated (best solution and 95PPU band) daily
flows at ten Phase 1 calibration areas for the time period 1995-2008.
98
viii
4.2
Relationship between upstream catchment area and goodness-of-fit
measures for best simulations obtained in SUFI-2 for ten analysed sites.
100
4.3
Nash-Sutcliffe Efficiency of best simulations, obtained using the PSO
technique in different phases of spatial calibration and validation for
calibration period (A) and temporal validation period (B)
101
4.4
Relationship between the model's performance measures and the area
upstream of the gauge in the calibration and temporal validation peri-
ods.
103
4.5
Observed and PSO-simulated daily flows at six example sites (one for
each calibration phase) for the joint time period of calibration and
validation (1989-2008).
104
4.6
Final parameter values in all calibration areas.
106
4.7
Scatter plot of Nash-Sutcliffe Efficiencies of the "old" model set-up
(NSE
1
) vs. the "new" set-up (NSE
2
) calculated for analysed gauges
for calibration and temporal validation periods.
108
4.8
SWAT reaches selected for further analysis.
110
4.9
Spatial diversity in RL
hof
(A) and RS
hof
(B), calculated using SWAT
for 52 reaches of interest. Numbers next to the names of the gauging
stations show bias, i.e. the difference between SWAT-based indicators
and indicators calculated using the observed discharge data.
112
4.10 Spatial diversity in RL
pike
(A) and RS
pike
(B), calculated using SWAT
for 52 reaches of interest. Numbers next to the names of the gauging
stations show bias, i.e. the difference between SWAT-based indicators
and indicators calculated using the observed discharge data.
113
4.11 Spatial diversity in RL
fvc
(A) and RS
fvc
(B), calculated using SWAT
for 52 reaches of interest. Numbers next to the names of the gauging
stations show bias, i.e. the difference between SWAT-based indicators
and indicators calculated using the observed discharge data.
114
4.12 Seasonal distribution of mean annual precipitation and snowfall under
two global change scenarios.
120
4.13 Seasonal distribution of runoff under single (A) and combined (B) scen-
arios.
121
4.14 Differences in mean monthly runoff between the decomposed scenarios
and the original scenarios.
123
4.15 Pie charts of different classes of impacts on the hands-off flow reliability
indicator HOF
rel
under all analysed scenarios in NRB sub-regions (pie
diameter is proportional to the number of reaches in a sub-region).
125
ix
4.16 Pie charts of different classes of impacts on the hands-off flow resilience
indicator HOF
res
under all analysed scenarios in NRB sub-regions (pie
diameter is proportional to the number of reaches in a sub-region).
126
4.17 Pie charts of different classes of impacts on the pike spawning reliability
indicator P IKE
rel
under all analysed scenarios in NRB sub-regions
(pie diameter is proportional to the number of reaches in a sub-region).
127
4.18 Pie charts of different classes of impacts on the pike spawning resilience
indicator P IKE
res
under all analysed scenarios in NRB sub-regions
(pie diameter is proportional to the number of reaches in a sub-region).
128
4.19 Pie charts of different classes of impacts on the floodplain vegetation
reliability indicator F V C
rel
under all analysed scenarios in NRB sub-
regions (pie diameter is proportional to the number of reaches in a
sub-region).
130
4.20 Pie charts of different classes of impacts on the floodplain vegetation
resilience indicator F V C
res
under all analysed scenarios in NRB sub-
regions (pie diameter is proportional to the number of reaches in a
sub-region).
131
4.21 Pie charts of different classes of impacts across all analysed reaches
and EF IIs for eight model experiments.
133
x
List of Tables
2.1
Basic socio-economic statistics for the NRB sub-regions in 2009 (GUS,
2011).
18
2.2
Gross Value Added (GVA) by type of activity and sub-regions as a
percentage of total GVA in 2009 (GUS, 2011).
19
3.1
GIS and monitoring datasets used to build the SWAT project in Piniewski
and Okruszko (2011).
24
3.2
Land use/land cover classes used as SWAT input.
26
3.3
Flow gauges used in calibration and validation.
29
3.4
Two example crop calendars applied in SWAT for spring wheat and
meadows.
32
3.5
Selected calibration parameters and their ranges.
39
3.6
Default and modified values and correction coefficients of maximum
stomatal conductance GSI and maximum leaf area index BLAI for
generic land cover types corresponding to the assumed CO
2
increase
by 1.48 for SRES A2 scenario.
45
3.7
Summary of storylines developed during the PA workshops.
54
3.8
Driving forces selected for converting storylines into model scenarios.
58
3.9
Stakeholder types participating in the fourth PA workshop.
59
3.10 Seven-level scale used for assigning qualitative changes to driving forces.
60
3.11 Summary of qualitative changes in driving forces for two scenarios.
62
3.12 Translation key developed using the MOM method.
64
3.13 Percentage of types of wetland plant communities in the NRB.
74
3.14 Mean values of the coefficient k used in the parametric version of the
Kostrzewa method of determining hands-off flows.
79
3.15 Categories of floodplain vegetation communities (FVC) determined
with respect to the optimal duration of inundation.
82
xi
3.16 Characteristic values of trapezoidal membership functions for reaches
with different inundation categories.
90
3.17 Summary of proposed reliability and resilience indicators.
91
3.18 Summary of proposed EF IIs.
92
3.19 Definition of thresholds for colour coding of the EF IIs representing
reliability: HOF
rel
, P IKE
rel
and F V C
rel
.
93
3.20 Definition of thresholds for colour coding of the EF IIs representing
resilience: HOF
res
, P IKE
res
and F V C
res
.
93
3.21 Experimental design for running scenario simulations in SWAT.
94
4.1
Summary of SUFI-2 results (cf. Table 3.3 and Figure 3.5 for gauge
codes and locations).
96
4.2
Summary of goodness-of-fit measures obtained, using the PSO tech-
nique in different calibration and validation areas during the calibra-
tion and temporal validation phases. Cases with
N SE < 0.4, R
2
< 0.5 or
|P BIAS| > 25
are marked in red.
99
4.3
Technical comparison of calibration processes conducted with the "old"
model set-up and with the "new" set-up.
109
4.4
Mean bias (i.e. difference between modelled and observed) and mean
absolute error of estimation of reliability and resilience indicators based
on validation with observed data from 18 gauging stations (cf. Tab.
3.17 for indicator description).
115
4.5
Mean annual water balance components as differences from the baseline
for all model experiments.
119
4.6
Decomposition of the model experiments.
122
4.7
Percentage of main land use types in the regional scenarios (as a per-
centage of total NRB area).
136
xii
Chapter 1
Introduction
1.1 Background
Globally, freshwater ecosystems are considered to be under severe threat from human
pressure and climate change (
Vörösmarty et al.
,
2010
).
Malmqvist and Rundle
(
2002
)
suggest that running water is the most impacted upon ecosystem on Earth due to
being surrounded by dense human settlements and exploited for domestic and indus-
trial water supply, irrigation, electricity generation and waste disposal. For example,
the progressive over-exploitation of surface water resources for irrigation and urban
uses in the Colorado River Basin has resulted most years in no runoff reaching the
river's delta (
Gleick
,
2003
).
Many aquatic ecologists perceive the flow regime to be the key driver of river
and floodplain wetland ecosystems (
Bunn and Arthington
,
2002
). Poetically, flow has
been praised as "the maestro that orchestrates pattern and process in rivers" (
Walker
et al.
,
1995
). Even if there is some exaggeration in this statement, there is indeed
enough evidence to support the role played by the natural flow regime in maintaining
river-dependent (i.e. aquatic and riparian) ecosystems in good health (
Junk et al.
,
1989
;
Poff et al.
,
1997
;
Richter et al.
,
1997
), as well as for the overwhelmingly negative
responses of these ecosystems to flow alterations (
Bunn and Arthington
,
2002
;
Poff
and Zimmerman
,
2010
).
The causes of flow alterations can be manifold. A comprehensive review of
Palmer et al.
(
2009
) reported multiple sources of stress for rivers and river-dependent
ecosystems. Stressors are commonly divided into natural (i.e. climatic) and human-
induced (anthropogenic). Global warming, driven by increased greenhouse gas emis-
sions, has been observed for decades and reported in various global (
IPCC
,
2001
,
2007
) and Polish (
Zmudzka
,
2009
;
Marszelewski and Skowron
,
2006
;
Maksymiuk
1
et al.
,
2008
) studies.
Palmer et al.
(
2009
) underlined that anticipating future condi-
tions of a river and a river-floodplain ecosystem in the face of climate change depends
largely on geographical location, as natural flow regimes vary by river size and by
differences in climate, geology, topography and vegetative cover (
Poff et al.
,
1997
),
as well as on the types of human pressures present in the river basin. In particular,
these studies feature the following human activities: water withdrawals for multiple
uses, damming rivers and land use changes. Furthermore,
Blöschl et al.
(
2007
) un-
derlined the importance of spatial scale: any impact of land cover change is likely to
decrease with catchment scale, whereas impacts of climate change are supposed to
be scale-invariant and consistent in a region.
Multiple examples showing global and regional environmental change nat-
urally raise concern about the future. A significant amount of effort from the sci-
entific community in recent years has been devoted to modelling the future. In
water resources, conducting this research has been partly enhanced by various stra-
tegic water policy documents. For example, the Water Framework Directive of the
European Union (
EU
,
2000
) imposed a timetable with specific requirements to be met
by European water bodies. Three six-year long cycles for river basin planning have
been implemented, and 2027 is the final year for meeting these objectives. As it was
noted by
van Griensven et al.
(
2006a
) that using catchment-scale hydrological models
is the most appropriate reference framework for WFD-oriented integrated modelling,
as watersheds form the physical borders for river basin management. These models
always have some semi-physical or physical representation of runoff generation pro-
cesses. They are applied in hydrology for different purposes, but one commonly used
model application type involves employing them to study the hydrological effects of
changes in various driving forces. Models are applied at all spatial scales: from hill-
slopes (
Ambroise et al.
,
1996
), small catchments (
Zehe et al.
,
2001
) and large river
basins (
Barthel et al.
,
2005
) through to continental and global applications (
Alcamo
et al.
,
2003
). The larger the scale, the more important the spatial heterogeneity of the
study area. Hence, models are nowadays often coupled with Geographical Informa-
tion Systems (GIS), which facilitate their use to a large extent. Distributed models
vary with respect to discretisation strategy: from fully-distributed, grid-element-
based models, such as Système Hydrologique Europèen (
Abbott et al.
,
1986
) and its
successors such as SHETRAN (
Bathurst et al.
,
1995
) and MIKE SHE (
Refsgaard
and Storm
,
1995
), to semi-distributed models built on the concept of hydrological
similarity, such as TOPMODEL (
Beven
,
2002
) or SWAT (
Arnold et al.
,
1998
).
As far as modelling used to quantify future impacts is concerned, a point
has to be made about assumptions regarding changes in the driving forces and time
horizons of anticipated changes. In this thesis the focus is only on long-term effects,
2
even reaching beyond the WFD time-line, namely the 2050s. Hence, it would be
naive to expect that extrapolating recently observed trends, be they in climate or
land use, would provide a meaningful way of unfolding such a distant future. In
the case of modelling the impacts of climate change, a well-established quantitative
method for estimating them involves employing projections from General or Regional
Circulation Models as the input for hydrological models. How to do it precisely is
a non-obvious question, though (
Fowler et al.
,
2007
). In the case of modelling an-
thropogenic change, there are undoubtedly more options, depending on the type of
driving force considered: projections of population growth (
Sun et al.
,
2008
), land use
change models (
Verburg et al.
,
2002
) and approaches involving stakeholders (
Alcamo
,
2008
). The latter seem to be particularly tempting, given the role that stakeholder
involvement has been given in European water resources planning by the WFD. In-
stead of trying to predict the future, it is better to sketch out "alternative futures",
the likely results of different choices (
Kundzewicz and Kindler
,
1995
). Such altern-
ative water futures have recently been developed for the Western Bug River Basin
(
Schanze et al.
,
2012
) and for the Narew River Basin (
Gielczewski et al.
,
2011
).
The relative significance of different driving forces for the flow regime is
spatially- and scale-dependent. Climate has appeared to be a stronger driver for
hydrological change than land use in several case studies in the United States (
Tu
,
2009
;
Sun et al.
,
2008
;
Choi
,
2008
;
Chang
,
2003
), South Korea (
Park et al.
,
2011
)
and Switzerland (
Wolf et al.
,
2012
). Only in the paper of
Tong et al.
(
2012
) was
the magnitude of impacts (in this case on mean annual runoff) comparable between
these two stressors; however, under the driest and the wettest climate change scen-
arios the impacts were higher than under the land use change scenario. Undoubtedly,
the results in all modelling efforts depend on scenario assumptions and, as mentioned
previously, on geographical setting. For this reason it is not recommended to gener-
alise findings from other studies. To the author's knowledge, no Polish publications
have attempted to weigh the projected long-term effects of natural and anthropogenic
driving forces on the flow regime. This thesis therefore seeks to bridge this gap.
Hereafter, natural and anthropogenic driving forces will be referred to as
global and regional driving forces, respectively. The future effects of these forces up
to the 2050s will be assessed in quantitative scenarios implemented in a hydrological
model. It is believed that using this nomenclature (i.e. global and regional instead of
natural and anthropogenic) better reflects considered environmental stressors, since
global-scale driving forces will include not only climatic change but also changes
in CO
2
, atmospheric carbon dioxide and plant physiological parameters, whereas
regional-scale driving forces will include changes in land use, agriculture development
and agricultural water management. Hence, the difference is that the first group of
3
driving forces acts globally and independently on the study area, whereas the second
group includes factors that are specific to the study area. Furthermore, in order
to expand on the title of this thesis, impacts in the present study will be assessed
not only on the flow regime as such, but also on its ecological functions, i.e. on
the environmental flow regime. This is motivated mainly by the the semi-natural
character of the study area, that is unique in Poland and in Europe, but it also
underlines the novelty of this thesis, as going beyond the pure impacts on the flow
regime in a scenario-modelling framework is rare in hydrological science, if achieved
at all.
1.2 Objective
The general aim of this thesis is to assess the impact of a multiple set of scenarios,
describing changes of global- and regional-scale driving forces on the ecological func-
tions of the flow regime. This general aim translates into three more specific research
objectives:
1. A comprehensive spatial validation of a large-scale, semi-distributed hydrolo-
gical model, to test its ability to assess spatial patterns in impacts on flow
regime indicators.
2. A broad description of the water requirements of selected river-dependent biota
and their parametrisation into indicators derivable using hydrological model
output.
3. A spatially-distributed analysis of future impacts on environmental flow indic-
ators under various model scenarios, derived from key local stakeholders' input,
as well as based on downscaled projections of global change.
The research undertaken to fulfil these objectives was carried out as part of a case
study of the Narew River Basin, NE Poland. There were two main reasons for choos-
ing this basin. Firstly, it is a semi-natural (near-pristine) river basin, where river
water use for domestic supply, industry and agriculture is minor compared to other
river basins of similar size in Poland. Secondly, this river basin was selected as the
pilot area catchment in the EU FP6 research project SCENES (Water Scenarios for
Europe and Neighbouring Countries), which was one of the motivations for this re-
search. Although it is believed that the findings of this thesis are important, especially
for the water management of the Narew River Basin, various approaches developed
and adapted in this thesis are not limited to this area only but are applicable else-
where.
4
Figure 1.1: Conceptual schematic overview of this thesis.
Figure
1.1
illustrates a schematic conceptual overview of this thesis. Driving
forces acting at both global and regional level influence catchment water balance,
so scenarios of changes in driving forces can be considered either separately (single
arrows in Figure
1.1
) or as combined (joining arrows). Changes in catchment water
balance imply flow alterations, which have environmental consequences for ecosystems
that depend on river water (i.e. aquatic/riparian biota). The parametrisation of
the water demand for particular biota enables one to develop a number of impact
indicators that can be quantified using altered flow time series. In this way future
impacts on the environmental water requirements of river-dependent biota are studied
in this thesis.
1.3 Scope
With respect to Figure
1.1
, illustrating the general concept of this thesis, it is clear
that the application of a distributed, physically-based, catchment-scale hydrological
model suitable for addressing water resource problems is, although probably not the
only one, certainly a good method for fulfilling the main objective. Nowadays numer-
ous hydrological models meeting the above-mentioned requirements exist, many of
which would be adequate tools in this thesis; consequently, at the preliminary stage
5
one tool had to be chosen. The Soil Water Assessment Tool (SWAT) was selected
due to several key features:
· Global popularity expressed by numerous published worldwide applications in
hydrological impact assessment studies;
· Its full integration with Geographical Information Systems (GIS);
· Being a public domain model;
· Using readily available input data.
An additional stimulus for applying SWAT was its relatively low popularity in Po-
land.
Figure
1.2
is an expansion of Figure
1.1
, schematically illustrating the scope
of this thesis. Five main zones are featured in different colours in this diagram,
each related to a specific part of the thesis. The diagram has four starting points
(numbered from 1 to 4 in Figure
1.2
) expanding into chains of boxes representing
different steps/processes/data and arrows representing conceptual flow. The scope of
this thesis is discussed below with reference to the diagram in Figure
1.2
.
The first (blue) chain begins with the set-up of the SWAT model, followed
by model calibration and validation. This chain is a prerequisite because only a
model which has passed calibration and validation criteria can be applied in scenario
analyses. Furthermore, due to the basin size, preparation the set-up of the model
is time-consuming, as the spatial calibration and validation approach is relatively
complex. A general description of SWAT, its set-up for the Narew River Basin and
the tools, approaches and parameters used for model calibration constitute section
3.1
, whereas the calibration and validation results belong to section
4.1
.
The second (red) chain illustrates the process involved in developing global
scenarios. In this thesis the term "global scenarios" refers to projected changes in
atmospheric CO
2
levels and climate. Very often only climate variables are altered
in climate change impact modelling studies focused on water resources, whereas,
in reality, climate change projections of General Circulation Models (GCMs) are
forced by elevated CO
2
concentrations according to different greenhouse gas (GHG)
emission scenarios. Furthermore, increased CO
2
is known to alter plant physiology,
which can also be considered in SWAT. Downscaling and bias correction of GCM
output (precipitation and temperature projections) are necessary steps preceding the
application of climate change data in a catchment model. Section
3.2
describes all
the data and methods used in developing global scenarios in this thesis.
The points of departure in the third (green) chain are SCENES scenarios,
namely Sustainability Eventually (SuE) and Economy First (EcF), which were de-
6
!
# $
!
%
' ( )(
*
+
,
-
'
) *
,
' (
( %
' .
/
'
'
)% *
) *
0 )% % .*
#( $
%
.
1
2
!
.
3 2 1
2 ' '
!
'
0
Figure 1.2: Schematic illustration of the scope of this thesis.
7
veloped using mainly qualitative and semi-quantitative scenario development meth-
ods during a series of stakeholder workshops carried out within SCENES. The green
chain illustrates the development of regional model scenarios. In this thesis the term
"regional scenarios" is understood as scenarios describing future changes in selected
driving forces specific to the Narew River Basin: land use, agricultural development
and the development of agricultural water management (drainage and irrigation).
The process of transferring qualitative knowledge (SCENES scenarios) into SWAT
model scenarios (modified parameters) was conducted with active stakeholder in-
volvement. In the first stage a qualitative description of driving forces for both
scenarios was produced by stakeholders in groups, which was followed by filling in
individual questionnaires asking about the quantitative meanings of previously de-
scribed changes. Post-processing of these data enabled the computation of trends
in driving forces, and finally, the modification of adequate SWAT input parameters.
The process involved in developing regional scenarios is described in section
3.3
, with
several supporting materials placed in the appendices.
The first three chains in Figure
1.2
end up in the "Validated SWAT" box.
In contrast, the fourth (purple) chain does not depend on SWAT. It originates with
a broad, mainly literature-based description of the water requirements of selected
representatives of the aquatic and riparian environment, namely fish and floodplain
vegetation communities. The process of quantifying these requirements is carried out
using the adapted building block approach, whereby three blocks are distinguished:
one for maintaining minimum (in-stream) flows, one for providing a spawning and
nursery habitat for pike and one for maintaining floodplain vegetation communities
in good health. Environmental flow requirements of selected biota and the applied
building block approach are described in section
3.4
.
The last (black) chain illustrates how the final results of this thesis are
achieved.
A juxtaposition of water supply (modelled flows) with water demand
(building blocks) enables the design of a set of indicators. Two types of indicat-
ors are distinguished: state indicators and impact indicators. State indicators show
the current state of a phenomenon under study, while impact indicators measure
changes to a phenomenon under study caused by a certain impact. The main fo-
cus of this thesis is on impact indicators, which are calculated by combining both
altered (under a given scenario) and natural (baseline) modelled flow time series with
the building blocks. Definitions of developed state and impact indicators belong to
section
3.5
, whereas the final results following the calculation of the Environmental
Flow Impact Indicators are given in section
4.3
. These indicators are spatially ex-
plicit, biota-specific and scenario-specific. In particular, they are quantified for all
possible combinations of global and regional scenarios.
8
Chapter 2
Study Area
The River Narew, situated in the north-east of Poland (Fig.
2.1
), is the right tributary
of the River Vistula, and the total drainage area upstream from its mouth is ca.
75,000 km
2
. However, in this thesis a part of the basin situated upstream of the
Zambski Kocielne gauging station is analysed. This part of the Narew River Basin
(which hereafter will be referred to as the NRB) is situated between the meridians
of longitude 20°21'E and 24°27'E and between the parallels of latitude 52°35'N and
54°16'N and occupies ca. 28,000 km
2
. The motivation for selecting Zambski Kocielne
as the main outlet is that it is the most downstream gauging station beyond the
reach of the artificial Lake Zegrzyskie's backwater effects. Approximately 4% of
the drainage area, in its upstream part, lies in western Belarus, i.e. outside of the
territory of the Republic of Poland.
River basin physiography (geology, topography, hydrography, soils, land cover,
climate and hydrology) is discussed in section
2.1
, while socio-economic characterist-
ics are given in section
2.2
.
2.1 Physiography
The NRB lies in the western edge of the East European Plain, the largest mountain-
free part of the European landscape. Present lowland relief of the NRB was shaped
by two last glaciation periods: Riss (Oder) and Würm (Vistulian), using Alpine
glaciation nomenclature. As shown in the map of geological units of the NRB (Fig.
2.2
) three classes occupy over 76% of the Polish part of the NRB. Tills, weathered
tills, glacial sands and gravels occupy ca. 31% of this area, covering most of the part
north of last glacial maximum line. The second largest class (ca. 29%) is outwash
sands and gravels, predominating especially in the extent of the Pomeranian phase
9
0
25
50
Km
Ø
ZAMBSKI
KOCIELNE
Orz
y
c
Om
ulew
Narew
Pi
sa
Sup
ral
Bi
eb
rza
Narew
Vistula
Bug
Bialystok
Loma
Elk
Ostrolka
Szczytno
Augustów
Bielsk Podlaski
Nar
ew
Lake
Zegrzyskie
24° E
23° E
22° E
21° E
20° E
54°
N
53°
N
Baltic
Sea
Od
er
Vistu
la
Byelarus
Czech
Republic
Germany
Lithuania
Slovakia
Poland
Russia
Ukraine
Warsaw
Ø
Main outlet
Country borders
Lakes
Main rivers
Cities with pop. 25,000
Natura 2000 SACs
Natura 2000 ASCs
Narew River Basin
0
100 200
Km
Figure 2.1: Map of the study area.
of Vistulian glaciation limit. Finally, fluvial sands, gravels, muds, peats and organic
silts are the third largest class covering ca. 17% of the NRB in Poland, filling the
river valley bottoms, in particular the Biebrza ice-marginal valley.
The NRB can be characterised by a mean altitude of 137 m.a.s.l. and a flat
topography (Fig.
2.3
) spatially correlated with geological features. The highest elev-
ated areas in the north correspond to till- and kame sand-covered end moraine hills,
such as the Szeskie Hills, with an elevation reaching 307 m.a.s.l. and a topographic
prominence surpassing 100 m. However, 99% of the drainage area lies below 200
m.a.s.l., while the mean basin slope calculated based on the contour DEM shown in
Figure
2.3
is equal to 2.1%.
The basin's present hydrographic network reflects its geology and relief,
which is fully developed and rich in tributaries, many of which are right-bank tribu-
taries draining post-glacial lakes located in the northern part of the basin (Fig.
2.3
).
The drainage area of these right bank tributaries is four times larger than that of
their left bank counterparts, which is as a result of the direction of ice sheet reces-
sion in the late Pleistocene. In the Lake District region there are more than 500
lakes greater than 1 ha, with the largest, Lake niardwy, occupying 102.4 km
2
. The
mean channel slope along the 363 km-long stretch of the River Narew, downstream
of the Siemianówka dam and upstream of Zambski Kocielne, is equal to 0.16 m/km,
which is a low value even for lowland rivers. Other important landscape features of
10
0
25
50
Km
Bug
NRB
Glaciation extent lines
Last Glacial Maximum
The Pomeranian Phase of Vistulian Glaciation limit
Main geological units
End moraine gravels, sands, boulders and tills
Eolian sands, locally in dunes
Fluvial sands, gravels and silts
Fluvial sands, gravels, muds, peats and organic silts
Ice-dam clays, silts and sands
Kame sands and silts
Lakes and main rivers
Outwash sands and gravels
Tills, weathered tills, glacial sands and gravels
Figure 2.2: Map of geological units and glaciation extent in the NRB (source: Geo-
logic Map of Poland from the Polish Geological Institute).
Orz
y
c
Om
ule
w
Pi
sa
Sup
ral
Bi
eb
rza
Narew
Bug
Rivers
Lakes
m asl
307
194
81
0
25
50
Km
Figure 2.3: Topography and hydrographic network of the NRB (hydrographic source:
Digital Map of Hydrological Division of Poland from the Institute of Met-
eorology and Water Management).
11
0
25
50
Km
Soil type
Sand
Loam
Peat
Water
NRB
Figure 2.4: Generalised soil types in the NRB (after
Gielczewski
,
2003
).
the NRB valley bottoms are large floodplains, very often still in connection with the
main channel.
Sandy soils of various types, but usually pure and loamy sands, dominate
in the NRB (Fig.
2.4
). Loamy soils that also exist are typically sandy loams; very
heavy impermeable soils (clay, clay loam, silt loam) are rare in the landscape. The
main valley bottoms were filled with peat deposits from the Holocene, and they are
still partly undrained at present.
Agriculture dominates land use in this area: 46% of land is used as arable
land and 17% as grassland, whereas 33% is occupied by forests (Fig.
2.5
). The
remaining 4% of land is covered by wetlands, lakes and urban areas. The upland of
the basin area is mainly used for growing crops, and the valley bottoms are used as
permanent meadows and pastures. The largest and most compact forest complexes
are located in the northern, western and eastern parts of the basin, while wetland
areas are concentrated mainly in the Biebrza valley bottom.
The NRB is the core part of the region known as the "Green Lungs of Poland"
(GLP), an initiative that was launched as an inter-province agreement aiming to
support eco-development in this exceptional region. There are three national parks
(ca. 750 km
2
), protecting wetland and forest ecosystems, and a number of other
protected areas, in particular Special Areas of Conservation (SACs, protected under
the EC Habitats Directive) and Special Protection Areas (SPAs, protected under the
12
0
25
50
Km
Bug
NRB
Land cover
Arable land
Forests
Grassland
Urban areas
Water
Wetlands
Figure 2.5: Land cover map of the NRB (source: CORINE Land Cover 2000 from
the Chief Inspectorate of Environment Protection).
EC Birds Directive) of the NATURA 2000 network (Fig.
2.1
).
The NRB is located in a temperate climatic zone in which marine and contin-
ental air masses collide. The latter shape the area's climate to a larger extent than in
central and western Poland. This region has moderately warm summers (July's mean
temperature equal to 18°C) and cool winters (January's mean temperature equal to
-2°C). Figure
2.6
illustrates spatial variability in mean annual January (A.) and July
(B.) temperatures, which was interpolated using the Local Polynomial method in
ArcGIS (only for demonstration purposes), based on data from 14 stations belonging
to the monitoring network of the Polish Institute of Meteorology and Water Man-
agement (IMGW). The rise in temperature can be observed in both cases from NE
to SW, in general. However, the spatial pattern in winter is more subject to cold
continental air and to the warming effect of lakes in the north, whereas the pattern
in summer is correlated more with latitude. Spatial variability in temperature has
a clear impact on the duration of the growing season, which will be discussed in
subsection
3.1.2
.
The annual basin-averaged precipitation yields ca. 600 mm, of which 35%
falls between June and August, and only 17% between January and March. Figure
2.7
illustrates spatial variability in mean annual precipitation, which was interpolated
using Radial Basis Functions method in ArcGIS (only for demonstration purpose),
based on data from 78 stations from the IMGW network. The lowest amounts of
13
Bug
-1.6
-1.8
-1
.8
-2
-2.2
Stations
Contours
Mean Jan. T
deg. C
-1.3
-2.7
Stations
Contours
Mean Jul. T
deg. C
18.8
17.8
A.
B.
Bug
18.6
18.4
18
18
.2
18.2
0
25
50
Km
Figure 2.6: Map of the interpolated mean January (A.) and July (B.) temperatures
for 1989-2008.
14
0
25
50
Km
550
525
650
625
62
5
55
0
62
5
600
575
575
600
55
0
600
62
5
60
0
57
5
600
60
0
55
0
650
6
25
625
55
0
575
60
0
Stations
Contours
Annual PCP
Value
703
604
505
Figure 2.7: Map of interpolated mean annual precipitation for 1989-2008.
annual precipitation are observed in the southern and central part of the basin. Two
directions of spatial variability can be observed: from south to north and from centre
to east. The highest precipitation is generally measured in the lake district in the
north. Apart from general spatial trends, several smaller scale features of mean annual
precipitation variability can be found, which proves the importance of maintaining
dense network of precipitation gauging stations. Visual comparison of Figure
2.7
with DEM in Figure
2.3
demonstrates that mean annual precipitation is correlated
to elevation in the NRB and elevation may sometimes explain small-scale variations.
Figure
2.8
shows discharge hydrographs at the Zambski Kocielne gauging
station for two example hydrological years, specifically a wet year in 1994 and a dry
year in 2003. In both cases hydrograph peaks are associated with snow-melt that
usually occurs in early spring or during warmer spells of winter. The magnitude of
floods can vary to a large extent between dry and wet years. Since evapotranspiration
is the dominant process in summer, floods occur very rarely in this season, even after
heavy rainfall events. Hence, the period between July and September is typically the
low flow period. The hydrographs shown in Figure
2.8
represent the whole basin,
and even though the flow regime of this area is rather uniform, some local variations
certainly exist. One of the main differentiating features is intra-annual flow variability
(or flashiness of the flow regime), which can be measured by the coefficient of daily
flow variation (CV , i.e. standard deviation divided by mean flow). For example, CV
for the River Pisa is equal to 0.37 at the Pisz gauge, whence CV for the upstream
15
Figure 2.8: Observed daily discharge hydrographs at the basin outlet (Zambski Ko-
cielne station on the Narew) during a wet year (1994) and a dry year
(2003).
part of the River Biebrza is equal to 1.31.
2.2 Population and economy
The Polish part of the NRB lies within three different provinces Podlasie, Warmia-
Masuria and Mazovia (Fig.
2.9
). The largest part of ca. 52% of the drainage basin
lies within Podlasie Province, followed by Warmia-Masuria Province and Mazovia
Province occupying ca. 28% and 20% of the area, respectively. A further subdivision
is provided by EUROSTAT for statistical purposes via NUTS (The Nomenclature of
Territorial Units for Statistics) at level 3. Six NUTS 3 sub-regions belong to the Polish
part of the NRB (actually there are eight of them, but two occupy a negligible por-
tion): Bialostocki, Lomyski, Suwalski, Elcki, Olsztyski and Ostrolcko-Siedlecki.
These six sub-regions cover a total area of 49,000 km
2
, which is larger than the ca.
27,000 km
2
of the Polish part of the NRB; however, they reasonably approximate its
present socio-economic conditions in terms of socio-economic statistics.
Table
2.1
specifies several socio-economic indicators for the NRB sub-regions
and combines them with similar statistics for Poland as for 2009 (
GUS
,
2011
). The
16
0
25
50
Km
Bialystok
Loma
Elk
Ostrolka
Szczytno
Augustów
Bielsk
Podlaski
Na
re
w
m a z o w i e c k i e
m a z o w i e c k i e
p o d l a s k i e
p o d l a s k i e
w a r m i s k o -
w a r m i s k o -
m a z u r s k i e
m a z u r s k i e
o s t r o l c k o - s i e d l e c k i
b i a l o s t o c k i
l
o
m
y
s
k
i
s u w a l s k i
o l s z t y s k i
e
l
c
k
i
Provinces
mazowieckie
podlaskie
warmisko-mazurskie
Cities with pop. 25,000
Sub-regions
NRB
Figure 2.9: Administrative subdivision in the NRB: NUTS 2 (provinces) and NUTS
3 (sub-regions) levels.
NRB area belongs to the most poorly populated Polish regions. Around 1.6 million
inhabitants are estimated to live in this region, at an average of 59.2 people per km
2
,
with three sub-regions below 50 people per km
2
, while in the rest of the country
average population density is more than twice as high at up to 122 people per km
2
.
More than half of the population (56.9%) live in urban areas, which is approximately
4.1% less than in the whole country. However, this statistic is over-estimated due to
the inclusion of the city of Olsztyn, which is outside the NRB. The largest city of
the NRB is Bialystok, the capital of Podlaskie province, with 285,000 inhabitants.
Other cities are definitely smaller and none of them exceeds 70,000 inhabitants (cf.
Fig.
2.1
), which means that ca. 69% of the urban population live in towns and
cities smaller than 70,000 inhabitants. All cities have sanitary sewerage systems,
transporting effluent to wastewater treatment plants, and storm drainage systems
that drain off precipitation water to the nearest receiving water source. In most of
the cities the sewerage network is distributive. Enterprises situated in rural areas
usually have their own effluent treatment plants (
Okruszko et al.
,
2012
).
The unemployment rate as of February 2012 in the NRB was around 4%
higher than the average for Poland at 13.5%. It was particularly high in the Elcki
sub-region, yielding 25.6%. GDP per capita for 2009 was also considerably (by 28.4%)
lower than the Polish average. This indicator is clearly correlated with the percentage
of the urban population. It is worth noting that the sub-regions Elcki, Lomyski
17
Table 2.1: Basic socio-economic statistics for the NRB sub-regions in 2009 (
GUS
,
2011
).
Sub-region
Area [km
2
]
Weight [%]
a
1
b
2
b
3
b
4
b
Bialostocki
5,144
17.7
98.2
75.0
15.1
30.4
Lomyski
8,838
19.2
46.3
46.1
13.5
22.5
Suwalski
6,233
15.3
44.3
54.5
16.9
23.1
Elcki
6,359
17.3
44.7
58.0
25.6
21.5
Olsztyski
10,339
11.6
59.4
61.6
19.5
29.0
Ostrolcko-Siedlecki
12,065
18.6
61.9
49.6
15.9
26.2
NRB approximation
100
59.2
56.9
17.5
25.2
Poland
312,685
122
61.0
13.5
35.2
a
weight is defined here as the ratio of the area of the sub-region part overlapping with the NRB to the area of the
NRB overlapping with all-subregions; this serves to calculate the "NRB approximation".
b
1 - Population density (people per km
2
);
2 - Urban population (%); 3 - Unemployment rate (%) as of February
2012;
4 - GDP per capita (thous. PLN).
and Suwalski are ranked 66th, 57th and 54-th, respectively, with respect to GDP per
capita across 66 Polish sub-regions. In contrast, the Bialostocki sub-region is ranked
24th.
Table
2.2
shows Gross Value Added (GVA) by type of activity in the NRB's
sub-regions. The most apparent feature of this region is its extraordinarily high
share of agriculture, forestry and fishing (in short: agriculture) industries compared
to country-wide data (11.7% in the NRB vs. 3.6% in Poland as a whole). This is
compensated mainly by a lower share in non-manufacturing industries (e.g. mining,
energy production and supply) and in a part by the services sector (trade, repair of
motor vehicles, etc. cf. Tab.
2.2
). The chief crops are cereals, potatoes and fodder,
and cattle breeding is important. The region is not rich in mineral resources and
there are no large heavy industry factories. Textiles, food processing and timber
are the major manufacturing industries, and the food industry is probably the most
dynamically developing branch in this region, with a growing number of factories
producing milk, meat, poultry, cereals, vegetables, fruits and beer (creameries also
being the most important at national level). Recently observed tourism development
has taken place, due to extraordinary natural values rather than to the development
of infrastructure (
Gielczewski et al.
,
2011
).
The presented statistics prove that the NRB distinguishes itself from the
rest of the Republic of Poland with respect to its socio-economic development, usu-
ally in a rather negative way.
Wójcik
(
2008
) analysed sub-regional and temporal
(1995-2005) variability of GDP per capita in Poland, demonstrating that the highest
economic growth was observed in this period for metropolitan areas. The sub-regional
diversification of relative GDP per capita grew between 1995 and 2005. The poorest
sub-regions in the country (including Lomyski, Suwalski, Elcki and Ostrolcko-
18
Table 2.2: Gross Value Added (GVA) by type of activity and sub-regions as a per-
centage of total GVA in 2009 (
GUS
,
2011
).
Sub-region
1
a
2
a
3
a
4
a
5
a
6
a
Bialostocki
5.0
18.4
15
7.6
31.6
37.4
Lomyski
16.1
19.3
16.5
7.4
25
32.1
Suwalski
14.4
22.4
19.9
6.6
24.4
32.3
Elcki
10.5
19.8
16.3
8.7
25.3
35.8
Olsztyski
6.3
21.8
18.6
7.9
27.9
36.1
Ostrolcko-Siedlecki
15.8
18.8
14.6
8.2
22.2
35
NRB approximation
11.7
19.8
16.6
7.7
25.9
34.6
Poland
3.6
24.6
18.7
7.8
30.1
33.9
a
1 - Agriculture, forestry, fishing; 2 - Industry (total); 3 - Industry (manufacturing); 4 - Construction; 5 - Trade,
repair of motor vehicles, transportation and storage, accommodation and catering, information and communication;
6 - Other services.
Siedlecki sub-regions) showed relative impoverishment due to the faster than normal
growth of the richest sub-regions (mainly Warsaw, Pozna and Kraków).
The advantage of lower economic development is generally lower pressure
on water resources than in the rest of the country, and hence better chances for
achieving a yet more sustainable status in the future.
Kozlowski
(
2006
) stated that
the European Union's strategy of sustainable development is largely based on the
creation of large spatial nature structures and Poland, most notably the GLP area,
plays an important role in the implementation of this strategy.
19
20
Chapter 3
Materials and Methods
In this chapter the materials used and methods applied to reach the objectives out-
lined previously are described. Section
3.1
presents basic features of the SWAT model,
as well as the model's set-up and calibration approaches. The following two sections
describe global change scenarios (section
3.2
) and regional changes (section
3.3
). In
section
3.4
an environmental flow method based on the building block approach is
outlined, which is then followed in section
3.5
by an introduction to state and impact
indicators measuring different aspects of providing sufficient amounts of water to the
environment.
3.1 SWAT model
Piniewski and Okruszko
(
2011
) undertook initial work to calibrate and validate the
hydrological component of SWAT for the NRB, by using multiple gauges and evaluat-
ing its ability to simulate flows on small spatial scales. In this thesis this preliminary
work is further developed by making several essential changes to the previously con-
structed model set-up, as well as the calibration and validation approaches. This
section begins with a description of basic SWAT model features and modelling con-
cepts in subsection
3.1.1
, followed by a summary of the model set-up developed by
Piniewski and Okruszko
(
2011
) and a description of its newer features in subsection
3.1.2
. Next, the tools applied for auto-calibration and uncertainty analysis are presen-
ted in subsection
3.1.3
, and the approach used for the calibration and validation of
SWAT for the NRB is described in subsection
3.1.4
. Finally, the parameters used in
the calibration process are presented in subsection
3.1.5
.
21
Figure 3.1: Visualisation of HRU delineation from an overlay of land use and soil
maps within sub-basins.
3.1.1 Model features
SWAT is a public domain, river basin scale model developed to quantify the impact
of land management practices in large, complex river basins (
Arnold et al.
,
1998
).
SWAT2009 rev. 481 model version (
Neitsch et al.
,
2011
;
Arnold et al.
,
2011
) un-
der ArcSWAT 2009.93.7 (
Winchell et al.
,
2010
), an ArcGIS-ArcView extension and
graphical user input interface for SWAT, was used in this study. SWAT is a phys-
ically based, semi-distributed, continuous time model that operates on a daily time
step and simulates the movement of water, sediment, nutrients, pesticides and bac-
teria on a catchment scale. The river basin can be partitioned into a desired number
of sub-basins based on the Digital Elevation Model (DEM) or optionally a stream
network layer to burn-in streams and a threshold that defines the minimum drainage
area required to form the origin of a stream. The smallest unit of discretisation is a
unique combination of land use, soil and slope overlay, referred to as a "hydrological
response unit" (HRU; Fig.
3.1
). Runoff is predicted separately for each HRU, and
then aggregated to the sub-basin level and routed through the stream network to the
main outlet, in order to obtain the total runoff for the river basin. One of the most
important features of SWAT is the fact that HRUs are lumped within sub-basins.
Key processes associated with the land phase of the hydrological cycle included in
SWAT can be found in (
Arnold et al.
,
1998
;
Neitsch et al.
,
2011
):
22
1. Snow-melt estimated using the degree-day method;
2. Surface runoff estimated using the modified SCS curve number method or the
Green-Ampt infiltration equation;
3. Lateral subsurface flow estimated using a kinematic storage model;
4. Redistribution in soil modelled using a layered storage routing technique;
5. Groundwater storage modelled using two aquifers: shallow unconfined and deep
confined;
6. Groundwater flow to streams from shallow aquifers modelled using a recession
constant method;
7. Potential evapotranspiration by the Hargreaves, Priestley-Taylor or Penman-
Monteith method;
8. Actual evapotranspiration (separate evaporation from soil and transpiration
from plants).
Main processes of the hydrological cycle's routing phase represented in SWAT are:
1. Channel routing modelled with the variable storage coefficient or the Muskingam
river routing method;
2. Transmission losses from streams;
3. Water storage and losses from ponds and reservoirs.
Figure
3.2
presents the schematic pathways of water movement between different
storage forms as conceptualised in SWAT.
In this thesis the hydrological component of SWAT is applied solely (i.e. the
water quality component is not applied). Fundamentally, the applied component will
be referred to hereafter simply as "SWAT", instead of the "hydrological component
of SWAT".
3.1.2 Model set-up of the Narew River Basin
The preliminary set-up of SWAT for the NRB was effected by
Piniewski and Okruszko
(
2011
). This section begins with a summary of the model set-up prepared therein,
which is then followed by a description of updates to the model set-up applied for
the purposes of this thesis.
Preliminary set-up
The main sources of SWAT input data used to set up the project for the NRB are
presented in Table
3.1
. The Digital Elevation Model (DEM) and the stream network
23
Figure 3.2: Schematic pathways of water movement in SWAT (
Neitsch et al.
,
2011
).
Table 3.1: GIS and monitoring datasets used to build the SWAT project in
Piniewski
and Okruszko
(
2011
).
Category
Name
Source
Digital
Elevation
Model
30m resolution contour DEM
covering 80% of the NRB and 90m
resolution DEM covering the
remaining part
Contour DEM: topographic maps; the
rest: NASA Shuttle Radar Topography
Mission (SRTM)
Stream
network
Digital Map of Hydrological
Division of Poland (MPHP)
Institute of Meteorology and Water
Management
Land use
CORINE Land Cover 2000 (CLC
2000)
Institute of Geodesy and Cartography
Soil
Point feature class of 3400
benchmark topsoil profiles
Institute of Soil Science and Plant
Cultivation
Climate
hydrology
12 rain gauge stations, 6 climate
gauge stations, 11 flow gauge
stations: daily data for time period
2001-2008
Institute of Meteorology and Water
Management
24
Figure 3.3: Preliminary conguration of SWAT for the NRB as in
Piniewski and Ok-
ruszko
(
2011
).
layer were used in ArcSWAT to partition the NRB into 151 sub-basins (Fig.
3.3
).
An average sub-basin area equals 182 km
2
and the average length of a reach in a
sub-basin measures 24 km. The land use map and the soil map were used to divide
each sub-basin into HRUs, whose total number in the catchment equalled 1,131.
Due to the catchment's rather flat topography,a single slope option was chosen when
creating the HRUs. Land use codes had to be reclassified from the CLC 2000 land
use classification into the classification used in the SWAT Land Cover/Plant Growth
database. Figure
2.5
shows the original generalised land use map, whereas eight land
use types from the SWAT database were used in the input land use map (Tab.
3.2
).
The three major land use classes were arable land (46.3%), evergreen forests (23.2%)
and grasslands (17.3%).
The soil map shown in Figure
3.4
was prepared using a stepwise procedure.
Firstly, a classification of soils from the benchmark soil profiles database, taken from
the Institute of Soil Science and Plant Cultivation, was made based on two main
characteristics: soil texture and organic matter content. The original Polish soil
texture types (BN-78/9180-11) used on Polish soil maps were retained.
PTG
(
2008
)
provides an approximate key for linking Polish soil texture types to those of the United
States Department of Agriculture (USDA). Secondly, a polygon layer was produced
from a point feature class, with the locations of classified benchmark soil profiles,
using the Thiessen polygon method. In a third step, an overlay of this map with the
25
Table 3.2: Land use/land cover classes used as SWAT input.
Name
SWAT code
SWAT name
Percent
Arable land
SWHT
Spring Wheat
46.3
Deciduous forests
FRSD
Forest-Deciduous
5.4
Evergreen forests
FRSE
Forest-Evergreen
23.2
Mixed forests
FRST
Forest-Mixed
4.4
Water
WATR
Water
1.9
Meadows and pastures
FESC
Tall Fescue
17.3
Urban land
URML
Residential-Med/Low Density
0.5
Wetlands
SWCH
Alamo Switchgrass
1.0
map showing the extent of hydrogenic soils was done and the result of this overlay
was the final map that was used as the SWAT input. In this final map, 27 soil classes
were distinguished. Sandy loams (PGMP1, PGMP2, GL1, GL2, GLP1, GLP2 in Fig.
3.4
), sands (PL1, PL2, PS1, PS2) and loamy sands (PGL1, PGL2, PGM1, PGM2,
PGLP1, PGLP2) represented the three dominating classes, occupying 26.7%, 25.3%
and 21.1% of the total basin area, respectively, while the percentage of peat soils was
considerably high as well (TORF, 16.9%). The values of physical soil parameters
were then set in the User Soils database based on available handbooks on Polish
soils (
Zawadzki
,
1999
;
Ilnicki
,
2002
) and pedotransfer functions (using soil texture
and organic matter content as explanatory variables). The parameters defined in the
User Soils database, as well as the default values of parameters from the SWAT Land
Cover/Plant Growth database, were automatically transferred to HRU files. Hence,
all HRUs with the same soil and land use type had the same set of parameters.
Daily meteorological data from 12 precipitation gauges and six climate gauges,
as well as hydrological data from 11 flow gauges, were used in the model set-up of
Piniewski and Okruszko
(
2011
) (Fig.
3.3
). Daily climate data covered the time period
of hydrological (i.e. beginning on 1 November) years 2001-2008, of which the first six
years were used for calibration and the last two years for validation. Climate data
consisted of minimum and maximum air temperature, relative humidity, wind speed
and solar radiation. This extensive climate dataset enabled the most physically based
Penman-Monteith method for modelling the evapotranspiration to be used. The pre-
cipitation data were interpolated from the gauging stations to sub-basins using the
Thiessen polygon method outside SWAT (as a data pre-processing step). This ap-
proach was an improvement over the default SWAT method of transferring data from
the closest station to every sub-basin.
Natural lakes or artificial reservoirs situated on the main stream network
can be represented in SWAT using "Reservoir" objects, for which water balance is
calculated separately (cf. Fig.
3.2
). There can be only one reservoir per sub-basin,
therefore it was tested which lakes could potentially have significant impacts on flow
26
0
25
50
Km
Bug
Benchmark profile
Soil codes
GL1
GL2
GLP1
GLP2
GS1
GS2
GSP1
MADY
MURSZ
PGL1
PGL2
PGLP1
PGLP2
PGM1
PGM2
PGMP1
PGMP2
PL1
PL2
PS1
PS2
PSP1
PSP2
PLZ1
PLZ2
TORF
WATER
Figure 3.4: SWAT input soil map of the NRB and benchmark soil profiles used to
create the map.
regime (based on features such as lake area and volume, average outflow from the
lake, etc.). Finally, only eight lakes (seven natural and one reservoir) were selected
to be modelled as reservoirs in SWAT (Fig.
3.3
).
New features in model set-up
Piniewski and Okruszko
(
2011
) showed the reasonable fit of simulated and observed
flows in most of the studied calibration and validation areas. However, the results
obtained for small catchments (i.e. with upstream areas below ca. 600 km
2
) were
worse than for larger catchments. Hence, they concluded that the applicability of
SWAT at small spatial scales, in a situation when the model has been set up for a
larger scale, would require improved input data, in particular:
1. More precipitation stations and more flow gauging stations for spatial validation
in small catchments;
2. Using a real soil map instead of transferring soil information from points to
polygons;
3. Daily reservoir/lake releases;
4. Including agricultural management practices such as drainage and fertilisation.
27
Figure 3.5: New SWAT model set-up for the NRB (cf. Table
3.3
for flow gauges
codes).
In the current study the model was improved in terms of points 1, 3 and 4 above.
For point 2, any improvement was constrained by data and time availability, and
thus this effort was abandoned. Moreover, a more complex approach towards model
calibration and validation than in the previous study was used, this time using a new
tool, SWAT-CUP, which will be described in section
3.1.3
. For point 3, no new input
data on reservoir releases were available; however a different approach to modelling
reservoir outflow was selected, which will be discussed further in this section.
In the updated SWAT project set-up the new climate dataset, supported by
the Institute of Meteorology and Water Management, was used. It covered in total 78
precipitation gauges and 14 climate (i.e. air temperature, relative humidity and wind
speed) gauges, with daily data covering the time period from 1986 to 2008 (Fig.
3.5
).
Since SWAT assigns the closest station to each sub-basin, it is beneficial to perform
a spatial interpolation of point data into areal data outside the model. As shown in
Masih et al.
(
2011
) for the precipitation data, this simple approach can considerably
reduce precipitation data-related uncertainty. Hence, the Thiessen polygon method
was used for precipitation, temperature, relative humidity and wind speed data to
better account for spatial variability in these variables. The last climate variable
used by SWAT, solar radiation, is not extensively measured in Poland. Therefore,
50
× 50 km resolution daily data from the freely available MARS-STAT dataset
were used (
van der Goot and Orlandi
,
2003
). These data were validated using point
28
Table 3.3: Flow gauges used in calibration and validation.
Id
Gauge name
Gauge code
River
Catchment area [km
2
]
1
Bialobrzeg Bliszy
BBl
Omulew
1,876
2
Bialobrzegi n. Nett
Bia
Netta
981
3
Burzyn
Bur
Biebrza
6,900
4
Czachy
Cza
Wissa
488
5
Czarnowo
Czr
Orz
529
6
Dobrylas
Dob
Pisa
4,061
7
Fasty
Fas
Supral
1,817
8
Gródek
Gro
Supral
208
9
Karpowicze
Kar
Brzozówka
650
10
Krasnosielc
Kra
Orzyc
1,268
11
Krukowo
Kru
Omulew
1,265
12
Maków Mazowiecki
MMa
Orzyc
1,948
13
Narew
Nar
Narew
1,978
14
Narewka
Nka
Narewka
635
15
Osowiec
Oso
Biebrza
4,365
16
Ostrolka
Ost
Narew
21,862
17
Pisz
Pis
Pisa
3,024
18
Przechody
Prz
Elk
1,452
19
Sokolda
Sok
Sokolda
464
20
Strkowa Góra
StG
Narew
7,181
21
Sura
Sur
Narew
3,377
22
Sztabin
Szt
Biebrza
846
23
Ukta
Ukt
Krutynia
635
24
Wizna
Wiz
Narew
14,308
25
Wonawie
Woz
Jegrznia
852
26
Zambski Kocielne
Zko
Narew
28,268
27
Zaruzie
Zar
Ru
309
observations of solar radiation in the Mikolajki station.
Daily river flow data from 27 gauges were used for calibration and validation
(Fig.
3.5
and Tab.
3.3
). In several cases, gauges were located upstream or down-
stream from the outlets of corresponding SWAT sub-basins. In such cases, data from
the gauges were rescaled using catchment area ratios to represent given outlets.
Eight reservoirs, the same as in
Piniewski and Okruszko
(
2011
), were in-
cluded in the new SWAT set-up (Fig.
3.5
). Only one of them, the Siemianówka
reservoir on the Upper Narew, is a managed reservoir, the rest being natural lakes.
As discussed by
Zambrano-Bigiarini et al.
(
2011
) and
Schuol and Abbaspour
(
2006
),
managed river structures such as dams are a large source of uncertainty in hydrolo-
gical modelling. Current management rules of the Siemianówka reservoir, established
by
BIPROMEL
(
1999
), allocate water between fixed goals, namely agriculture, en-
ergy production, fisheries and the natural ecosystem, depending on actual storage,
29
Figure 3.6: Daily outflow from the Siemianówka reservoir measured at the Bondary
gauging station on the Narew for the time period 1990-2008.
time of year and inflow forecast (
Kiczko et al.
,
2011
). Currently available options of
modelling reservoirs in SWAT do not allow the researcher to mimic these kinds of
management rules. Indeed, the release policy used throughout the time period from
1990, when reservoir operations started, to 2008 was variable through time, as illus-
trated in Figure
3.6
. Hence, the option of using observed daily flows as a reservoir
outflow simulation method was selected. Data from the Bondary station situated at
the very outflow of the reservoir were used for this purpose.
The target release for controlled reservoirs option was used for the reservoir
outflow simulation for the seven remaining (unmanaged) reservoirs. In this option,
a reservoir releases water as a function of the desired target storage (
Neitsch et al.
,
2011
). Although this function was designed by SWAT developers to simulate out-
flow from managed reservoirs, it can also be used for natural impoundments situated
along the river network, as in
Bosch
(
2008
), who investigated the hydrological ef-
fects of impoundments in two catchments draining into Lake Erie in Michigan. The
main advantage of this approach is the possibility to account for flattening of the
impoundment outflow hydrograph by using a parameter defined as the number of
days required for the reservoir to reach target storage. The default value of 15 days
was kept in the initial SWAT set-up for all modelled reservoirs.
Finally, in the new model set-up, various agricultural management practices
available for modelling in SWAT were applied. Management practices in SWAT,
which are defined at the HRU level, can be divided into two groups: general (per-
manent) practices and scheduled practices (
Arnold et al.
,
2011
). Practices from the
first group never change during simulation, while those from the second group occur
at specific times of year. The time of occurrence of a given practice can be based on
date or heat units, although it is recommended that whenever dates are available,
they should be used instead of heat units; consequently, this general guideline was
followed herein.
30
From the first group, a tile drainage function was used to simulate water
transport in open drainage ditches on grasslands, as well as subsurface tiles on arable
land. This function was originally designed for subsurface drains on heavy soils, but
it can also approximate the operation of open drainage ditches, which are much more
frequent in the NRB. The grassland HRUs, to which the tile drainage function was
applied, were identified using the numerical Map of Hydrographic Division of Poland
by IMGW. Mean ditch density on grassland HRUs was calculated for each SWAT sub-
basin and a threshold of 50 m/km
2
was used to select sub-basins where the presence
of ditches was significant. Since drainage is present mostly in organic soils, the use of
the tile drainage function was restricted to these soils only. As no spatial data on the
location of subsurface tile drains on arable land were available, the selection of HRUs
in this case was approximate. A spatial correlation between the proportion of drained
arable land in Polish provinces (from the previous administrative division of Poland)
and the proportion of heavy soils was identified based on data from
WZMiUW
(
2008
)
and
Siuta and ukowski
(
2009
). Hence, tile drainage on arable land HRUs was defined
on selected soil types - clay, clay loam and loam, which are scarce in the NRB (cf.
Fig.
3.4
).
From the second group, the following practices were used to build the typical
crop calendars for different crops:
1. Plant/begin growing season;
2. Fertiliser application;
3. Harvest and kill;
4. Tillage;
5. Harvest only;
6. Kill/end of growing season;
7. Grazing;
8. Auto irrigation initialisation.
Detailed information on the parametrisation of these practices can be found in
Arnold
et al.
(
2011
) and
Neitsch et al.
(
2011
). There are two major reasons why considering
these practices is important in a hydrological model. First of all, they allow the crop
growth sub-model to use inputs that are close to reality, which makes it possible to
simulate correctly crop yields, and thus water uptake by plants. Secondly, using these
practices in the baseline (control) period allows for their consideration in modelling
scenarios of regional change in section
3.3
.
Scheduled agricultural practices concern two major land uses in the NRB: ar-
able land and grassland (cf. Fig.
2.5
). As shown in Table
3.2
, SWHT (spring wheat)
31
Table 3.4: Two example crop calendars applied in SWAT for spring wheat and mead-
ows.
Id
Date
Operation (spring wheat,
East)
Date
Operation (meadows,
West)
1
5 April
Mineral fertiliser: 40 kg
N, 8 kg P
1 April
Beginning of growing
season
2
10 April
Spring tillage
1 April
Fertiliser: 10t solid
manure
3
15 April
Planting
20 April
Mineral fertiliser: 10kg P
4
27 May
Mineral fertiliser: 30 kg N
25 May
First harvest
5
30 July
Harvest and kill
26 May
Fertiliser: 10t solid
manure
6
5 August
Post-harvest tillage
30 July
Second harvest
7
18 October
Winter tillage
20 September
Third harvest (only 20%
of farmers)
8
1 October
Fertiliser: 10t solid
manure
9
20 October
End of growing season
was selected in the initial phase of project development to represent all arable land
in the NRB. Next, HRUs with SWHT were partitioned into three classes: those with
spring wheat (SWHT) representing spring cereals, with rye (SWAT code: RYE) rep-
resenting winter cereals and with potatoes (POTA) representing root plants. These
crops were spatially distributed in the basin using statistical data on crop cultivation
structures at district level (
GUS
,
2002
).
Since there is a general paucity of published data on agricultural manage-
ment practices in the NRB in the form required by the SWAT model, crop calen-
dars were built in cooperation with agricultural experts (Stefan Pietrzak
1
and Jerzy
Barszczewski
2
, pers. comm.). Due to basin-wide climatic variability, the NRB was
divided into two regions, east and west, for the purpose of crop calendar definition.
The difference in the length of the growing season between the north-eastern and
south-western parts of the basin is approximately 15 days (
Demidowicz et al.
,
1998
).
Table
3.4
illustrates two example crop calendars, one for spring wheat grown in the
eastern part of the basin and the second for meadows grown in the western part of
the basin. Dates and the parametrisation of planting, harvests, fertiliser applications
and tillage were defined approximately for each of the crops mentioned above.
It was problematic defining irrigation management in the SWAT model of the
NRB properly. Although the NRB has a considerable amount of permanent meadows
and pastures, and many of them are equipped with open drainage systems, only a few
1
Department of Water Quality Protection, Institute of Technology and Life Sciences in Falenty
2
Department of Grasslands, Institute of Technology and Life Sciences in Falenty
32
of these systems are working gravitational sub-irrigation systems. Irrigation of crops
grown on arable land is hardly observed, and therefore can be neglected. Moreover,
a rapid decline in the irrigated area has been observed in recent years in Poland
(
Labdzki
,
2007
) and in the NRB in particular (
WZMiUW
,
2008
). Nonetheless, HRUs
where irrigation operation should be applied were identified using available spatial
data, e.g. a thematic layer of dams supported by the National Water Management
Authority and a thematic layer of sub-irrigated areas in the Upper Narew Basin
developed in
Puslowska-Tyszewska et al.
(
2005
). An automatic irrigation (auto-
irrigation) option instead of a "conventional" irrigation option was used, since fixed
days of irrigation application are largely uncertain. In the automatic option, water
diversion from a reach is triggered by one of the water stress identifiers observed in
a given HRU, plant water demand or soil water content. If enough water is available
in the source, the model adds water to the soil until it is at field capacity (
Arnold
et al.
,
2011
).
3.1.3 Auto-calibration tools
Piniewski and Okruszko
(
2011
) performed a spatially distributed flow calibration us-
ing SWAT2005 Auto-calibration Tools incorporated in ArcSWAT, including a sens-
itivity analysis tool and an auto-calibration tool. Parameter sensitivities were es-
timated using the Latin Hypercube-One-factor-At-a-Time (LH-OAT) technique (
van
Griensven et al.
,
2006b
). Automatic calibration was carried out using the Para-
Sol tool (
van Griensven and Meixner
,
2007
), which is based on the Shuffled Com-
plex Evolution-University of Arizona (SCE-UA) optimisation algorithm (
Duan et al.
,
1992
). The conclusion was that although these tools proved to be effective, the
amount of time needed to perform auto-calibration at multiple sites was great. In-
deed, with eight years of daily data from 11 stations, 8D parameter space, 151 sub-
basins and 1,131 HRUs and most of the SCE-UA parameters kept as defaults, the
total amount of time used for auto-calibration yielded approximately 85 days of com-
puter work (pure model runs) on an Intel Core2 Duo 3GHz processor using a Windows
XP OS. Hence, a more time-efficient tool would be more desirable.
The purpose of this part of thesis was to recalibrate the updated model using
experience gained during the previous study, a new auto-calibration tool and a new
approach towards spatial calibration and the validation of SWAT.
SWAT-CUP software (
Abbaspour et al.
,
2007a
;
Abbaspour
,
2008
;
Rostamian
et al.
,
2008
), i.e. SWAT - Calibration and Uncertainty Programs
3
, was applied un-
der version 3.1.3. SWAT-CUP is a public domain interface which links five different
3
http://www.eawag.ch/forschung/siam/software/swat/index
33
optimisation algorithms (SUFI-2, GLUE, ParaSol, MCMC and PSO) to SWAT. The
main function of an interface is to provide a link between the input/output of a cal-
ibration program and the model. Hence, SWAT-CUP facilitates sensitivity analysis,
calibration, validation and uncertainty analysis of the SWAT model. In the current
study two algorithms were applied: SUFI-2 and PSO.
The SUFI-2, Sequential Uncertainty Fitting Version 2, combines optimisa-
tion with uncertainty analysis.
Abbaspour et al.
(
2007b
) and
Abbaspour
(
2008
)
provide a detailed description of this algorithm, which is summarised below. The
user has to select a number of model parameters to be included in the procedure,
usually after assessing their sensitivities, and defines their ranges. The user also has
to input the measured data time series, select the objective function and define sev-
eral internal parameters. Furthermore, the user can choose from among seven types
of objective functions described in
Abbaspour
(
2008
), two of which are particularly
popular among hydrological modellers (
Moriasi et al.
,
2007
): the Nash-Sutcliffe Ef-
ficiency coefficient (NSE) and the coefficient of determination R
2
.The program first
performs Latin Hypercube sampling across the defined parameter space, assuming
uniform distribution, leading to n parameter combinations, where n is the number of
desired simulations. Next, SWAT is run n times and saves n time series of simulated
output variables (in this case river flow). In the last step, measures employed to
assess uncertainties are calculated. Since SUFI-2 is a stochastic procedure, it does
not evaluate "standard" goodness-of-fit measures based on two signals; instead, it
calculates the degree to which all uncertainties (input, parameters, structural and
measured variables) are accounted for, which is further quantified by a measure re-
ferred to as the P -factor, i.e. the percentage of measured data bracketed by a 95%
prediction uncertainty (95PPU). The second uncertainty measure used in SUFI-2 is
the so-called R-factor, which quantifies the average thickness of the 95PPU uncer-
tainty band divided by the standard deviation of the measured data. As a general
rule, it is advisable to keep as large a percentage of observations bracketed by the
prediction uncertainty (i.e. a P -factor close to 1), while at the same time narrowing
the thickness of the uncertainty band (i.e. an R-factor as small as possible).
Even though SUFI-2, being a stochastic procedure, does not converge with
any "best simulation", it quantifies standard goodness-of-fit measures such as R
2
and
NSE for each of n model runs. Hence, it indicates the "best simulation" among all
performed runs, which is the run with the highest/lowest value of the earlier defined
objective function.
The second applied algorithm, PSO (the Particle Swarm Optimisation), is
a population-based stochastic optimisation technique developed by
Eberhart and
Kennedy
(
1995
) and inspired by the social behaviour of birds flocking or fish school-
34
Excerpt out of 207 pages
- scroll top
Details
- Quote paper
- , 2014, Scenario-based impact assessment of global and regional change on the semi-natural flow regime, Hamburg, Bedey Media GmbH, https://www.anchor-publishing.com/document/312734
Comments