Drainage for agricultural purposes; how to distinguish and incorporate it in model structures?

Euser, Tanja

Abstract:

1. Introduction

A lumped parsimonious model does not represent detailed catchment processes, but rather the average hydrological behaviour of a catchment. These models have shown their value for operation forecasts, as they can be calibrated relatively easy and produce reasonable forecasts, as long as the climatic conditions do not differ strongly from the calibration period. However, more detailed process representation is required to increase our understanding of the hydrological behaviour of catchments and to increase the predictive power of models under different climatic conditions. More detailed and distributed process representation can be done without making the model unnecessary complex, by using available data and expert knowledge in a systematic way.

Previous studies have shown that the realism of models improves, if processes are only modelled in those parts of the catchment where they are dominant (e.g. Beven and Freer, 2001; Uhlenbrook et al., 2004); topography appears to be a good indicator for the occurrence of specific runoff processes (e.g. Detty and McGuire, 2010; Rodhe and Seibert, 1999). A specific study in the Ourthe catchment in Belgium (Euser et al., 2014) showed that for this catchment the realism was increased by incorporating a differentiation between wetlands (areas with saturation overland flow as dominant mechanism; Savenije, 2010) and the remainder of the catchment (see Figure 1 for used model structure). However, still some shortcomings could be identified. For example, a trade-off existed between the representation of the peak distribution and other hydrological signatures.

Drainage for agricultural purposes occurs in different areas around the world; this artificial drainage can significantly change the hydrological behaviour of a catchment. Therefore, it is useful to understand how this drainage works and how we can incorporate it in our models (cf. Brauer et al., 2014). In the Ourthe catchment the majority of the wetlands is artificially drained. In the first case study (Euser et al., 2014), these areas are included in the remainder of the catchment, as it is more plausible that the dominant runoff mechanism is quick subsurface flow than saturation overland flow. However, it is expected that artificial drainage functions differently from a natural hillslope; therefore, for this study drained wetland is distinguished as well. This study aims at answering the research question if incorporating artificial drainage for agricultural purposes can help to increase model performance and consistency. Model performance is defined as the ability of a model to reproduce a specific signature; model consistency is defined as the ability of a model to reasonably reproduce multiple signatures with the same parameter set (Euser et al., 2013).

2. Approach

2.1 Study area

For this study, one of the two upstream tributaries of the Ourthe is used: Ourthe Occidentale. An upstream tributary is selected instead of the entire catchment, because the upstream part of the catchment mainly consists of relative flat areas, so a lot of drainage can be expected. About 60% of the area is used for agriculture.

2.2 Identification of drained wetlands

Before drainage can be included in the model structure, the areas where drainage occurs need to be identified. Two assumptions are used to identify drainage; first, only flat areas are being drained; second, subsurface drainage only occurs in land used for crops or forestry, as subsurface drainage for pasture land would not be economical feasible. So, the classification was based on both topography and land use (cf. Gao et al., 2014). Areas with a Height Above Nearest Stream (HAND; Gharari et al., 2011; Rennó et al., 2008) smaller than 5.9m are classified as wetland. Areas in these wetlands covered with arable land or forest are following classified as drained wetland.

Not only the areas with a HAND lower than 5.9m have low slopes; areas with a high HAND and low slopes occur in the catchment as well. When these areas are used for agriculture, it is possible they are (slightly) drained as well. This will be investigated in a next step, to be able to identify the added value of different model adaptations.

2.3 Conceptualisation of drained wetlands

Different model conceptualisations can be configured to represent the drained areas, depending on the strength of the drainage and the dominant runoff mechanisms of the areas before drainage. The conceptualisation used in this study consists of an interception reservoir and a combined unsaturated and saturated zone. The drainage is represented by an linear outflow just above the lowest level of the reservoir; further, runoff is generated from the wetland by an overflow function: once the unsaturated zone is full, the wetland starts discharging. For the drained and non-drained wetlands the same model configurations will be used in parallel, together with a parallel model structure for the remainder of the catchment (Figure 1). The dominant runoff mechanism in the remainder of the catchment is assumed to be quick subsurface flow; therefore, these areas are referred to as 'hillslopes' (Savenije, 2010).

2.4 Conditioning of model configurations

The use of parallel model structures increases the amount of parameters which need to be determined before the model can be evaluated and compared with other model configurations. However, as the difference in model structure and behaviour are compared and the models are not used for forecasting studies, a set of behavioural parameter sets can be used. Behavioural can be defined differently for each study; here it is defined as satisfying all relevant constraints.

Two kind of constraints are used: parameter and process constraints (Gharari et al., 2013). Parameter constraints are applied on the parameters before a model run and describe the expected relation between the parameters of two classes. For example, the interception storage capacity in a wetland (covered with grass) is expected to be smaller than that of a hillslope (covered with forest). The process constraints are applied on the modelled fluxes and can describe a more general behaviour or the expected relation between the fluxes of two classes as well. For example, the modelled yearly runoff coefficient should be bounded by the annual minimum and maximum observed runoff coefficient over a period around the modelling period, or the overflow from the drained wetland (Q_o,WD) should always be smaller than the overflow from the non-drained wetland (Q_o,W).

Table 1| Parameter and process constraints; during further evaluation the number of constraints can be extended

Parameter constraints	Process constraints
S_usmax2,WD - S_usmax1,WD < S_umax,H	Q_o,WD < _Qo,W
S_usmax2,W < S_umax,H	Q_o,WD > 0, max 2-3 events per year
I_max,H > I_max,W	C_r,max = 0.55
I_max,H > I_max,WD	C_r,min = 0.4
K_f,H (h) > K_d,WD(h)
L_p,H < L_p,W
L_p,H < L_p,WD

2.5 Evaluation of model configurations

A model configuration with two classes already functioned well, so mainly small differences can be expected in comparison with the three class version. Therefore, a comparison based on a couple of performance criteria might not reveal any differences. The differences will probably be visible in the modelled hydrographs, however, also here they will be small. The stages and fluxes from the different classes will show a clear difference, as an additional class was added and the conceptualisation of the non-drained wetland class was adapted. Thus, some of the states and internal fluxes were used for a first comparison.

3. Preliminary results

Figure 2 shows a first comparison between a model configuration with two (blue lines) and with three (red lines) parallel model structures for the same parameter values where possible. The comparison of the states shows that the difference for the wetland is limited: the storage only becomes a bit more peaky. For the drained wetland, however, the storage is not really comparable with the wetland or hillslope behaviour of the two class configuration: the storage in the drained wetland shows a larger annual component. The comparison of the internal fluxes shows that the overflow from the wetland in the three class version is much more concentrated than the fast runoff from the wetland in the two class version. For the drained wetland it can be seen that the drainage flow is earlier activated than the fast runoff from the wetland in the two class version and that the drainage flow is much larger than the fast runoff from the hillslope in the two class version.

The changed behaviour of the quick flow from the non-drained wetland could help to better capture the highest peaks. The difference in behaviour between the drained wetland and the hillslope, could help to better represent the peak distribution. However, to fully assess the value of the additional drained wetland class a more thorough analysis is required.

4. Discussion and outlook

The results presented above are a first comparison; a more thorough analysis is required to really assess the value of an additional drained wetland class. An important aspect in this analysis is to evaluate whether a possible increase in performance and consistency is worth the additional complexity. Previous studies have shown that parallel model structures have a higher complexity, but also a higher degree of realism (Gao et al., 2014; Gharari et al., 2013).

Possibilities for further analysis are evaluation of the entire hydrograph and the contributions of fluxes to the total discharge: different contributions are expected for different periods through the year. Following, the FARM method (Euser et al., 2013) can be used to assess both performance and consistency. For this analysis different signatures should be used, of which a couple can be selected on specifically describing wetland behaviour. Finally, the final model configurations should be tested on a different period and on another catchment or tributary.

The model configuration presented above only accounts for drainage in areas close to the stream. However, flat areas higher above the stream can be drained for agricultural purposes as well. If the above described configuration functions well, it is actually expected that the higher flat areas are drained as well, as the majority of the remainder of the catchment has a low slope. In addition, for this study, the distinction between wetland and remainder of the catchment is set at a HAND of 5.9m; however, it could be the case that this is not a fixed value, but dependent on the degree of saturation in the catchment (cf. Beven and Kirkby, 1979). This can be tested together with the drainage of the areas higher above the stream.

References

Beven, K. and Freer, J.: A dynamic TOPMODEL, Hydrol. Process., 15(10), 1993-2011, doi:10.1002/hyp.252, 2001.

Beven, K. J. and Kirkby, M. J.: A physically based, variable contributing area model of basin hydrology, Hydrol. Sci. Bull., 24(1), 43-69, doi:10.1080/02626667909491834, 1979.

Brauer, C. C., Teuling, A. J., Torfs, P. J. J. F. and Uijlenhoet, R.: The Wageningen Lowland Runoff Simulator (WALRUS): a lumped rainfall-runoff model for catchments with shallow groundwater, Geosci. Model Dev. Discuss., 7(1), 1357-1411, doi:10.5194/gmdd-7-1357-2014, 2014.

Detty, J. M. and McGuire, K. J.: Topographic controls on shallow groundwater dynamics: implications of hydrologic connectivity between hillslopes and riparian zones in a till mantled catchment, Hydrol. Process., 24(16), 2222-2236, doi:10.1002/hyp.7656, 2010.

<p"argin-bottom: line-height:="line-height:" .0001pt;=".0001pt;" normal;"="normal;"">Euser, T., Winsemius, H.C., Hrachowitz, H., Savenije, H.H.G.: Effect of spatial forcing data and landscape heterogeneity on performance and consistency of model structures, Geophysical Research Abstracts, 16, EGU2014-5025, 2014.Euser, T., Winsemius, H. C., Hrachowitz, M., Fenicia, F., Uhlenbrook, S. and Savenije, H. H. G.: A framework to assess the realism of model structures using hydrological signatures, Hydrol. Earth Syst. Sci., 17(5), 1893–1912, doi:10.5194/hess-17-1893-2013, 2013.

Gao, H., Hrachowitz, M., Fenicia, F., Gharari, S. and Savenije, H. H. G.: Testing the realism of a topography-driven model (FLEX-Topo) in the nested catchments of the Upper Heihe, China, Hydrol. Earth Syst. Sci., 18(5), 1895-1915, doi:10.5194/hess-18-1895-2014, 2014.

Gharari, S., Hrachowitz, M., Fenicia, F., Gao, H. and Savenije, H. H. G.: Using expert knowledge to increase realism in environmental system models can dramatically reduce the need for calibration, Hydrol. Earth Syst. Sci. Discuss., 10(12), 14801-14855, doi:10.5194/hessd-10-14801-2013, 2013.

Gharari, S., Hrachowitz, M., Fenicia, F. and Savenije, H. H. G.: Hydrological landscape classification: investigating the performance of HAND based landscape classifications in a central European meso-scale catchment, Hydrol. Earth Syst. Sci., 15(11), 3275-3291, doi:10.5194/hess-15-3275-2011, 2011.

Rennó, C. D., Nobre, A. D., Cuartas, L. A., Soares, J. V., Hodnett, M. G., Tomasella, J. and Waterloo, M. J.: HAND, a new terrain descriptor using SRTM-DEM: Mapping terra-firme rainforest environments in Amazonia, Remote Sens. Environ., 112(9), 3469-3481, doi:10.1016/j.rse.2008.03.018, 2008.

Rodhe, A. and Seibert, J.: Wetland occurrence in relation to topography: a test of topographic indices as moisture indicators, Agric. For. Meteorol., 98-99, 325-340, doi:10.1016/S0168-1923(99)00104-5, 1999.

Savenije, H. H. G.: HESS Opinions "Topography driven conceptual modelling (FLEX-Topo)", Hydrol. Earth Syst. Sci., 14(12), 2681-2692, doi:10.5194/hess-14-2681-2010, 2010.

Uhlenbrook, S., Roser, S. and Tilch, N.: Hydrological process representation at the meso-scale: the potential of a distributed, conceptual catchment model, J. Hydrol., 291(3-4), 278-296, doi:10.1016/j.jhydrol.2003.12.038, 2004. <p"argin-bottom: line-height:="line-height:" .0001pt;=".0001pt;" normal;"="normal;"">Euser, T., Winsemius, H.C., Hrachowitz, H., Savenije, H.H.G.: Effect of spatial forcing data and landscape heterogeneity on performance and consistency of model structures, Geophysical Research Abstracts, 16, EGU2014-5025, 2014.