From meso- to macro-scale dynamic water quality modelling for the assessment of land use change scenarios

Abstract

The implementation of the European Water Framework Directive requires reliable tools to predict the water quality situations in streams caused by planned land use changes at the scale of large regional river basins. This paper presents the results of modelling the in-stream nitrogen load and concentration within the macro-scale basin of the Saale river (24,167 km²) using a semi-distributed process-based ecohydrological dynamic model SWIM (Soil and Water Integrated Model). The simulated load and concentration at the last gauge of the basin show that SWIM is capable to provide a satisfactory result for a large basin. The uncertainty analysis indicates the importance of realistic input data for agricultural management, and that the calibration of parameters can compensate the uncertainty in the input data to a certain extent. A hypothesis about the distributed nutrient retention parameters for macro-scale basins was tested aimed in improvement of the simulation results at the intermediate gauges and the outlet. To verify the hypothesis, the retention parameters were firstly proved to have a reasonable representation of the denitrification conditions in six meso-scale catchments. The area of the Saale region was classified depending on denitrification conditions in soil and groundwater into three classes (poor, neutral and good), and the distributed parameters were applied. However, the hypothesis about the usefulness of distributed retention parameters for macro-scale basins was not confirmed. Since the agricultural management is different in the sub-regions of the Saale basin, land use change scenarios were evaluated for two meso-scale subbasins of the Saale. The scenario results show that the optimal agricultural land use and management are essential for the reduction in nutrient load and improvement of water quality to meet the objectives of the European Water Framework Directive and in view of the regional development plans for future.

Introduction

The increasing input of nutrients in rivers induced or caused by point sources (sewage treatment plants, industrial enterprises and municipal wastewater) and diffuse sources (agricultural land, atmosphere) often leads to pollution and degradation of water quality and can cause ecological changes in freshwater. As the discharge of polluted wastewater has been notably reduced by wastewater treatment in the recent decades, agriculture is generally perceived as the main source of nutrient input to rivers in Europe (De Wit et al., 2002). In recognition of the importance of water quality problem and the need for integrated management in river basins, the European Water Framework Directive (WFD) was adopted in 2000, aiming to restore the “good ecological status” at the scale of large river systems until 2015 (EC, 2000).

Water quality models could provide an important and valuable support for the assessment and analysis of pollution loads in river basins and possible measures to improve water quality, and therefore they could be a useful instrument for fulfilling the requirements of the Water Framework Directive. Hence, various water quality models for the basin scale were developed and tested in the last decades (Arheimer and Brandt, 1998, Whitehead et al., 1998, Bicknell et al., 1997, Arnold et al., 1998, Krysanova et al., 1998). These models vary in the level of complexity, from statistical and conceptual models based on statistical and empirical relationships to process-based and physically based models derived from physical and physicochemical laws and including also some equations based on empirical knowledge. Due to the lack of description of important physical processes, the simplified conceptual models have limited applicability, and are not appropriate for simulation of some essential spatially distributed processes. The physically based models are by definition fully distributed accounting for spatial variations in all variables. However, these models are often considered too complex without a guarantee of better performance (Beven, 1989, Beven, 1996). The requirement of large amount of input data and computation resources for such models is also a particulate concern, especially for large basin simulations (Lunn et al., 1996). This indicates that using the process-based models (Krysanova et al., 2005a) of intermediate complexity for the basin-scale water quality assessment may be sufficient and even more promising. Numerous studies have proven that the process-based models are spatially explicit and sufficiently adequate to represent major hydrological, biogeochemical and vegetation growth processes at the catchment scale (Arnold and Allen, 1996, Chaplot et al., 2004, Krysanova et al., 2005b, Stewart et al., 2006, Hattermann et al., 2006). The models SWAT (Arnold et al., 1998), HSPF (Bicknell et al., 1997), SWIM (Krysanova et al., 1998) and DWSM (Borah et al., 2004) belong to the process-based modelling tools for river basins. However, in the most of the published studies the validation is performed using the catchment outlet data only, which provides no information on how they perform spatially (Cherry et al., 2008).

Besides, most of the current catchment scale studies with dynamic process-based models focus on micro-scale and small catchments ranging from 1 to 100 km² (Du et al., 2006, Chaplot et al., 2004, Eisele et al., 2001 and Bogena et al., 2003) and meso-scale basins with the drainage area from 100 to 5000 km² (Abbaspour et al., 2007, Saleh et al., 2000 and Volk et al., 2008). The process-based modelling at the macro-scale is very rare in literature, where the statistical and conceptual riverine load models are usually applied. One example is by Even et al. (2007), who simulated water quality conditions in the Seine River basin (78650 km²) using four coupled deterministic models with satisfactory seasonal results at different stations. Since the WFD requires the assessment on large river basin scale, more efforts are needed on improving the capabilities of describing nutrient fluxes in large basins.

As all necessary input data, such as crop rotations and fertilization schedule, and parameters of the current process-based models usually cannot be precisely defined at the basin scale, there are considerable uncertainties in the water quality modelling. Moreover, the process-based water quality models often confront the problem of over-parameterization, which results in the equal or very similar model result using various parameter combinations (problem of equifinality) (Oreskes et al., 1994). There are already many published studies on uncertainty in water quality modelling related to physical data such as rainfall (Bertoni, 2001), spatial data such as land use maps (Payraudeau et al., 2004), and parameters (Abbaspour et al., 2007, Hattermann et al., 2006). The uncertainty analysis allows to find the most important factors controlling the model behaviour, and to represent the modelling results in a more reliable way.

Recent studies on land use change impacts have demonstrated that the process-based water quality models are effective tools to simulate the hypothetic land use scenarios. They are helpful to either evaluate the long-term impacts of implementation of the current management practices (Santhi et al., 2005), or to investigate the optimal solutions in order to reduce the nutrients loads in rivers (Zammit et al., 2005, Hesse et al., 2008 and Volk et al., 2008). However, these studies did not account for some new tendencies in land use, like the influence of introducing energy plants on the water quality aspect in the future. In this study the land use scenarios are based on various measures of controlling nutrient emissions, and include the potential agricultural practice changes due to the development of the energy market.

Hence, the objectives of the study were:

• to test the applicability of the process-based ecohydrological model SWIM for simulating nitrogen dynamics in a macro-scale basin considering the outlet and intermediate gauges as validation points,

• to investigate the uncertainties related to input data and parametrization,

• to test the hypothesis about the usefulness of distributed retention parameters for nitrogen simulation in a macro-scale basin, and to establish ranges of retention parameters depending on the catchment characteristics, and

• to evaluate impacts on nitrate nitrogen load under the potential land use changes.

In this study, both input data and a set of most sensitive parameters for nitrogen dynamics were included in the uncertainty analysis. The most effective factors revealed in the analysis could explain the difficulties in simulating nutrient loads simultaneously in a large basin and its subbasins, and to suggest the ways for improvement.