Review of the St. Johns River Water Supply Impact Study: Report 3 (2010)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Chapter 2 Review of Modeling Methods and Results This chapter briefly describes the major components of the hydrologic and hydrodynamic modeling, including groundwater modeling. Some discussion of improvements that would strengthen the analysis of the District is included, but the bulk of recommendations in this regard are found in Chapter 3. In addition, the results of model runs for particular scenarios are presented. LAND USE/LAND COVER 1995 land use/land cover data, developed from interpretation of color-infrared aerial photography, was used as the baseline for the WSIS because it is the basis for current water supply planning conducted by the District. The 1995 land use data, together with data sets on precipitation, evaporation, and other drivers of the models from 1995-2005, were used to calibrate the HSPF models. The District adopted 15 land use/land cover categories for modeling, as shown in Table 2-1. Runoff and other coefficients used in the hydrologic modeling were selected to reflect the differing characteristics of these land use/land cover categories. For use in the HSPF modeling, the District grouped the land-use categories according to similarities in hydrologic response— usually corresponding to degrees of imperviousness. The model employs a lumped parameter approach, such that all of the areas for a given land use within a subwatershed are summed and multiplied by the discharge expected from that type of land use. Population growth is expected to be the main driver for changes in land use through the planning period (to the year 2030). Projected land use for the year 2030 (see Table 1-1) was developed by expanding 1995 conditions in relation to expected population growth. County- level population forecasts were prepared by the University of Florida Bureau of Economic and Business Research, and a consulting firm, GIS Associates, Inc., was retained to develop detailed spatial disaggregations based on the county-level forecasts (SJRWMD, 2009) in order to project future populations, and thus water demand, for public water supply utility service areas. Special algorithms were used at the parcel level to reallocate growth from high growth parcels having spatial constraints to lower growth parcels, while still conforming to the county-level projections. As a result of the projected population growth, residential and commercial/industrial land use areas increased over the time period of interest, while open, range, forest, and agricultural land areas in the drainage basin decreased. Wetlands and water areas were generally kept at 12 P R E P U B L I C A T I O N C O P Y 

Review of Modeling Methods and Results 13 TABLE 2-1 Hydrologic Model Land Use Categories Low density residential Open land and barren land Forest Medium density residential Pasture Water High density residential Agriculture general Wetlands Industrial and commercial Agriculture tree crops Forest regeneration Mining Rangeland Non-riparian wetlands Note that there are extra land use categories here compared to Table 1-1. The committee found that the District was not consistent with respect to the number of land use categories presented in various documents. 1995 values2. As shown in Figure 2-1 for the lower St. Johns River basin, projections were made in five-year increments from 2005 to 2030. The District has indicated that because of the recent economic downturn, the 2030 population estimates actually may be more accurate for 2033 or 2035, but this affects only the timing of growth and not the growth itself. Land use forecasts, however, are a function not only of population forecasts but also the nature of the population’s impact on the landscape (e.g., the mix of housing densities and the degree and characteristics of imperviousness in developed lands). The relationships developed for the current WSIS may not hold in the future, especially if the actual rate of population increase or its impact on the hydrologic response of the resulting change in land use is significantly different from the current forecast. Indeed, the land use shown for 2004 in Figure 2-1 does not seem to reflect projected changes in population growth. Thus, although the Committee does not have any technical issues concerning the approach used by the District to estimate 2030 land use/land cover conditions, it recommends that the District revisit and update the population and resulting land use projections in future periodic reviews (e.g., the District’s water supply plans and the water supply assessments made by the District every five years). 2 The committee cautions that this assumption of wetlands acreage remaining the same is uncertain. This is because the predicted increases in stormwater discharge that will accompany the 2030 land use condition could result in stream channel deterioration that would reduce the surface area of riparian wetlands (discussed in Chapter 3). These losses would have to be offset by wetlands mitigation required under the Clean Water Act and/or restoration of the upper basin for the assumption to hold true. P R E P U B L I C A T I O N C O P Y 

14 Review of the St. Johns River Water Supply Impact Study: Report 3 Population/Land Use in LSJRB FIGURE 2-1 Population and Resulting Land Use Projections in the Lower St Johns River Basin for the Planning Period 1995–2030. SOURCE: Cera et al. (2010). METEOROLOGY For the District’s analyses, the required meteorological data consist of rainfall and potential evaporation (for a detailed description, see pp. 12-23 of Cera et al., 2010). Rainfall distribution across the catchment was modeled for historical data by taking National Weather Service (NWS) rain gage data and distributing them spatially using a network of Thiessen polygons. Gage data for rainfall accumulations for time periods larger than the hydrologic model time step (1 hour) were disaggregated using either nearby hourly data or (where no hourly data were available) by assuming that rain events of less than a half inch (1.3 cm) occur within an hour and that larger events were distributed with a triangular weighting over three to five hours. Missing data were estimated from the nearest available measured data. Where only daily rainfall records were available, disaggregation from daily to hourly values was accomplished using the software package WDMUtil, which is part of the EPA BASINS package and can be considered state-of-the-science. Although more comprehensive Doppler rainfall data are available, the District chose to use the rain gage data due to their longer period of availability. Potential evaporation was modeled using the Hargreaves method, which requires only historical temperature data. Observed temperature data at meteorological stations were distributed across the basin using Thiessen polygons. P R E P U B L I C A T I O N C O P Y 

Review of Modeling Methods and Results 15 Model Calibration and Uncertainty The use of NWS data for rainfall does not require (or allow) any formal calibration. Distribution of rainfall across a watershed from point-based rain gage data introduces some uncertainty into the rainfall model that is difficult to quantify with only the point-based gages, but this is a problem inherent to hydrologic modeling at the watershed level and not a specific criticism of the District’s approach. The Hargreaves estimate of potential evaporation was calibrated based on satellite-based estimates of potential evaporation computed for three sites using the Penman method. Model Capabilities and Limitations The modeling approach is limited in its ability to represent small-scale convective storms and flashy rain events. This limitation, which is inherent in models that rely on data from point- based gages with no better than hourly data, may not be important for present conditions because the St. Johns is a heavily dampened system. It could become more important as the landscape urbanizes, however, because the model will be limited in its ability to predict runoff from the catchment during flashy events. In the committee’s opinion, the assumptions in the models presented are reasonable. We agree with the District that that Doppler and point rain gage data cannot be intermingled in a single model data set to drive a watershed hydrologic model. However, the existing Doppler radar sets that the District did not use in their hydrologic modeling efforts could be used to help quantify the uncertainty associated with distribution of point-gage rainfall data across the Thiessen network. A further criticism regarding the use of NWS rain gages is discussed below relative to the calibration of the watershed hydrology model. WATERSHED HYDROLOGY The watershed hydrology for the entire upper basin (including river reaches) and the middle and lower basins (excluding the main river stem) were modeled with the HSPF hydrological model. This model is a recognized, EPA-supported “lumped parameter” model that represents the relationship between infiltration, surface runoff, subsurface flow, and transpiration based on the fractions of a sub-watershed with pervious and impervious cover. “Lumped parameter” refers to models that use a set of calibrated coefficients to represent hydrologic behavior in each subbasin based on land use or land cover. This approach can be contrasted with other “distributed parameter” models that use small-scale grid cells to directly represent individual regions of pervious or impervious cover, their mechanical characteristics, and their spatially distributed connections. The District chose a reasonable approach in using the lumped parameter model. Although this approach renders the model highly dependent on calibration, a mechanistic distributed parameter model could not reasonably be supported for a basin the size of the St. Johns and with the available data. P R E P U B L I C A T I O N C O P Y 

16 Review of the St. Johns River Water Supply Impact Study: Report 3 Model Calibration and Uncertainty Model calibration was extensively presented in Cera et al. (2010). The calibration approach used historical rainfall and potential evaporation models (described above) for the period 1995 to 2006 and land-use data based on a 1995 study. The accepted “Parameter ESTimation” (PEST) software was used for calibration of eight of the 19 parameters used in the HSPF Common Logic Table (Appendix C, p. 209, Cera et al., 2010) and the “special actions” associated with a wetland storage-outflow model. Other parameters were fixed. The HSPF calibration modeling reported to date has been conducted carefully and thoughtfully, and the District is commended on this effort. The model has been demonstrated to provide reasonable results during the calibration period. The District chose to use data from the NWS rainfall gages rather than NEXRAD Doppler radar data for their calibration study because they are using a 32-year NWS rainfall gage record for scenario analysis. However, to limit the influence of land use changes, model calibration was based only on the 1995-2006 period rather than the 32-year period. This approach is reasonable with some limitations. It should be recognized that the model response to rainfall outside the calibrated period is uncertain, and so statistics gathered from the model based on the larger 32-year data set will be reliable only if the weather patterns are similar to those in the 11-year data set. There is little reason to assume that a lumped parameter model will respond correctly to rainfall events that are significantly different than those in the calibration data set. The calibration data set thus should contain a reasonable approximation of all expected weather over the scenario testing data set. If there are significantly different weather patterns in the 32- year set than in the 11-year set, then the calibration approach may need to be reconsidered. The District has not yet presented or proposed any methods to estimate the uncertainty associated with the HSPF model. Model Capabilities and Limitations The hydrologic model appears to provide reasonable approximations of the major fluxes through the landscape. The runoff values predicted to the river main stem are reasonable approximations of the observed flows. The principal limitation appears to be whether or not the approach to modeling wetlands (FTABLES) provides enough accuracy for evaluation of ecological impacts. Lumped-parameter hydrological models inherently perform better under high-flow conditions, where the modeled physical processes are strongly affected by the calibration coefficients. Under low-flow conditions that are often important to ecology, a lumped-parameter model may have significant hydrological flux uncertainty because of the difficulty of obtaining data and calibrating for such conditions. Thus, the model’s ability to represent water retention times for short events (e.g., convective thunderstorms), flux rates through wetlands, and saturated soil fluxes in areas undergoing periodic immersion/drying cycles is more problematic. The District may not be taking advantage of the best available rainfall data by relying on the NWS gage data rather than using NEXRAD Doppler data for scenario testing. The principal reason for the modeling work is to predict likely future catchment characteristics. The District likely will want to use this model for adaptive management and to monitor how well its decisions hold up under rainfall and land-use conditions that actually occur in the future. Thus, using the P R E P U B L I C A T I O N C O P Y 

Review of Modeling Methods and Results 17 best available rainfall data (NEXRAD Doppler) is appropriate. As noted in the outside peer- review of modeling commissioned by the District (INTERA, 2009, page 16): The District should consider calibrating the sub-watershed models with the shorter 1995 to present NEXRAD data set using the same approach and objective functions... [This] would provide the District an indication of how much the model underperformance is due to issues with the major forcing variable (rainfall) versus uncertainty in the model parameters such as vegetative cover and evaporation. The committee agrees with the above comment and recommends that the District consider conducting a secondary set of calibration runs. The issue of how accurately wetlands are modeled under drying conditions is a principal concern and is discussed below in the section “Applicability of HSPF to Wetlands Hydrology. The report section on “Scenario Simulations for Water Quantity” (Cera et al., 2010) is difficult to follow and understand. Different naming and numbering conventions were used to describe the scenarios and to present them in figures and tables. The withdrawal scenario for Taylor Creek Reservoir (p. 177) is particularly difficult to follow, and is presented without any reasoning for its development. RIVER HYDRODYNAMICS The District provided an extensive PowerPoint presentation on the river hydrodynamic model development and calibration. It is clear that they are proceeding carefully and analyzing how model implementation choices affect results. The Committee appreciates the level of detail provided and the development of graphs that clearly illustrate what the model can and cannot achieve. For the lower SJR, where tidal oscillations are a dominant forcing, the District has conducted grid resolution tests to determine whether their 1X (coarse) model grid is consistent with results produced by a 4X (medium) and 16X (fine) model grid. This effort is a standard modeling exercise to establish that the coarser, more practical grid provides results that are sufficiently similar to the fine grid. The results are reasonable, showing very small percentage differences in the tidal amplitude in the lower reaches (where the amplitudes are larger). The only points that seem questionable are those near Lake George, where the absolute difference in the tidal amplitude between the models is small, but the percentage difference is large because the Lake George tidal amplitude is quite small. This problem may not be resolvable through calibration at the 1X grid scale, and so the District should analyze the differences to understand their effects on model predictions. It is likely that the differences are irrelevant at the upstream locations. Tidal phase differences were also examined by the District. Phase differences between results on the different grids are expected for this type of model, and the results shown appear well within the acceptable bounds. Overall, the grid resolution tests show that the coarse resolution model is adequate for this stage of the project and the available computer power. However, as computational power increases over the next decade, continued modeling for adaptive management should plan on using the 4X or perhaps even the 16X grid. It can be expected that as more data become available, the ability of the 1X model to distinguish finer detail will limit the model’s utility. The District analyzed oscillation frequencies in Crescent Lake using the hydrodynamic model and a field-deployed drifter, which provided confidence that the large-scale surface P R E P U B L I C A T I O N C O P Y 

18 Review of the St. Johns River Water Supply Impact Study: Report 3 seiching of this lake is reasonably modeled, both in velocity magnitude and decay. The District is congratulated for this effort; numerical error in many models artificially damps the surface seiche and prevents representation of this behavior. The District used the PEST software to calibrate the model by adjusting the bottom roughness height. In the main stem of the tidal river, the calibrated bottom roughness heights are sub-millimeter, which is an order of magnitude or more smaller than expected physical roughness heights. This result is not entirely surprising for the model using the 1X grid. It is well known that coarser model grids induce a damping error – “numerical dissipation” – that has the same effect on the model as the turbulent dissipation produced by the bottom boundary. The small bottom roughness height produced by calibration indicates that numerical dissipation is of the same order of magnitude or larger than physical dissipation produced by boundary-driven turbulence. From an academic modeling standpoint, this is not an ideal model formulation. It clearly would be better to have the energy dissipation developed through bottom boundary dissipation rather than the numerical dissipation. However, engineering efforts must always balance the ideal against the practical. In this case, the additional computational expense of moving to a 4X grid simply to decrease the numerical dissipation is not justified. The grid resolution study shows that the 1X model captures the bulk physical transport fairly well. Thus, it is unlikely that using a 4X grid to obtain better resolution of bottom turbulence will significantly alter the overall model results. However, the District should keep in mind that physical processes associated with the bottom boundary layer are poorly represented; it follows that any conclusions that depend on the vertical distributions in the water column driven by turbulent mixing may be suspect. In particular, these effects may limit the ability of the model to represent the physical/ecological dynamics driven by salinity gradients and resulting stratification in the Lower St. Johns River. The model predictions of salinity provided to the committee have reasonable comparison to field data; however, the amount of data available on vertical stratification is fairly sparse. The hydrodynamic model results for the middle St. Johns River are quite satisfactory. The calibrated model captured daily discharge at various stations along the river. The model was calibrated based on a one-year subset of a 10-year data record, and performed equally well in the nine years outside of the calibration period. Not only did the model perform well over the longer time scale, but shorter time-scale (daily and hourly) discharge features were also reasonably represented. At the hourly scale, the model output has errors in phase and amplitude of smaller- scale oscillations that are not unexpected, but the overall characteristics are very similar to the observations. Modeled salinity in the Middle St. Johns River also compared well with both observed data and model results from HSPF. The District provided some preliminary PowerPoint slides (Sucsy, 2010) and a draft report (Sucsy et al., 2010)3 on sensitivity and uncertainty analyses. The combination of First- Order Error Analysis and Monte Carlo simulations were used to understand the major contributors to model sensitivity and uncertainty. The approach appears to be reasonable and well-considered. 3 This document was received too late in the report preparation process to be thoroughly reviewed by the committee, but it appears to support the conclusions drawn from the District’s PowerPoint presentation in March 2010. P R E P U B L I C A T I O N C O P Y 

Review of Modeling Methods and Results 19 GROUNDWATER Multiple interactions in the hydrologic cycle within the St. Johns basin involve groundwater. Along the main stem, the St. Johns River collects runoff, and shallow groundwater contributes to the base flow. At the same time, a multi-layered groundwater flow system has been grouped into three distinct hydrogeologic systems: the surficial aquifer (SAS), the Intermediate Confining Unit (ICU), and the Floridan Aquifer System. Recharge to the SAS is primarily from rainfall, although infiltration due to irrigation, reclaimed water, seepage from surface waterbodies, and septic tanks also make contributions. Discharge from the SAS occurs through evapotranspiration, from seepage into surface waterbodies, and from well withdrawals. In addition, the Floridan Aquifer System recharges to the SAS in areas where the potentiometric surface of the Upper Floridan Aquifer (UFA) is higher than the water table (otherwise, downward leakage from the SAS to the underlying UFA occurs). Springs discharging from the UFA exist throughout the basin and exert a significant influence upon the base flow in the river. Some of the above interactions are understood well and amenable to quantitative modeling (e.g., water movement in the Floridan Aquifer) or measurement (spring discharges), but others are poorly understood and difficult to model at the current state of understanding (e.g., exchange of water between the unconfined shallow aquifer and wetlands). Using a single model to represent these hydrologic interactions would be a complicated task, to say the least, and consequently the District is using several independent models to represent the system: a GIS tool to identify wetlands, HSPF for surface water runoff from the watershed, the mainstem river hydrodynamic model EFDC, and the Eastern Central Florida (ECF) groundwater model, which is based on MODFLOW, a widely used groundwater flow model of the U.S. Geological Survey. One of the main criticisms of this approach is the limited extent to which HSPF can model wetland hydrology, including groundwater–surface water interactions in the riparian wetlands. This issue is discussed in Chapter 3. This section focuses on resolving comments made previously by the Committee about the assumption of constant chloride concentrations in groundwater input to the river channel over time and about the use of a steady state, non-density-dependent groundwater model. Groundwater Salinity An earlier criticism of the District’s analysis of groundwater was their assumption of a constant average chloride concentration value for a given location in their groundwater modeling. As discussed in the Phase 1 report (SJRWMD, 2009), the District calculated the chloride load into the river by obtaining chloride concentration data from monitoring wells and from a chloride concentration map produced by the District. Then they used a GIS feature to distribute those concentration values to each ECF model grid. Those values were then multiplied by the groundwater flux from MODFLOW to get the chloride load contributed by the groundwater. Upon being asked to show data that demonstrate the stability of the chloride concentration at more locations, District scientists presented chloride concentration data from eight wells from 1990 to 2009. The chloride concentrations across the eight wells ranged from a mean of about 5,300 mg/L over a 22-year period to a mean of only 9 mg/L. For any individual well, the highest temporal variability about the mean was 2.2 percent. Wet vs. dry years P R E P U B L I C A T I O N C O P Y 

20 Review of the St. Johns River Water Supply Impact Study: Report 3 appeared to make little difference. Based on these findings, the committee no longer is concerned with the use of a temporally constant chloride concentration. Density and Temporal Dependence of Models In NRC (2009a), the District was criticized for employing a steady-state groundwater model (ECF) that was not density-dependent, which limits its ability to predict changes in salinity that might result from water withdrawals. District scientists considered both assumptions (steady state and density independence) to be valid, and they were asked to provide evidence for their opinions. To prove density independence, the District analyzed salt concentrations at three wells (S0025, V0083, and BR1526) in the Lake Harney area with high chloride concentrations and showed that converting real head data from these wells to their freshwater equivalent did not lead to a significant change in the potentiometric surface. When asked to address what criteria were used to determine whether changes were “significant,” the District presented data suggesting that a change in head of 0.3 m or a change in chloride flux of 0.02 m3/s was not significant. In Lake Harney, which has one of the highest chloride concentrations, the salinity change due to density stratification is only 0.012 practical salinity units or PSU (or 6.6 mg/L of chloride ion). To determine that transience in groundwater flux is not important enough to forgo the use of steady state values in the modeling, the District analyzed how sensitive the river hydrodynamics model (EFDC) is to using transient vs. steady state groundwater input. Using well level and river stage data from four locations in the middle basin, the EFDC model was run with both the steady state and transient groundwater flux (using 10 years of transient groundwater discharge data). The change in water level or salinity predicted by the model when using steady state groundwater flux vs. transient flux was small (i.e., 95 percent of the time there was less than half a centimeter change in water level and less than 0.035 PSU change in salinity). As a second line of evidence, the District argued that in absolute value terms river water storage and surface water discharge are much larger than groundwater discharge, such that the impact of groundwater on overall flow is not large. The committee accepts this statement as being generally correct, but notes that there are likely to be some occasions when groundwater is significant because there are times of zero surface water discharge. Of course, if such occasions (where groundwater is a major contributor to flow) happen once every 10 years, they are less important than if they occur several times a year. SIMULATIONS FOR VARIOUS SCENARIOS AND FUTURE CONDITIONS Results of the hydrologic/hydrodynamic simulations, expressed in terms of changes in flow and water stage, were generated for the 12 scenarios described earlier for two upper basin locations—Christmas and Cocoa. As shown in Table 2-2, several major results became evident. First, substantially more flow (51 MGD) was generated with the 2030 land use baseline scenario than with the 1995 land use baseline scenario because of the increase in impervious area over time. Second, the surface water restoration projects in the upper basin also increased the mean flow (by 11 MGD for case with no withdrawal) and stage over the 1995 baseline condition. Third, the effects of full and half withdrawal were more pronounced upstream at Cocoa than P R E P U B L I C A T I O N C O P Y 

Review of Modeling Methods and Results 21 TABLE 2-2 Stage/Flow Modeling Results at Cocoa and Christmas for 9 Withdrawal Scenarios Scenarios Cocoa Cocoa % Change Christmas Christmas %Change in Mean stage Mean flow in Flow Mean stage Mean flow Flow from (ft NGVD) (mgd) from 1995 (ft NGVD) (mgd) 1995 Base Base Case Case 1995 Base 11.88 568 NA 6.73 721 NA 1995 + Half 11.77 545 -4.0% 6.65 693 -3.9% SJR Withdrawal 1995 + Full 11.7 524 -7.7% 6.57 669 -7.2% SJR Withdrawal 1995 + USJB 12.09 579 2.0% 6.85 732 1.5% Projects 1995 + USJB 11.97 556 -2.0% 6.76 705 -2.2% Projects + Half SJR Withdrawal 1995 + USJB 11.9 537 -5.5% 6.69 683 -5.4% Projects + Full SJR Withdrawal 2030 Base + 12.22 619 9.1% 7.02 793 9.9% USJB 2030 + USJB 12.12 596 5.0% 6.93 765 6.1% + Half SJR Withdrawal 2030 + USJB 12.04 576 1.5% 6.87 741 2.8% + Full SJR Withdrawal downstream at Christmas. Thus, despite the initial (and intuitive) assumption that water withdrawals would reduce river flows and stage, the modeling revealed that both variables in fact would increase under the full withdrawal condition, assuming management of the upper basin to bring water back into the system and the 2030 land use condition. These results should not be interpreted to mean that stage and flow would never fall below the 1995 baseline condition, but rather that on average both variables would increase. Temporal changes in stage and flow were represented as the difference in stage at Cocoa between the 1995 land use baseline case for zero withdrawal vs. half or full withdrawal (see Figure 2-2). Each dot in the figure represents a day, such that lines of dots indicate “events.” Figure 2-2 indicates that a full withdrawal could, at times, lead to as much as 0.7 ft drop in stage around the 10-11 ft water level. It should be noted that there are smaller stage and flow differences on either side of the 10-11 ft mark because (1) when there is less water in the river, the full withdrawal is not permitted, and (2) when there is more water in the river, the effect of a withdrawal becomes less important on a relative scale. P R E P U B L I C A T I O N C O P Y 

22 Review of the St. Johns River Water Supply Impact Study: Report 3 FIGURE 2-2 Changes in Stage Difference at Cocoa Under Various Water Withdrawal Scenarios assuming 1995 Land Use. SOURCE: Jobes (2010). The analysis shown in Figure 2-2 was repeated with the 2030 land use scenario, sea level rise, and upper basin project management included (Figure 2-3). This scenario brought so much water into the river that the effects on stage were reversed compared to Figure 2-2. For example, at Cocoa there was a 1.5-ft rise in stage at the 11-12 ft mark because the upper basin project and the 2030 land use more than compensated for any loss in stage from water withdrawal. The modeled results indicate that the effect of the upper basin projects would taper out as stage rises, because the projects will store water in the upper basin at high flows. As seen in Figure 2-3, there still would be some events where the water level drops below the baseline condition, but most of the curves are around zero. The same results were observed at Christmas, although the graphs are “noisier” at this location. P R E P U B L I C A T I O N C O P Y 

Review of Modeling Methods and Results 23 FIGURE 2-3 Changes in Stage Difference at Cocoa Under Various Water Withdrawal Scenarios assuming 2030 Land Use and Upper Basin projects. SOURCE: Jobes (2010). It should be noted that the modeling studies predicted that full withdrawal plus the Ocklawaha withdrawal would lead to a 9.3 percent decrease in flow in the Ocklawaha River basin compared to the 1995 baseline condition. However, the impacts of the reduction in flow to the Ocklawaha River at the Rodman Reservoir are not part of the WSIS and consequently are not analyzed further in this report. P R E P U B L I C A T I O N C O P Y