AUDITING COMPLEX LONG-TERM SOLUTE TRANSPORT MODELS USING REAL DATA

Successful design and operation of a brackish-water desalination plant is greatly dependent on the long-term stability of the feedwater salinity. When the feedwater salinity exceeds the ability of the process design to treat the water, the plant fails. Most of these facilities utilize a groundwater source of raw water supply, which is commonly a leaky aquifer that is prone to pumping-induced changes in water quality. Since the operating life expectancy of a BWRO desalination plant is typically 20 to 30 years, it is common to use a solute transport model to predict changes in the total dissolved solids (TDS) concentration (or dissolved chloride concentration) for a 30-year period prior to the process design. The modeling takes into account the starting TDS concentration in all test, monitoring, and production wells, the configuration of the wellfield with expansions based on water use projections, the aquifer hydraulic properties, and the distribution of TDS within, above, and below the production aquifer. Many of the wellfields that have been modeled in the past have now been in operation between 10 and 35 years. The wellfield and plant operators continuously monitor the total pumpage from each production well on a monthly basis and measure the dissolved chloride concentration in water from each production well monthly. The conductivity and the TDS of the aggregated feedwater into the plant are continuously monitored. Therefore, the ability exists to use measured dissolved chloride concentration data from each production well to create a regression analysis of the data from the wellfield to compare it to the predictions made in the initial solute transport model. Hence, long-term solute transport models can be audited to assess their accuracy, and to evaluate the use of the initial conceptual model and overall model design. In six solute transport models audited from BWRO plant wellfields in Florida, two models showed very accurate predictions within a few percent of the real change in water quality, two models showed very acceptable results within 10 to 25% of real, one model severely over-predicted the salinity change, and one model severely under-predicted the salinity change, resulting in plant failure. The model that over-predicted the salinity changed was affected by the utility adding feedwater from a secondary aquifer source that was not used in the initial modeling. The model that under-predicted the salinity change used a faulty conceptual model and incorrect boundary conditions in the model grid.