Forecasting Water Demands in South Florida in the Context of Everglades Restoration: Retrospective

Two contradictory hypotheses mark the starting point of the research reviewed in this paper. Each of these hypotheses was proposed in the late 1990s by the principal investigator when Everglades restoration plans were first taking form (Governor’s Commission for a Sustainable South Florida 1995, 1996), and each hypothesis has since been tested by different economic methodologies. The first hypothesis asserted that the $6 billion to be spent on Everglades restoration would undermine the restoration efforts due to the economic stimulus of that expenditure on the built environment (USACE 1999b). The second hypothesis is broader and more comprehensive and is actually the reverse of the first: the hypergrowth of South Florida, independent of restoration spending (that is, the growth in its cities, farms, tourist activities, and construction), would create such an increase in demand for water and land that efforts to restore the Everglades would be futile (Weisskoff 2000, 2003, 2005). Prior to the articulation and testing of these hypotheses, there had been no comprehensive quantitative analysis of all the economic pieces that had been cited in the narratives and blamed for creating the Everglades crisis in the first place (Douglas 1947, 1997).

The economy of South Florida had been carved out of the swamps, marshes, and lagoons of the natural ecosystem, and even the Atlantic coast’s once impenetrable mangrove forests had been cleared to reveal the white, sandy beaches that were to become America’s winter playground (Foster 2000). Despite resource managers and environmentalists’ attempts to arrest ecological degradation, methods to restore the fragile remnants of the natural system remained elusive, especially while the economy expanded at hyperspeed.

An early editorial in this journal addressed the need to enact comprehensive reforms in those economic elements most involved in US water management policy (Stakhiv 2003). In his critique of Everglades restoration planning, Stakhiv (2003) suggested that reducing subsidies to South Florida’s sugar industry could yield greater environmental benefits than the restoration plan itself. He inquired rhetorically,

Which institution or agency is responsible for raising this issue as part of the restoration planning process? How are all the nonstructural, institutional and consumption issues that have direct bearing on resource use in the Everglades region dealt with in the context of a more straightforward “plumbing problem”? (Stakhiv 2003).

Stakhiv was asking for the integration of the issues of the built system together with the natural system, given that both were now to be protected and promoted by public policy.

Early work by Weisskoff in 1998 had tested the first hypothesis using an IMpact analysis for PLANning (IMPLAN) model to trace the impacts of restoration spending on the economy of 12 South Florida counties over the 50-year planning period (methodology and detailed results are presented in “Regional Economic Impacts,” Appendix E of USACE 1999b). This study was a part of the Central and Southern Florida Comprehensive Review Study (USACE 1999a), which became the Comprehensive Everglades Restoration Plan (CERP) with the passage of the Water Resource Development Act of 2000 (Grunwald 2006).

The results of this early IMPLAN study (Appendix E of USACE 1999b) led to a resounding rejection of the first hypothesis. At most, around 8,000 jobs (0.27% of regional employment) and $700 million or 0.31% of regional output would be generated from restoration, changes unlikely to affect overall water demand. Moreover, the $6 to $8 billion of anticipated CERP spending was to be spread over 20 years, in comparison to the $13 billion already being spent each year on public and private construction in the region.

To test the second hypothesis—the impact of the economy on the environment—a broader model would be needed. The IMPLAN model is frozen in time, a single, in-depth snapshot. We needed a model of a growing economy and population for 30 years into the future, and with that, changes in the concomitant water and land requirements. Would there be enough water for the Everglades restoration under different expectations of growth?

The technology and models (IWR MAIN) for water forecasting in the late 1990s, as practiced by the Army Corps consultants (Gulf Engineers 1996), depended primarily on forecasts of population and the coefficients of water consumption per capita, plus other factors such as technology, consumer habits, weather, and housing patterns as they affect water consumption per person. However, the demographic projections computed by the state agencies had consistently understated population in the South Florida counties due to their extraordinary growth and immigration. Therefore, an economic as well as demographic model of South Florida was needed to capture the effects of an expanding economy, drawing in and holding a population much larger than that computed by the routine procedures using births, deaths, and historical migration.

IMPLAN is a widely used static input-output model developed by the US Forest Service to assess the impact of large projects on broad regional economies [Carlson et al. (1995) and Black et al. (1999) on the Upper Mississippi River]. It is based on the Leontief system that distinguishes Total Output $[\mathbf{x}]$ from Total Income [y] by adding the value of Intermediate inputs-and-outputs [X], which, in normal economic accounting, are omitted, because their value is seen as incorporated into the value of the Total Income, also commonly known as Final Demand. It is Final Demand that has become the central pillar of modern macroeconomics and recession control, but it is the growth of Intermediate Output that reveals the depth and intricacies of economic linkages within the regional economy. In simplified termswhere $\mathbf{x}=m\times 1$ vector of total outputs of $m$ sectors ($m=528$); $\mathbf{Q}$ = square $m\times m$ matrix of inter-industry transactions; and $\mathbf{y}$ = vector final demand.

Let $\mathbf{A}$ be the square coefficient matrix of total direct input requirements, defined as $\mathbf{A}=\mathbf{Q}\times {\mathbf{x}}^{-1}$, such that $\mathbf{Q}=\mathbf{A}\mathbf{x}$. Then the balancing equation isor, more commonlywhere ${[\mathit{I}-\mathit{A}]}^{-\mathbf{1}}$ = Leontief inverse that computes the total value of output $\mathbf{x}$ that is required to produce an income (final demand) of $\mathbf{y}$, andwhere a change in the values of the vector $\mathbf{y}$ results in a change in total output by sector. Employment follows fromwhere $\mathbf{N}$ = diagonalized matrix of jobs per dollar of output for the 528 industries; and $\mathbf{L}$ = vector of jobs associated with the output levels of $\mathit{x}$.

Thereforewhere a change in jobs is associated with a change in output.

The Regional Economic Modeling Inc. (REMI) model is a dynamic, comprehensive model, also widely used by governments and private enterprise, that builds a variety of economic and demographic variables around the input-output in Eq. (3), and, on the basis of equations fitted with data for the past 20 years, projects the regional economy forward into the next 30 years. The application of this model to South Florida, it was thought, would give more reliable projections of population for the fast-growing region and hence improved water forecasts. By comparison, the prepackaged IMPLAN model gives great detail for inputs and outputs for 528 producing sectors for any region for a single year, while our prepackaged REMI model uses hundreds of equations and other economic and demographic variables for only 14 sectors and permits the regions to interact and affect one another. (13 full counties comprise the four regions of our REMI model and correspond roughly to the four planning regions of the South Florida Water Management District; see Fig. 1.)

We shall describe the REMI model and then explain how the detailed, single-year IMPLAN sectors were hybridized into the broader and dynamic REMI sectors. In REMI, the values for previous known historical years are used to forecast future values through a system of lagged variables that incorporate feedbacks and multivariable impacts. A summary and simplification of the REMI equations are listed subsequently. The blocks to which these equations refer are sketched in Fig. 2, with some of their linkages indicated by arrows. [See Treyz (1993) for complete specification.]

The REMI model begins with Eq. (3), using the 13 nonfarm industrial-and-service sectors, and continueswhere $Y$ consists of the usual components of final demand; $C$ = final consumption; $I$ = investment for 3 categories; $G$ = government (local and state); $X$ = all exports from the region; and $[\mathbf{RPC}]$ = variable matrix of regional purchase coefficients, a key part of this model, that designates the share of inputs that are purchased within the region.

Blocks 1 and 5: Output Linkages, Market Shares

Blocks 2 and 4: Capital and Labor Demand; Wage, Price, Profit

Block 3: Population and Labor Supply

The purpose of this model was to provide future population estimates that would be superior to and more accurate than the state’s demographic forecasts. The first step to using this model was to refit it with new, locally generated employment data. Nevertheless, four major pieces were still missing from the off-the-shelf model, and these would require hybridizing both the REMI and IMPLAN in addition to adding several new layers to the basic data. First, agriculture, a key sector in all four South Florida regions, was poorly developed in REMI. Twenty-three agricultural sectors from IMPLAN were grafted onto REMI with annual totals supplied by the Regional Economic Information System (REIS) and crop forecasts from the US Department of Agriculture (USDA). Second, annual investment data from the Dodge Value-in-Place series and the Bureau of Economic and Business Research (BEBR) Series on Value of Building Permit Activity were needed to add an exogenous component to investment that captured speculative construction unrelated to ongoing regional activity, because the normal REMI model computes investment as adjustment to the internal capital stock. Third, tourism, a major economic generator, had to be modeled separately on the basis of two Visitor Spending Surveys and inserted into the model.

Fourth, the expected expenditures for Everglades Restoration—construction, operation and maintenance, planning, loss of crop lands for reservoirs, etc.—had to be programmed into the appropriate region, and taxes subtracted from each region according to the federal-state cost-sharing agreement in the funding and spending plan. Construction costs were allocated to the appropriate industrial sector on the basis of the components of the seven prototypical structures (aquifer storage recovery, seepage barriers, pumping stations, levees, canals, spillways, and culverts) that comprised the 298 individual projects in the initial $6 billion construction budget. See Weisskoff (2005), chapters 11–15 and insightful comments in Agthe (2005) and Cunningham (2007).

It was expected that this effort to hybridize the two regional economic models–IMPLAN and REMI–into a single simulation forecasting framework, and to add the necessary missing pieces, would yield superior population, income, and output estimates for 2010 through 2030 for each region. Once computed, how would these estimates from the economic system be linked to water demand?

The connection of such sophisticated economic modeling with the ecosystem and its services is challenging. Environmental factors had been integrated into input-output models of the US (Leontief 1970; Leontief and Ford 1972) and global scale (Duchin and Lange 1994; Duchin 1998). But in the case of Everglades restoration planning, the natural system had been compartmentalized, as if a firewall separated it from the built system. Yet the latter has clearly been the cause for the precarious condition of the former. The descriptive and historical literature on the Everglades has narrated those impacts ever since draining, dredging, and diking of the region began (e.g., Douglas 1947, 1997; McCluney 1969; Buffalo Tiger 2002; Grunwald 2006; Swihart 2011), yet the importance of these ecological-economic links had not been explicitly recognized in restoration planning.

By the early 2000s, the Science Working Group of the South Florida Ecosystem Restoration Task Force had created a series of flowchart-like conceptual models of the 9 (later, 11) prototypical ecological habitats found in the Everglades region [South Florida Ecosystem Restoration Task Force 2006; see Weisskoff (2005) and Ogden et al. (2005) for background and Rudnick et al. (2005) on Florida Bay].

These laid out detailed interactions between the fish, birds, grasses, soil, and water that make the Everglades so unique. At the very top of the flowchart for each habitat (see Fig. 3 for example), economic and societal factors are shown. In other words, influences from the human or built environment affect everything in the natural system. These are almost entirely economic: road and railroad construction, water management, agricultural and residential development, recreation, and fisheries. Despite the compartmentalization of the Army Corps’ mission to save the Everglades primarily through hydrological replumbing, it became clear that the economy had to be included in the Everglades Plan. Although the CERP plan did briefly examine how its activities would benefit regional economics (the hexagons at the very bottom of Fig. 3), the economic and societal stressors (the top rectangles) were never addressed.

Inspired by the conceptual models of scientists, our goal was, first, to create the hybrid economic model of the region that addressed the extreme openness of this region to outside factors as well as its unique cluster of activities, such as tourism, speculative housing investment, and farming. Second, the forecasts of the economic variables, such as employment, population, and income were connected to water usage for each region in order to forecast the water needs to 2030. It was thought that if the results of this linkage showed a significant impact on the Everglades ecosystem, then the water managers, the environmental community, and the general public would support radical change in water use. Indeed, the solution for Everglades restoration lay not entirely in the replumbing of the Everglades, but in concomitant changes in urban and farm growth, land use policy, and public water consumption practices.

Many current studies of water demand still use population as the major driver. The traditional technique of depending almost exclusively on population (Billings and Jones 1996) was challenged early by Gleick (2002) and Wolff and Gleick (2002), and has, in more recent years, given way in Miami-Dade County, for example, to the use of microeconomic regions, zip code files, and traffic zones to identify economic activities and their particular needs for water (Goldenberg 2012; Heberger et al. 2016). However, it is not clear what factors determine the growth of these microunits. By comparison, the South Florida Water Management District (SFWMD) regional plans still rely on the BEBR population forecasts for urban water use and on conventional factors such as weather, specific crop, and irrigated acreage for projecting agricultural water use and similarly in the Texas Water Plans (see Wurbs 2014; Texas Water Development Board 2019).

The projections of a regional growth model, encompassing all the sectors and such variables as prices, government budgets, population, and immigration, give policy makers the tools and the freedom to simulate the desirable as well as unthinkable futures, from moderate to hypergrowth, economic stagnation or collapse, hurricane calamity, and sea level rise. Based on these simulations, the policy makers are then in a better position to prioritize their infrastructure needs and crisis planning.

Was the model successful? The purpose of this paper is to examine how the forecasts of population and water usage, made nearly 20 years ago, compare with the actual measured values for the years 2005 through 2015. When the forecasts prove to be accurate (i.e., coincide with the actual counts), the model is then thought to be credible and reliable. But what does it mean when the forecasts deviate widely from the actual?

Evidence of the systematic underforecasting of the official BEBR demographic models is presented in Table 1 by percentage and absolute differences between the early forecasts made in 1982, 1995, and 2000, each for the subsequent and later censuses. In all but one case, the percent of error for each South Florida forecast is higher than the corresponding percent error for the entire state forecast. The absolute forecast for South Florida made in 1982 for 2010 was short 1.254 million people compared to the actual census and the forecast made in 2000 for 2010 was short by 377,000.

The gaps between the earlier population forecasts and their later actual census counts are shown in Fig. 4 as the distance between the topmost Census line and the lower lines, all for the terminal year of 2010. Under conditions of hypergrowth, BEBR consistently undershot the South Florida counties, despite its better record for the other, more-northern counties (Smith and Rayer 2011; Rayer and Smith 2014).

The regional population forecasts in Table 2 compare the actual 2010 census counts (column 3) to the BEBR and the hybrid model forecasts for 2010 (columns 4–5) in absolute numbers and in percentage differences (columns 6–7). The hybrid model estimates for the first three regions overshoot the actual population by 5.1% to 7.6%. The estimate for the fourth region, the Kissimmee Valley (KSV), is, however, exact. By contrast, the BEBR forecasts all underestimate the three smaller regions by $-4\%$ to $-17\%$ but overestimates the Lower East Coast (LEC) population by 5.3%. BEBR’s over and underestimated regions balance one another so their total Florida forecast appears accurate (0.3%), in comparison to our hybrid model that overshot the state total by 9.4%.

The hybrid model measures its outcomes in terms of dollars, jobs, and people. To connect the economic model to water usage requires the juxtaposition of the water usage series with the appropriate economic time series, i.e., in our case, employment, population, and output, respectively, for each of the two comprehensive water-use categories: public supply (PS), i.e., water withdrawn by utilities, and the rest of the fresh (RF), which includes all the other freshwater not drawn by utilities, namely, self-supplied agricultural (by far, the largest category), and the other self-supplied categories: domestic, commercial-industrial-mining, recreational, landscape irrigation, and power generation. PS and RF both draw from surface and ground sources. Actual water use data for all categories were available at five-year intervals from 1970 to 2000 for the four regions.

For the most recent five-year periods, we computed the elasticity of water demand with respect to each economic variable (employment, population, and output) for each region and water category, elasticity being defined as the percentage change in water use relative to the percentage change in the economic variable. Thus a trend was built into the future water coefficients based on their immediately prior historical behavior. The percentage change in the exogenous economic variables that were added as the missing pieces to the hybrid model were run at three different rates of growth (farming at 1%, 2%, 4%; investment at 0.25%, 0.5%, and 1.0%; and tourism at 1%, 2%, and 3%). Then the final values for population, employment, and output were applied to the corresponding historical elasticity to give the projected water use for 2010 through 2030. How well did the model perform? (See Tables 3 and 4 for the 2010/2015 forecasts for the four regions and three water categories using population as the major driver. See Figs. 5 and 6 for sketches of the forecasted water use of the regions through 2030 using population, employment, and output.)

The forecast of the hybrid model for the 2010 Public Supply for the Upper East Coast (UEC) region was extremely accurate, only 0.2% from the actual water use, and the forecast for the 2010 RF for the KSV region was 7.2% from the actual (Table 3). The forecasts for the 2015 PS usage for the UEC and KSV, respectively, were 3.3% to 4.0% above their actual usage (Table 4).

For the other regions, the divergence of the hybrid’s estimates from the actuals was much greater. However, the hybrid’s forecasts are actually closer to the actuals than the forecasts made by the SFWMD and US Geological Survey (USGS). For 2010, in the PS category, the hybrid ranks first (i.e., closest to the actual) in only one (UEC) of the four regions, but in the RF category, the hybrid’s forecasts rank first in three of the four regions (LEC, UEC, and KSV). In the All Water category, the hybrid ranks first in two of the four regions (LEC, UEC) and ties with the SFWMD for the third region, KSV (Table 3).

For the 2015 estimates, the hybrid ranks first for PS in two of the four regions (LEC, KSV). For RF and All Water categories, the hybrid ranks first in three of the four regions (LEC, UEC, KSV; Table 4). In summary, the hybrid’s estimates are closer to actual water use than the forecasts of the Water Management District and the USGS in 14 cases (or 58%) out a possible 24 cases ($4\mathrm{regions}\times 3\mathrm{water}\mathrm{categories}\times 2\mathrm{forecast}\mathrm{years}$).

Figs. 5 and 6 give better pictures of the divergence between the actual and the forecast. In the LEC (Fig. 5, top), public supply rose continuously from 1970 to 2005 and then fell only from 2005 to 2010. This contrasts with the RF (mostly agriculture), which began its decline from 2000 and, like PS, stabilized by 2015.

Two main patterns are apparent across the four regions for total freshwater use in the most recent years (Fig. 6). Two regions, KSV and Lower West Coast (LWC), maintained steady levels of water use from 2000 to 2010 before KSV plunged and LWC declined only moderately by 2015. By contrast, water demands fell both in LEC and UEC from 2000 to 2010. Water use in both LEC and UEC stabilized or rose slightly by 2015.

The hybrid model connects the economic variables to water use through the historically estimated elasticities (percentage change ratios), which themselves reflect the water use per person for the corresponding year of the economic variable. The most extreme decline in the coefficients of actual water use per person occurred in the UEC from 2000 to 2015 by 37.2% and lesser reductions of 18%, 19%, and 22% in the KSV, LWC, and LEC, respectively (Table 5, column 6).

The fall in our hybrid model’s coefficients are more modest because they are based on pre-2000 history. They exceed the actual coefficients (Table 5, righthand column) by 16% for LWC and 22%, 29%, and 41% for the LEC, UEC and KSV, respectively.

Note in Fig. 7 the two patterns for the trajectories: the water/person coefficients fell in all regions from 2000 to 2010, but the coefficients for the LWC and KSV rose from 2010 to 2015, while the LEC and UEC continued to decline. The most recent water supply plans for the water management district (see SFWMD-A 2012, 2016) suggest that the economic recovery may restore water consumption to its predrought trajectory.

The county data (Table 5, lines 8–24) provide the underlying detail of the changes going on within the regions. The heavily agricultural counties of Glades, Hendry, and Martin experienced the most dramatic declines in water use per person. Okeechobee is the lone county showing an increase (12.9%) increase in water use per person, reflecting perhaps its very low coefficient in the 2000 base year.

The dramatic changes in these water coefficients document nothing less than a major paradigm shift in urban water use that is consistent with the decoupling of water use from economic and population growth observed elsewhere (Gleick 2002; Wolff and Gleick 2002). The enormous effort to create and calibrate the hybrid economic model, adding the missing pieces for each region to the REMI model, disaggregating the agricultural sectors with IMPLAN sector data and REIS time series data, varying their growth rates, and then connecting the range of population, employment, and output to varying water coefficients in order to compute a more realistic set of water demands, all resulted in overestimates of 62.8% for 2010 and 77.0% for 2015 of the total freshwater demand for South Florida (see Table 3, line 19, column 8, and Table 4, line 19, column 6).

The economic model and its derived water demands state unambiguously what the society would have needed had the existing trends and structure continued, even at different rates of growth. After all, the purpose of the whole exercise was to answer the question, “Would there be enough water for the continued growth of the cities and farms—and for nature, even as a residual?”

But other forces were at work. First, the seemingly unbridled growth continued from 2001 to 2006, for the opening years of the decade—and then collapsed. The growth of jobs in the LEC, the economic engine of the region, which had averaged 2.53% per year from 2001 to 2006, fell to an average rate of zero for the next five years. The growth rate of jobs in the other regions, which averaged over 5% per year for the 2001–2006 period, fell to an average annual rate of $-1.5\%$ in the LWC, $-0.9\%$ in UEC, and $+0.7\%$ in the KSV. In all the regions, the actual number of employed people peaked in 2007, not to be surpassed until 2012, based on our computations aggregating county employment totals from BEBR (2017).

The reasons for the decline in PS and especially RF lie in the effects of the two-year drought that affected crop production, on both irrigated and nonirrigated lands from 2007 to 2009 (see SFWMD 2009), and also brought water shortage restrictions from the Water Management District to the public utilities and directly to the people (FL Department of Agriculture 2007; FL Division of Emergency Management CIEM 2007). Water use declined due to the planting of less water-intensive crops, reduction in cattle herds, changes in the housing mix to apartments, duplexes, and condos with smaller or even negligible landscaping, the widespread and utility-incentivized diffusion of water-saving technologies in homes and businesses, and strict enforcement of water restrictions. In many regions of South Florida, the utilities deliberately maintained their traditional user charges and relied on voluntary restrictions (with the threat of fines) to reduce demand. Even after the initial stressors passed, the restrictions on landscaping irrigation were kept and a new level of water frugality maintained.

Another unexpected benefit of the reduced water demand was the “savings” in capital expenditure that was planned but no longer needed. The Miami-Dade Water and Sewer Department reported a saving (or postponing) of $1.6 billion expenditure for a two-phased expansion of a reverse-osmosis (RO) plant and a new water reclamation plant (Fritsche et al. 2012). Similar observations have been reported elsewhere. See Dziegielewski and Chowdury (2012) on growth scenarios for the Chicago region, and Ghimire et al. (2015) on the importance of pricing during drought on Oklahoma City. Contrast the Donkor et al. (2014) review of urban water demand and the necessity of 20–30 year forecasting, with Fullerton and Cardenas (2016) on short-term forecasting for Phoenix. Compare the methodologies of Zhi et al. (2015) using scale, technology, and input-output techniques during Beijing’s hypergrowth, with demand estimation by Hester and Larson (2016) during periods of economic crash, rebirth, and drought in three North Carolina cities. See Avri et al. (2015) for demand modeling in Israel by clustering different types of users according to homogeneous consumption patterns. See Chabet-Ferret et al. (2019) on the results of experiments in nudging French farmers to save water.

For economic models to improve forecasting, additional and more current variables must also be introduced, e.g., factors to keep up with shorter and more erratic business cycles, climate change, weather alterations, and the dynamic international situation which affects migration, food prices, international trade, and sharp changes in capital flows. The key finding is what the model tells us about how both economy and water use respond to factors such as recession, water restrictions, drought, and new technology, not the supposed accuracy of the model’s forecasts. Thus, the comparison of our model’s forecasts to the actual is a measure of the water quantity saved due to these and other factors. This speaks to the priority of the soft path over the traditional economic view that price alone should be the major tool to alter demand. Our model measures the distance between what would have happened if both business and nature had continued to run their normal courses during these two decades.

Another striking feature in the findings is the surprising decline in the RF water category in SF from $128.2{\mathrm{m}}^3/\text{s}$ (2.9 billion g/d) in 2000 to $74.2{\mathrm{m}}^3/\text{s}$ (1.7 billion g/d) by 2015. During the same period, PS for South Florida (SF) held steady around $46-47{\mathrm{m}}^3/\text{s}$ (1 billion g/d) (Tables 3 and 4). Thus the fall in the All Water category is almost totally due to the decline in RF, of which agricultural demand is the major part.

This striking finding might be subject to revision due to three ongoing competing studies by three different agencies. First, USGS maintains a long-term series based on a consistent methodology involving crop-specific irrigated acreage and weather variables (Marella 2014, 2015; Marella and Dixon 2018). Second, SFWMD estimates agricultural water use as part of its Water Supply Plans, and these are based partly on survey responses and permit data. However, their annual data do not form a comparable long-term series due to changing methodologies (see SFWMD-B 2016, 2017a, b, 2018). Third, the Florida Department of Agriculture recently commissioned the Balmoral Group (2014, 2015, 2017, 2018), a private consultant, to develop projections for a consistent water use series for all Florida counties based on the USDA agricultural census database of irrigated acreage and crop choice.

But recent findings by the USGS raise questions about all of the preceding methodologies. Marella and Dixon (2014) and Marella et al. (2018a, b, c, d, e, f), in their pioneering field verification studies of irrigated farmland, crop choice, and irrigation equipment in South Florida counties, found 21.8% fewer irrigated hectares [that is, 36,584 ha (90,400 acres)] than those listed in the Florida Statewide Irrigation Demand (FSAID) Irrigation Lands Geodatabase (ILG), version 4, used by the Balmoral Group studies, and 17.4% more [that is, 19,492 ha (48,167 acres)] irrigated acres than reported in the 2017 US Agricultural Census for those 6 counties (USDA 2019). Continuous monitoring is needed to verify actual agriculture water use.

Was South Florida’s water use decoupled from population growth earlier or later than the US trends? Coupling implies that changes in water use are linked to population. The results in Table 6 suggest the following trends for the period 1970 to 2015. For the US, the five-year average rates of change of PS water use (line 7) do rise and fall with the changes in US population growth rates (line 2), except for two of the eight periods studied, namely, from 1990–1995, when US population growth rate fell and PS growth rate rose, and in 2010–2015, when the US population growth rate stabilized but the PS growth rate declined.

In South Florida, however, the decoupling of the PS growth rate (line 10) from its corresponding population growth rate (line 4) began in the 1990–1995 period and continues to the present, as evidenced by a different pattern in each of the most recent five-year periods: (1) from 1990–1995, the population growth rate was stable but PS growth rate rose; (2) from 1995–2000, the population growth rate fell but the PS growth rate was stable; (3) from 2000–2005, the population growth rate stabilized, but the PS growth rate fell; (4) from 2005–2010, the population growth rate fell by 7 percentage points but the PS growth rate fell by 17 percentage points; and (5) finally, from 2010–2015, the population growth rate rose by 1.6 percentage points but PS growth rate jumped by 14.5 percentage points. In short, factors other than population must explain PS water use in South Florida.

The growth of the RF (mostly agricultural) (Table 6) should not be coupled with population due to the importance of weather, crop choice, prices, irrigation, and other factors. The US RF water use peaked in 1980, stabilized until 2005, and has fallen from 2010 to 2015. By contrast, the South Florida RF water use peaked in 2000 and also fell from 2005 to 2015 at negative rates far greater than the rates of decline of the RF for the US series.

Regional economic models may be superior to purely demographic models if they account for special features of the region such as agriculture, migration, exogenous investment, and unique federal programs. Water use is connected to the economic model through a series of elasticities that pair water use to economic variables such as population, employment, and income. Future water use can then be forecast on the basis of the projected values of the economic variables at different rates of growth—or decline. We have seen, however, how past performance of the South Florida hybrid model provides little guidance for the future. New variables must be introduced into the model that include changes in housing patterns, consumer habits, family composition, bottled water use, and also private landscaping wells.

These forecasts, originally computed in 2000 for the 2010–2030 period, while overestimating the actual water use in 2010 and 2015, actually provide a scorecard of each region’s achievement in reducing its water use. These reductions, which are measured by the difference between the forecast and actual water use, are the sum of each region’s reactions to economic factors (such as recession, housing patterns, and loss of farm acreage), government policies (such as water restrictions, diffusion of water-saving technology), and social factors (such as home management practices, water conservation, reduction in lawn size, use of bottled water).

How do our findings on changing water use in South Florida affect progress in Everglades restoration, which, indeed, was the very motivation for this study? On one level, the two systems—the built and the natural—have been so successfully compartmentalized that the most recent scientific review by the National Academies of Science, Engineering, and Medicine on Everglades Restoration (NASEM 2018) mentions only in passing that “future population growth and development…has important implications for land and water use and will add challenges associated with flood management and water quality” (p. 134). The report goes no further.

To this end, our hybrid model has come of age, to simulate the impact of anticipated growth on region-wide job-creation, land-use, and all the “hidden” infrastructure—roads, schools, utilities, police, fire prevention—needed for a growing society. The demand for water provided by the public utilities, we have seen, has been stable or falling despite the growth in the number of people and the value of commerce and industry. In the agricultural sector, new irrigation practices and field-verified data will provide more realistic estimates of the true decline in that sector’s water use.

In the past, the cry of water shortage, based on alarmist but authoritative forecasts, led to regional conflicts over water rights, e.g., the proposed piping of Central Florida spring water down to South Florida (see Florida Council, 2003; Goodnough 2003). More recently, Carr and Zwick (2016), in a study for the 1000 Friends of Florida, proposed aggressive water conservation and zoning changes on the basis of the Balmoral Group’s water forecasts (see Holt 2015; Staletovich 2016). Today, the water utilities of South Florida face challenges, not of water quantity, but of aquifer depletion, salt water intrusion of coastal wells, sea level rise, and wastewater treatment and disposal. The once-controversial soft-path strategies of conserving water have become business as usual in South Florida since the drought-recession period.

The conflict between freshwater use by cities and farms, on the one hand, and a restored Everglades, on the other hand, is not the hypothesized competition over water quantity per se, but over the issues implicit in the development package being promoted in the current growth spurt, e.g., water quality, waste-water disposal, air and water pollution, intensification of the farmlands, run-off from insecticides and chemical fertilizers, the transformation of farmland into residential estates, and the expansion inland of the coastal counties—all against the Everglades frontier. New megamalls, entertainment and shopping centers, sprawling residential developments, and oil exploration on Florida panther reserves all enter the planning stages. Nutrient-rich discharges from Lake Okeechobee trigger bursts of toxic algae blooms on the Atlantic coast estuaries.

Just as the soft path was forced upon the cities and farms by the confluence of natural (drought) and national (recession) events in South Florida, so too the Everglades restoration recommended by the most recent National Academies Report (2018) is already directing the Restoration path away from its early goal of “Getting the Water Right” (quality, quantity, timing, distribution, and flows) by the “hard” path, that is, replumbing parts of the historic Everglades (McVoy et al 2011) with pumps, canals, levees, reservoirs, storage treatment areas (STAs), and aquifer storage and recovery (ASRs). The Everglades is still a human-controlled system in which the natural habitat struggles to survive within compressed boundaries and managed water flows.

The soft path for Everglades restoration might require less but purer water, more organic water-tolerant crops demanding less drainage in the surrounding agricultural areas. The soft path for CERP might utilize its adaptive management authority to reduce the projected $40 billion construction bill, including the lure of the 10.69-m (35-ft)-deep and 0.81-m (32-in)-wide underground cement-bentonite slurry wall intended to stabilize differing water levels on either side of the protective levees (National Academies 2018). The reported success of these seepage barriers could encourage even more intensive construction on the city-side of the boundary while supposedly maintaining year-round flood levels on the Everglades side of the boundary.

Our hybrid economic model can also estimate the comprehensive effects on the new inland towns, shopping malls, office complexes, and hotels planned to be built against the Everglades by forecasting traffic, water pollution, and air quality, and the need for police, fire protection, schools, parks, and open space. How will these new stressors undermine the Everglades?

In the early years of Everglades restoration, South Florida looked briefly to the Brazilian Pantanal, a wetland similar to the Everglades but 12 times larger (Junk et al. 2011), as a living, somewhat idyllic model of the predrainage Everglades. But the Pantanal too is confronted by the pollution and drainage of expanding farmlands on its fringes. Its rain-driven pulsating waters flow through a complex system of open tributaries in contrast to the Everglades “catch and store” restoration efforts and the CERP’s hard path that grows out of the early flood-control program set up when free-flowing water was viewed as the enemy of city and farm.

The paradigm shift already occurring in water use in the city and countryside and the very absence of the once-projected water shortages, all document the changes toward the soft path underway in Everglades restoration. The large-scale removal of levees and dikes, the partial raising of the major east-west corridor through the swamp (Tamiami Trail) to allow the Everglades to flow more freely from north to south; all these pro-nature adjustments must compete with extraordinary development pressures when hyper-growth prosperity returns to South Florida. Stakhiv’s (2003) question, quoted in the opening section of this paper, (“Who is responsible for planning resource use in the Everglades region?”) must now be juxtaposed with the more recent question raised by Zhang and Balay (2014), also in this journal, regarding growth and development: “How much is too much?”