The stage of Lake Brooklyn, in southwestern Clay County, Florida, has varied over a range of 27 feet since measurements by the U.S. Geological Survey began in July 1957. The large stage changes have been attributed to the relation between highly transient surface-water inflow to the lake and subsurface conduits of karstic origin that permit a high rate of leakage from the lake to the Upper Floridan aquifer. After the most recent and severe stage decline (1990-1994), the U.S. Geological Survey began a study that entailed the use of numerical ground-water flow models to simulate the interaction of the lake with the Upper Floridan aquifer and the large fluctuations of stage that were a part of that process. A package (set of computer programs) designed to represent lake/aquifer interaction in the U.S. Geological Survey Modular Finite-Difference Ground-Water Flow Model (MODFLOW-96) and the Three-Dimensional Method-of-Characteristics Solute-Transport Model (MOC3D) simulators was prepared as part of this study, and a demonstration of its capability was a primary objective of the study. (Although the official names are Brooklyn Lake and Magnolia Lake (Florida Geographic Names), in this report the local names, Lake Brooklyn and Lake Magnolia, are used.)

In the simulator of lake/aquifer interaction used in this investigation, the stage of each lake in a simulation is updated in successive time steps by a budget process that takes into account ground-water seepage, precipitation upon and evaporation from the lake surface, stream inflows and outflows, overland runoff inflows, and augmentation or depletion by artificial means. The simulator was given the capability to simulate both the division of a lake into separate pools as lake stage falls and the coalescence of several pools into a single lake as the stage rises. This representational capability was required to simulate Lake Brooklyn, which can divide into as many as 10 separate pools at sufficiently low stage.

In the first of two calibrated models, recharge to the water table, specified as a monthly rate, was set equal to 40 percent of the monthly rainfall rate. The specified rate of inflow to the uppermost stream segment was set equal to outflows from Lake Lowry estimated from lake stage and the 1994-97 rating table. Leakage to the intermediate and Upper Floridan aquifers was assumed to occur from the surficial aquifer system through the confining layers directly beneath deeper parts of the lake bottom. A leakance coefficient value of 0.001 feet per day per foot of thickness was used beneath Lake Magnolia, and a value of 0.005 feet per day per foot of thickness was used beneath most of Lake Brooklyn. With these values, the conductance through the confining layers beneath Lake Brooklyn was about 19 times that beneath Lake Magnolia.

The simulated stages of Lake Brooklyn matched the measured stages reasonably well in the early (1957-72) and later (1990-98) parts of the simulation time period, but the match was unsatisfactory in an intermediate time period (1973-89). To resolve this discrepancy, the hypothesis was proposed that undocumented losses of water from Alligator Creek upstream from Lake Brooklyn or from the lake itself occurred between 1973 and 1989 when there was sufficient streamflow. The resulting simulation of lake stages matched the measured lake stages accurately during the entire simulation time period. The model was then revised to incorporate the assumption that only 20 percent of precipitation recharged the water table (the second calibrated model). Recalibration of the model required that leakance values for the confining units under deeper parts of the lakes also be reduced by nearly 50 percent. The stages simulated with the new parameter assumptions, but retaining the assumption of surface-water losses, were an excellent match of the measured values. The stage of Lake Magnolia was also simulated accurately. The results of sensitivity analyses show that simulated streamflow between Lakes Magnolia and Brooklyn tends to be water-budget controlled, and is not appreciably affected by the specified outflow altitude or channel characteristics of the receiving stream.

To match heads measured in observation wells of the surficial aquifer network, the assigned hydraulic conductivity values were zoned, and ranged from a minimum of 4 feet per day to a maximum of 400 feet per day in the first calibrated model. These values were reduced by about 50 percent in the second calibrated model. Differences between observation wells were noted in the abruptness of changes of measured head values, and in the relation of the timing of peak measured heads and simulated peak heads. These differences seemed to be correlated with the depth of the water table below land surface. Spatially uniform values of transmissivity were specified for the intermediate (10,000 feet squared per day) and Upper Floridan (100,000 feet squared per day) aquifers. Simulated heads in the Upper Floridan aquifer layer follow the trend of the heads measured in a long-term observation well with data beginning in 1960. This result suggests that the observed head decline could be explained entirely in terms of the stage decline in Lake Brooklyn and may not indicate a regional trend.