Purpose and Scope

This report describes the development and results of the SWB model used to estimate potential recharge from 1900 to 2019 for the Long Island, New York regional aquifer system. Two SWB simulations were conducted for this study, with land cover being the factor that differentiated them. The first simulation used existing land-cover datasets (Sohl, Reker, and others, 2018; Sohl, Sayler, and others, 2018) to estimate the annual potential recharge with changing land-cover patterns from 1900 to 2019 (referred to herein as the postdevelopment simulation). The second simulation changed all land-cover categories including developed, mechanically disturbed, barren land, pasture and hay, and cultivated crops to the land cover category mixed forest to represent a predevelopment land-cover condition annually from 1900 to 2019 (referred to herein as the predevelopment simulation). This simulation was designed to calculate potential recharge for a predevelopment setting, without any human-induced changes to the land cover. These simulations were designed to calculate the amount of water affected by impervious surfaces in urbanized areas (referred to herein as rejected recharge or potential rejected recharge) by subtracting the predevelopment from the postdevelopment simulations for the equivalent year.

Meteorological Data Inputs

The meteorological data inputs needed for this SWB model were (1) total daily precipitation, (2) daily maximum temperature, and (3) daily minimum temperature. Gridded meteorological data were downloaded from the Oak Ridge National Laboratory (ORNL; Thornton and others, 2014), for the years 1980–2019. Gridded meteorological data prior to 1980 were created by interpolating daily weather-station data from the National Oceanic and Atmospheric Administration (NOAA; National Oceanic and Atmospheric Administration, 2020) using a geographic information system (GIS).

The SWB model required the climate data to be in the same format for the entirety of each incremental model. As stated previously, the SWB model required a precondition year to be run, which meant the model would have to start 1 year before the simulation period, in 1899. The year 1980 acted as the pivot year between the two meteorological datasets. The SWB model outputs for 1980 were modeled as a part of the 1970–80 incremental model, which used the interpolated NOAA weather-station data. ORNL data was used as the input for 1980 (and every year after), which acted as the precondition year for the 1980–90 incremental model.

National Oceanic and Atmospheric Station Data for 1899–1980

Meteorological data were available from a network of 150 NOAA meteorological stations during the period 1899–1980 (National Oceanic and Atmospheric Administration, 2020). Although the data collected at all the stations were not continuous throughout the period, and the number of stations with available data changed throughout the period, these were the best available data to use for generating raster grids representing the meteorological data necessary for the SWB model.

To ensure that the entire modeled area was represented in the daily interpolations, the selection of meteorological stations extended beyond Long Island. Data were collected from stations in southern New York, Connecticut, Rhode Island, Massachusetts, and northeastern New Jersey (fig. 4). Table 1 (in back of report) lists the locations of the stations along with the years for which data were used.

For this study, the GIS interpolation tool Spline with Barriers (Esri, 2019) was used for generating the daily grids from the NOAA data. This tool interpolated a continuous surface between data points. The tool had several mandatory fields: (1) input point features (the latitude and longitude position), (2) a Z-value for each input point, and (3) an output raster. There were several optional fields: (1) input barrier features (features used to constrain the interpolation), (2) output cell size, and (3) smoothing factor. The daily meteorological station data points were the input point features. The Z-values included the daily measurements of either precipitation (in inches) or the maximum or minimum temperatures (in degrees Fahrenheit). The output rasters were either the daily precipitation (PRECIP) or the daily maximum (TMAX) and minimum (TMIN) temperatures, depending on the Z-value used as input. The output cell size was set to 1,000 meters, the smoothing factor input was set to 0.9, and the input barrier features were not used. The daily PRECIP, TMAX and TMIN output rasters were created by using the different input Z-values and executing the spline interpolation tool three individual times. A conditional tool was also used on the PRECIP raster to remove any negative interpolated values created by the spline interpolator. These daily gridded raster outputs became the daily meteorological inputs of the SWB model from 1899 to 1980.

Both the interpolated NOAA and Daymet datasets for Long Island were used to create the daily meteorological data input for 1899–2019 required by the SWB model code. A graph of the spatially averaged temperature and total precipitation for each year from 1900 to 2019 for Long Island is shown in figure 5. The temperature trend line shows an increase of 2.7 degrees Fahrenheit during the 120-year period. An increase of 4.6 inches (in.) can be seen in the precipitation trend line, despite a major drought that occurred on Long Island during the 1960s.

Hydrologic Soil Groups and Available Water-Capacity Inputs

Soil-related datasets were additional gridded input requirements for the SWB model code. They were the (1) hydrologic soil group dataset and (2) the available water capacity dataset. Both datasets were available through the Natural Resources Conservation Service Soil Data Development Toolbox extension in ArcMap (Peaslee, 2020). A hydrologic soil group category and an available water capacity value were assigned to every SWB model grid cell. The categories and values were unchanged throughout the 120-year period and did not vary between the pre- and postdevelopment SWB model simulations.

The soils were categorized and grouped using a letter designation of A through D on the basis of the soil infiltration rate (Peaslee, 2020). Soils in group A had the highest infiltration rate (greater than 0.30 inches per hour) and consisted mostly of sandy soils, whereas those in group D had the lowest rate (less than 0.05 inches per hour) and consisted mostly of clayey soil (Natural Resources Conservation Service, 2019). Soils in group A are most common in Nassau and Suffolk Counties, whereas soils in group B are most common in Kings and Queens Counties (fig. 6; table 2).

The available water capacity is defined as the amount of water that a soil can store for potential use by plants (Smith and Westenbroek, 2015). Figure 7 shows the available water capacities for soils on Long Island (Peaslee, 2020). These data are expressed in inches of water per foot of soil thickness. Generally, available water capacity varies depending on the organic matter content, texture, and structure of the soil (Natural Resources Conservation Service, 2019). In general, coarse-grained soils have a lower water capacity than fine-grained soils because of the larger pore spaces between the coarse grains, which allow the water to drain more freely.

Land-Cover Inputs

Annual land-cover grids for inputs to the SWB model are available from 1938 to 1992 in Sohl, Reker, and others (2018) and from 1992 to 2005 in Sohl, Sayler, and others (2018). These gridded data have a spatial resolution of 250-m pixels with 14 unique land-cover categories, although not all of those categories applied to Long Island. The published data were arranged in an Albers Equal Area Conic projection and were reprojected to a Universal Transverse Mercator (UTM) projection for the SWB model.

Land-Cover Data for 1900–37 and 2006–19

The 1938 and 2005 land-cover datasets from Sohl, Reker, and others (2018) were used as a starting point to create the land-cover inputs for 1900–37 and 2006–19, respectively. The 1900–37 data use the same land-cover categories as 1938 across Long Island. From 1900 to 1937, the only grid cells that changed categories were those of developed land (low-, medium-, and high-intensity development), which were dependent on the U.S. Census population within a grid cell for each year (described in the previous section). The land-cover datasets for 2006 to 2019 had the same categorical spatial distribution as that published for 2005. The development intensity was again adjusted to reflect the population changes during the 2006–19 period and those associated categorical changes were in turn made to the model grid. Figure 8 shows the modified SWB model land-cover data for Long Island during 1900 and 2019.

Interpretation of Land-Cover Inputs Used in Soil-Water-Balance Model Simulations

The two SWB simulations completed for this study required the land-cover inputs to be represented differently. The postdevelopment SWB simulation land-cover inputs depict the land cover as it changed over time. Figure 8B illustrates the SWB model land-cover input used for the postdevelopment simulation for 2019. Figure 9 illustrates the SWB model land-cover input used for the predevelopment simulation for 2019. The predevelopment SWB model simulated land-cover inputs depict land cover in a simulation with many of the land-cover categories changed to properties of the mixed forest land cover (fig. 9). The result simulates a development-free landscape for Long Island. Adjusting the recharge properties of the land-cover categories for the predevelopment simulation was also done through adjusting the predevelopment lookup table (table 4).

Postdevelopment and Predevelopment Simulation Results

The SWB model simulated both postdevelopment and predevelopment conditions from 1900 to 2019. The postdevelopment simulation shows that the mean annual potential recharge rate across Long Island varied from a low of 7.57 inches per year (in/yr) in 1965 to a maximum of 34.44 in/yr in 1983 (fig. 10). The predevelopment simulation results show that the mean annual potential recharge rate ranged from 6.43 in/yr in 1965 to 34.28 in/yr in 1983 (fig. 10). These simulation results indicated that—on average—40.7 percent of precipitation was recharged to the aquifers under postdevelopment conditions and 37.5 percent of precipitation was recharged under predevelopment conditions. The mean annual potential recharge rate across Long Island was 19.24 in/yr (1,293 million gallons per day [Mgal/d]) and 17.81 in/yr (1,197 Mgal/d) for the post- and predevelopment simulations, respectively. Despite the increase in impervious surfaces during the postdevelopment simulation, the overall increase in the island-wide recharge rate likely reflects deforestation in the eastern part of the island.

The deforestation that has occurred on Long Island has reduced evapotranspiration, which allows for more precipitation to infiltrate the ground (Yang and others, 2014). Despite the impervious surfaces that are present, precipitation infiltrated the ground at a higher rate in the postdevelopment simulation than in the predevelopment simulation, in which forested conditions and low- to medium-intensity developed areas were prevalent; however, for the high-intensity development areas, where impervious surfaces are much more widespread, infiltration rates were lower and the rejected recharge amounts were higher, thus increasing overland surface runoff. A breakdown of the data by county helps illustrate these relations and is discussed in the following section. When comparing the spatial distribution of recharge for the postdevelopment and predevelopment simulations in 2019, potential recharge is higher in western Long Island in the predevelopment simulation, and higher in eastern Long Island in the postdevelopment simulation (fig. 12). This was a common pattern in each year of the predevelopment simulation.

Most of Kings and Queens Counties in western Long Island was simulated to receive only 1 to 2 in. of potential recharge in the postdevelopment model because of the influence of high-intensity development and the land-cover lookup table inputs used in table 4. The predevelopment model, using forested land cover, provided an approach to estimate the amount of water that would have recharged into the aquifer system under natural conditions. This approach was used to determine the potential rejected groundwater recharge (lost recharge) that would be captured in storm drains, retention basins, and—in the case of Kings and Queens Counties—the combined sewer system (fig. 13). The difference between the pre- and postdevelopment SWB model simulations estimated a potential rejected recharge of 19.24 in. (around 167 Mgal/d) for 2019 in Kings and Queens Counties combined. No rejected recharge amounts are shown where there was more potential recharge occurring during the postdevelopment SWB model simulation (much of Nassau and Suffolk Counties).

The overall postdevelopment simulated potential recharge (fig. 12A) was greater than the predevelopment simulated potential recharge (fig. 12B) in the low- to medium-intensity developed areas, which were mostly located in Suffolk and Nassau Counties. The predevelopment simulation results with forested land cover showed that more evapotranspiration occurred in what are now low-to medium-density developed areas, meaning less water was available to infiltrate and recharge the aquifer (fig. 14).

Recharge Totals by County

The potential recharge amounts simulated and averaged for all of Long Island (fig. 10) do not provide insight into the spatial variability of potential recharge seen across the island (fig. 12). A graph of the mean annual potential recharge for the postdevelopment land-cover simulation by county (fig. 11) shows recharge for Kings and Queens Counties diverges from Nassau and Suffolk Counties over time. This divergence between the two sets of counties is largely the result of impervious surface development in western relative to eastern Long Island.

Both pre- and postdevelopment land-cover simulations for Nassau and Suffolk Counties indicate that the annual differences in recharge rates are relatively small (2.8 and 2.56 in. respectively) compared with Kings and Queens Counties (10 and 5.76 in. respectively). The mean annual potential recharge for Suffolk County is 21.16 in. for the postdevelopment and 18.36 in. for the predevelopment simulation. Similarly, the mean annual potential rechange for Nassau County is 19.13 in. for the postdevelopment and 16.57 in. for the predevelopment simulation. Although the differences are small, recharge rates are almost always greater during the postdevelopment simulation (fig. 15) and are likely attributed to the loss of forested land cover in what are now low- and medium-intensity developed areas throughout both Nassau and Suffolk Counties.

In Kings and Queens Counties, the differences between pre- and postdevelopment simulated potential recharge rates are much more pronounced than those for Nassau and Suffolk. The mean annual potential recharge for Kings County is 7.52 in. for the postdevelopment and 17.52 in. for the predevelopment simulation. Similarly, the mean annual potential rechange for Queens County is 10.92 in. for the postdevelopment and 16.68 in. for the predevelopment simulation. The large differences between the post- and predevelopment simulations for the two counties highlight the effects of the high-intensity development and the impervious surfaces in these highly urbanized areas.

Comparison to Previous Studies

Lusczynski (1952) did not provide a quantitative estimate of recharge for Kings County; however, he indicated that the development and increase of impervious surfaces that occurred in Kings County from the early 1900s through 1930 coupled with lower-than-average precipitation rates in the early 1930s caused a reduction in natural recharge. These factors—along with the increased groundwater withdrawal that accompanied population growth—caused a substantial depression in the water table. Although this study did not evaluate groundwater levels, the decline in natural recharge from 1900 through the 1930s, as discussed by Lusczynski (1952), is consistent with the mean annual potential recharge simulated by the SWB model (fig. 15A).

Soren (1971) estimated that the average recharge rate from precipitation in Queens County during the mid-1900s was about 55 Mgal/d (10.41 in/yr) and that the recharge rate from precipitation during the drought years of 1962–66 declined to about 40 Mgal/d (7.57 in/yr) or less. The mean potential recharge rate for 1930 to 1970 in Queens County estimated by the SWB model is 9.32 in/yr; this time period includes the drought years for which Soren (1971) estimated recharge. The SWB model estimated that the potential recharge rate for the drought years 1962–66 was about 5.00 in/yr for Queens County, about 2.5 in. less than the rate estimated by Soren (1971).

For the period 1968–75, Peterson (1987) estimated the regional groundwater recharge rate for Nassau County to be 20.6 in/yr (47.6 percent of precipitation) and the evapotranspiration rate to be 21.8 in/yr (50.3 percent of precipitation). For Suffolk County, Peterson (1987) estimated the recharge rate to be 23.5 in/yr (51.2 percent of precipitation) and the evapotranspiration rate to be 22.1 in/yr (48.1 percent of precipitation). The precipitation rates used by Peterson (1987) were 43.3 in/yr for Nassau County and 45.9 in/yr for Suffolk County. At the time, these values validated using 50 percent of precipitation to estimate recharge for Long Island.

For the period 1900–2019 for Nassau County, the SWB model estimated that the average recharge rate was 19.13 in/yr (41.1 percent of precipitation) and the average evapotranspiration rate was 23.10 in/yr (49.6 percent of precipitation). For the same period for Suffolk County, the SWB model estimated that the average recharge rate was 21.16 in/yr (45.2 percent of precipitation) and the average evapotranspiration rate was 22.68 in/yr (48.4 percent of precipitation). The average precipitation rate for 1900–2019 was 46.6 in/yr for Nassau County and 46.8 in/yr for Suffolk County. For the period 1968–75 (the period studied by Peterson, 1987) for Nassau County, the SWB model estimated that the average recharge rate was 19.63 in/yr (42.3 percent of precipitation) and the average evapotranspiration rate was 21.19 in/yr (45.7 percent of precipitation). For the same period for Suffolk County, the SWB model estimated that the average recharge rate was 21.61 in/yr (46.9 percent of precipitation) and the average evapotranspiration rate was 21.69 in/yr (47.1 percent of precipitation). All of these calculations indicate that using an assumed recharge rate of 50 percent of precipitation may have resulted in an overestimation of recharge.

Lastly, Masterson and others (2013) used a SWB model to calculate potential recharge for 2005–9 for the Northern Atlantic Coastal Plain aquifer system, in which Long Island was a part. Recharge rates were not broken down by State or county; however, they indicated that the mean annual potential recharge rate on eastern Long Island reached as high as 37.3 in/yr, which was the highest throughout their entire study area. For the same time period for eastern Long Island, the SWB model used in this study estimated that the mean annual potential recharge rate to be slightly higher, at around 39 in/yr.

Limitations of Soil-Water-Balance Model Analysis

This SWB model analysis has a number of limitations and uncertainties related to the input datasets. The greatest are related to (1) the spatial distribution of the U.S. Census population, which was used to classify low-, medium-, and high-intensity development; (2) the interpolation of meteorological data to generate daily precipitation and maximum and minimum temperature grids for the years 1900–80, as well as the interpolation processes used by the ORNL for the Daymet datasets for 1980–2019; and (3) modifications of the predevelopment simulation lookup table to assume a widespread forested land-cover type. All recharge grids generated as a part of this analysis are representative of potential recharge and do not yield actual recharge values.

The U.S. Census collects data on 100 percent of the population every decade. The population totals are grouped and summed into different census shapes (tracts, block groups, blocks). These shapes are not consistent and vary from decade to decade. To standardize the population counts for the SWB model, all decadal population counts were distributed to the Long Island model grid used for the groundwater-flow model (Walter and others, 2020). A linear interpolation between decadal census years was applied to each grid cell so that a continuous annual population dataset from 1900 to 2019 could be used to identify spatial differences in population intensity. Because the 2020 census was not available at the time of this analysis, the same rate of population change was applied to years 2011–19 as was seen for years 2000–10.

The annual population totals were then applied to the land-cover grids and reclassified as either low-, medium-, or high-intensity development. A population between 1 and 50 was designated as low, a population between 51 and 100 was medium, and any grid cell with a population greater than 100 was high. Altering these population bins changed all land-cover inputs to the SWB model; however, it was a necessary measure to add more detail to the land-cover inputs because, in their published state, only one population category was present. The reclassification was a way to show the variability of population across Long Island during the entire period.

Daily climate data were not available in a gridded data structure before 1980; therefore, an interpolation tool (Spline with Barriers) was used to generate the daily climate grids in ArcGIS Desktop. The tool interpolated a continuous surface between the NOAA weather stations (fig. 4) for precipitation, maximum temperature, and minimum temperature. The output raster grids are the climate parameters necessary to run the SWB model.

Additional limiting assumptions included changes made to the land-cover lookup table to group various land-cover types. For example, in the predevelopment simulation, most of the land cover was set to evergreen forest conditions because the land-cover properties for evergreen forest are an averaged value of all three forested land-cover types present on Long Island.

It is also important to note that for the purposes of this application, both soil coverages (hydrologic soil group and available water capacity) were assumed to be constant throughout the simulation period. And lastly, surface-water flow direction was not included as part of the SWB model. This option is sometimes selected to channel overland flow between cells within the model domain (Smith and Westenbroek, 2015); however, its exclusion in this SWB model means that any rejected recharge did not get redirected to neighboring SWB model grid cells.