David A. Horn1, Peter Yeates1, Jorg Imberger1 and
Angelos N. Findikakis2
1Centre for Water Research, The University of Western Australia, Crawley
W.A., Australia.
2Bechtel Systems & Infrastructure Inc., San Francisco, California, U.S.A.
Correspondence: Dr David Horn, Centre for Water Research, The University of Western Australia, 35 Stirling Highway, Crawley W.A., Australia, 6009.
Telephone: +61-8-93801684; Fax: +61-8-93801015; E-mail: horn@cwr.uwa.edu.au.
Abstract: The one-dimensional water quality model DYRESM-WQ was applied to Lake Maracaibo. The model was calibrated and validated against field data and shown to reproduce the observed salinity stratification and profiles of key water quality parameters. DYRESM-WQ was then used to assess the long-term effects of a number of remediation options. The simulation results confirm that the salinity stratification is a major influence on the oxygen and nutrient dynamics. Primary production in Lake Maracaibo is not limited by the nutrient flux from the bottom sediments, but is completely dominated by the nutrients coming from the rivers.
Keywords: Lake Maracaibo, DYRESM, water quality, computer modelling
This is one of a series of five papers describing a recent study on the environmental remediation of Lake Maracaibo, Venezuela (for an overview of the study and a description of the Lake Maracaibo system see Findikakis et al., 2001). One of the aims of the study was to develop analytical modelling tools that would allow the evaluation of alternative remediation strategies for the lake. Because of the different time- and space-scales of the processes affecting environmental quality in the Lake and the computational requirements for their proper simulation, a combination of three- and one-dimensional models was used. This paper describes the one-dimensional water quality modelling of the lake using DYRESM-WQ.
DYRESM-WQ is a coupled one-dimensional hydrodynamic and water quality model for lakes and reservoirs (Imberger and Patterson, 1981; and Hamilton and Schladow, 1997). It is used to predict the variation of water temperature, salinity, dissolved oxygen, nutrients, pH, dissolved metals and biota (including phytoplankton, macroalgae, zooplankton, fish, seagrass and jellyfish) with depth and time. DYRESM-WQ is process-based and explicitly models key processes that affect water quality over seasonal time-scales. The hydrodynamic component does not require calibration, but the water quality component of the model has many parameters that require calibration.
The model DYRESM is
based on the assumption that the variations in the vertical play a more
important role than those in the horizontal direction. This gives rise to the
layer construction, in which the lake is represented as a series of horizontal
layers. In the model there is no horizontal variation within the layers, and the
vertical profile of any property is obtained from the property values from each
layer. In DYRESM these layers are of different thickness; as inflows and
outflows enter or leave the lake, the affected layers expand or contract and
those above move up or down to accommodate the volume change. The vertical
movement of layers is accompanied by a thickness change as the area changes with
vertical position. Mixing is modeled by amalgamation of adjacent layers, and the
layer thickness is dynamically set internally by the model to ensure that for
each process, an adequate resolution is obtained. The one-dimensional assumption
allowed DYRESM-WQ to quickly assess long-term effects of various remediation
options (for example, ten-year simulations routinely take less than one hour).
DYRESM-WQ was used to model the main body of Lake Maracaibo, with the southern end of the Strait forming the northern boundary of the model domain. The model was calibrated against available data for the period August 1997 - September 1998. It was then validated using data from the period November 1998 – March 1999. The calibrated and validated model was used to assess the long-term (10 to 20 years) effects caused by changes in the exchange flow in the Strait and changes to the nutrient loadings, and to perform a sensitivity analysis.
Meteorological data (short wave radiation, cloud cover, air temperature, relative humidity, wind speed and rainfall) used to force the simulations was derived from data collected at three locations around Lake Maracaibo: La Chinita Airport (near Maracaibo), Santa Barbara (south-west of lake) and Canal VOC (east shore of lake).
Although there are thirty catchments draining into Lake Maracaibo, 81% of the total run-off is accounted for by just four major rivers, with the largest of these, Rio Catatumbo, contributing approximately 60% of the total inflow to the lake. Since these rivers all flow into the south-western part of Lake Maracaibo, DYRESM-WQ modelled the river inflow as a single source. The time-series of river inflow volume was based on synthesised flows for the eleven largest rivers.
The exchange flow through the Strait was modelled as two separate inflows: a moderately saline (approximately 5-12 psu) periodic underflow and a less saline (approximately 1-4 psu) wind- and tide-driven surface exchange. The definition of these two saline inflows was based on data provided by the MIKE-3 regional model output (Hansen et al., 2001a). The MIKE-3 data were provided as a two-dimensional cross-section of velocity and salinity at hourly intervals from the western bank of the lake entrance (71.8153W, 10.2478N) to the eastern bank (71.3892W, 10.1834N).
The river inflow properties (salinity and water quality variables) were calculated as volume weighted averages of all the rivers for which data (measured or synthesised) was available. Lumped river nutrient loadings into Lake Maracaibo were available for the 1997 period and nutrient loadings for 1998 were generated using regression analysis of the 1997 data. The nutrient loading of the saline underflow and surface overflow into the lake from the strait was estimated using data from the ICLAM 1997 intensive campaign. Station NO-2 was selected to represent the characteristics of the saline inflows.
In addition to nutrient loadings from the river and saline inflows, there are significant nutrient contributions from point sources, run-off from the cattle zones on the western shore and from atmospheric loading. These additional nutrient inputs were included by increasing the nutrient concentrations of the river inflow to produce the total nutrient load to the lake. In the absence of specific information about the timing of the nutrient inputs from the other sources, it was assumed that the nutrient inflow is correlated with run-off.
DYRESM-WQ was initially calibrated against available field data for the period August 1997 - September 1998. Since DYRESM is one-dimensional the properties in each layer of DYRESM-WQ represent basin-wide horizontal averages. For the purposes of calibration and validation of the model, the DYRESM-WQ output was compared with data from station C-11, located near the centre of Lake Maracaibo. (A map showing the location of all water quality stations, including C-11 is given in Findikakis, et al. 2001). Initially the model was run without the water quality modules (i.e., as DYRESM only) to simulate the observed temperature and salinity stratification., and then it was run with the water quality modules (i.e. DYRESM-WQ) to fix the rate coefficients for the biogeochemical processes. After the calibration the model was validated, without changes to any coefficients, against field data from November 1998 to March 1999.
Figure 1 compares the field data salinity profiles with the DYRESM simulation for the 1997-98 calibration period. The simulated salt stratification is considered acceptable given the potential errors in the simulated inflow data and the assumption of one-dimensionality. The model reproduces the vertical salinity structure well over this period, with the exception of the profile in November 1997. The difference in the simulated and observed salinity structure in November 1997 could be possibly due to the phase differences in the estimated saline underflow due to the lack of adequate data to define the boundary and forcing conditions in the Gulf of Venezuela. These errors are reflected in the DYRESM output. The salinity profiles measured on 19 November 1997 suggest that a major inflow of saline water may have occurred sometime before that date, causing an increase in the salinity of the hypolimnion to approximately 10 psu at that time. Such an event is not shown in the MIKE 3 simulation for the performance test period.
Calibration of the water quality module of DYRESM-WQ is an iterative process and was initiated from previously used values of the water quality parameters from tropical lakes. The processes that required calibration were:
l sediment release rates for nitrogen and phosphorous
l biochemical reaction rates for nitrification and denitrification
l sediment oxygen demand coefficient
l phytoplankton parameters and rate coefficients
l chlorophyll-a induced light attenuation coefficient
Despite some differences between the values of observed water quality variables and those simulated by the calibrated model, DYRESM-WQ appears to reproduce the important nutrient dynamics sufficiently well to allow an assessment of scenarios based on hydraulic works and/or management options on the water quality in Lake Maracaibo.
Selected results of the water quality calibration simulations are shown in Figures 2–4 (DO, dissolved reactive phosporus, and NH4 are shown) and show that the model successfully reproduced profiles of water quality parameters. Dissolved oxygen profiles generated by DYRESM-WQ compare favorably with the field observations in relation to magnitude and vertical distribution. The hypolimnion remains anoxic throughout the simulation profiles, but oxygen is shown to penetrate to the bottom in the field data collected in March 1998. It is important to note that the field data shows no disturbance to the salt stratification during March 1998 period and hence the mechanism by which the hypolimnion became oxygenated is not clear. It is possible that the dissolved oxygen is brought into the Lake in the saline underflow. In the present DYRESM-WQ simulations it was assumed that the dissolved oxygen concentration of the saline underflow is negligible and so the model may underestimate the oxygen flux from this source. However, it is expected that the oxygen demand of the hypolimnion and the sediments will quickly consume any oxygen brought in by the underflow coming into the Lake from the Gulf of Venezuela.
The calibration and validation simulations confirm that the salinity stratification is a major influence on the oxygen and nutrient dynamics. The stratification reduces vertical fluxes across the halocline so that the hypolimnion remains anoxic for the duration of the simulation and nutrient concentrations (NH4 and PO4) increase slightly before reaching a steady state. However, the nutrient and chlorophyll-a concentrations in the surface layer remain relatively constant.
In addition to the 1998-99 validation simulation, a 10-year simulation was made by using the 1997-98 forcing data to (a) confirm the long-term stability of the model, and (b) serve as a base-case against which to compare the results of simulations of potential remediation options. Some time-series (salinity, DO, Chl-a and PO4) from this base-case simulations are shown in Figure 5.
The calibrated DYRESM-WQ was validated using a series of water quality profiles collected at station C-11, at the center of the Lake, every 2 to 4 weeks from November 1998 to March 1999. Figure 6 compares the field data salinity profiles with the DYRESM simulation for the validation period. Given the potential errors in the simulated inflow data and the assumption of one-dimensionality, the model predicts the salinity profile well. The importance of correctly simulating the halocline and resultant suppression in vertical fluxes is illustrated in the dissolved oxygen profiles shown in Figure 7. Both the field and simulated dissolved oxygen profiles show a permanently anoxic hypolimnion during the validation period and the depth of the anoxic layer is predicted well. Interestingly, DYRESM does not reproduce the observed depression in dissolved oxygen in the surface layer of Lake Maracaibo in January 1999. The observed depression in oxygen may have been an artifact of spatial patchiness of biological processes. The sampling of a local patch of low chlorophyll, a collapsing bloom or high local ammonium concentrations could each contribute to local reductions in dissolved oxygen and none of these features should be expected in the simulations results from a basin-wide one-dimensional model.
The coupled one-dimensional hydrodynamic and water quality model DYRESM-WQ was calibrated against field data collected near the center of Lake Maracaibo over the period August 1997 - September 1998. The model reproduces adequately the major hydrodynamic and water quality features of the Lake. In particular, DYRESM reproduces the observed salinity stratification which plays an important role in determining the vertical fluxes of oxygen, nutrients and other biological agents within the Lake. The water quality module was calibrated and shown to reproduce the seasonal patterns and vertical structure observed in the oxygen, nutrient and chlorophyll-a profiles. Although the simulations showed some phase errors, the model reproduced the range of observed concentrations of all variables, both at the surface and at depth.
The simulation results confirm that the salinity stratification is a major influence on the oxygen and nutrient dynamics. The stratification reduces vertical fluxes across the halocline so that the hypolimnion remains anoxic for the duration of the simulation and nutrient concentrations (ammonium and phosphate) increase slightly before reaching a steady state. However, the nutrient and chlorophyll-a concentrations in the surface layer remain relatively constant.
Acknowledgements
The work presented in this paper was part of the Inegral Study for the Environmental Restoration of Lake Maracaibo, conducted for PDVSA by Bechtel International. Luis Delgado and Nelson Corrie managed the study for PDVSA. We acknowledge the contributions made to the project by all members of the project team, including the PVDSA Technical Committee. This paper is CWR reference ED 1146 DH.
Imberger, J. and Patterson, J.C. 1981 A dynamic reservoir simulation model-DYRESM:. in: Transport Models for Inland and Coastal Waters, H.B. Fischer (ed). Academic Press, New York, pp. 310-361.
Findikakis, Angelos N., Jőrg Imberger, Ian Sehested Hansen, and Erich Gundlach, 2001: A Study Of Environmental Remediation Options, For Lake Maracaibo, Venezuela, Proceedings XXIX IAHR Congress, Bejing, China, September 17-21, 2001.
Hansen, Ian Sehested, Jacob Steen Møller, and Angelos N. Findikakis, 2001a:3 Dimensional Hydrodynamic Modelling of the Maracaibo System, Venezuela, Proceedings XXIX IAHR Congress, Bejing, China, September 17-21, 2001.
Schladow, S. G. and Hamilton, D. P. 1997 Prediction of water quality in lakes and reservoirs: Part II: Model calibration, sensitivity analysis and application. Ecological Modelling 96: 111-123.
Fig.1 Salinity time series for
the 400 day calibration simulation.
The DYRESM simulated profiles are plotted as solid lines, and the measured
profiles as dotted lines.
Fig.2 Dissolved
oxygen profiles for the 400-day calibration simulation.
The DYRESM simulated profiles are plotted as solid lines, and the measured profiles as dotted lines.
Fig.3 Dissolved
reactive phosphorus profiles for the 400-day calibration simulation.
The DYRESM simulated profiles are plotted as solid lines, and the measured
profiles as dotted lines.
Fig.4 Ammonium
profiles for the 400-day calibration simulation.
The DYRESM simulated profiles are plotted as solid lines, and the measured
profiles as dotted lines.
Fig.5 Ten year simulation
using existing bathymetry. Panels show contours of salinity,
DO, Chl-a and PO4 plotted against depth and time.
Fig.6 Salinity
time series for the validation simulation.
The DYRESM simulated profiles are plotted as solid lines, and the measured
profiles as dotted lines.
Fig.7 Dissolved
oxygen time series for the validation simulation.
The DYRESM simulated profiles are plotted as solid lines, and the measured
profiles as dotted lines.