Nutrient Transport Modelling in the Daugava River Basin Andrea Wallin Examensarbete 20 p

Nutrient Transport Modelling in the Daugava River Basin Andrea Wallin Examensarbete 20 p

UPTEC W05 009

Examensarbete 20 p

Februari 2005

Nutrient Transport Modelling in the Daugava River Basin

Andrea Wallin

ii

ABSTRACT

Nutrient transport modelling in the Daugava River basin

Andrea Wallin

Eutrophication is one of the most serious threats to the Baltic Sea environment. Nutrient loading into the sea therefore needs to be quantified by available mathematical models.

The Generalised Watershed Loading Functions (GWLF), a lumped-parameter model that predicts hydrology and monthly nutrient loads, was applied to the Daugava River

Basin, discharging into the Baltic Sea. The aim of the study was to model historic transport of nutrients into the Baltic Sea and thereby produce estimates of parameters and input data needed for a spatial extension of the GWLF to surrounding river basins.

nitrogen loads were modelled with an R s and validation data from the 1980’

2

value of 0.78 for the calibration period.

Predicted yearly nitrogen loads for the validation period were about 30 % lower than reported values, probably depending on decreasing groundwater and runoff compared to reported values, the main reason probably being the exclusion of septic systems and too low reported point sources.

Modifications of the model are suggested for long-term predictions of nutrient loads and the need for harmonised, up-to-date and generally accessible data for nutrient transport modelling discussed

Keywords: runoff, nutrient, modelling, GWLF, Baltic Sea drainage basin

REFERAT

Modellering av näringsämnestransport i Daugavas avrinningsområde

Andrea Wallin

Övergödning utgör ett av de allvarligaste hoten mot Östersjöns miljö. Storleken av näringsbelastningen till havet behöver därför bestämmas med hjälp av tillgängliga matematiska modeller. Modellen ”Generalised Watershed Loading Functions” (GWLF), en icke-distribuerad parametermodell som uppskattar hydrologi och månatlig näringsbelastning, tillämpades på avrinningsområdet till Daugava som mynnar i

Östersjön. Syftet med studien var att genom modellering av historisk transport av näringsämnen till Östersjön ta fram parametrar och indata som sedan kan användas vid applicering av GWLF på omkringliggande avrinningsområden. Data från 1990-talet användes för kalibrering av modellen och data från 1980-talet för validering. Årlig kvävebelastning modellerades med R

2

-värdet 0,78 för kalibreringsperioden. Modellerad

årlig kvävebelastning för valideringsperioden underskattades med ungefär 30 % vilket troligen beror på att kvävekoncentrationer i grundvatten och ytavrinning minskade mellan 1980- och 1990-talen. Fosforbelastningen underskattades jämfört med rapporterade värden vilket troligen beror på att enskilda avlopp inte inkluderades och att rapporterade punktutsläpp är för låga.

Modifikationer av modellen föreslås för prediktion av näringsbelastningar under lång tid och behovet av harmoniserad, uppdaterad och lättillgänglig data för näringstransportsmodellering diskuteras.

iii

Nyckelord: avrinning, näringsämnen, modellering, GWLF, Östersjöns avrinningsområde

Department of Earth Sciences, Uppsala University, Villavägen 16, 752 36 Uppsala

ISSN 1401-5765 iv

PREFACE

This thesis is part of a Master of Science degree in Environmental and Aquatic

Engineering at Uppsala University. Supervisors were Fredrik Hannerz and Jerker Jarsjö,

Department of Physical Geography and Quaternary Geology, Stockholm University.

My supervisor in Uppsala was Lars-Christer Lundin, Department of Earth Sciences, Air and Water Science, Uppsala University.

The study is part of the MARE (Marine Research on Eutrophication) project at

Stockholm University. The MARE project aims to develop a decision support system,

Nest, which will help decision makers in the Baltic region to find the most cost efficient measures to take against eutrophication of the Baltic Sea. In this study the runoff model that will be used in Nest has been applied to the river basin of Daugava, one of the largest rivers in the Baltic Sea region.

The work was conducted parallel and in collaboration with Evelyn Dahlberg, student of

Department of Physical Geography and Quarternary Geology, Stockholm University, who has applied the same model to the river basin of Nemunas.

I would like to thank my supervisors, Fredrik and Jerker, for support and guidance during the project. Additionally I would like to thank Georgia Destouni, Department of

Physical Geography and Quaternary Geology, Stockholm University, who contributed with guidance and expert knowledge. I would also like to thank Magnus Mörth,

Department of Geology and Geochemistry, Stockholm University, for answering questions regarding the simulation system and the model in general.

Stockholm, February 2005

Andrea Wallin

Copyright © Andrea Wallin and Department of Earth Sciences, Air and Water Science,

Uppsala University

UPTEC W 05 009, ISSN 1401-5765

Printed at Department of Earth Sciences, Geotryckeriet, Uppsala University, Uppsala,

2005.

v

TABLE OF CONTENTS

1. INTRODUCTION...............................................................................................................1

2. BACKGROUND .................................................................................................................2

2.1 EUTROPHICATION OF THE BALTIC SEA ................................................................2

2.2 FEATURES OF THE NITROGEN AND PHOSPHORUS CYCLES .............................3

2.2.1 N

ITROGEN RETENTION AND LEAKAGE

...............................................................................3

2.2.2 P

HOSPHORUS LEAKAGE AND EROSION

..............................................................................3

2.3 GENERAL MODEL DESCRIPTION..............................................................................3

3. MATERIALS AND METHODS ........................................................................................4

3.1 SITE DESCRIPTION – THE DAUGAVA RIVER BASIN.............................................4

3.1.1 G

EOGRAPHY AND HYDROGRAPHY

....................................................................................4

3.1.2 H

YDROLOGY AND CLIMATE

..............................................................................................5

3.1.3 G

EOLOGY AND LAND USE

.................................................................................................6

3.1.4 P

OPULATION AND ANTHROPOGENIC INFLUENCE

...............................................................7

3.1.5 I

NTERNATIONAL COOPERATION IN THE

D

AUGAVA BASIN

..................................................7

3.2 MODEL DESCRIPTION .................................................................................................8

3.2.1 A

N OVERVIEW OF THE MODEL

..........................................................................................8

3.2.2 W

ATER FLOW

...................................................................................................................9

3.2.3 N

UTRIENTS

.....................................................................................................................10

3.3 AVAILABLE OBSERVATION DATA USED IN THE STUDY ..................................12

3.3.1 C

LIMATE DATA

...............................................................................................................12

3.3.2 S

TREAMFLOW AND NUTRIENT LOAD DATA

......................................................................12

3.4 CALIBRATION AND VALIDATION...........................................................................13

3.5 IMPACT OF THE CURVE NUMBER ON NUTRIENT SOURCE

APPORTIONMENT.............................................................................................................14

3.5 SENSITIVITY ANALYSIS ............................................................................................14

4. RESULTS..........................................................................................................................14

4.1 CALIBRATION AND VALIDATION...........................................................................14

4.2 IMPACT OF THE CURVE NUMBER ON NUTRIENT SOURCE

APPORTIONMENT.............................................................................................................18

4.3 SENSITIVITY ANALYSIS ............................................................................................18

5. DISCUSSION....................................................................................................................19

6. CONCLUSIONS ...............................................................................................................22

7. REFERENCES..................................................................................................................23

APPENDIX A. INPUT DATA AND PARAMETERS USED BY THE GWLF MODEL...26

APPENDIX B. DETERMINATION OF SURFACE RUNOFF ..........................................31

vi

APPENDIX C. DETERMINATION OF POTENTIAL EVAPOTRANSPIRATION ........32

APPENDIX D. DETERMINATION OF MONTHLY SEDIMENT YIELD ......................33

APPENDIX E. DETERMINATION OF URBAN RUNOFF NUTRIENT LOADS ...........34

APPENDIX F. REGRESSION LINES OF REPORTED AND MODELLED MONTHLY

AND YEARLY VALUES .....................................................................................................36

vii

1. INTRODUCTION

During the 20 th

century and especially during the last 50 years, the loads of nutrients

(nitrogen and phosphorus) to the Baltic Sea have increased dramatically (SWEPA,

2005). In order to master the problems of eutrophication in the Baltic Sea, proper water management is needed. A decision support system called Nest is presently being developed at Stockholm University within the frameworks of the scientific projects

MARE (Marine Research on Eutrophication) and ELME (European Lifestyles and

Marine Ecosystems). The aim of Nest is to help decision makers and managers working with Baltic Sea environmental issues to take the most cost effective measures against the eutrophication of the Baltic Sea. The main user target groups are the decision makers within the HELCOM (Helsinki Commission) and people in the region working on the implementation of the EU WFD (Water Framework Directive)

.

Nest consists of interlinked models that combine knowledge in ecology, physical oceanography, biogeochemistry, and economics (MARE, 2005). However, the sub-model for prediction of nutrient loads from the river basins of the Baltic drainage area is not satisfactory. Therefore it has been decided that a better watershed model for the determination of nutrient loads to the Baltic Sea shall be incorporated in the system. The chosen model is GWLF (Generalized Watershed Loading Functions). A number of river basins have been selected in order to test this model in the Baltic Sea region, among these is the river basin of Daugava.

Runoff from the entire drainage basin of the Baltic Sea has previously been modelled with the use of the HBV model. The HBV model is like the GWLF model a conceptual, lumped parameter model and it has been concluded that such models can be useful for macro scale hydrological modelling (Bergström and Graham 1998). Another conceptual runoff model much like the GWLF, the METQ98 model, has been successfully applied to the Daugava basin by Ziverts and Jauja (1999). The basin was then divided into 22 sub-basins that were modelled separately (Ziverts and Jauja, 1999).

The temporal trends in nitrogen and phosphorus concentrations in the Daugava River have been analysed by Stålnacke et al. (2003). A small but significant (one-sided test at the 5 % level) decrease in dissolved inorganic nitrogen was found during the years

1987-1998, but no significant decrease in total phosphorus was found. It could seem remarkable that the decrease in nutrient concentration was not greater considering the dramatic decrease in use of fertilisers and the change in agricultural practices due to the collapse of the Soviet Union. Stålnacke et al. (2003) suggest two main explanations for this; the first one is a substantial time lag due to long water-transit times and the other one is mineralization of large pools of organic nitrogen that have accumulated in the soil during the years of immense application of fertilisers.

Modelling of nutrient loads from various drainage basins of the Baltic Sea has also been done. The PolFlow model has for instance been applied to the basins of Lake Mälaren

(Darracq et al., 2005) and lake Peipsi (Mourad et al., 2004), which both fall into the

Baltic Sea. However, modelling of nutrient loads from the entire Baltic Sea drainage basin has so far not been done.

The purpose of this study was to apply the GWLF to the river basin of Daugava, which falls to the Baltic Sea, and thereby produce estimates of site specific parameters and input data needed by the model. The modelling was done using the Windows-based

1

simulation system CatchmentSim. The purpose of the study is also to evaluate the model and the simulation system and point out strengths and weaknesses that should be considered when applying the model to the rest of the Baltic basin.

2. BACKGROUND

2.1 EUTROPHICATION OF THE BALTIC SEA

Nutrients such as nitrogen and phosphorus are vital to life on land and in water.

Moderate input of nutrients to water systems may be beneficial for fish catches since it will increase the growth of macrophytes. However, an over-fertilization will lead to an excessive growth and the bacterial decay of organic matter may lead to oxygen depletion, especially at the deep bottoms of the sea. Due to this there will be an alteration of the ecological community structure. Some species will be favoured while others will be disadvantaged in the altered environment. This is called eutrophication

(Clark, 2001). Phosphorus is usually the limiting nutrient in fresh water but also in some parts of the Baltic Sea, for instance the Bothnian Bay. In saline water, nitrogen is generally the limiting nutrient, which is the case in most parts of the Baltic Sea

(BOING, 2002). The anthropogenic sources of nitrogen and phosphorus are leakage from agricultural land and forested land, point source discharges from waste water treatment plants and industries, leakage from septic systems, deposition from the atmosphere and release of phosphorus from the sediments during anoxic conditions

(SMHI, 2005).

Eutrophication has long been considered to be one of the most serious threats to the

Baltic Sea environment. The nitrate concentration in the Baltic Sea south of the Åland

Sea increased almost threefold from the 1960’s to the 1980’s. A pronounced increase of plankton production has been noticed in this part of the Baltic Sea and the area of anoxic bottoms has likewise increased. Nowadays, about one third of the bottom area of the Baltic Sea is anoxic (SWEPA, 2003) and anoxic bottoms are found in more shallow areas than earlier. Since the 1970’s, when the problem of eutrophication of the Baltic

Sea became apparent, a lot of efforts have been made by the countries surrounding the sea in order to master the problem and reduce the nutrient loads (HELCOM, 2004). For example, in Sweden the regulations for storage and spreading of manure and for the proportion of cultivated land that has to be vegetation-covered during fall and spring have been sharpened. Artificial wetlands have also been constructed in agricultural areas and waste water treatment plants in southern Sweden have introduced nitrogen purification (SWEPA, 2003). However, in spite of all the efforts the nutrient loads from land have not decreased significantly. One reason for this is probably the long retention time of nutrients in agricultural soils. It is evident that it will take a long time until we will see an improvement of the situation in the Baltic Sea, but by active and coordinated measures taken by the countries in the region this time can be shortened (HELCOM,

2004a).

2

2.2 FEATURES OF THE NITROGEN AND PHOSPHORUS

CYCLES

2.2.1 Nitrogen retention and leakage

Nitrogen interacts with the atmosphere through nitrogen fixation and denitrification.

Nitrogen fixation is the process where certain microorganisms convert nitrogen gas of the atmosphere to nitrogen-containing organic compounds. Denitrification, on the other hand, is the process where nitrate is converted to gaseous forms of nitrogen by a series of biochemical reactions. Nitrogen in the watershed may thus leave the system through denitrification before it reaches the mouth of the river. It is difficult to predict the magnitude of denitrification, but 5 to 20 % of the nitrogen in some streams may be lost by nitrification. Due to this nitrogen retention the net amount of nitrogen that reaches the sea is smaller than the gross amount that has been added in the watershed (Brady and Weil, 2002).

Nitrate is generally the predominant mineral form of nitrogen in most soils. Since the nitrate ion is negatively charged it is not adsorbed to the negatively charged colloids that dominate most soils. Therefore, nitrate ions easily move downward with drainage water and are readily leached from the soil into the groundwater. The leaching rate depends on the rate of percolation through the soil, the concentration of nitrogen in the drainage water and the texture and structure of the soil. Great nitrogen losses may come from agricultural land if the input of nitrogen exceeds the amount removed by harvest, but with proper management the losses can be reduced (Ibid).

2.2.2 Phosphorus leakage and erosion

Unlike nitrogen, phosphorus is not lost from the soil in gaseous form. Leaching losses are also generally very low since soluble, inorganic forms of phosphorus are strongly adsorbed by mineral surfaces. However, leaching losses may still be large enough to contribute to eutrophication in downstream waters. The more common pathway of phosphorus leaving the soils is through erosion of phosphorus-carrying soil particles

(Ibid).

2.3 GENERAL MODEL DESCRIPTION

The Generalized Watershed Loading Functions (GWLF) is a mathematical model that estimates monthly loads of nitrogen and phosphorus from watersheds. The mathematical structure of the model is simple. Surface runoff, groundwater flow and point sources of nutrients are included in the model. Examples of required input to the model are climate data and information on land use and soil types in the basin. The original GWLF model was described by Haith and Shoemaker in 1987 (Dai et al.,

2000). An overview of the hydrological part of the GWLF model is shown in figure 1.

In this study the computer program CatchmentSim has been used. CatchmentSim is a

Windows based simulation system for simulating nutrient yields from watersheds. The system is based on a modified version of the GWLF model and it has been developed at

Stockholm University (Mörth, 2004). From now on in this report GWLF will refer to the modified version of the GWLF that has been used in the study if nothing else is stated.

3

Figure 1. The hydrological part of the GWLF model.

3. MATERIALS AND METHODS

3.1 SITE DESCRIPTION – THE DAUGAVA RIVER BASIN

3.1.1 Geography and hydrography

The Daugava River is the third largest river in the Baltic Sea basin (Lundin, 2000). The river begins in the western part of Russia and flows through Belarus and Latvia where it enters the Baltic Sea in the Gulf of Riga (figure 2). The total length of the river is 1 005 km (Klavins et al., 2002). The river basin has a small difference of elevation within the basin. Only a few hills are higher than 250 m above sea level (Ziverts, 1999). The catchment area is 85 900 km

2

(Hannerz and Destouni, 2005) and it is thus by definition a large river basin (Klavins et al., 2002). 32 % of the area belongs to Russia, 39 % to

Belarus and 27 % to Latvia. The remaining 2 % of the basin belong to Estonia and

Lithuania (Hannerz and Destouni, 2005).

4

Figure 2. The Daugava River Basin.

The Daugava basin borders with the catchments of River Gauja in the northwest, Lake

Peipsi and Lake Ilmen in the north, the Rivers Volga and Dniepr in the east, River

Nemunas in the south and southeast and River Lielupe in the west. There are 5 000 tributaries and 95 % of them are less than 5 km. The width of the river varies between

12 and 300 m (Jourakova, 2003). Lake Dvineca in Russia is the origin of the river and here the river is only about six to eight metres wide (Rozins, 2001).

3.1.2 Hydrology and climate

The Daugava River has a mixed feeding from rain, snowmelt water and groundwater.

Snowmelt water is the most important contributor as it composes 50-55 % of the river discharge. Accordingly more than half of the river discharge takes place during spring flood (Klavins et al., 2002). Daugava contributes about 5 % of the total water inflow to the Baltic Sea. The annual precipitation decreases from west to east and varies between

600 and 800 mm/year (Ziverts, 1999). 75 % of the part of precipitation that falls as rain falls between April and October (Jouravkova, 2003). The monthly average temperature varies between about -5 °C in January to about 17 °C in July (BALTEX Hydrological

Data Centre data base, 2004). The climate can be characterised as humid and precipitation usually exceeds evapotranspiration (Klavins et al., 2002). In figure 3 the monthly average temperature and precipitation in the basin is displayed.

5

Temperature Precipitation

20.00

15.00

10.00

5.00

0.00

-5.00

Jan Feb M ar

A pr

M ay

Jun

Jul Aug Sep O ct

Nov Dec

-10.00

120

100

80

60

40

20

0

Jan

Feb M ar

Apr M ay

Jun

Jul Aug Sep O ct

N ov

Dec

Figure 3. Monthly average temperature and precipitation in the Daugava basin, 1980-1999 (BALTEX Hydrological Data Centre data base, 2004).

3.1.3 Geology and land use

The Daugava River basin is covered by glacial deposits to depths of 200 m. Most of these deposits consist of Weichselian (Latvian) Glacial characterised by till, sandy till and glaciolacustrine sand. There are also alluvial deposits from after the ice age

(Rozins, 2001).

About half of the basin area consists of agricultural land. The Belarusian part of the basin has the largest area of cultivated land. In the Russian part of the basin there is also a considerable amount of agricultural land, but more than half of the land in this area is forested. In the Latvian part of the basin about half of the land is agricultural and 30 % is covered with forests (Hannerz and Destouni, 2005). The percentage of different land uses in the part of each country that belongs to the Daugava Basin is presented in table 1. In figure 2 the distribution of different types of land use in the basin is shown.

This is the GIS data that were used to determine the area of each land use which is required as model input by the GWLF. Artificial surfaces include urban areas and roads.

Cultivated land consists of managed areas and mosaic areas of crops and trees.

Herbaceous land is natural scrub, pasture, steppe and heath. Lichens and wetlands is a quite heterogeneous land use category and include lichen, moss, bog, peat and flooded herbaceous areas. Forests are divided into three land use categories; coniferous forest, deciduous forest and mixed forest (both coniferous and deciduous trees). The last land use category represented in the basin is water, which includes all open water surfaces like lakes and rivers (Ibid.).

Table 1. Percent of each land use in the part of each country that belongs to the Daugava River basin and for the entire basin (Hannerz and Destouni, 2005)

Land use category

Artificial surfaces

Cultivated land

Herbaceous land

Lichens and wetlands

Forest

Water

Area percentage for each country and for the entire basin

Latvia Belarus Estonia Lithuania Russia Daugava Basin

1

47

18

0

31

2

0

63

1

0

34

1

0

38

2

0

57

3

0

41

19

1

32

7

0

42

0

1

54

1

1

51

6

1

40

2

6

3.1.4 Population and anthropogenic influence

The population in the area is about two million, of which 1.4 million live in the Latvian part of the basin. About 83 % of the Latvian population is living in urban areas. The major cities in the basin are Riga (800 000 inhabitants) in Latvia and Vitsyebsk

(400 000 inhabitants) in Belarus. The population of the Russian part of the basin is only

250 000. There are tendencies of a decrease in population in the upper (Russian) parts of the basin and an increase in the lower (Latvian) parts. There is also an ongoing shift of people from rural to urban areas (Gooch et al., 2002).

The economic development is well correlated with the distribution of the population in the basin. In the Russian part of the river basin there are few industries and only one hydroelectric power station. Agriculture is rather extensive (Ibid.).

The industry in the Belarusian part of the basin is more developed. The main branches of industry are energy production, oil processing facilities, light industry and food production. The city of Navapolatsk is a major source of pollution of the Daugava due to its oil processing, refinery plants and developed chemical industry. Other important sources of pollution are thermoelectric power stations and municipal waste water treatment plants. Agriculture is also quite well developed in the region (Ibid.).

In the Latvian part of the region there are many different branches of industry, for example food and energy production, textile and wood industries and different kinds of machinery and chemical industries. There are three Latvian hydroelectric power plants along the river; Plyavins, Kegums and Riga. After the construction of hydroelectric power stations, the Daugava lost its importance as a transport route. Today the river is only navigable at certain stretches and the intensity of navigation is low. However, sea transport is still important and the port in Riga is one of the biggest ports around the

Baltic Sea (Ibid.).

3.1.5 International cooperation in the Daugava basin

Since 1997 the Swedish Environmental Protection Agency (SWEPA) has supported the work of bringing about a trilateral agreement between Latvia, Belarus and Russia for managing the Daugava basin. In 2003 a draft international cooperation agreement on managing the river basin according to the EU Water Framework Directive (WFD) was compiled and accepted by the three countries. A complete agreement has also been written and will be signed by all three countries as soon as Russia has finished the national preparation (SWEPA, 2004).

A bilateral Latvian-Swedish project with the aim to develop a river basin district management plan for Daugava according to the EU WFD was completed in 2003

(Vattenresurs AB, 2004). However, the Daugava River Basin District only covers the

Latvian part of the Daugava River basin and thus this management plan will not be sufficient for proper management of the basin. It is of great importance that also Belarus and Russia become involved in the river basin management process.

7

3.2 MODEL DESCRIPTION

3.2.1 An overview of the model

The GWLF model simulates monthly stream flow and monthly loads of phosphorus and nitrogen from any watershed. Hydrology is the driving force of the model. There are three types of water storage; soil water (unsaturated zone), shallow groundwater

(groundwater box 1) and deeper groundwater (groundwater box 2). The amount of water in these three storages characterises the system at each moment and determines the evapotranspiration, surface runoff and groundwater flows. The catchment is divided into land use categories with different characteristics (different parameter values).

Surface runoff from each land use area and groundwater flow from the entire catchment to the river are predicted from data on daily precipitation and temperature. The user has to state the nutrient concentration in surface runoff, groundwater and in eroded material.

The nutrient load is obtained by multiplication of the nutrient concentration by the water flow. Point sources of nutrients are added in order to obtain the total nutrient load. In the original GWLF model the nutrient load from septic systems in the catchment is included. This part of the model has been excluded in the study due to difficulties with finding data on the number of people that have different kinds of septic systems. Figure

4 shows a schematic picture of the model.

Evapotranspiration

Precipitation

Land use characteristics

Temperature

Input data

Infiltration Snowmelt

Function

Soil water

(unsaturated zone)

Nutr.

conc.in

eroded material

Erosion

Surface runoff

Nutr. conc. in surface runoff

(land use specific)

Water storage

Output

Nutr. conc. in box 1

Percolation

Groundwater box 1

Percolation r1

Point sources

Nutr. conc. in box 2

Groundwater box 2 r2

Stream channel

Streamflow Nutrient load

Figure 4. The structure of the modified GWLF model used in this study.

The model calculates flows and nutrient loads for each land use and the contributions from the different land uses are summed up in order to obtain the total flow/load from the watershed. In Appendix A there is a list of all the input data and parameter values used in this study.

8

3.2.2 Water flow

The structure of the hydrological part of the GWLF is shown in figure 1. The input needed for this part of the model is data on precipitation and temperature as well as the areas of each land use. Streamflow is computed as the sum of groundwater flow and surface runoff, as shown in equation 1. All the flows are divided by the watershed area which is how the unit centimetre is obtained.

Streamflow t

= Q t

+ G1 t

+ G2 t

where

Streamflow t

= streamflow on day t [cm]

Q t

= surface runoff to the stream channel on day t [cm]

G1 t

= flow from groundwater box 1 to the stream channel on day t [cm]

G2 t

= flow from groundwater box 2 to the stream channel on day t [cm]

(1)

Surface runoff is computed from weather data by the U.S. Soil Conservation Service’s

Curve Number Equation. This is further described in Appendix B.

Water that does not contribute to surface runoff will infiltrate the soil. Infiltrated water will then either be lost to the atmosphere through evapotranspiration or percolate to the groundwater zone. Evapotranspiration is limited by available soil moisture in the unsaturated zone and by potential evapotranspiration:

E t

=

Min

(

CV t

PE t

;

U t

+

R t

+

M t

Q t

) (2) where

CV t

= evapotranspiration cover coefficient

PE t

= potential evapotranspiration [cm]

U t

= unsaturated zone moisture [cm]

R t

= rainfall on day t [cm]

M t

= snowmelt on day t [cm]

Q t

= surface runoff from land use k on day t [cm]

The evapotranspiration cover cover coefficient CV t

is dependent on land use and season. It has a higher value during the growing season since the transpiration of the vegetation is then greater. CV t

values for different land uses and for dormant season and growing season respectively are given in The BasinSim 1.0 User’s Guide (Dai et al,

2000). Potential evapotranspiration for each day is calculated from number of daylight hours, saturated water vapour pressure and daily mean air temperature. The equations for determining potential evapotranspiration are given in Appendix C.

Percolation is assumed to occur each day, so that every day all the unsaturated zone moisture percolates into the saturated zone. It is thus assumed that the unsaturated zone is completely emptied each day. If there is no moisture in the unsaturated zone, there will be no percolation:

PC t

= Max(0 ; U t

+R t

+M t

-Q t

-E t

)

(3)

9

Groundwater discharge to the river channel and percolation from groundwater zone 1 to groundwater zone 2 is calculated from recession coefficients, so that for each day

G

1

t

=

r

G

2

t

G t

=

=

gr

1

r

2

S

1

t

S

S

1

t

2

t

(4)

(5)

(6) where

G1 t

= flow from groundwater box 1 to the stream channel [cm/day]

G2 t

=flow from groundwater box 2 to the stream channel [cm/day]

r1 = recession coefficient from box 1 [day

-1

]

r2 = recession coefficient from box 2 [day

-1

]

S1 t

= soil moisture in groundwater box 1 [cm]

S2 t

= soil moisture in groundwater box 2 [cm]

G t

=flow from groundwater box 1 to groundwater box 2 [cm/day]

gr = coefficient for groundwater flow from box 1 to box 2 [day

-1

]

In the model, there is also a possibility to include deep seepage, which is interpreted as water leaving the system by percolation from groundwater box 2 to a deep aquifer. This water will not contribute to the streamflow. However, assuming that all water entering the system as precipitation will leave the system either as evapotranspiration or end up in the stream channel the deep seepage has been set to zero in this study.

3.2.3 Nutrients

In the GWLF model loads of dissolved and solid nitrogen and phosphorus are estimated for each day. Daily values are summed to provide monthly estimates of nutrient loads. It is assumed that streamflow travel times are much less that one month. Dissolved nutrients are assumed to origin from surface runoff, point sources (as known, constant mass flows) and groundwater discharge to the stream. Thus, monthly loads of dissolved nitrogen or phosphorus in streamflow are:

LD m

=DP m

+DR m

+DG m

where

LD m

= total dissolved nutrient load in month m [kg]

DP m

= point source dissolved nutrient load in month m [kg]

DR m

= rural runoff dissolved nutrient load in month m [kg]

DG m

= groundwater dissolved nutrient load in month m [kg]

(7)

The sources of solid-phase nutrients in the model consist of rural soil erosion and wash off of material from urban surfaces:

LS m

= SR m

+SU m

where

LS m

= total solid phase nutrient load in month m [kg]

SR m

= solid phase rural runoff nutrient load in month m [kg]

SU m

= solid phase urban runoff in month m [kg]

(8)

10

Rural runoff nutrient loads

Dissolved rural runoff loads. The dissolved rural runoff nutrient load for each land use is obtained by multiplying runoff by nutrient concentration. Nutrient concentration in runoff from different land uses is thus an input needed by the model. By summing daily contributions from different land uses the monthly load from the total watershed is obtained:

DR m

=

0 .

10 *

dm

∑∑

k t

=

1

(

Cd k

Q kt

AR k

)

(9) where

Cd k

= nutrient concentration in surface runoff from land use k [mg/l]

Q kt

= surface runoff from land use k on day t [cm] (from equation B1)

AR k

= area of land use k [ha]

dm = number of days in month m

0.10 = dimensional factor associated with the units of the rest of the factors in the equation

Solid-phase rural runoff loads. The solid-phase rural nutrient loads (SR m

) are given by the product of monthly sediment yields from the watershed and average sediment nutrient concentration:

SR m

=

0 .

001

Cs

Y m

(10) where

Cs = average sediment nutrient concentration [mg/kg]

Y m

= monthly watershed sediment yield [ton]

The definition of sediment yield is “The total amount of eroded material which does complete the journey from source to a downstream control point” (Chow, 1964). The monthly watershed sediment yield is inter alia depending on soil and land use characteristics and rainfall. Equations for calculating the monthly sediment yield are given in Appendix D.

Urban runoff nutrient loads

Urban surfaces are assumed to be impermeable. Nutrients accumulate on urban surfaces over time and are washed off by runoff events. With increasing time the accumulated amount of nutrients on the surface approaches an asymptotic value as the depletion rate approaches the accumulation rate. During the next rainfall event nutrients will then be washed off. Equations used and assumptions made for calculating this load are given in

Appendix E.

11

Groundwater nutrient loads

The groundwater nutrient load to the stream is obtained by multiplying the flow from each groundwater box by the groundwater nutrient concentration. The monthly groundwater nutrient load is thus calculated as follows:

DG m

=

0 .

1 *

AT

(

C

1

g t dm

=

1

G

1

t

+

C

2

g t dm

=

1

G

2

g

) (11) where

C1 g

= nutrient concentration in groundwater box 1 [mg/l]

C2 g

= nutrient concentration in groundwater box 2 [mg/l]

AT = watershed area [ha]

3.3 AVAILABLE OBSERVATION DATA USED IN THE STUDY

3.3.1 Climate data

Climate data (daily precipitation and mean air temperature) for the years 1980 to 2000 were taken from the BALTEX Hydrological Data Centre data base administrated by

SMHI. The value of daily temperature and precipitation is a spatial and temporal mean of several measuring points in the basin and several times of measurement during the day (Danielsson, 2005).

3.3.2 Streamflow and nutrient load data

The reported data consist of monthly means of streamflow, total nitrogen load and total phosphorus load from 1970-2000 (Wulff and Rahm, 1990; Stålnacke 1996) with some nitrogen and phosphorus from the Daugava are displayed. Data from 1970-1990 were compiled by Stålnacke (1996) with help from data obtained from the Latvian

Hydrometeorological Agency in Riga. Gaps in the data series were filled in by the use of statistical interpolation and extrapolation methods. Different sampling sites were used and it is not clear where the sampling sites are located, but they are probably situated a considerable distance from the mouth of the river. Since the areas of different land uses in the basin have been calculated from the mouth of the river this means that the drainage area used as model input is overestimated with respect to the drainage area of the sample stations. Also, the impact of Riga is not included in reported data, since

Riga is situated at the mouth of the Daugava. Considering the data from the 90’s, there is little information on how the data series were compiled and where the sample station is located.

12

a) Streamflow

10000

8000

6000

4000

2000

0

Ja n-

70

Jan-

71

Ja n-

72

Ja n-

73

Jan-

74

Jan-

75

Ja n-

76

Ja n-

77

Jan-

78

Ja n-

79

Ja n-

80

Ja n-

81

Jan-

82

Ja n-

83

Ja n-

84

Jan-

85

Jan-

86

Jan-

87

Ja n-

88

Jan-

89

Ja n-

90

Jan-

91

Ja n-

92

Jan-

93

Ja n-

94

Ja n-

95

Ja n-

96

Ja n-

97

Jan-

98

Ja n-

99

Ja n-

00

b) Nitrogen load

25000.00

20000.00

15000.00

10000.00

5000.00

0.00

Jan-

70

Ja n-

71

Ja n-

72

Jan-

73

Jan-

74

Ja n-

75

Jan-

76

Jan-

77

Ja n-

78

Ja n-

79

Jan-

80

Jan-

81

Ja n-

82

Jan-

83

Ja n-

84

Ja n-

85

Jan-

86

Jan-

87

Ja n-

88

Ja n-

89

Jan-

90

Ja n-

91

Jan-

92

Jan-

93

Ja n-

94

Ja n-

95

Jan-

96

Jan-

97

Ja n-

98

Jan-

99

Jan-

00

c) Phosphorus load

1000.00

800.00

600.00

400.00

200.00

0.00

Ja n-

70

Ja n-

71

Ja n-

72

Ja n-

73

Jan-

74

Ja n-

75

Ja n-

76

Jan-

77

Ja n-

78

Jan-

79

Ja n-

80

Ja n-

81

Jan-

82

Ja n-

83

Ja n-

84

Jan-

85

Ja n-

86

Ja n-

87

Ja n-

88

Ja n-

89

Ja n-

90

Ja n-

91

Ja n-

92

Ja n-

93

Ja n-

94

Ja n-

95

Jan-

96

Ja n-

97

Ja n-

98

Jan-

99

Ja n-

00

Figure 5. Reported monthly streamflow (a) and load of total nitrogen (b) and phosphorus (c) from the Daugava River for the years 1970-2000.

3.4 CALIBRATION AND VALIDATION

The model was applied to the Daugava drainage basin with the input data and parameter values presented in Appendix A. In the appendix it is also described which parameters were calibrated.

(1993-1999) were used for calibration. They were chosen because they were the most recent years and the desire was to achieve a calibration as close to the present situation as possible. As a first step calibration was done for the hydrological part of the model.

The evapotranspiration cover coefficients (CV t

), curve numbers (CN2 k

) and coefficients for groundwater flow (r1, r2 and gr) were used as calibration parameters. In order to obtain the best fit several aspects were taken into consideration. It was a trade off between obtaining an average value over the seven years of the predicted streamflow as close to the observed average value as possible, minimising the average percentage error and reaching the observed top flows. The process was iterative and since there is no built-in calibration procedure in CatchmentSim the work had to be done manually.

13

There is therefore no guarantee that the calibrated values are the optimal values. When the calibration was complete the validation was done on data from the 1980’s.

When the hydrological part of the model was calibrated, the next step was to add nutrient concentrations in groundwater and surface runoff, erosion of solid nutrients and nutrient loads from point sources. The concentration in the groundwater boxes 1 and 2 were used as calibration parameters. Focus was placed on achieving the best fit for loads, but concentrations were also studied. The calibration was done in the same way as for the hydrological part; by studying the quotient between predicted and observed average values, the average percentage error and how well top flows/concentrations were predicted.

As for streamflow, validation was done on data from the 1980’s.

3.5 IMPACT OF THE CURVE NUMBER ON NUTRIENT SOURCE

APPORTIONMENT

The curve number (see Appendix B) determines the fraction of precipitation that will go to surface runoff. The curve number varies between 0 and 100. Curve number 0 means that all precipitation will infiltrate and curve number 100 means that all precipitation will go to surface runoff. There are different views on how to choose the curve number for each land use. In the original GWLF manual curve numbers are suggested for different land use types. However, in communication with Stefan Löfgren (2005) it was decided that in this study the same curve number should be used for all vegetationcovered surfaces. Since the curve number determines surface runoff and since surface runoff is the main contributor to nutrient loads from a certain land use, the curve number will have a large impact on the nutrient source apportionment from different land uses. In order to illustrate this, the model was run both with the calibrated curve numbers that were used in the rest of the study (Approach A) and with the curve numbers suggested by the original GWLF manual (Approach B). The nutrient contributions from each land use were studied for the two cases.

3.5 SENSITIVITY ANALYSIS

A sensitivity analysis was conducted in order to determine which model parameters have the greatest influence on the output. The parameters chosen for the sensitivity analysis were recession coefficients 1 and 2 (r1 and r2), the coefficient for groundwater flow between box 1 and 2 (gr) and the nutrient concentration in groundwater box 1 and

2 (C1 g

N and C2 g

N). The output that was studied was the average monthly load of nutrients. The calibrated values of the chosen parameters were multiplied by a fixed factor one at a time and the impact on the average load of nutrients was studied. The factors chosen were 0.5, 0.75, 1.25 and 1.5.

4. RESULTS

4.1 CALIBRATION AND VALIDATION

The result of the modelling can be seen in figure 6, where observed and modelled monthly values for streamflow and nutrient loads are presented for the calibration period (1993-1999) and the validation period (1980-1989). For the calibration period the model predicts streamflow and nitrogen rather well although top flows are sometimes underestimated. Streamflow is also fairly well predicted for the validation

14

period. Nitrogen loads are underestimated for this period, especially during peak flows.

When it comes to phosphorus the model greatly underestimates monthly loads for both the calibration period and the validation period.

a) Streamflow, calibration period

6

4

2

0

12

10

8

Ap r-93

Ap r-94

Ap r-95

Ap r-96

Apr

-97

Ap r-98

Apr

-99

c) Nitrogen load, calibration period

14000

12000

10000

8000

6000

4000

2000

0

Apr

-93

Apr

-94

Ap r-95

Ap r-96

Ap r-97

Ap r-98

Apr

-99

b) Streamflow, validation period

Reported

12

10

8

6

4

2

0

Apr

-80

Ap r-81

Apr

-82

Apr

-83

Apr

-84

Apr

-85

Apr

-86

Ap r-87

Apr

-88

Apr

-89

Modelled

d) Nitrogen load, validation period

25000

20000

15000

10000

5000

0

A pr

-80

A pr-8

1

A pr-8

2

A pr

-83

Apr

-84

A pr

-85

A pr-8

6

A pr

-87

A pr

-88

Apr

-89

e) Phosphorus load, calibration period f) Phosphorus load, validation period

1000

800

600

400

200

0

Apr

-93

Apr

-94

Apr

-95

Apr

-96

Apr

-97

Apr

-98

Apr

-99

600

500

400

300

200

100

0

Apr

-80

Ap r-81

Ap r-

82

Ap r-83

Ap r-84

Apr

-85

Ap r-86

A pr-8

7

Ap r-88

Ap r-89

Figure 6. Reported and modelled values of monthly means of streamflow, nitrogen loads and phosphorus loads from the Daugava River for the calibration period

(1993-2000) and the validation period (1980-1990).

The model output is monthly means of loads and streamflow, but since the purpose of the model is to estimate long-term loads it is of interest to see how well the model predicts yearly means. In figure 7 the reported and modelled yearly means of streamflow and nutrient loads are presented. The model seems to estimate yearly streamflow quite well, both for the calibration period and the validation period. There are, however, tendencies towards a higher variance for modelled yearly flows than for observed yearly flows. Yearly nitrogen loads are well estimated by the model for the calibration period, but for the validation period nitrogen loads are underestimated .

15

a) Streamflow, calibration period

40

35

30

25

20

15

10

5

0

19

93

1994 19

95

199

6

1997 199

8

1999

c) Nitrogen load, calibration period

50000

40000

30000

20000

10000

0

1993 19

94

1995 19

96

1997 19

98

1999

b) Streamflow, validation period

35

30

25

20

15

10

5

0

19

80

19

81

1982 19

83

19

84

1985 19

86

19

87

1988 19

89

Reported

Modelled

d) Nitrogen load, validation period

60000

50000

40000

30000

20000

10000

0

1980 1981 198

2

198

3

19

84

198

5

19

86

1987 19

88

1989

e) Phosphorus load, calibration period f) Phosphorus load, validation period

2000

1500

1000

500

0

19

93

19

94

19

95

19

96

19

97

1998 19

99

2000

1500

1000

500

0

1980 1981 1982 198

3

198

4

19

85

198

6

198

7

19

88

19

89

Figure 7. Reported and modelled values of yearly means of streamflow, nitrogen loads and phosphorus loads from the Daugava River for the calibration period

(1993-2000) and the validation period (1980-1990).

Table 2 shows a statistical summary of the comparison of observed and predicted values. Generally, the prediction of yearly values is a lot better than the prediction of monthly values. Regressions lines were made between reported and modelled values.

The R

2

value of the regression of modelled and reported yearly nitrogen loads is 0.78

and the corresponding slope of regression is 1.17. The R

2

value for yearly phosphorus loads is 0.5, but since the regression slope is only 0.14 the ability of the model to predict phosphorus loads must be considered to be very low. The regression slope for monthly means is generally much lower than for yearly means. This is due to the fact that the model underestimates peak flows. For yearly values the impact of peak flows is not so significant, which is why the regression slope for yearly means is higher than for monthly means. Plots of the regressions are found in Appendix F.

16

Table 2. Comparison of reported and modelled values

Calibration period,

1993-1999

Monthly means

Yearly means

Validation period,

1980-1989

Monthly means

Yearly means

Stream flow:

Modelled mean/reported mean for entire simulation period

Mean absolute percentage error

Slope of regression between reported and modelled values

R

2 of regression between reported and modelled values

Nitrogen load:

Modelled mean/reported mean for entire simulation period

Mean absolute percentage error

Slope of regression between reported and modelled values

R

2 of regression between reported and modelled values

Phosphorus load:

Modelled mean/reported mean for entire simulation period

Mean absolute percentage error

Slope of regression between reported and modelled values

R

2 of regression between reported and modelled values

1

45

0.48

0.36

1

59

0.47

0.32

0.16

80

0.016

0.006

1

18

1.21

0.67

1

18

1.17

0.78

0.16

84

0.14

0.5

0.99

48

0.7

0.53

0.67

44

0.41

0.44

0.2

83

0.18

0.18

0.99

13

1.58

0.82

0.67

35

0.57

0.75

0.2

81

0.24

0.47

The contribution to the modelled streamflow is divided by flow from surface runoff, groundwater box 1 and groundwater box 2 respectively. In figure 8 the distribution of these three flows over two years is shown. Flow from groundwater box 1 is the main contributor to streamflow. The largest flows occur during spring. Surface runoff only occurs during periods of rainfall and/or snowmelt, see Appendix B. The contribution from groundwater box 2 is generally lower than from groundwater box 1. On the other hand, the flow from box 2 is more constant over the year, and during fall and winter the contribution from box 2 at times exceeds the contribution from box 1.

.2

.1

.

.4

.3

.7

.6

.5

Ja n-95

Surface runoff

Flow from box1

Flow from box 2

Jul

-95

Ja n-

96

Jul

-96

Figure 8. The predicted daily contribution to streamflow from surface runoff and groundwater box 1 and 2 and the total streamflow, 1995-1996.

17

4.2 IMPACT OF THE CURVE NUMBER ON NUTRIENT SOURCE

APPORTIONMENT

The model was run with different choices of curve numbers according to Approach A and B. The curve numbers are presented in table 3.

Table 3. The two sets of curve numbers used

Land Use

Artificial surfaces

Cultivated land

Herbaceous land

Lichens and wetlands

Coniferous forest

Mixed forest

Deciduous forest

Water

Curve number

Approach A Approach B

100

80

80

80

80

80

80

100 100

100

70

70

70

60

60

60

Figure 9 illustrates the impact of the curve number on the distribution of total nitrogen loads from different land uses. The total nitrogen load includes surface runoff and erosion from each land use. Water surfaces and cultivated land are the main contributors. With approach A cultivated land makes up most of the nutrient loads

(71 %) and water surfaces make up 23 %. With approach B the relative impact of water surfaces is enlarged and makes up 54 % of the total load while the relative impact of cultivated land is lessened (46 %). The impact of forested land is smaller for approach B than for approach A since the curve number is the same for forested land and cultivated land in approach A, but smaller for forested land than for cultivated land in approach B.

The phosphorus source apportionment is not displayed since the model fails to predict phosphorus loads.

Approach A

Approach B

Cultivated land

Herbaceous land

Lichens and w etlands

Coniferous forest

Mixed forest

Deciduous forest

Water

Cultivated land

Herbaceous land

Lichens and w etlands

Conif erous forest

Mixed forest

Deciduous forest

Water

Figure 9. The relative contribution to nitrogen loads from each land use with different methods for choosing the curve number.

4.3 SENSITIVITY ANALYSIS

The result of the sensitivity analysis is presented in figure 10. The nitrogen concentration in groundwater box 1, C1 g

N, is the one of the analysed parameters whose value has the greatest impact on the nutrient load. A change of this value by 50 % will change the nitrogen load by 27 %. The variable that has the least impact on the nitrogen load is the recession coefficient for box 2, r2. A change of the value of this coefficient by 50 % will change the nitrogen load by less than 1 %. It should be noted that the change in nutrient load will be directly proportional to the change in the nitrogen

18

concentration in box 1 and 2 due to the linearity of the model equations that determine nutrient loads from the groundwater.

140

130

120

110

100

0.5

0.75

1

1.25

1.5

90

80

70

60 r1 r2 gr C1gN C2gN

Figure 10. Result of the sensitivity analysis. On the x-axis are the analysed parameters. The y-axis shows the average change in monthly nitrogen load over 10 years when the parameter values have been multiplied by the factors 0.5, 0.75,

1, 1.25 and 1.5, respectively.

5. DISCUSSION

The aim of the model is to predict long-term nutrient loads from a watershed. The model predicts yearly nitrogen loads from the Daugava basin quite well. The mean absolute percentage error of estimations of yearly nitrogen loads during the years 1993-

1999 is 18 %, and the corresponding R

2

-value is 0.78. Considering the simplicity of the model and the defectiveness of input data this can be considered a quite good result.

During the 1980’s (validation period) the model underestimates nitrogen loads by approximately 30 % even though streamflow is quite well predicted during this period.

A reason for this may be that the surface runoff and groundwater nitrogen since the use of fertilisers, especially in the Latvian part of the basin, dropped

There has been a negative trend in nitrogen loads from the Daugava during the years

1987-1998 (Stålnacke et al., 2003). Since the nutrient concentration in the GWLF model is constant, the model can never predict trends in nutrient concentration due to changes in the groundwater or surface runoff concentrations. In that sense the model is static and not suitable for long-term nutrient load predictions.

The GWLF model, as implemented, fails to predict phosphorus loads. The modelled phosphorus loads are about 90 % less than reported loads. One reason could be that phosphorus concentrations in surface runoff and groundwater used were too low.

However, in order to reach the observed levels, the input concentrations in surface runoff and groundwater needed to be increased almost by a magnitude, which does not seem reasonable. A possible cause to this discrepancy could instead be an underestimation of phosphorus loads from erosion. The erosion factor, the rainfall erosivity and the sediment delivery ratio are dependent on parameters such as agricultural management practices and crop coverage. Since no such data were available

19

the chosen value is a very rough approximation. The value of rainfall erosivity is likewise an approximation and the sediment delivery ratio is an extrapolation since values were not given for watersheds as large as the Daugava basin. All together, the predicted erosion may be considerably erroneous. Never the less, a more probable explanation for the underestimation of phosphorus loads is that the reported values of point source phosphorus loads in the basin are a lot less than the actual value and that phosphorus from septic systems, which are not included in the modelling, contributes substantially to the total load. According to HELCOM (2004b), about 70 % of diffuse phosphorus loads to the Gulf of Riga origin from other sources than agriculture and forestry, for instance from septic systems. Looking at Daugava the figure is probably something similar since water from the Daugava is the main contributor to the Gulf of

Riga. Gren et al. (1997) state that the lack of sewage treatment in the Gulf of Riga region contributes to very large input of phosphorus to the sea. Since agriculture and forestry are the main sources of phosphorus according to the model and since septic systems are not included it is therefore not surprising that the model fails to predict phosphorus loads.

An indication of point source phosphorus pollution is that concentrations are higher at lower flows and lower at higher flows. In figure 11 the observed phosphorus concentration and streamflow are displayed.

Stream flow Phosphorus concentration

12

10

8

6

4

2

0.14

0.12

0.1

0.08

0.06

0.04

0.02

0 0

A pr-

80

A pr

-81

A pr

-82

A pr

-83

A pr

-84

A pr-

85

A pr-

86

A pr-

87

A pr

-88

A pr

-89

A pr

-90

A pr

-91

A pr-

92

A pr-

93

A pr-

94

A pr

-95

A pr

-96

A pr

-97

A pr

-98

A pr-

99

Figure 11.

Reported streamflow and phosphorus concentration,

1980-2000.

It seems like the peaks in phosphorus concentration occur at low flows which could be an indication of point sources of phosphorus, but in order to say something definitely one would need to conduct a statistical analysis of the two time series.

Of the parameters used in the sensitivity analysis the nitrogen concentration in groundwater box 1 is the one that affects average loads the most. This is not surprising since streamflow mainly consists of water from groundwater box 1. It is therefore important to calibrate this parameter carefully. The impact of the recession coefficients and the coefficient determining flow between the two groundwater boxes appears to be very small, looking at the result of the sensitivity analysis. The reason for this is the small difference between the nitrogen concentration in the two boxes (1 and 1.3 mg/l, respectively). With a more pronounced difference in concentration between the boxes the impact on the nitrogen load of the recession coefficients and the coefficient determining flow between the two boxes would have been greater.

20

The choice of curve numbers that determine surface runoff has a great impact on the nutrient source apportionment. Since no data were available on site specific nutrient contributions from different land uses, it was not possible to calibrate the curve numbers in order to fit observed apportionments. The values suggested in the manual for the original GWLF (Dai et al., 2001) are developed for North American conditions and it is difficult to say much about their relevance in this area. It is important to be consistent when choosing curve numbers in the rest of the Baltic Sea basin and to bear in mind that the source apportionment from different land uses contains substantial uncertainties.

When looking at the source apportionment it should be remembered that only surface runoff and erosion is included; nutrient contributions from the groundwater flow is not considered. Due to nitrogen leakage a substantial part of nitrogen in agricultural soils will be transported down to the groundwater and thus result in increased nitrogen concentrations. The total nitrogen load is thus largely affected by nitrogen from the groundwater. Thus, the actual contribution from agricultural land is greater than what is suggested by the source apportionment analysis.

In a drainage basin as large as Daugava the amount of nitrogen that is lost on the way to the sea is substantial, but nitrogen retention is not included in the model. The retention affects the model calibration in such a way that the calibrated groundwater nitrogen concentrations are lower than they would have been, had retention not existed. The groundwater nitrogen concentration in the model is thus probably lower than in reality.

The inability of the model to estimate nitrogen retention might affect the usefulness of the model for scenario analysis. For example, the construction of wetlands may help decrease nitrogen loads from water passing through the wetland since nitrogen is lost by denitrification. In this way, wetlands can help decrease nitrogen loads from for instance agricultural lands. But in the GWLF the nitrogen load from agriculture will not be affected by the increase in wetland areas. Also, since retention is not included in the model, the impact of land anywhere in the basin will have the same impact, a situation that is unlikely to occur in reality.

The quality of data is of major importance for the result of any modelling. In the

Daugava basin it is difficult to find reliable data. The reported streamflow and nutrient loads can also be questioned. Little information is given regarding what reported data are observed and what are calculated. In order for modelling results of nutrient flows from the entire Baltic Sea basin to be trustworthy more reliable data are therefore needed. It is of utmost importance that such data are easily accessed and homogenous in order to support sustainable water management in the Baltic Sea Region (Hannerz et al.

2005).

In order to use the GWLF for long-term prediction of nutrient loads, modifications of the model are suggested. For instance, nutrient concentrations should be variable with time. For simulations over short time periods it is probable that the groundwater concentration remains more or less the same. However, considering long time spans or periods with sudden changes in management, like for instance the collapse of the Soviet

Union and its agricultural policy, it is not likely that the groundwater concentration will be constant. As stated above, it must be remembered that the model is static in this sense. Another change that could be made is to divide the existing land cover categories into several subcategories depending on more site-specific land use. As an example, presently it would be impossible to handle a future scenario where a certain fraction of

21

all cultivated land will have winter crops. In order to simulate such a scenario it would be necessary to divide cultivated land into two categories; cultivated land with winter crops and cultivated land without winter crops. This way it would be easier to handle scenarios where new agricultural policies affect management practices.

6. CONCLUSIONS

The main conclusions that can be drawn from the study are:

Yearly nitrogen loads are fairly well estimated by the model during the calibration period, but since the loads during the validation period generally were higher than during the calibration period it was not possible to validate the ability of the model to predict nitrogen loads.

The model significantly underestimates phosphorus loads. Possible reasons for this are that septic systems, which were not included in the model, are a main contributor to phosphorus loads and that reported point sources are underestimated.

Modifications of the model need to be done in order to use the model for longterm predictions. As an example it should be possible to vary the groundwater nutrient concentration with time.

The lack of good quality data makes it difficult to model nutrient loads in parts of the Baltic Sea basin. It should be of high priority to make data more readily available and to create more homogeneous databases for the region. In order to do this successfully cooperation between all the countries in the region is needed.

22

7. REFERENCES

BALTEX Hydrological Data Centre data base, 2004. SMHI. URL: http://www.smhi.se/sgn0102/bhdc/ (16-02-2004)

Bergström, S. and Graham, L.P., 1998. On the scale problem on hydrological modelling. In: Journal of Hydrology 211 (1998), pp. 253-265.

BOING – Baltic on-line Interactive Geographical and Environmental Information

Service, 2002. Basic facts about eutrophication as an environmental problem in the

Baltic Sea. URL http://jolly.fimr.fi/index.html

(22-02-2002)

Brady, N.C. and Weil, R.R., 2002. The Nature and Properties of Soils. Pearson

Education Inc., Upper Saddle River, New Jersey.

Chow, V.T., 1964. Handbook of applied hydrology. McGraw-Hill, New York, NY.

Clark, R.B., 2001. Marine Pollution. Bath Press Ltd., Bath, Avon.

Dai, T., Wetzel, R.L., Christensen, T. R. L. and Lewis, E.A., 2000. The BasinSim 1.0

Users Guide. URL: http://www.vims.edu/bio/models/bsdownload.html (05-11-2004)

Danielsson, Åsa, Institutionen för Tema, Linköpings universitet, 2005. Personal communication.

Darracq A., Greffe F., Hannerz F., Destouni G. and Cvetkovitc V., 2005. Nutrient transport scenarios in a changing Stockholm and Mälaren valley region. Water science and technology. (in press)

Daugava River Basin Project. URL: www.daugava.lv (20-12-2004)

Gooch, G.D., Höglund, P., Roll, G., Lopman, E. and Aliakseyeva, N., 2002. Review of existing structures, models and practices for transboundary water management. Report no D1b. URL: http://www.mantraeast.org/pdf/D1b.pdf (05-11-2004)

Gren, I-M., Söderqvist, T. and Wulff, F., 1997. Nutrient Reductions to the Baltic Sea:

Ecology, Costs and Benefits. In: Journal of Environmental Management 51 (1997), pp.

123-143.

Hannerz, F. and Destouni, G., 2005. New definition and update of spatial data and statistics for the Baltic Sea drainage basins. Ambio. (in review)

Hannerz, F., Destouni, G., Cvetkovic, V., Frostell, B. and Hultman, B. 2005. A flowchart for integrated river basin planning with the Water Framework Directive.

European Water Management Online. (in press)

23

HELCOM, 2004a. 30 years of protecting the Baltic Sea, HELCOM 1974-2004 URL: www.helcom.fi/publications/30_years_jubilee.pdf

(08-12-2004)

HELCOM, 2004b. The forth Baltic Sea Pollution Load Compilation (PLC-4) Helsinki

Commission, 2003, Pro No. 93. Erneko Painotuote Oy, Finland, ISSN 0357-2994.

Jouravkova, O.N., 2003. On problems and perspectives of GIS applicationsfor transboundary water management in Europe: a case study of the Daugava/Zapadnaya

Dvina. TRITA-LWR Master Thesis LWR-EX-2003-03. ISSN 1651-064X.

Klavins, M., Rodinovs, V. and Kokorite, I., 2002. Chemistry of Surface Waters in

Latvia. LU, Riga.

Latvian Environment Agency, 2002. Environmental indicators in Latvia 2002. URL: http://www.grida.no/enrin/htmls/latvia/udeni/eitro/baseini_ind.htm

(25-11-2004)

Lundin, L-C. (editor), 2000. The Waterscape. Uppsala Universitet, The Baltic

University Programme.

MARE. URL: http://www.mare.su.se

(05-01-05)

Mörth, Magnus, Institutionen för geologi och geokemi, Stockholms universitet, 2004.

Personal communication.

Mourad, D. and van der Perk, M., 2004. Modelling nutrient fluxes from diffuse and point emissions to river loads: the Estonian part of the transboundary Lake

Peipsi/Chudskoe drainage basin. In: Water Science and Technology 49 (2004), pp. 21-

28.

Orchide Culture. URL: http://www.orchidculture.com/COD/daylength.html#55N

(05-01-11)

Rozins, E., 2001. The Ecological characteristics of the water bodies near Riga. URL: http://www.valt.helsinki.fi/projects/enviro/articles/a-rig.pdf (05-11-2004)

SWEPA (Swedish Environmental Protection Agency), 2003. URL: http://www.naturvardsverket.se

(Last update 03-10-2003)

SMHI, 2004. SMHI Oceanografi, Information och Data, Källor till övergödningen?.

URL: http://www.smhi.se/oceanografi/oce_info_data/eutrophication/sources_sv.html

(21-09-2004)

Stålnacke, P., 1996. Nutrient loads to the Baltic Sea. Linköping University, Department of Water and Environmental Studies.

Stålnacke, P., Grimvall, A., Libiseller, C., Laznik, M. and Kokorite, I., 2003. Trends in nutrient concentations in Latvian rivers and the response to the dramatic change in agriculture. In: Journal of Hydrology 283 (2003), pp. 184-205.

Vattenresurs AB, 2004. URL: http://www.vattenresurs.se

(10-2004)

24

Wulff, F. and L. Rahm, 1990. A database and its tools. Chapter 13 in: Wulff, F. (editor).

Large-scale environmental effects and ecological processes in the Baltic Sea. Research programme for the period 1990-1995 and background documents. SNV REPORT 3849.

Stockholm. pp. 217-225.

Ziverts, A. and Jauja, I., 1999. Mathematical Model of Hydrological Processes

METQ98 and its Applications. In: Nordic Hydrology 30(2) (1999), pp. 109-128.

25

Appendix A. Input data and parameters used by the GWLF model

Description of parameter/input

Recession coefficient box 1

Recession coefficient box 2

Groundwater transfer from box 1 to box 2

Seepage coefficient

Initial unsaturated storage

Initial saturated storage box 1

Initial saturated storage box 2

Initial snow cover

Sediment delivery ratio

Unsaturated water capacity

ET cover coefficient, oct-apr

ET cover coefficient may-sept

Day length APRIL

Day length MAY

Day length JUNE

Day length JULY

Day length AUG

Day length SEPT

Day length OCT

Day length NOV

Day length DEC

Day length JAN

Name

R1

R2 gr s

U(0)

S1(0)

S2(0)

Ht

Ht

Ht

Ht

Ht

Ht

Ht

DR

U*

CVt

CVt

Ht

Ht

Ht

Unit

day

-1 day

-1

Initial value

0.0025

0.001

day

-1 day

-1

0.001

0 cm 0 cm 18.6

cm 23.6

cm 20

0.009

cm 0

0.3

1

H 13.82

h 15.87

h 17.06

h h h 16.57

h 14.74

h 12.43

h 10.15

h 8.05

6.93

7.5

Calibrated

Value

0.03

0.01

0.01

2.7

1.7

1.5

0.35

0.83

26

Reference

Mörth, 2004

Mörth, 2004

Mörth, 2004

Average predicted value in March

Average predicted value in March

Average predicted value in March

Dai et al., 2000

Mörth, 2004

Dai et al., 2000

Dai et al., 2000

Orchid culture

Unit

h h

Initial value

9.37

11.46

May-Sept

0.25

0.06

ha

50819

0

4409294

510356

44831

0

1482044

1429719

509744

148463

Calibrated

Value

100

75

75

75

75

100

75

80

80

80

80

80

Reference

BALTEX Hydrological Data Centre data base

Dai et al., 2000

Hannerz and Destouni, 2005

Löfgren, 2004

27

Description of parameter/input

Mixed forest

Deciduous forest

Water

Soil erodibility factor

Topographic factor

Artificial surfaces

Bare areas

Cultivated areas

Herbaceous+shrub

Wetland

Snow and ice

Coniferous

Mixed forest

Deciduous forest

Water

Cover and management factor

Bare areas

Cultivated areas

Herbaceous+shrub

Wetland

Snow and ice

Coniferous

Mixed forest

Deciduous forest

Name

K

(LS)

C k k k

Unit Initial value

75

75

100

0.25

Calibrated

Value

80

80

2.233

5.594

3.680

2.094

3.862

4.325

3.587

2.388

0

0.01

0.01

0.01

0

0.001

0.001

0.001

28

Reference

Dai et al., 2000

Hannerz and Destouni, 2005

Dai et al., 2000

Description of parameter/input

Water

Supporting practice factor

Sediment N concentration

Sediment P concentration

N concentration groundwater box 1

N concentration groundwater box 2

P concentration groundwater box 1

P concentration groundwater box 2

Number of manured land uses

N concentration in runoff

Bare areas

Cultivated areas

Lichens and wetland

Snow and ice

Coniferous

Mixed forest

Deciduous forest

Water

P concentration in runoff

Artificial surfaces

Bare areas

Cultivated land

Herbaceous land

Name

P k

Cs

C s

C1gN

C2gN

C1gP

C2gP

Cd

Cd k k

Unit Initial value

0

0 mg/kg 2800 mg/kg 1276 mg/l mg/l

3

1 mg/l mg/l

0.005

0.002

0

Calibrated

Value

1.3

1 mg/l mg/l

1.6

6.5

1

0

0.5

0.5

0.5

1.6

0

0.01

0.17

0.01

29

Reference

Dai et al., 2000

Mörth, 2004

Mörth, 2004

Löfgren, 2004

Löfgren, 2004

Löfgren, 2004

Löfgren, 2004

Löfgren, 2004

Löfgren, 2004

Description of parameter/input

Lichens and wetlands

Snow and ice

Coniferous forest

Mixed forest

Deciduous forest

Water

Artificial surfaces

N build-up rate on artificial surfaces

P build-up rate on artificial surfaces

Point sources

Point N

Point P

Name

n k n k

DP m

DP m

Unit Initial value

0.02

0

0.009

0.009

0.01

0.01

Calibrated

Value

kg/ha/day

kg/ha/day kg/month 91000 kg/month 17000

0.101

0.0112

Reference

Dai et al., 2000

Dai et al., 2001

Latvian Environment Agency, 2002

Latvian Environment Agency, 2002

30

Appendix B. Determination of surface runoff

Surface runoff is computed from weather data by the U.S. Soil Conservation Service’s

Curve Number Equation (equation B1) (Dai et al., 2001).

Q kt

=

(

R t

R t

+

+

M t

M t

+

0 .

2

0 .

8

DS kt

DS kt

)

2

(B1) where

Q kt

= surface runoff from land use k on day t [cm]

R t

= rainfall on day t [cm]

M t

= snowmelt on day t [cm]

DS kt

= detention parameter for land use k on day t [cm]

Precipitation is assumed to be rain when the daily mean air temperature is more than

0 °C.

If there is a snow cover, snowmelt is computed by a degree-day equation:

M t

=

TC

T t

for T

t

> 0 °C (B2) where

TC = temperature coefficient for snowmelt [cm/°C]

T t

= mean air temperature on day t [°C]

The detention parameter DS kt

is determined from a curve number CN kt

:

DS kt

=

2540

25 .

4

CN kt

(B3)

Curve numbers are determined from a function of antecedent moisture as shown in figure B1. Curve numbers for antecedent moisture conditions 1 (driest), 2 (average) and

3 (wettest) are CN1 k

, CN2 k

and CN3 k

respectively.

31

Figure B1. Selection of curve number as a function of antecedent moisture.

Recommended values of the break points AM1 and AM2 have been used (Dai et al,

2000). The 5-day antecedent precipitation is computed as follows:

A t

=

n

=

t

1

t

5

(

R n

+

M n

) (B4)

CN2 k

values for different land uses are given in the BasinSim 1.0 User’s Guide (Dai et al, 2000).Values of CN1 k

and CN3 k are computed from CN2 k

:

CN

1

k

=

2 .

334

CN

2

k

0 , 01334

CN

2

k

(B5)

CN

3

k

=

0 .

4036

CN

2

k

0 .

0059

CN

2

k

(B6)

Appendix C. Determination of potential evapotranspiration

Potential evapotranspiration is given by the following equation (Dai et al., 2001):

PE t

=

0 .

021

T t

+

H t

273

e t

(C1) where

H t

= number of daylight hours per day during the month containing day t

e t

= saturated water vapour pressure on day t [mbar]

T t

= daily mean air temperature, average value for entire basin [°C]

The saturated water vapour pressure is estimated from daily mean air temperature:

32

e t

=

33 .

8639

[

(

0 , 00738

T t

+

0 .

8072

)

8

0 .

000019

(

1 .

8

T t

+

48

)

+

0 .

001316 ,

]

T t

0 (C2)

Appendix D. Determination of monthly sediment yield

The monthly watershed sediment yield in the GWLF is determined from a model that is based on three assumptions:

(i)

(ii)

(iii)

Sediment originates from sheet and rill erosion (gully and stream bank erosion are neglected)

Sediment transport capacity is proportional to runoff to the

5/3 power

Sediment yields are produced from soil which erodes in the current year (no carryover of sediment supply from one year to the next)

Accordingly, the sediment yield in a certain month will consist of contributions from all other months during that year. The contribution from each month will be equal to the sediment yield for that month multiplied by the fraction of total transport capacity in that month (Dai et al., 2001).

The total watershed sediment supply generated in month j is:

X j

=

DR d

∑∑

k t j

=

1

X kt

(D1) where

DR = watershed sediment delivery ratio

X kt

= erosion from land use k on day t [Mg]

d j

= number of days in month j

Erosion from land use k on day t is given by:

X kt

=

0 .

132

RE t

K k

(

LS

)

k

C k

P k

AR k

(D2) where

RE t

= rainfall erosivity on day t [(MJ*mm)/(ha/h)]

K k

= soil erodibility factor

(LS) k

= topographic factor

C k

= cover and management factor

P k

= supporting practice factor

The product K*LS*C*P is called the erosion factor. Standard values for K, C and P for

(Dai et al., 2000). LS is determined from data on topography. Rainfall erosivity is estimated by the following empirical equation:

RE t

=

64 .

6

a t

R t

1 .

81

(D3)

33

where

a t

= rainfall erosivity coefficient

R t

= rainfall on day t

The rainfall erosivity coefficient varies with season and geographic location.

In order to determine the transport of sediments in each month, a transport factor is defined as:

TR j

=

t d

j

=

1

Q t

5 / 3

(D4)

The sediment produced in month j will be allocated to months j, j+1, … 12 in proportion to the transport capacity of each month. The total transport capacity for months j, j+1, … 12 is proportional to the sum of the transport factors during these months, defined as B j

.

B j

=

h

12

=

j

TR h

(D5)

For each month, the fraction of available sediment X j

which contributes to Y m

, is

TR m

/B j

. The total monthly yield is the sum of all contributions from preceding months:

Y m

=

TR m j m

=

1

(

X j

/

B j

) (D6)

Appendix E. Determination of urban runoff nutrient loads

Urban surfaces are assumed to be impermeable. Nutrients accumulate on urban surfaces over time and are washed off by runoff events. The accumulation rate during dry periods is:

dN k dt

=

n k

− β ⋅

N k

where

N k

(t)= accumulated nutrient load on land use k on day t [kg/ha]

n k

= constant accumulation rate [kg/(ha*day)]

= depletion rate constant [day

-1

]

By solving equation E1 we obtain:

N k

(

t

)

=

N k

0

e

β

t

+

(

n k

/

β

)( 1

e

β

t

)

(E1)

(E2)

34

where

N k0

= N

k

(0)

With increasing time, equation E2 approaches an asymptotic value as the depletion rate approaches the accumulation rate:

N k

, max

=

t

lim

→ ∞

N k

(

t

)

=

n k

/

β

(E3)

Assuming that N k

(t) reaches 90 % of its maximum value in 20 days, the equations can be solved for . The result is = 0.12, which is the value that is used here.

Equation E2 can be written for a time interval of one day, which is the form that is used by the GWLF model:

N k

,

t

+

1

=

N kt e

0 .

12

+

(

n k

/ 0 .

12 )( 1

e

0 .

12

) (E4)

By adding a negative term to the right hand side of this equation, the effect of wash off is included:

N k

,

t

+

1

=

N kt e

0 .

12

+

(

n k

/ 0 .

12 )( 1

e

0 .

12

)

W kt

(E5) where

W kt

= runoff load from land use k on day t [kg/ha]

The runoff nutrient load at a wash off event is a function of the amount of accumulated nutrients at the time and the runoff flow:

W kt

=

w kt

[

N kt e

0 .

12

+

(

n k

/ 0 .

12 )( 1

e

0 .

12

)] (E6) where

w kt

=

1

e

1 .

81

Q kt

(E7)

Equation E7 is based on the assumption that 1.27 cm of runoff will wash off 90 % of accumulated pollutants.

Monthly runoff loads of urban nutrients are thus given by:

SU m

=

dm

∑∑

k t

=

1

(

W kt

AR k

)

(E8)

35

Appendix F. Regression lines of reported and modelled monthly and yearly values

7

a) Monthly streamflow, calibration period (cm/month)

y = 0,4771x + 1,0175

R

2

= 0,3627

6

5

4

3

2

1

0

0 2 4 6

Reported

8 10 12

7

6

5

4

9

8

3

2

1

0

b) M onthly stream flow , validation period (cm /m onth)

y = 0,6978x + 0,6178

R

2

= 0,5334

0 2 4 6 8 10

Reported

12

10000

c) Monthly nitrogen load, calibration period

(ton/m onth)

y = 0,4623x + 1245,9

R

2

= 0,3177

8000

6000

4000

2000

0

0 2000 4000 6000 8000

Reported

10000 12000 14000

d) Monthly nitrogen load, validation period (ton/m onth)

y = 0,4125x + 992,1

14000

R

2

= 0,4437

12000

10000

8000

6000

4000

2000

0

0 5000 10000 15000

Reported

20000 25000

100

80

60

40

140

120

e) Monthly phosphor us load, calibration period

(ton/m onth)

y = 0,0146x + 16,562

R

2

= 0,0058

20

0

0 200 400 600

Reported

800 1000

250

f) Monthly phosphorus load, validation period

(ton/m onth)

y = 0,1834x + 1,7934

R

2

= 0,1837

200

150

100

50

0

0 100 200 300

Reported

400 500

Figure F1. Regressions of monthly reported and modelled values of streamflow, nitrogen load and phosphorus loads for calibration period and validation period respectively.

600

36

25

20

15

10

5

0

40

35

30

0

a) Streamflow , calibration period (cm/year)

y = 1,2114x - 5,0907

R 2 = 0,6728

5 10 15 20

Reported

25 30 35

50000

45000

40000

35000

30000

25000

20000

15000

10000

5000

0

c) Nitrogen load, calibration per iod (ton/year)

y = 1,1719x - 4889,3

R 2 = 0,7828

0 10000 20000 30000 40000

Reported

50000

20

15

10

5

35

30

25

0

0

b) Stream flow , validation period (cm /year)

y = 1,5776x - 15,117

R

2

= 0,8156

5 10 15 20

Reported

25 30 35

45000

40000

35000

30000

25000

20000

15000

10000

5000

0

d) Nitrogen load, validation period (ton/year)

y = 0,7452x - 3601,1

R

2

= 0,5713

0 10000 20000 30000 40000 50000

Reported

60000

350

e) Phosphorus load, calibration per iod (ton/year)

y = 0,144x + 17,002

R

2

= 0,4992

300

250

200

150

100

50

0

0 500 1000

Reported

1500 2000

400

350

300

250

200

150

100

50

0

f) Phosphorus load, validation period (ton/year)

y = 0,2375x - 55,489

R

2

= 0,4702

0 500 1000 1500

Reported

2000

Figure F2. Regressions of yearly reported and modelled values of streamflow, nitrogen load and phosphorus loads for calibration period and validation period respectively.

37

Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Related manuals