2.1. Study Area
Catalina Island is one of the eight Channel Islands, located about 74 km south-southwest of Los Angeles, California (
Figure 1a). The island encompasses an area of 194 km
2, of which 88% are controlled by the Catalina Island Conservancy, and are completely undeveloped. Elevations reach as high as 639 m above sea level at Mt. Orizaba, with nearly 90% of the elevation drop occurring in the first 1–2 km from the central ridge that bisects the island. The island represents one of several exposed ridge crests of the California Continental Borderland geomorphic province and consists of Mesozoic metamorphic basement (i.e., the Catalina Schist) overlain by Miocene igneous rocks (primarily andesitic lava flows and quartz diorites of the Catalina Island Pluton) and Paleogene to Neogene terrestrial and marine sediments [
45].
Catalina Island falls within a Köppen–Geiger climate classification of Mediterranean (
Csb) and is marked by warm, dry summers and mild, wet winters. Monthly temperature averages range from 11.9 °C in January to 21.7 °C in August at Avalon Airport for the years 1948 to 2016. The precipitation average for the same period was ~300 mm annually and varies seasonally and spatially depending on orographic factors, with Little Harbor on the southwest side receiving 200 mm and Avalon on the southeast receiving 350 mm per year [
38]. There are no perennial streams on the island, although several days of runoff can occur in the larger watersheds, such as Middle Canyon, after intense rainfall events. None of Catalina’s streams are currently instrumented with discharge gauges, which prohibits a more detailed analysis of surface runoff.
According to the NCLD land cover classification, the island is dominated by chaparral shrub/scrub vegetation (82%) with minor proportions of herbaceous grassland (10%;
Figure 1b). This classification is not corroborated by field studies [
38,
39], which highlight a dominance of herbaceous grassland (>80% of cover) with isolated occurrences of chaparral vegetation, coastal sage scrub, and prickly pear.
Despite the recent drought and mandatory water rationing, very little information exists on the island’s hydrogeology. Groundwater elevations roughly follow topography [
46], and increased rainfall with elevation implies increased recharge in topographically higher areas. All of the supply wells, which are maintained and operated by an electricity supply company, extract water from the alluvium that is hydraulically connected to the bedrock aquifer; the strength of this connection, however, is undefined [
46]. None of these wells were accessible for sampling in this study, but the State Water Board Groundwater Ambient Monitoring and Assessment (GAMA) database reports groundwater geochemistry data for 18 wells and lists well construction data (i.e., screen depth and length) for 11 of those. The wells are generally shallow, with depths to the well screen bottom ranging from 6.1 m to 34.4 m below ground surface. To the authors’ knowledge, data on hydrostratigraphy, hydraulic conductivities, and the water storage capacities of the alluvial aquifers are not yet publicly available. One problem associated with Catalina groundwater is high concentration of total dissolved solids [
47]. Many water supply wells installed near Avalon in the early 20th century were abandoned due to salt water intrusion [
47], and many wells along the west coast of the island were abandoned in the late 1980s/early 1990s due to high groundwater salinities. The majority of the island’s drinking water is currently derived from supply wells in the immediate vicinity of Thompson Reservoir, a 1.42 × 10
6 m
3 capacity storage reservoir located in Middle Creek Canyon, about 10 km west of Avalon.
2.2. The SWB Model
SWB computes potential groundwater recharge at a daily frequency on a grid-by-grid cell basis. The model follows the approach of Thornthwaite and Mather [
48] and quantifies recharge below the root zone as the residual in a mass balance equation (Equation (1)):
where
R, P, I. SNmelt, DRin, DRout, ETsm, and ∆
S correspond to recharge, gross precipitation, interception, snowmelt, direct runoff into the grid cell from upslope grid cells, direct runoff out of the grid cell, soil moisture evapotranspiration (
ET), and change in soil moisture, respectively. The term for soil moisture
ET,
ETsm, is used to account for soil moisture evaporative losses and plant transpiration [
43]. Thus, total
ET,
ETtot, may be computed as interception,
I, plus soil moisture
ET (
ETtot =
I +
ETsm).
Recharge from irrigation and/or cloud water interception (i.e., “fog drip”) are not accounted for in the model as these parameters are often difficult to constrain [
49,
50]. In this study, the grid cell dimension was set to 30 m × 30 m. Gross precipitation was estimated using a natural neighbor algorithm and daily measurements obtained from the Desert Research Institute from 13 climate stations on Catalina and one climate station on nearby San Nicholas Island for the time period 1 January 2008 to 31 December 2014. The lack of concurrent data over the entire island prior to 2008 and from March 2015 onward (
Table 1) prevented the analysis of recharge over longer time frames. Natural neighbor was chosen over inverse distance weighting and spline algorithms because it presented a smoother output, gave highest
R2 values in linear regression analysis, and produced lower root mean square errors (RMSEs;
Figure 2). Geostatistical approaches, such as kriging, were not pursued in this study to eliminate the need for daily variogram. Linear regression analysis was performed to fill in days where climate data was not recorded.
In SWB, gross precipitation (
P) must exceed maximum assigned interception
(I) amounts before the model assumes that net precipitation (
Pnet) has reached the soil surface. The authors are not aware of measurements of interception losses on Catalina. Therefore, target interception loss rates for the different types of land cover found on Catalina for both growing season and nongrowing seasons were approximated based on results from comparable settings elsewhere [
27,
40,
51,
52,
53,
54]. These rates are listed in
Table 2. This study relied on the NLCD classification of land cover and assumed there is a dominance of chaparral and coastal scrub over herbaceous grassland/weed vegetation. The alternate scenario of grassland dominance, as highlighted in field studies by, e.g., Minnich [
38,
39], was tested for in the subsequent sensitivity analysis. The assigned growing season lengths (GSLs) of 1 November to 1 June for the chaparral/coastal shrub and from 1 November until 1 May for the herbaceous grasses are based on reported results from studies conducted elsewhere in California [
37,
55,
56].
A flow direction grid was created using the ArcGIS D8 algorithm performed on a digital elevation model (DEM) to simulate runoff. SWB iteratively routes runoff downslope (
DRout) to an adjoining cell, where it may be added as a potential infiltration source (
DRin), or continues to be redirected until all runoff is infiltrated and/or reaches the boundary of the study area and is removed from the process. This approach is considered an improvement over more traditional water balance approaches where rainfall is considered the sole recharge source [
57,
58]. Direct runoff was estimated in SWB using the curve number method for the 13 land cover classes and four hydrologic soil groups (HSGs) mapped on Catalina. The HSG input (
Figure 3a) stems from the Soil Survey Geographic Database (SSURGO) obtained from the USDA NRCS Geospatial Data Gateway. Assigned curve numbers were based on published values by Hjelmfeld et al. [
59] and Westenbroek et al. [
27] and are listed in
Table 3.
Maximum infiltration rates (MIRs) are used in SWB to specify a maximum daily recharge rate for each of the four HSGs (e.g., [
40]). In this study, MIRs were estimated from the range of saturated hydraulic conductivity (
Ksat) data reported by the USDA NRCS Soil Survey of Santa Catalina Island (
Figure 3b). First, the
Ksat of each soil profile of a map unit, such as “Tongva”, “Freeboard”, and “Starbright” (
Table 4) found in map unit “156” (see, e.g.,
Figure 3b), was calculated as the harmonic mean using layer-specific
Ksat and depth data (e.g.,
Table 4). Topsoil layers with
Ksat ≥ 42 µm/sec were excluded from this analysis as they were assumed to represent leaf litter associated with interception rather than infiltration in the soil zone. Next, the
Ksat values of the map units (e.g., “156”) were averaged based on the profile-percentage makeup within each complex. Map unit
Ksat values were then converted to MIRs of HSGs by weight averaging them according to individual HSG areas (
Table 5).
SWB allows for the application of five separate potential evapotranspiration (
ETpot) estimation methods [
48,
60,
61,
62,
63]. In this study, the Hargreaves–Samani (H–S) [
61] method was chosen because it was developed with, and tested against, datasets obtained from coastal Californian regions (e.g., the town of Lompoc in Santa Barbara county) that are similar in climate and vegetation to those studied herein. Furthermore, the method is the only available in SWB that is capable of producing spatially distributed output grids rather than just one uniform value to be applied over the entire island. The application of the H–S method requires gridded data of maximum and minimum daily temperature (
Tmax,
Tmin), which were generated following the approach of Mair et al. [
43] and Hagedorn et al. [
64] using temperature lapse rates applied to a 30-m DEM. Temperature at any grid cell of the DEM was extrapolated from daily temperature data recorded at the Parsons Landing (PL) climate station. PL was selected as the reference station for
T extrapolation because its records contained the least amount of missing data between the years 2008 and 2014.
ETsm was computed from
ETpot for each grid cell as follows: (1) when
Pnet −
ETpot ≥ 0, then
ETsm =
ETpot; (2) when
Pnet −
ETpot < 0, then
ETsm equates to only the amount of water that can be extracted from the soil via
ET, a value computed via the soil moisture retention tables of Thornthwaite and Mather [
48] and modified by Westenbroek et al. [
27]. Estimates of maximum soil moisture storage capacity needed to use the soil moisture retention tables were computed as the product of the available water soil capacity (AWC) multiplied by the root zone depth. AWC data (
Figure 3c) were obtained from the USDA NRCS Web Soil Survey (WSS). Root zone depths (
Figure 3d;
Table 5) were assigned by accessing “restrictive layer” soil depth data from USDA NRCS WSS and weight-averaging those across both land cover and hydrologic soil types. No depth values were obtained for “Developed, High Intensity Urban” (24) HSG A, “Woody Wetlands” (90) HSG B-D, and for “Emergent Herbaceous Wetlands” (95) HSG B-C. In those instances, reference values reported by Westenbroek et al. [
27] were applied.
An initial amount of soil moisture (SM) is needed in SWB to allow for the potential of soil saturation and subsequent infiltration/recharge or evapotranspiration experienced on day 1 of the study period. Two methods may establish the initial SM. The first requires an extra year of data to prime SM for the following (initial) year. This “primer” year may be incorporated into the statistical analysis, albeit at a limited initial accuracy. The second and easier to apply method is to utilize the control file input and assign an estimated initial percentage of AWC. To determine the appropriate initial percentage of AWC, a “phantom” year of 2007 climate data was generated using the National Oceanic and Atmospheric Administration (NOAA)’s Climate at a Glance portal (
https://www.ncdc.noaa.gov/cag/). It was determined that the year 2013 was most similar to 2007 for precipitation amounts, while 2009 most resembled maximum and minimum temperature for 2007. Using these datasets and including the phantom year 2007 in the analysis yielded recharge results that closely resemble those obtained from an initial AWC percentage of 80%. Therefore, the assigned initial AWC of 80% was applied, and 2008 was included in the discussion of recharge results.
All the input rasters contained some minor areas of missing data, particularly along the steep and undeveloped southern shoreline of the island (shown in black/grey in
Figure 3). The reasons for that lack of data are not clear. All areas of missing data exhibited by any of the input rasters were excluded from the recharge analysis.
2.3. Corroboration of Recharge Estimates
To produce reliable results, it is critical that modeled recharge values are calibrated against independent estimates of recharge and/or measured water balance parameter data. Previous soil water balance studies have relied on various datasets for calibration, including measurements of runoff [
43], water table fluctuations [
65], or groundwater age tracers [
66]. One issue pertaining to Catalina Island, however, is that such datasets are not available. None of the streams on the island are instrumented, and none of the groundwater wells were accessible for sampling for this study. Groundwater level data were also not publicly available, neither from the utility company serving the island nor from resources such as the State Water Board Groundwater Ambient Monitoring and Assessment (GAMA) database, the State Department of Water Resources Groundwater Information Center, or the U.S. Geological Survey (USGS) Groundwater Watch portal. Nevertheless, there were some chemical data available from the GAMA portal that were used to corroborate SWB recharge estimates presented herein.
Groundwater age dating tracer data, specifically concentrations of tritium (
3H), were available for only two of Catalina’s wells: well 12 sampled in 2014 and well 30 sampled in 2004 (see
Figure 4a for well locations). Both wells exhibited
3H levels below analytical detection limits, although the detection limits for the individual analyses were not explicitly stated. Nevertheless, measurements at or below values of 0.2 tritium units (TUs) should be considered indicative of “old” (i.e., >60 year old) water typically encountered in low-recharge areas [
20,
67]. The available data on CFC-11 and CFC-12 concentrations from the GAMA database (all below detection limits) cannot be used for groundwater age dating or recharge analysis because they all reflect detection limits (5 μg/L for CFC-11 and 1 μg/L for CFC-12) that are too high for a reliable comparison with atmospheric input since the mid-20th century [
68]. Additional age dating data, such as measurements of
14C,
85Kr, SF
6, CFC-13, and/or CFC-113, were not available in any of the examined databases. Reasonable estimates of groundwater transit times derived from Darcy’s law could also not be obtained due to the lack of hydraulic head and conductivity data.
More extensive data exist on Catalina’s dissolved Cl concentrations of groundwater (
Figure 4), which allow for a more in-depth assessment of recharge at wells using the traditional chloride mass balance (CMB) method:
Here,
P,
ClP, and
ClGW correspond to mean rainfall, the precipitation weighted mean Cl concentration of precipitation (including dry deposition), and the Cl concentration of groundwater at a particular well, respectively. The method assumes the following: there is no direct runoff; all of the
ClGW is derived from evapotranspiration of atmospheric water; Cl is an inert tracer; flow is one-dimensional, vertical downward, piston type; groundwater is well-mixed; and water and tracer mass fluxes are steady.
P values for Equation (2) (316–357 mm/year) were estimated for each well from the Parameter–Elevation Relationships on Independent Slopes Model [
69] and represent 30-year means (1981–2010). Given that residence times of island groundwater can be on the scale of decades [
70], we found this to be a better approach than using the
P grids estimated herein that extend back only until 2008. No
ClP data were available for Catalina, but for the CA85 Channel Islands National Park station on Anacapa Island for the years 1980 through 1982 [
71]. The reported annual Cl deposition for CA85 (located about 120 km NW of Avalon) ranged from 3.48 to 6.90 kg/ha and translates, using measured rainfall rates, to an equivalent
ClP range of 3.06 to 13.6 mg/L. These values exceed those estimated for other regions of Southern California, such as 2.33 mg/L in La Conchita Ranch, Ventura County [
72], or 2.60 mg/L in the Simi Hills of Ventura County [
73], and likely reflect the proximity of the climate station to the ocean, as has been observed in coastal settings elsewhere [
74]. To represent uncertainty in the measured data, both
ClP end-members for Anacapa Island were applied in Equation (2). Assigning
ClGW is more challenging because sample size and value ranges differ greatly among wells (
Figure 4b). There are also limited concurrent datasets available because many of the supply wells (i.e., wells 02, 07, 09, 13, 19, and 22) were abandoned in the late 1980s/early 1990s due to high groundwater salinity. Also problematic is the fact that the supply well cluster 16, 18, 30, and 31 is located only about 300–400 m downgradient of Thompson Reservoir dam and, as such, draws primarily from impounded surface water [
47]. This scenario violates the assumption of Equation (2) that all groundwater Cl is rainfall and not rainfall + surface water-derived [
18]. In the absence of reliable Cl and flow information of the reservoir water that discharges to the wells, data from this well cluster was not included in the analysis. Many previous studies have established
ClGW as the arithmetic or geometric mean of measured groundwater data at a well [
73,
75,
76], but this approach was not considered useful for wells with a small sample size and Cl values that varied by more than 100%. As an alternative in this study,
ClGW was based on the availability of concurrent datasets. For each of the years 1992, 1995, 1998, 2001, 2004, 2010, 2011, 2012, and 2015, the GAMA database revealed ≥6 concurrent and spatially distributed groundwater Cl data points. Accordingly, these datasets were used for a multiyear CMB analysis.
The mean SWB recharge value for each well for which CMB data was available was calculated following the approach of Johnson and Belitz [
77] as the mean of recharge within a 500-m buffer around the well location. Uncertainty in the SWB recharge estimate around each well was represented by the standard deviation of values of all cells within the buffer.
CMB recharge estimates reveal a statistically significant positive correlation (Pearson
r ≥ 0.5;
P < 0.1) with SWB recharge estimates for all but one of the concurrent datasets (
Figure 5). Even though this may be interpreted as some indication for corroboration of spatial recharge patterns obtained by the SWB model, the limiting assumptions inherent in both recharge estimation methods, uncertainty of input data, and the different temporal scales at which these methods apply indicate that a comparison of the different recharge estimates should be treated with caution. It is, however, interesting that the mean SWB recharge values most resemble the upper limit (i.e., high
ClP) CMB recharge estimates (
Figure 5). Assuming the applied high
ClP end-member to reflect a positive outlier, a reasonable assumption based on the lower reported
ClP values for other sites in coastal California (e.g., Reference [
73]), the depicted trends in
Figure 5 suggest either an overestimate of recharge by the SWB model or an underestimate of recharge by the CMB method. A similar discrepancy between CMB and SWB results was also observed for a non-irrigated site elsewhere in Southern California [
72] and was attributed to halite dissolution in the subsurface and the resulting potential underestimation of actual chloride inputs to groundwater. Additionally, intrusion by seawater could explain the large Cl range observed in well 12 as this well is located <500 m away from the shoreline (
Figure 4b). Leakage of trapped, connate water, a process that has been documented in nearby San Diego County [
78], could also explain high Cl and low CMB values. Additional data, particularly concentrations of dissolved Br and B, as well as values of
87Sr/
86Sr,
3H/
3He,
14C, and δ
11B, have shown to be valuable salinity tracers in comparable semiarid settings [
78,
79,
80,
81] and are needed to further identify the water Cl sources on Catalina. Still, in the absence of any other potential recharge calibration data, we consider the CMB results, particularly the agreement of spatial trends with SWB estimates, not to contradict but to generally support the SWB model outputs.