Modelling the hydrological impacts of channelization on streamflow characteristics in a Northern Ireland catchment

Transcription

1 Modelling and Management of Sustainable Basin-scale Water Resource Systems (Proceedings of a Boulder Symposium, July 1995). IAHS Publ. no. 231, Modelling the hydrological impacts of channelization on streamflow characteristics in a Northern Ireland catchment DAVID WILCOCK & FRANCES WILCOCK School of Environmental Studies, University of Ulster, Coleraine, County Londonderry, Northern Ireland, UK Abstract A lumped catchment model designed to predict daily flows is calibrated for a 200 km 2 catchment in Northern Ireland using data collected prior to a major channelization scheme in The calibrated model is used to predict a two-year sequence of post-channelization flows for comparison with observed post-channelization flows. Channelization appears to systematically increase flood peaks and diminish low flows. The analysis is repeated for an upstream, tributary, catchment on which channelization did not take place. Model parameters calibrated for a three-year period on this catchment before 1984 are used to predict hydrographs for the post-1984 period. These conform in most respects with observed hydrographs. INTRODUCTION Applied to rivers, the concept of sustainable development ultimately requires the ability to predict the impacts of contemporary catchment management on streamflow characteristics as these provide the most rational criteria for managing water abstraction rates, pollution loads and, arguably, aquatic ecology (Bovee, 1982; Bullock & Johnson, 1991). The most widespread cultural impact on the hydrology of Northern Ireland's rivers in the last 50 years has been brought about by channelization. Initial studies on the hydrological impacts of channelization in Northern Ireland have been able to focus only on such summary statistics as the annual and monthly water balances, mean daily flows, flow duration curves and flood peaks (Essery & Wilcock, 1990a; 1990b). The development of catchment modelling packages now facilitates study of real-time hydrological impacts. Using HYRROM, a rainfall-runoff modelling package developed by the UK Institute of Hydrology, this paper describes an attempt to identify the effects of channelization on the continuous annual hydrograph of a recently channelized river. THE MODELLING PACKAGE - HYRROM HYRROM (Institute of Hydrology, 1989) is a lumped catchment rainfall-runoff model (Blackie & Eeles, 1985; Eeles et al., 1990) containing nine parameters. These parameters are available for optimization (either interactive or automatic) since they are the most sensitive to change. They are complemented by a further ten parameters which

2 42 David Wilcock & Frances Wilcock have fixed values pre-set empirically in the model computing code. Input to the model is in the form of catchment daily rainfall. Estimates of catchment daily evaporation form the model's "loss" function. Daily rainfall and evaporation estimates are incorporated into the model datafileas input, and output from these is processed to give the simulated hydrograph. Modelling proceeds by optimizing the full set of nine parameter values over a length of record for which observed hydrographs are available. In the present study an interactive optimization was undertaken. The process involves varying the nine parameters to produce a succession of simulated hydrographs which are graphically overlaid on the observed hydrograph. When the match between observed and simulated hydrograph is as good as the observer requires, the nine parameters are presumed to characterize the catchment under consideration and can be used to predict hydrographs over other periods of time for which daily rainfall and evaporation values are available. An Objective Function quantifies levels of agreement between predicted and observed hydrographs. This is the function which is minimized by the optimization process. It is defined (in m 3 s" 1 ) as follows: i _ where Q 0 is observed daily flow, Q p predicted daily flow and Nthe number of days used to calibrate the parameters. The first of the nine parameters in HYRROM (SS) controls the size (in mm) of the vegetation and surface detention stores. This parameter is free to vary between 0 and 5.0. A second parameter (RQ defines the proportion of rainfall entering the surface routing store and is allowed to vary between 0 and 1.0. Low values of RC imply relatively large amounts of infiltration to the soil moisture store. High values imply large amounts of water entering the river via surface stores. Three parameters determine how the surface routing store contributes to runoff. RDEL measures the time difference, in days, between water leaving the surface store and leaving the catchment as streamflow. RDEL must exceed 0. Two parameters, RK and RX, are respectively coefficient and power function in the equation: Runoff = RK (Current volume in storage)** RX must exceed 1.0 and RK can vary only between 0 and 1.0. A sixth parameter, FC, defines the relative importance of évapotranspiration and is allowed to vary between 0.3 and 1.0. Higher FC values imply more évapotranspiration. A final set of three factors define the contribution of the groundwater store. GDEL measures the length of time in days plus one, between water leaving the groundwater store and leaving the catchment as runoff. Its function in the model is similar to that of RDEL on the surface routing store. GSU and GSP define the relationship between runoff and the groundwater store in the following equation: Groundwater runoff = (Current volume in store/gsu) Gsp GSU is arbitrarily constrained to above 30 by the model and GSP must exceed 1.0.

3 Modelling hydrological impacts of channelization on streamflow characteristics 43 Lumped catchment models like HYRROM need to be used with care (Bevan, 1989). They imply a uniformity in parameter values across a single catchment and the precise physical processes represented by each of the parameters is often difficult to define. Parameter values may also be interdependent, and model calibration is heavily dependent on accurate input data. Despite these theoretical weaknesses, lumped catchment models have been found to work quite well in different situations. THE STUDY AREA Channelization on the River Main above Dunminning began in April 1984, reaching Dunloy in 1987 (Fig. 1). At the Dunminning gauge, 80 m above sea-level, the River Main catchment area is km 2. The unchannelized Clogh catchment provides "control" data for the study and has a catchment area of 95 km 2. Extensive areas of Fig. 1 The study area.

4 44 David Wilcock & Frances Wilcock blanket peat cover the upper slopes of both drainage basins, which reach 543 m at their highest point. Impermeable tertiary basalts underlie the study area, covered by impermeable boulder clays and peats, though glacial gravels protrude through alluvium in the Main floodplain above Dunminning. Extensive areas of raised bog occur in the lower Main valley and 730 ha of flood plain here was subject to frequent flooding before channelization. DATA COLLECTION Proposals by the Northern Ireland Department of Agriculture to channelize the River Main were known in advance and the catchment above Dunminning was instrumented with rainfall, streamflow and groundwater level recorders from 1978, though streamflow gauges on tributary catchments only became operational in October The rainfall record throughout the study period has been interrupted by changes in cooperating personnel, and the streamflow gauging section at Dunminning relocated twice, in 1981 and 1982, because of sluice gate installation work. Runoff records on the River Clogh are continuous from October The application of HYRROM to the River Main catchment upstream of the Dunminning gauge is described for two 2-year periods: a pre-channelization period from April 1979 to March 1981 and a post-channelization period between October 1984 and September These are the pre-channelization periods for which daily streamflow records are considered most accurate and suitable for use in a daily-flow modelling exercise. The Clogh has been unaffected by channelization or any other catchment works in the period 1980 to Optimization of HYRROM's nine parameters for this catchment is undertaken using data, corresponding to the period before the River Main was channelized. The model's calibration is checked against runoff data from 1983 to 1986 when post-channelization conditions existed on the Main. Rainfall HYRROM accepts daily data for only 3 rain gauges. For the modelling period on the River Main, the rainfall stations used were those at Altnahinch, Cloghmills and Broughshane. Data from Patterson's gauge replaced those for Cloghmills in the post-channelization modelling period, Modelling on the Clogh catchment uses rainfall data from Altnahinch, Broughshane and Patterson's for the whole 6-year period. Thiessen weighting factors are used to estimate daily catchment rainfall. Streamflow Drainage areas at the two streamflow gauges on the Main and Clogh are km 2 and 95.0 km 2 respectively. Daily flows for both catchments are derived from water level recordings at 3 hourly intervals and stage-discharge rating curves, each constructed from at least 18 current-metered flow measurements. HYRROM's optimization procedure requires a continuous run of daily data, and 22 days of missing streamflow were interpolated using crude water balance estimates.

5 Modelling hydrological impacts of channelization on streamflow characteristics 45 Evapotranspiration Monthly Penman potential évapotranspiration (PT) values are available for Altnahinch (altitude250 m) and Lisnafillan (altitude38 m). Mean annual P7between 1978 and 1986 is similar at both sites mm at Altnahinch and mm at Lisnafillan - and mean daily PT at the two sites was, therefore, calculated for each month of record, averaged for the two sites, and used as input into HYRROM for both catchments. RESULTS River Main The following adjustments were made, iteratively and incrementally, to HYRROM's default parameter values to produce the match between predicted and observed prechannelization flows illustrated in Fig. 2(a). SS was increased from 3.1 to 5.0, the maximum value allowed by the model, to reflect flood plain storage available in prechannelization conditions. RC was decreased, from 0.7 to 0.61, on the grounds thatprechannelization flood plain storage also recharged groundwater aquifers, thus diverting precipitation to the groundwater store. RDEL was increased from 0.2 to 0.91 to reflect the large catchment area of the River Main relative to those for which the model was designed. RX was reduced, from 2.6 to 1.47, again primarily to reflect the relatively large size of the River Main catchment and the relatively low frequencies with which all surface routing stores are filled simultaneously. A final RK value of 0.05 indicates a high sensitivity to changes in the surface routing store content even with low store volumes. This is taken to reflect the large flood plain surface storage immediately adjacent to the river and the high soil impermeabilities in the catchment generally. The évapotranspiration parameter, FC, was reduced from 0.7 to FC is permitted to range only between 0.3 and 1.0, low values being associated with low annual évapotranspiration amounts. Northern Ireland experiences some of the lowest évapotranspiration rates in the UK and a weighting factor towards the lower part of the permitted spectrum of values seems appropriate. GSU was reduced to 40.0 from a default value of 90.4 on the logic of a low capacity to sustain groundwater flow for any length of time in most of the catchment. GSP was reduced, from 1.7 to 1.0, to take into account the large size of the Main catchment and the low hydraulic conductivity of most of the catchment soils. A value of 1.0 is the lowest permitted value for GSP in the HYRROM model. The groundwater delay parameter, GDEL, was increased from 0.5 to 4.1 on similar logic. These changes produced an Objective Function of and a predicted total flow for the two years only 0.8% greater than that observed. The periods of worst agreement between observed and predicted hydrographs usually correspond: (a) with the two periods of estimated observed daily flow; and (b) with short periods of snowfall and/or frost in winter for which HYRROM is not designed to cope and which represent insignificant events in the River Main's regular hydrological calendar. The logic of the optimization exercise is presented above despite warnings in the literature about attaching too much precise physical significance to individual parameters. Manual optimization must proceed by some hydrological logic, however, and

6 David Wilcock & Frances Wilcock Apr Jul Nov Oct Jan Jun Fig. 2 Observed and predicted hydrographs on the River Main at Dunminning: (a) for before channelization and (b) for after channelization. Note that the predicted hydrograph for is derived from parameters optimized on the data. a set of parameters can only be accepted if they appear internally coherent and justifiable on hydrological grounds. In the case of the pre-channelization data the predicted hydrograph appears sensitive to individual rainfall events and closely reflects the timing and magnitude of all but the most extreme changes in the observed hydrograph. Parameter values derived from optimization of the pre-channelization data on the River Main were used to predict post-channelization hydrographs (Fig. 2(b)). This analysis produced an Objective Function of and predicted total flows exceeded observed flows by 2.62%. It is frequently argued that the effect of channelization on river flows is to increase high flows and reduce low flows, principally through the elimination of flood plain storage and improved hydraulic efficiency in the new channel. Both effects appear to be systematically displayed in Fig. 2(b). Increased high flows resulting from channelization in Ireland have been demonstrated by Bailey & Bree (1981), among others, but a systematic drawdown of low flows such as that shown in Fig. 2(b) has been more difficult to demonstrate (Essery & Wilcock, 1990). The River Clogh Optimization of HYRROM parameters for the River Clogh data followed the same principles as described for the River Main but took into account the River Clogh's smaller catchment area, steeper slopes, higher altitude, and lack of extensivefloodplain storage. An Objective Function of could not be improved and was brought about by increases to the RC, RX and GSP parameters, and by decreases in SS, RDEL, FX, GSU and GDEL. The visual fit between observed and predicted flows (Fig. 3(a)) is good, though the highest flood peaks tend to be underestimated. This is probably

7 Modelling hydrological impacts of channelization on stream/low characteristics (b) Observed flow Predicted flow Oct Apr Sep Oct Apr Sep Fig. 3 Observed and predicted hydrographs on the River Clogh (a) for and (b) for Note that the predicted hydrograph for the period is derived from parameters optimized on the data. because they overtop the channel banks at the Clogh gauge and are exaggerated by extrapolation of the stage-discharge rating curve beyond the bankfull stage. The difference between total observed and predicted flows for the data on the Clogh is 4.49%. Use of the parameter values to predict the three-year hydrograph for the Clogh data produces an Objective Function of 1.63 and an under-prediction of total observed flows of 4.9%, presumably resulting, again, primarily from measurement errors at high flow. Visual agreement elsewhere on the hydrographs is close, particularly for low flows and recession curves (Fig. 3(b)). The successful prediction of a three-year hydrograph on the Clogh for the years using parameters derived from the period suggests parameter stability over the whole 6 year period on this unmodified catchment, in contrast to the situation on the River Main where parameter stability appears to have been interrupted by channelization. CONCLUSIONS (a) HYRROM is a flexible modelling package which can be successfully applied to impervious catchments in Ireland up to 200 km 2 in area. (b) The model is sensitive enough to identify systematic discontinuities in runoff responses to rainfall produced by channelization. (c) Similar discontinuities are not present on an adjacent "control" catchment and, therefore, do not appear to be produced by changes to model inputs.

8 48 David Wilcock & Frances Wilcock Acknowledgements We thank C. W. O. Eeles, senior hydrological consultant for SINCAT/Atkins, for his comments and advice on the first draft of the manuscript, and Killian McDaid and Mark Millar of our own School for the diagrams. REFERENCES Bailey, A. D. & Brec, T. (1981) The effect of improved land drainage on river flood flows. In: The Flood Studies Report, Five Years On, Institution of Civil Engineers, London. Bevan, K. (1989) Changing ideas in hydrology - the case of physically-based models. J. Hydro!. 105, Blackie, J. R. & Eeles, C. W. O (1985) Lumped catchment models. In: Hydrological Forecasting (ed. by M. G. Anderson &T. P. Burt), Wiley. Bovee, K. D. (1982) A guide to stream habitat analysis using the instream flow incremental methodology. Instream Flow Information Paper no. 12, US Dept. of the Interior, Fish and Wildlife Service, Office of Biological Serv., FWS/OBS- 82/26. Bullock, A. & Johnson, I. (1991) Towards the settingof ecologically acceptable flow regimes with IFIM. Proc. 3rd National Hydrology Symp. (Southampton), British Hydrological Soc. Eeles, C W. O., Robinson, M. & Ward, R. C. (1990) Experimental basins and experimental models. In: Hydrological Research Basins and the Environment (ed. by J. C. Hooghart, C. W. S. Posthumus & P. M. M. Warmerdam), The Netherlands Organization for Applied Scientific Research (TNO), The Hague. Essery, C. I. & Wilcock, D. N. (1990a) The impact of channelization on the hydrology of the upper River Main, County Antrim, Northern Ireland - a long-term case study. Regulated Rivers: Research & Management 5, Essery, C. I. & Wilcock, D. N. (1990b) Quantifying the hydrological impacts of a major arterial drainage scheme on a 200 km 2 river basin. In: Hydrological Research Basins and the Environment (ed. by J. C. Hooghart, C. W. S. Posthumus&P. M. M. Warmerdam), The Netherlands Organization for Applied Scientific Research (TNO), The Hague. Institute of Hydrology (1989) HYRROM Operation Manual, Rainfall Runoff Modelling System Version 3.1. Institute of Hydrology, Wallingford, Oxfordshire, UK.