Assessment of Managed Aquifer Recharge Sites Using GIS and Numerical Modeling

3 Assessment of managed aquifer recharge sites using GIS and numerical modeling

5 Tess A. Russo*1,2, Andrew T. Fisher1, Brian S. Lockwood3, Randall T. Hanson4 (?)

6 1Department of Earth and Planetary Sciences, University of California, Santa Cruz, CA

7 2Columbia Water Center, Earth Institute, Columbia University, NY

8 3Pajaro Valley Water Management Agency, Watsonville, CA

9 4United States Geological Survey, San Diego, CA

10 *[email protected], (347) 913-6835

1 13 Abstract 14 We completed a geographic information system (GIS) analysis to assess suitability for managed 15 aquifer recharge (MAR) using the Pajaro Valley Groundwater Basin, central coastal California 16 (PVGB), as a case study. With results from the GIS study, we used a groundwater model to 17 assess the hydrologic impact of potential MAR operating scenarios, illustrating how a 18 comprehensive analysis of MAR suitability can help with regional water supply planning. The 19 GIS analyses used topographic, land use, surficial geology, soil infiltration capacity, aquifer and 20 associated confining layer locations, properties, thicknesses, and historical changes in water 21 levels. A map of MAR site suitability and comparison with an existing project suggests that 22 about 7% (15 km2) of the basin may be highly suitable for MAR. Model results show simulated 23 MAR projects in locations identified as “highly suitable” for MAR reduce seawater intrusion 24 more than projects simulated in “unsuitable” locations, supporting the GIS analysis results. 25 Results from the model also illustrate the variability in seawater intrusion reduction and head 26 level changes throughout the basin and over time. Projects distributed throughout the PVGB 27 were more effective at reducing seawater intrusion than projects along the coast, at the time scale 28 of several decades. Collectively, these studies introduce a novel data integration method and help 29 to evaluate management options for improving long-term groundwater conditions throughout the 30 PVGB.

32 1. Introduction

33 Groundwater overdraft can occur when the sum of extraction and other aquifer outputs exceeds 34 the sum of inputs over the long term. Groundwater overdraft can lead to numerous undesirable 35 consequences, including increased pumping costs associated with lowering of aquifer water 36 levels; drying of streams, lakes and wetlands; land subsidence and an associated loss of storage 37 capacity; and seawater intrusion. As water demand rises, unsustainable groundwater use is 38 expected to continue to increase around the world, especially in developing nations, making 39 groundwater management increasingly important (Foster and Chilton 2003; Giordano 2009; 40 Konikow and Kendy 2005; Rosegrant and Cai 2009). 41 Strategies for mitigating groundwater overdraft include limiting extraction (water 42 conservation, pumping moratoriums), and enhanced aquifer inputs. The latter can be

2 43 accomplished with injection wells, aquifer storage and recovery (ASR, with injection and 44 extraction through the same wells), and managed aquifer recharge (MAR, using surface 45 infiltration systems). Enhanced recharge has become a common and frequently effective method 46 for water resource management (Dillon et al. 2009; Ma and Spalding 1997; Maliva and Missimer 47 2012). ASR can help to reduce overdraft, (Shammas 2007), but must be planned carefully, as it 48 can have high energy requirements and requires creation and maintenance of conveyance and 49 pumping systems (Bouwer 2002). In contrast, MAR can be part of a more passive operational 50 system, potentially involving less engineering and lower operating costs. Water is diverted to a 51 natural depression or constructed retention area, where it infiltrates into the subsurface over time. 52 MAR projects have also demonstrated improvements in water quality through denitrification 53 during the infiltration process (Fryar et al. 2000; Ma and Spalding 1997; Missimer et al. 2011; 54 Rauch-Williams et al. 2010; Schmidt et al. 2011). These improvements can be particularly 55 important for sites lacking reliable access to pristine surplus surface water supplies, such as 56 basins in which there is extensive agricultural development or widespread use of septic systems, 57 resulting in elevated nutrient levels. The primary disadvantages of MAR include relatively large 58 land area requirements, and the challenge in identifying locations with amenable surface and 59 subsurface conditions for infiltration to an unconfined aquifer. 60 Identifying areas suitable for MAR projects and estimating the influence of these projects 61 on groundwater levels and fluxes are challenging problems with numerous solutions. The first 62 step is to locate regions where surface water can infiltrate and flow to available space within an 63 aquifer. These assessments are often made on a regional basis, within which there may be limited 64 data on complex surface and subsurface geology. In addition, there is a need to determine how 65 the benefits of managed recharge could vary with project location, size, and operating conditions. 66 Some of these questions can be resolved through field testing, but computational tools can play 67 an important role in evaluating project scenarios and screening potential MAR sites, on the basis 68 of their broader hydrologic impact. 69 Studies of the spatial distribution of recharge have been developed and applied to assess 70 groundwater vulnerability to contamination. Aller et al. (1987) developed a method for 71 evaluating the potential for groundwater degradation, DRASTIC, which uses a relative ranking 72 system. The method combines multiple datasets related to groundwater infiltration, including net 73 recharge, aquifer and soil properties, and impact of infiltrating water through the vadose zone on

3 74 water quality. The parameters for each dataset are classified (generally to values 1 to 10), then 75 multiplied by a dataset weight (1 to 5). The product of value and weight for each area (or 76 subarea) are then summed for all the datasets, resulting in a relative vulnerability index for each 77 area in the study region. This method provides the basis for identifying recharge areas using 78 geographic information system (GIS) based integration. We follow the general structure of this 79 method, but use a different ensemble of datasets, different classification and integration 80 techniques, and different parameter weights. 81 Many hydrologic applications, including identification of locations for potential MAR 82 projects, are well suited for GIS analysis (Jha et al. 2007). Several studies have used GIS-based 83 integration of spatial coverages pertinent to groundwater recharge, with various data values 84 being classified and weighted before combining (Adham et al. 2010; Chenini et al. 2010; 85 Chitsazan and Akhtari 2009; Jasrotia et al. 2007; Murray and Mcdaniel 2003; Piscopo 2001; 86 Saraf and Choudhury 1998; Shankar and Mohan 1998; Yeh et al. 2009). Methods used for 87 classification and weighting generally differ from study to study, due to variations in data 88 availability, local geology, and perceived level of dataset importance to groundwater recharge. 89 Chowdhury et al. (2010) polled a group of geologists and hydrogeologists to determine a 90 weighting system for their GIS-based recharge location assessment, and found that half the group 91 thought equal weighting was appropriate while the other half agreed on a variable weighting 92 method. As a practical matter, all classification schemes of this kind are somewhat arbitrary, but 93 initial approaches and values can be refined over time as new data becomes available and 94 individual recharge projects are tested and implemented. Some studies have used GIS with a 95 multi-criteria decision analysis that accounts for local preferences, and attempt to reduce the 96 arbitrary nature of weight assignment by using an analytical hierarchy process (Chowdhury et al. 97 2010; Rahman et al. 2012). Though more rigorous, this decision analysis method still requires 98 (largely heuristic) estimation of the relative importance of each parameter. 99 Numerical modeling can also help to identify sites amenable for MAR, and be used for 100 estimating the potential benefits of MAR projects on regional hydrologic conditions during a 101 range of future climatic, water use, and management scenarios(Munevar and Marino 1999). 102 Groundwater models may be combined with an optimization algorithm to test water management 103 strategies, including artificial recharge (Abarca et al. 2006). These models tend to use simple 104 governing equations and highly generalized aquifer properties. Another option for reducing

4 105 computation time is to employ an ensemble of analytical models based on simplified lumped 106 parameters (Smith and Pollock 2012). Combination of the GIS-based integration methods with 107 numerical modeling allows a more detailed and quantitative assessment of MAR opportunities 108 and impacts, and takes advantage of overlapping data requirements for GIS and numerical 109 modeling studies (for example, aquifer geometry and depth and soil properties) (Chenini and 110 Mammou 2010). The use of numerical modeling as a follow-up to a GIS-based study also 111 provides an opportunity to assess the MAR location suitability analysis developed using GIS, 112 and on time requirements to see specific improvements to resource conditions. 113 True assessment of MAR location suitability requires field testing to determine how 114 project placement influences local and regional hydrologic conditions. Ultimately this requires 115 field-scale implementation of MAR projects, but budgetary and time constraints generally limit 116 opportunities for large-scale installations purely for testing purposes. Thus, numerical modeling 117 has an important role to play in pre-implementation evaluation of project options, based on a 118 MAR suitability analysis, helping to reduce the number of choices made in selecting appropriate 119 management strategies. Similarly, evaluation of actual hydrologic responses to implementation 120 of MAR projects can be used to "validate" individual and ensemble groups of groundwater 121 models, contributing to a better understanding of system function and improved basin-wide 122 management of scarce resources. 123 The present study combines GIS and numerical analyses to address the following 124 questions, as applied to the Pajaro Valley Groundwater Basin (PVGB), central coastal California 125 (Figure 1): 1) How should surface and subsurface information datasets be combined to assess 126 MAR site suitability? 2) How does MAR suitability vary within the basin? 3) How might 127 hypothetical MAR operating scenarios influence groundwater conditions in the basin over the 34 128 year model simulation? This project limits analysis to MAR options for the PVGB rather than 129 exploring a more comprehensive assessment of basin management options and anticipated 130 changes to water usage. An extensive technical and public process is currently underway in the 131 PVGB to evaluate a wide range of supply and conservation options, and develop a new basin 132 management plan, in an effort to improve groundwater conditions in the Pajaro Valley in coming 133 decades. We limit analyses in this study to assessing the spatial distribution of MAR suitability 134 and potential hydrologic impacts of several MAR options. 135

5 136 2. Study area 137 The PVGB (Figure 1) underlies a 322 km2 area managed by the Pajaro Valley Water 138 Management Agency (PVWMA). The region relies almost entirely on groundwater to satisfy 139 agricultural and municipal/domestic needs (83% and 17%, respectively). Precipitation in the area 140 has averaged ~560 mm/yr in recent decades, ranging from 480 mm/yr in the south to 1270 141 mm/yr in the northern high elevation area, with considerable year-to-year variability. 142 Precipitation is highly seasonal, with a majority falling between December and April, resulting in 143 distinct dry and wet seasons. Historic changes in precipitation intensity and air temperature in 144 central coastal California (Dettinger 2005; Russo et al. 2013), suggest that natural recharge rates 145 and crop water requirements are not stable and may contribute to groundwater stress (Gleeson et 146 al. 2010; Taylor et al. 2013). 147 The PVGB is a coastal basin, bounded to the west by Monterey Bay and to the east by the 148 San Andreas Fault. Northern and southern boundaries are political rather than (hydro)geologic, 149 with key aquifer units extending beyond the area managed by the PVWMA. Much of the PVGB 150 corresponds to the lower drainage basin of the Pajaro River, which flows into the valley from the 151 east at an average discharge of 1.3 x 108 m3/yr (USGS Gage #11159000), after draining an 152 upstream area of 3.1 x 103 km2. Corralitos Creek, a tributary of the Pajaro River having a 153 drainage area contained entirely within the PVGB, contributes an additional 1.4 x 107 m3/yr of 154 discharge (USGS Gage #11159200), and the Watsonville Sloughs also drain into the lower 155 Pajaro River before it discharges into Monterey Bay (Figure 1). 156 The PVGB has six hydrogeologic units that comprise aquifer and confining layers: 157 alluvium, alluvial clay, the Upper Aromas Aquifer, the Aromas confining unit, the Lower 158 Aromas Aquifer, and the Purisima Aquifer (Dupre 1990; Hanson 2003; Muir 1972). These layers 159 are herein referred to as A1, C1, A2, C2, A3 and A4, respectively, where A signifies an aquifer 160 and C signifies a confining unit (Table 1). The six layers are underlain by hydrogeologic 161 basement rocks consisting of granite and Oligocene-aged deposits. 162 Groundwater extraction from the basin currently averages 6.2 x 107 m3/yr, with the 163 majority of water pumped from Layers A1 and A2 (Hanson et al. 2013; PVWMA 2013). Over 164 time, total groundwater outflows (including extraction) have exceeded recharge and inflow rates; 165 the current estimated overdraft in the PVGB is approximately 1.5 x 107 m3/yr (Hanson et al. 166 2013). This annual overdraft is approximately equivalent to 24% of annual pumpage and 10% of

6 167 precipitation falling on the PVGB. Long-term monitoring in recent decades indicates that water 168 levels within more than half of the PVGB are below sea level, particularly during the dry part of 169 the water year, with the greatest depression of water levels below the City of Watsonville 170 (PVWMA 2013). Due to chronic over-extraction, a zone of seawater intrusion extends up to 5 171 km inland and is advancing at ~80 m/yr along much of the coastal edge of the basin (Hanson et 172 al. 2008; Hanson 2003; Wallace and Lockwood 2009) (Figure 1).

173 Table 1. Model layer IDs and geologic information Layer ID Layer Name Thickness1 (m) Aquifer lithology2 A1 Alluvial aquifer 0 to 116 Unconsolidated, moderately sorted silt, sand, and gravel with discontinuous lenses of clay and silty clay C1 Alluvial clay 0 to 16 -- A2 Upper Aromas aquifer 0 to 153 Sequence of eolian and fluvial sand, silt, clay and gravel C2 Aromas clay 0 to 35 -- A3 Lower Aromas aquifer 0 to 319 Semiconsolidated, fine-grained, oxidized sand and silt A4 Purisima aquifer 0 to 500 Thick bedded tuffaceous and diatomaceous siltstone with interbeds of fine- grained sandstone 174 1Layer thickness obtained from the Pajaro Valley Hydrologic Model (Hanson et al. 2013) 175 2Aquifer lithology summarized from USGS geologic maps (Brabb et al. 1997; Clark et al. 1997) 176 177 In 199X, the PVWMA constructed the Harkins Slough MAR system which is permitted 178 to divert up to 2.5 x 106 m3/yr (~2,000 ac-ft/yr) from near the confluence of Harkins and 179 Watsonville sloughs when flows and water quality are sufficiently high. Diverted water passes 180 through a sand pack filter and is pumped through a 1.5 km pipeline to a 7 acre infiltration pond. 181 Some of this percolated water is subsequently recovered and blended with other water supplies, 182 then delivered using a coastal delivery pipeline to local farms and ranches. In addition, the

7 183 PVWMA and the City of Watsonville jointly developed and operate a water recycling plant that 184 contributes water to the coastal delivery pipeline, as do inland groundwater wells, allowing 185 project water to be blended to achieve quality and supply goals. 186 The PVWMA is currently working with regional stakeholders to update their basin 187 management plan in an effort to bring the basin back into hydrologic balance (PVWMA 2013). 188 Several projects have been implemented, including the MAR system and recycled water plant, 189 and additional projects are planned, including enhanced water conservation, surface storage, and 190 MAR efforts. The overall goal is to develop about 1.5 x 107 m3/yr of demand reduction and 191 additional supply. Local stakeholders are exploring options for development of distributed 192 (smaller-scale) MAR projects that could benefit from capture of stormwater flows during the 193 rainy seasons. Collectively these options raise questions about how best to identify MAR project 194 locations, and operate and maintain these projects for benefit of basin as a whole. 195 196 3. Methods 197 3.1. GIS analysis 198 We used GIS for data management, manipulation, and analysis of eleven surface and subsurface 199 data sets to generate a basin-wide map of "MAR suitability". As defined for this study, high 200 MAR suitability indicates that, if a water supply of sufficient quantity and quality is available, 201 surface and subsurface conditions could be favorable to developing one or more MAR projects. 202 Surface and subsurface property data sets were initially analyzed separately, and then were 203 combined to produce a final map. For surface analyses, primary data included: (1) surficial 204 geology, (2) soil infiltration capacity, (3) land use, (4) elevation (topographic slope), and (5) 205 verified (measured) infiltration and recharge rates from observational studies. For subsurface 206 analyses, primary data included: (6) aquifer thickness, (7) aquifer hydraulic conductivity, (8) 207 confining layer thickness, (9) aquifer storativity, (10) vadose zone thickness, and (11) historical 208 changes in water table height. 209 Surficial geology data were obtained from 1:62,500-scale geologic maps of Santa Cruz 210 and Monterey Counties (Brabb et al. 1997; Clark et al. 1997). Lithologic descriptions were used 211 to classify individual geologic units in terms of whether or not they corresponded to PVGB 212 aquifers, or if fine-grained sediment (clay and silt) would be likely to reduce direct connection to 213 underlying aquifers. Higher MAR suitability is associated with outcropping aquifers units. Soil

8 214 infiltration capacity data were obtained from the Natural Resources Conservation Service 215 SSURGO database (NRCS 2010a; NRCS 2010b). Infiltration capacity of basin soils was mapped 216 in irregular polygons having values ranging from 0.2 to 12 m/d. Land use classifications were 217 developed by the PVWMA and the USGS in collaboration with regional stakeholders, based on 218 field visits and parcel-specific reports of crops grown, as part of a broader effort to develop a 219 regional hydrogeologic framework and groundwater model (Hanson 2003; Hydrometrics 2012; 220 Hanson et al. 2013). Land use classifications include native vegetation, urban, and agricultural 221 areas designated by crop type or presence of a nursery. Land surface slope values were 222 calculated from the 10-m resolution USGS National Elevation Dataset (ned.usgs.gov). Locations 223 of measured seepage rates along losing sections of the Pajaro River were reported in earlier 224 studies based on differential gauging and streambed geothermometry (Hatch et al. 2010; Ruehl et 225 al. 2006). 226 Subsurface data sets were prepared initially during development of a regional 227 hydrogeologic model (Hanson et al. 2013; Hanson 2003), and modified as needed for use in our 228 GIS-based analysis. Aquifer properties, including layer thicknesses, hydraulic conductivity, and 229 storativity, were assembled using data from >300 well logs distributed throughout the basin, and 230 compiled on a grid having horizontal resolution of 250 x 250 m and variable cell thickness. The 231 present unsaturated zone thickness was calculated by subtracting the interpolated water table 232 elevations, using data collected in 2010, from the ground elevation. Water levels in the basin 233 were compared using 1998 and 2010 data sets, to quantify decadal changes in groundwater 234 levels. There were sparse water level data from earlier times, in many cases indicating greater 235 absolute changes in water levels, but 1998 was the earliest year for which data covered a 236 majority of the study area. Nevertheless, this data set had the smallest spatial coverage of all data 237 sets used in this study, and thus defined the spatial extent of the final MAR suitability map. 238 The common approach for combining GIS data sets such as these requires reclassifying 239 relevant datasets to a common scale (e.g., relative values of 1 to 5) and then assigning a weight to 240 each dataset in proportion to its perceived importance for the condition or process being 241 evaluated. For each grid cell in the analysis, an index is calculated by summing the products of 242 value and weight for each dataset: 243

244 (1)

246 where n is the total number of data sets, vi is the classified value for data set i at location (x,y),

247 and wi is the weight assigned to data set i. We defined a weighting scheme for use in the present 248 study based on (a) a review of published recharge mapping studies that used a similar GIS-based 249 approach, (b) consideration of available data sets for the PVGB, and (c) inferences as to how 250 groundwater recharge in this region might be influenced by coexisting factors (Figure 2). 251 Our approach differs in several respects from methods applied in earlier GIS-based 252 studies of natural recharge and potential for development of MAR projects. Most significantly, 253 we evaluated surface data (sets 1 to 4) and subsurface data (sets 6 to 11) independently, with the 254 former indicating the ease of surface water infiltration, and the latter indicating the ease of 255 subsurface transport and extent of available storage. In addition, rather than simply combining all 256 available datasets as independent indicators through a process of weighted summation (as with 257 equation 1), we used some data sets as modifiers for other data sets before combining individual 258 coverages to derive a final assessment of MAR suitability (described below). Finally, locations 259 for which there were direct measurements of recharge rates (set 5) were subsequently assigned 260 MAR suitability values based entirely on field observations, which are considered to be the most 261 reliable of available data types. 262 263 3.1.1. Data classification 264 We standardized several of the datasets by classifying values or properties on a scale of 1 to 5, 265 where 1 represents an unfavorable attribute for MAR suitability, and 5 represents a favorable 266 attribute. Both numerical and non-numerical datasets (e.g., soil infiltration capacity and surficial 267 geology, respectively) were used in this study, requiring different methods for classification 268 before data could be combined. We used three approaches for classifying numerical datasets: (1) 269 classify values based on knowledge of field properties and past MAR operations, (2) classify 270 values using the Jenks optimization method based on the distribution of property values (Jenks 271 1967), and (3) operate on raw data. The first method was applied to soil infiltration capacity and 272 locations with stream seepage rates measured in the field (Table 2). The second method uses the 273 ArcMap modification of the Fisher-Jenks algorithm, and was applied to specific yield, 274 unsaturated zone thickness, and historical changes in water table height. The third method was 275 applied to surface slope values. Non-numerical datasets, including surficial geology and land

10 276 use, were classified based on interpretation of associated properties that could influence MAR. 277 For surficial geology, we assigned each lithologic unit a value based on whether the mapped 278 lithology and texture (Brabb et al. 1997; Clark et al. 1997) corresponded to a known aquifer or 279 would likely be connected to a known aquifer. This reduced 54 and 85 (for Santa Cruz and 280 Monterey Counties, respectively) lithologic categories to three possible connection classes 281 (Table 2). For land use, we classified descriptions based on associated roughness coefficient 282 values (Chow 1959) (Table 2), where roughness coefficients range from 14 to 100, for 283 nursery/pavement to forested/native vegetation. Roughness coefficients were assigned to 284 agricultural areas according to crops grown in rows, fields, or pasture. 285 286

11 287 Table 2. Classification of data based on physical properties

Soil infiltration Stream seepage Aquifer storage Surficial geology Land use capacity Rate Value Rate Value1 Value Connection to Value Description Roughness [m/d] [m/d] [m] aquifer coefficient2 >3 5 >1 80 64.01-131 5 Good 5 Forest/ 100 Nat. veg. 1.2 4 0.2 to 1 60 40.01-64 4 Moderate 3 Pasture 40 0.6 3 22.01-40 3 Poor 1 Field crop 38 0.2 2 8.01-22 2 Row crop 35 0 1 0-8 1 Fallow 30 Turf 27 Pavement 14 288 1Stream seepage rates were determined from direct observations and assigned values that 289 represent highly suitable locations for MAR. For locations where L is measured, the MAR 290 suitability index = L (equation 7) 291 2Roughness coefficients modified from Chow [1959] are used in equation 2 292 293 3.1.2. Data integration 294 Earlier studies of recharge potential have treated infiltration capacity, slope, and/or land use as 295 independent variables (e.g., Jasrotia et al. 2007; Yeh et al. 2009). We reasoned that the primary

296 influence of slope and land use should be to modify soil infiltration capacity (IC), and developed 297 an equation to facilitate this approach. The equation incorporates a dependence on land slope (s) 298 and roughness (n), similar to those used with the Manning equation for calculating mean runoff

299 velocity in open channels, to generate an effective infiltration capacity (IE): 300 301 (2) 302

303 where IC is infiltration capacity (based only on soil type), n is a surface roughness coefficient 304 (with values ranging from 14 to 100, based on land use classificaiton), and s is slope in radians 305 (extracted from the regional digital elevation model). The second term in Eqn. 2 is intended to 306 be proportional to the quantity of water that will not infiltrate. Because the product of square- 307 root-slope and surface roughness is normalized by the maximum (optimal) conditions for the 308 region, the second term is ≤ 0. Calculated effective infiltration values are thus dependent on the

12 309 soil infiltration capacity, but modified by the relative likeliness of water running off the site 310 rather than infiltrating, based on surface slope and roughness (Figure 3). For example, if the soil 311 infiltration capacity is low, the influence of low-slope and native vegetation (high roughness 312 coefficient), which might help to maintain good infiltration conditions through a permeable soil, 313 becomes negligible. Conversely, a soil having a high infiltration capacity located in a region with

314 high slope and a land use of turf grass (low roughness coefficient), will have an IE value that is

315 lower than the IC value associated with the land use alone. IE will equal IC for optimal surface and 316 slope conditions (the second term goes to 0 as the term in brackets goes to 1), but otherwise the

317 IC value will be reduced by the second term, so that IE ≤ IC. 318 Transmissivity (T) is in important subsurface parameter and can be difficult to estimate 319 across a large spatial area. For operating a MAR system, high transmissivity is necessary for 320 avoiding excessive mounding (which could lead to waterlogging of the root zone of crops or 321 surface flooding), and for allowing infiltrated water to flow to nearby recovery wells. The 322 primary constraints on transmissivity with respect to MAR are aquifer hydraulic conductivity (K) 323 and thickness (b) and the presence or absence of confining layers between the ground surface 324 and the underlying aquifer (three separate subsurface data coverages). To account for spatially 325 variable K and b and the presence of confining layers in the subsurface, we use the following

326 equation to calculate an effective transmissivity (TE) as it applies to MAR suitability: 327

328 (3)

329 for (4)

330 for (5) 331

332 where F1 and F2 are confining unit factors that affect the influence of underlying aquifer units. F1

333 and F2 scale linearly between 1 and 0 for confining unit thicknesses ranging between 1 and 10 m, 334 respectively. Thus, the transmissivities of multiple aquifer layers can be combined (at least in 335 part), using an arithmetic mean rule, if confining layers between separate aquifer layers are <10 336 m in thickness. The vertical integration accounts for noncontinuity of thin confining layers that 337 were readily apparent in hundreds of well logs and drilling records from across the basin.

338 Calculated TE values were subsequently classified on a scale of 1 to 5 per method 2, described in 339 §3.1.1.

13 340 Available storage space (V) was assessed by multiplying aquifer specific yield (Sy) by the

341 unsaturated (vadose zone) thickness (Tu) of each cell: . Specific storage is likely to be negligible

342 in comparison to Sy, especially with respect to MAR surface infiltration, which was the focus of 343 this study. MAR suitability was additionally enhanced in areas where there has been a large drop 344 in water levels during the period of 1998 to 2010. 345 Following calculations and classifications, each dataset was assigned a weight based on 346 the perceived importance of individual properties and conditions to positioning of potential MAR 347 projects. The normalized weights used in this study are comparable to those obtained from a 348 review of similar peer-reviewed studies (Figure 2), although there is considerable variability 349 between studies depending on the number and type of available datasets and local hydrogeology 350 for each study. We note that in all of the earlier studies shown for comparison, individual 351 parameters were added as independent variables on a cell by cell basis. As described earlier, we 352 used land use and topographic slope data to modify the MAR suitability implied by soil property 353 data sets, rather than applying land use and slope data independently. Values shown for these 354 parameters in Figure 2 are the means of weights applied in the present study when calculating 355 effective infiltration (Equation 2). In general, the relative weight of each data set is lower in the 356 present study than other studies, but this is mainly because we used more data sets than most 357 other studies, distributing the assessment of MAR suitability across more coverages. 358 A final map of MAR suitability was created by summing the weighted, classified values 359 (all varying from 1 to 5, from least to most suitable for MAR) for every 10-by-10 m grid cell in 360 the basin for which all data sets existed: 361 362 (6)

363 If L exists, MAR suitability index = L (7)

364 365 where: G is surficial geology, D is an historic change in water table height, and L is the index for 366 a losing stream reach within which recharge rates have been measured and indicate high MAR

14 367 suitability. The weights applied to the various indices are somewhat arbitrary, but generally align 368 with the perceived importance of each data set from other studies (Figure 2). We reasoned that 369 effective infiltration properties and the volume of storage space should be weighted most 370 strongly at 5, with formation transmissivity and outcropping of primary aquifers weight at 4. The 371 historic change in water level was given the least weight (2) because of the uncertainty 372 associated with interpolating from only a few measurements. The full process was constructed 373 using ArcGIS ModelBuilder, which can be modified as additional datasets become available, 374 field data are collected to test GIS predictions, or weighting methods are changed based on 375 availability of new information. 376 377 3.2. Numerical modeling of MAR scenarios 378 To model the relative hydrologic impact of hypothetical MAR projects, and the importance of 379 project placement and operational parameters, we modified a hydrogeologic model developed to 380 assess a range of conditions and management options for the Pajaro Valley. The details of 381 developing a hydrogeologic framework for the Pajaro Valley, and of creating and applying a 382 complex model for assessing historical groundwater extraction and conditions, are presented 383 elsewhere (Hanson, 2003; Hydrometrics, 2012, PVWMA 2013, and Hanson et al., 2013), and 384 summarized briefly herein. Surface and subsurface hydrologic processes were simulated using 385 MODFLOW-2005 (Harbaugh 2005). The model domain extends from the back of the basin 386 (bounded by the San Andreas Fault) to >10 km offshore (Figure 4A), with grid resolution of 250 387 x 250 m. The six model layers vary in thickness across the basin, corresponding to aquifer and 388 confining layer thicknesses (Figure 4B). The model includes nearly 1000 active production 389 (agricultural, municipal, domestic) groundwater wells, and uses the Farm Process (Hanson et al. 390 2010; Schmid and Hanson 2009) which modifies agricultural groundwater pumping rates during 391 the simulation based on changes in land-use, climate, and groundwater availability. The 392 simulations used in the present study represent 34 years (nominally conditions from 1976 to 393 2009) divided into 408 (monthly) stress periods, each having two time steps. 394 We worked with a Basecase simulation developed to represent a 34 year time period 395 beginning nominally in 2009 (Hydrometrics, 2012). Climate conditions for the Basecase 396 simulation were constructed to be a "mirror image" of climate during the preceding 34 years, and 397 land use in the simulation was fixed to be that in 2009. This approach allowed us to assess the

15 398 influence of MAR operating scenarios in the context of a historically realistic climatological 399 scenario. After this simulation was completed, we ran 31 additional simulations, each with a 400 different combination of hypothetical MAR projects adding water in different locations and at 401 different rates around the basin (Table 3). Differences in simulated water levels and the extent of 402 seawater intrusion, as compared to results from the Basecase model, are considered to be 403 simulated MAR "benefit."

404 Table 3. Description of MAR scenario model simulations

Total Total MAR MAR Inf. Rate Inf. Rate Active Water Water Run # Location (m3/d) (ac-ft/yr) Quantity (mo/yr) (m3/yr) (ac-ft/yr) 1 Coastal 505 50 5 4 3.1x105 250 2 Coastal 505 50 10 4 6.2x105 500 3 Coastal 169 50 10 12 6.2x105 500 4 Coastal 8085 800 5 4 4.9x106 4000 5 Coastal 8085 800 10 4 9.9x106 8000 6 Coastal 505 150 10 12 1.9x106 1500 7 Coastal 674 200 10 12 2.5x106 2000 8 Coastal 1347 400 10 12 4.9x106 4000 9 Coastal 2695 800 10 12 9.9x106 8000 10 Coastal 4043 1200 10 12 1.5x107 12000 11 Back-basin 505 50 5 4 3.1x105 250 12 Back-basin 505 50 10 4 6.2x105 500 13 Back-basin 8085 800 5 4 4.9x106 4000 14 Back-basin 8085 800 10 4 9.9x106 8000 15 Back-basin 2695 800 10 12 9.9x106 8000 16 Good 505 50 5 4 3.1x105 250 17 Good 505 50 10 4 6.2x105 500 18 Good 505 150 10 12 1.8x106 1500 19 Good 674 200 10 12 2.5x106 2000 20 Good 8085 800 5 4 4.9x106 4000 21 Good 2695 800 5 12 4.9x106 4000 22 Good 1347 400 10 12 4.9x106 4000 23 Good 8085 800 10 4 9.9x106 8000 24 Good 2695 800 10 12 9.9x106 8000 25 Good 4043 1200 10 12 1.5x107 12000 26 Poor 505 50 5 4 3.1x105 250 27 Poor 505 50 10 4 6.2x105 500 28 Poor 8088 800 5 4 4.9x106 4000 29 Poor 2695 800 5 12 4.9x106 4000

16 30 Poor 8085 800 10 4 9.9x106 8000 31 Poor 2695 800 10 12 9.9x106 8000 405 406 MAR projects were simulated by adding water to the surface aquifer layer using a head- 407 independent boundary condition. It was assumed that each MAR project existed within a single 408 model cell (6.3 hectares,15.6 acres). Adding water directly to the subsurface did not allow 409 evaluation of how surface properties (slope, land use, and soil infiltration capacity) influenced 410 recharge dynamics, but subsurface storativity, transmissivity and the presence of confining units 411 did govern flow after infiltration. 412 MAR project scenarios had four variables: (1) project locations, (2) number of projects, 413 (3) quantity of applied water per project (and in total), and (4) duration of activity during each 414 year. We evaluated the influence of locating MAR projects in four general regions: coastal area 415 (“Coastal”), back (eastern side) of the basin (“Back-basin”), areas identified as being particularly 416 suitable for MAR (“Good”) and areas identified as being considerably less suitable for MAR 417 (“Poor”). We expected that the MAR project distance from the coast would have a significant 418 influence on seawater intrusion, so locations for good and poor simulations were selected in pairs 419 such that the sites in each pair were equidistant from the coast. MAR sites in each location group 420 recharge to different layers, depending on which aquifer is exposed at the surface in each 421 location. For example, sites used for MAR in the back of the basin (simulation group Back- 422 basin) recharge directly to aquifer layer A4 (Purisima Formation), whereas sites used for MAR 423 projects based on the most suitable conditions (simulation group Good) are located over a mix of 424 aquifer layers A1, A2 and A4. 425 Each modeling scenario had either 5 or 10 MAR projects. The rate of MAR-associated 426 recharge applied at individual project sites ranged from 6.2 x 104 m3/yr (50 ac-ft/yr) to 1.5 x 106 427 m3/yr (1200 ac-ft/yr), comparable to the amount of water that might be applied based on 428 stormwater capture of runoff (near the lower end) or based on diversion from major aquatic 429 systems (near the higher end). Water was applied evenly during periods of either 4 or 12 430 months/yr. The 4-month MAR operation was intended to represent projects that run only during 431 the wet season, when runoff or diversion from other surface water supplies is most likely to be 432 available. MAR projects that use water supplied by a water recycling plant, or water conveyed 433 from higher in the river basin using the Pajaro River, might theoretically operate throughout the 434 water year. This set of model scenarios was not intended to be exhaustive or representative of

17 435 actual basin management plan scenarios under consideration by local stakeholders (PVWMA, 436 2012). Rather, the goal was to assess from a purely hydrologic perspective how MAR project 437 number, placement, and operations could influence groundwater conditions over several decades. 438 To analyze MAR scenario results, we compared model output of head levels and flows 439 from the ocean into the aquifer below the Pajaro Valley. Changes in head levels were quantified 440 in two ways: 1) for a given time over the entire basin, and 2) at a given location over the duration 441 of the model simulation. The first method was applied to compare MAR scenario groundwater 442 head levels from layer A2 (the volumetrically most significant aquifer layer in the region) during 443 the final time-step to the respective head levels in the Basecase simulation. Using the second 444 method, we selected a single grid cell and extracted the head values in each aquifer layer during 445 six stress periods. Flux of water from the offshore zone to coastal zones was classified as 446 seawater intrusion. With seawater intrusion as an active concern in the study region, it was 447 natural to use coastal flux as a metric for comparing MAR scenarios to the Basecase model. 448 Model coastal flux values were calculated for each stress period, and then summed to provide 449 flux per year over the entire duration of the model run. Flux values are given for the six model 450 layers combined either as seawater intrusion (flow inland from the ocean) or flow to the offshore 451 zone (loss of water to the ocean). 452 453 4. Results 454 4.1. Distribution of Classified Properties and MAR Suitability 455 Results from classification of six of the surface and subsurface properties are shown in Figure 5. 456 The majority of the surficial geology in the PVGB indicates moderate to good connectivity to 457 shallow local aquifers, except on the floodplain of the Pajaro River system, where there are

458 significant shallow silt and clay layers (Figure 5A). Effective infiltration (IE) shows similar 459 breadth of suitable areas, except in urban areas or nurseries, and on the floodplain of the Pajaro

460 River (Figure 5B). As discussed earlier, IE was calculated from soil infiltration capacity (IC), 461 land use (roughness) and elevation (slope) (Figure 3). The roughness coefficient, which is based 462 on land use, varies throughout the basin, with urban and turf areas concentrated in and around 463 Watsonville, CA, near the center of the basin. Urban and turf areas account for 21% of the total 464 area, whereas agricultural fields and pastures account for 41%. The remaining 38% is native 465 vegetation and unfarmed land, predominately located in the higher sloped northern and north-

18 466 western edges of the basin. Classified values of effective transmissivity (Te) are highly variable 467 across the basin, illustrating considerable heterogeneity in aquifer properties at a fine scale 468 (Figure 5C). The lower values generally correspond to locations with thicker alluvial clay

469 layers. Te values represent three datasets: aquifer layer thickness, aquifer hydraulic conductivity, 470 and confining layer thickness. Available storage (V) is low for much of the lower valley and 471 coastal region, with higher values in the northwest and southeast higher elevation areas (Figure 472 5D). Groundwater levels are generally lower near the coast and along the most northern and 473 western parts of the basin, relative to water levels in 1998, but there is a band of higher 474 groundwater levels that runs north-south through the center and to the southwestern side of the 475 basin (Figure 5E). Classification of measured infiltration rates are shown for two reaches on the 476 Pajaro River and Corralitos Creek (Figure 5F). 477 These and other coverages were combined to generate a map of MAR suitability across 478 the PVGB (Figure 6). This map has a nominal resolution of 10 x 10 m, although resolution of 479 the individual datasets varies considerably (Figure 5). The full spatial extent of the MAR 480 suitability map is limited by the intersection of the extents of all data sets used in the analyses 481 (228 km2). The normalized weights used to integrate the classified datasets are generally low 482 compared to weights for similar data sets used in other peer-reviewed studies (Figure 2). The 483 low weights in the present study result mainly from use of more datasets than were used in the 484 other studies, which reduces the relative influence of any individual dataset. 485 Calculated MAR suitability index values from across the PVGB range from 6 to 97 (low 486 to high suitability) and appear to follow a roughly normal distribution, with a mean of 22 and a 487 standard deviation of 52 (Figure 7). The upper quartile of this range, comprising land areas 488 being the most suitable for MAR, accounts for 7% of the analyzed land area in the PVGB (15 489 km2). These areas are located throughout the basin, but are particularly concentrated along the 490 coast north and south of the Pajaro River, inland south of the Pajaro River, and along the eastern 491 side (back) of the basin (Figure 6). The site of the Harkins Slough MAR project (Figure 6), 492 which is permitted to recharge up to 2.5 x 106 m3/yr diverted water to a perched aquifer, has a 493 MAR suitability index of 78. 494 495 4.2. Modeling the Influence of Distributed MAR Projects Options on Resource Conditions

19 496 Thirty one MAR scenarios (Table 3) illustrate how these projects could help to raise aquifer 497 water levels and reduce (or reverse) seawater intrusion, relative to the 34 year Basecase model. 498 Unsurprisingly, groundwater levels increased the most in locations closest to simulated MAR 499 projects (e.g. Figure 9). The Good location scenario (Figure 9A) shows the greatest increase in 500 the northwest part of the PVGB, and produces >1 m head level increase in over 80% of the 501 onshore area. The Coastal location scenario (Figure 9B) raises the head levels mostly along the 502 coast, on the western side of the PVGB, and produces a >1 m head level increase across ~60% of 503 the onshore area. There are significantly greater head level increases offshore with MAR projects 504 located in Coastal positions, compared to Good positions. 505 Simulated benefits to water levels within the aquifer layers vary based on MAR location 506 because water is applied to the exposed surface layer, which differs in properties and geometry 507 (thickness, extent of connection to deeper aquifer layers) throughout the basin. In general, and at 508 the mid-basin location (Figure 8), head levels increase by similar amounts in the shallowest 509 aquifer layers, A1 and A2 (Figure 10A and 10B), with coastal and MAR-good model scenarios 510 showing the most long-term improvement. Increases in head within the deepest aquifer layer, 511 A4, are more similar for the four sets of MAR project locations (Figure 10C), and net benefit is 512 generally lower than seen in shallower aquifers. 513 For all tested scenarios, simulated MAR projects reduced seawater intrusion compared to 514 the Basecase, with the benefit increasing overall with time (Figure 11). There is a period of 515 abrupt reduction in the extent of seawater intrusion, between simulation years 21 and 27, which 516 coincides with a period of increased precipitation and reduced groundwater pumping. The 517 location of simulated MAR projects has a notable effect on the magnitude of reductions in 518 seawater intrusion. The greatest benefit is achieved by simulating MAR projects in the MAR- 519 good locations, followed by Back-basin, Poor, and Coastal, in order of decreasing benefit 520 (Figure 11A). Though the rate of change of benefit varies with time, the benefit in the Good 521 scenarios increases approximately twice as quickly as that for the Coastal scenario over the 34 522 year simulation. MAR project location also affects the quantity of flow from the aquifer to the 523 ocean, and the rate of change over time (Figure 11B). The increase in flow to the ocean is 524 greatest early in the model runs and subsequently decreases over time, as head levels rise and/or 525 more water is extracted from the basin by pumping. The Coastal scenario results in the greatest

20 526 increase in flow to the ocean compared to the Basecase, followed in order by Good, Poor and 527 Back-basin scenarios. 528 The reduction of seawater intrusion varies with the amount of water applied in simulated 529 MAR projects, and as a function of project location and the passage of time. As expected, 530 increasing water applied to the system through simulated MAR projects results in larger 531 reductions of seawater intrusion along the coast (Figure 12). But increasing the amount of MAR 532 water applied at Coastal locations (Figure 12A) produces a smaller reduction in seawater 533 intrusion than does increasing the amount of MAR water applied at Good locations (Figure 534 12B). The difference in benefit between the two MAR project locations groups is minimal for the 535 lowest applied water rate (1.5x106 m3/yr), and increases to ~50% greater benefit at the Good 536 MAR locations for the highest applied water rate (1.5x107 m3/yr). 537 We divide reduction in seawater intrusion by the total applied water to measure MAR 538 efficiency (Figures 12C and 12D). For a given MAR location group, the efficiency is 539 approximately equal for the full range of applied water quantities at time = 1 yr. In other words, 540 the initial benefit is linearly related to the amount of MAR water applied to the surface. For 541 example, with MAR projects in Good locations, the seawater intrusion reduction efficiency is 542 approximately 1.5% of the total water applied for any given amount of water (Figure 12D). Over 543 time, the efficiency increases at a rate dependent on the amount of applied water, where 544 scenarios with less applied water show the greatest increase in efficiency. 545 Changing the number of MAR projects from 5 to 10 appears to have an influence similar 546 to that of doubling the total applied water, although locations of the additional 5 MAR projects 547 are likely to influence specific results. If the additional projects have a different average 548 proximity to the coast and/or MAR suitability index, then their influence on seawater intrusion 549 will be different than simply doubling the total applied water. 550 The intra-annual duration of MAR operations has minimal effect on the reduction of 551 seawater intrusion over the full (34 year) simulations. The scenarios active for 4 months and 12 552 months per year have nearly identical influence on seawater intrusion for the first 20 years of the 553 model simulation, then the projects active year-round tend to have an impact ~5 to 8% greater 554 than do the projects operating only 4 months per year (assuming the same amount of total MAR 555 water is applied). 556

21 557 5. Discussion 558 5.1. Classification and merging of GIS data sets 559 There are no "standard" practices for classifying indices used to assess suitability for managed 560 recharge. Most peer-reviewed, GIS-based studies completed to assess recharge properties and 561 processes have emphasized natural or incidental recharge, rather than MAR (Figure 2). Each of 562 these studies used a different weighting system for combining disparate datasets, and few earlier 563 studies attempted to test the results of GIS-based analyses for accuracy or applicability. We 564 attempted to address this latter issue, in part, by linking the GIS analysis to deterministic 565 modeling, discussed below, though this approach cannot confirm the "correctness" of regional 566 interpretations. 567 One approach for development of a suitable weighting system for applying GIS data is to 568 generate a suitability map that follows a desired distribution (e.g., normal, log-normal). If the 569 fundamental goal is to distinguish between the relative suitability of candidate field sites within a 570 basin, this approach could be useful, allowing clear delineation of land areas having 571 characteristics of a desired percentile of analyses (top 10%, best 100 hectares, etc). On the other 572 hand, application of different data sets and methods for combining them could lead to challenges 573 in comparing results from multiple basins, particularly if there are fundamental differences in 574 subsurface geology, weather patterns, land use, and other factors. In either case, using the GIS 575 analyses can guide or inform (rather than dictate) MAR placement as a component of critical 576 water resource decisions. The wide distribution of areas amenable for MAR projects may 577 encourage more landowners to participate in distributed recharge enhancement efforts, not just 578 those who are experiencing the consequences of aquifer overdraft. 579 Our data integration approach differed from those taken in earlier studies, in that we 580 combined several data sets to generate interim interpretations of effective properties. Effective 581 infiltration capacity encompasses the relationship between traditional soil infiltration capacity, 582 ground slope, and surface roughness. We reasoned that a greater slope and smoother land surface 583 would serve mainly to reduce the relative rate of infiltration, given intrinsic soil properties. Slope 584 and roughness should have less influence for soils that have a low infiltration capacity, but these 585 factors could result in a larger reduction in infiltration through highly permeable soils. Similarly, 586 we calculated effective transmissivity values for a series of aquifer layers, by summing all (or 587 part of) the values of individual layers from the surface downwards until a significant confining

22 588 unit was encountered. A more traditional approach for calculating transmissivity could either 589 under-represent effective values of areas where there are multiple (partly confined) aquifer units, 590 but only the shallowest is assessed, or over-estimate transmissivity if the presence of shallow 591 confining layers were ignored. Our calculations suggest the highest transmissivity values, with 592 respect to potential MAR projects, are located around the margins of the PVGB, but there are 593 also areas of moderately high transmissivity near the center of the basin, north of the Pajaro 594 River (Figure 5C). Different aquifer layers are responsible for these high transmissivity values: 595 the shallow alluvial and Aromas aquifers (A1 to A3) near the western and central parts of the 596 basin, and deeper Purisima (A4) aquifer to the west and north (Figures 4 and 5). 597 598 5.2 Integration of GIS analyses and numerical modeling 599 This study was designed from the start to link GIS-based assessment of MAR suitability to 600 results from a regional numerical model. Several of the surface and subsurface datasets used for 601 the GIS analysis were created originally as part of regional model development. This was 602 fortunate both in terms of the effort that would have been required to gather and classify each 603 dataset independently, and to co-register spatial data and modeling domains, but also for making 604 sure that data and assumptions applied for the GIS analyses were not violated when moving to 605 the modeling stage of the study. Resource managers and stakeholders in many groundwater 606 basins have access to similar datasets, although their resolution, accuracy, and completeness is 607 highly variable. The availability of a detailed and up-to-date regional groundwater model that 608 can be run on the basis of a GIS-based analysis of MAR is more unusual. 609 Modification of the PVGB model to include MAR projects facilitated evaluation of the 610 relative influence of major MAR characteristics, including project location, number of projects, 611 amount of water applied, and duration of operation through the year. The original model 612 included the Harkins Slough MAR project, and was calibrated to PVGB conditions under past 613 and existing climatological and water use conditions. Results presented in the current study, 614 evaluating hypothetical MAR projects in comparison to a baseline model, are thus most useful in 615 assessing relative impacts. 616 617 5.3. Implications for MAR in the Pajaro Valley

23 618 Our GIS-based analysis of MAR suitability shows considerable variability throughout the 619 PVGB, on the basis of eleven physical characteristics (Figure 6). The most prominent feature in 620 the final MAR suitability map is the Pajaro River floodplain, which has relatively low MAR 621 suitability primarily due to soil infiltration and surficial geology classifications. This is not 622 surprising given that floodplain lithology tends to comprise silt and clay lithologies which limit 623 surface infiltration. The GIS analysis would result in assigning similar properties to the bed of 624 the Pajaro River, but recent field studies documented streambed losses on the order of 1 m/d 625 along part of the river near the back of the basin (Ruehl et al. 2006; Hatch et al. 2010). This 626 discrepancy illustrates another limitation of the GIS-based analysis, within which the assessment 627 of soil properties is based on surveys near, rather than in, the active channel. In this case, there is 628 heterogeneity in lithology and in hydrologic behavior that is not captured by regional soil 629 property maps. 630 We can assess GIS results with respect to an active MAR project in the PVGB operated 631 by the PVWMA which recharges approximately 106 m3/yr (Racz et al. 2011; Schmidt et al. 632 2011). We define this as a successful MAR project, and therefore characterize the location as 633 highly suitable. The projected MAR suitability index based on the GIS analysis for this site is 78 634 (Figure 6), an index value met or exceeded by 4% of the basin (8.7 km2). These highly suitable 635 areas are distributed throughout the basin. There would need to be ~15 similarly performing 636 MAR projects to offset annual overdraft in the PVGB (PVWMA, 2012), but this would equate 637 to only ~4.8% of the land area that was classified as equal to or more suitable for MAR than the 638 Harkins Slough recharge project, or 0.19% of the total area analyzed. Of course, this assessment 639 does not account for surface water availability to supply MAR projects, and that may be the most 640 important limiting factor in applying this approach. The new Basin Management Plan 641 emphasizes conservation and water recycling, in additional to MAR, and development of a 642 portfolio of approaches is most likely to be successful in this setting. Nevertheless, knowing 643 about the potential for development of MAR projects, based on an assessment of surface and 644 subsurface conditions, can assist in efforts to bring the basin back into hydrologic balance. 645 Model results showed that MAR project location, amount of applied water, and years of 646 operation affect groundwater conditions in different ways. Projects located close to the coast 647 provide the greatest immediate benefit through reduction of seawater intrusion, but after ~3 648 years, seawater intrusion reduction is greatest for scenarios that place MAR projects throughout

24 649 the PVGB or in the back (East side) of the basin. Also, as the amount of water recharged 650 increases over time, the project efficiency (defined by reduction in seawater intrusion per unit of 651 water recharged) decreases due to flows to the ocean. These offshore flows may help to improve 652 groundwater quality, a parameter not assessed in this study. 653 Results show that the benefit from MAR projects varies depending on which evaluation 654 metric is used (groundwater rise or seawater intrusion reduction), and for the former, where the 655 metric is applied in the basin. MAR projects located at coastal sites result in the largest 656 groundwater head increase along the coast (Figure 9), but also the lowest long-term seawater 657 intrusion reduction (Figure 11A). MAR projects located where there is the highest suitability 658 based on the GIS analysis are most effective at reducing seawater intrusion, and on average are 659 farther from the coast than Coastal MAR projects. This illustrates the importance of assessing 660 both surface and subsurface properties and conditions when comparing locations for MAR 661 projects. 662 Comparing results from the Good and Poor scenarios provides confidence in the 663 applicability of the MAR suitability map. There was negligible difference in aquifer head levels 664 for the two simulated location groups of MAR projects. However, on average, there was a 25% 665 greater reduction of seawater intrusion for MAR projects located in areas identified as highly 666 suitable in the GIS analysis (Figure 11). The inefficiency of Poor MAR projects located in 667 unsuitable locations might be partially because they tended to recharge mainly layer A1. With 668 seawater intrusion occurring in all aquifer layers in the PVGB, it is beneficial to distribute MAR 669 projects among the regions where each aquifer is unconfined, in addition to selection by 670 suitability index. For example, if all MAR projects recharge to layer A1, where the presence of 671 underlying confining units restrict downward flow, layers A2 and A4 might only have limited 672 reduction of seawater intrusion. Note that Poor sites were not intentionally selected in areas 673 where layer A1 is unconfined; rather, sites with poor suitability often correspond to locations 674 where the alluvial layer contains low permeability flood plain deposits. 675 Modeled groundwater levels indicate that confining units do not restrict all flow between 676 aquifer layers, though flow between geologic layers is limited in some parts of the basin. While 677 groundwater head levels increase the most in the layer being recharged, modeled head increases 678 also occur in under- and over-lying aquifers. For example, the Back-basin MAR projects 679 recharge to A2, A3 and A4, but the head levels also increase by ~0.3 m in A1 near the City of

25 680 Watsonville over the 34 year model simulation (Figure 10A). With the majority of groundwater 681 extraction occurring in layer A2 (Hanson et al. 2013), having discontinuous confining units to 682 accommodate fluid flow between aquifers is critical, because it allows MAR projects in all areas 683 of the basin, taking advantage of available water supplies. 684 685 5.4. Study limitations and next steps 686 Several critical factors are not accounted for in the GIS and modeling analyses, including 687 water availability, solute sources and transport, unsaturated zone transport, site access, land use 688 and climate change, sea level rise, and proximity to areas that are already intruded by seawater 689 intrusion. These factors should be considered as evaluations are done to identify locations for 690 field and pilot testing. The GIS analyses were not intended to be the primary basis for making 691 placement and operational decisions for MAR project sites. 692 This study does not formally evaluate water availability for MAR projects. The analysis 693 shows many areas in the PVGB that are highly suitable for MAR, especially near the coast, 694 which coincides with the blending and distribution system developed for use of water from the 695 recycling plant. Locations in the Back-basin could use runoff from adjacent hills as a recharge 696 water source. 697 Numerical model results suggest that placement of MAR projects according to the GIS 698 suitability index provides the greatest reduction of seawater intrusion along the coast. The 699 quantity of water applied using MAR is proportional to the long-term benefit. However, in this 700 water-stressed area, it will be necessary to optimize the quantity of water applied with respect to 701 desired reduction in seawater intrusion. Larger applied quantities of water will provide a greater 702 benefit, though at a lower efficiency than smaller applied quantities of water. Water availability 703 will likely govern the quantity of applied water on an annual basis. 704 The current model does not include solute sources or advection, and therefore cannot 705 estimate the influence of recharge on salinity of the seawater intruded areas or overall water 706 quality elsewhere in the basin. Future studies could add solute transport capabilities to the MAR 707 suitability assessment. For example, placing MAR projects within the seawater intruded area 708 might be a feasible option for reducing the rate of future intrusion, but might not have a strong 709 enough influence on water quality benefit to allow extraction from areas that are already 710 intruded. Conversely, recharging onto a local perched aquifer above the seawater intruded area

26 711 can provide an alternate source for users, allowing coastal farming to continue and reducing 712 demand on overdrafted aquifers below (Racz et al. 2011; Schmidt et al. 2011). Recharging to a 713 perched aquifer was not considered an advantage, as in this particular case, in the GIS analysis. 714 The model does not include unsaturated flow, which is important for understanding 715 aquifer recharge. Within the model, water applied to the surface moves immediately down to the 716 uppermost saturated cell. Results from studies of the Harkins Slough MAR project suggest that 717 infiltrating water may recharge the shallow aquifer in just a few days (Schmidt et al. 2011). At 718 least in this location, the assumption of instantaneous transport through the vadoze zone may be 719 reasonable when using month-long stress periods. However, this approximation could lead to 720 errors in assessing MAR projects located where infiltration rates are slower. 721 The groundwater model uses the Farm Process (Schmid and Hanson 2009), which varies 722 the amount of water pumped based on land use, climate and water availability. As MAR water 723 recharges the aquifer, groundwater availability increases, and this allows an increase in pumping. 724 The rate at which pumping increases is modest relative to the rate of recharge. However, if 725 higher head levels due to MAR cause pumping in the model to increase beyond realistic 726 projections, then estimations of seawater intrusion reduction would be conservative. One could 727 disable the Farm Process and fix factors such as natural recharge and pumping, but this could 728 result in model scenarios that are less realistic; land use and climate are expected to vary year by 729 year, and these changes influence the locations and rates of groundwater extraction. For example, 730 there is an eight-year increase in seawater intrusion reduction starting in model year 21 for all 731 MAR scenarios (Figures 11 and 12), largely in response to a modeled wet climate period and 732 associated reductions in water use. This model response also shows how climate and changes to 733 pumping can compound the benefit provided by simulated MAR projects. But for these reasons, 734 we consider model results mainly in the context of relative benefits from MAR scenarios, rather 735 than quantitative predictions. Indeed, the scenarios tested in this study deviate considerably from 736 those explored in the most recent Basin Management Plan (PVWMA, 2013). The latter focused 737 on realistic project alternatives based on physical, chemical, economic, and social factors, many 738 of which were not explored in this paper. Narrowly focused scenarios presented in this paper 739 should not be viewed as alternatives to those proposed in the Basin Management Plan. 740 The next step in determining where to implement MAR projects is field testing soil 741 infiltration properties at locations that have been identified as suitable for MAR by the GIS and

27 742 numerical modeling analyses. Assessing sites for MAR is a complicated problem, as evidenced 743 by the number of studies and variety of methods. We produced a MAR suitability index map, 744 and show relative impacts on seawater intrusion and groundwater levels, though this must be an 745 iterative process. Field testing and MAR implementation are required both to reduce 746 groundwater overdraft and to help calibrate the suitability mapping method. With more field 747 data, and comparisons to similar results from other study areas, the GIS-based integration 748 method will become more robust for use in this region and in other groundwater basins. 749 750 6. Conclusions 751 This paper proposed a physically based, GIS integration method for identifying locations that 752 may be suitable for MAR projects, and quantified the relative benefit of such projects using a 753 numerical model. We developed a method that allows data to be combined using traditional 754 approaches (overlying coverages and adding indices) and by allowing some properties to operate 755 on other properties before coverages are combined. We propose that this method can provide a 756 more accurate understanding of relationships between geology, hydrology, and managed 757 groundwater recharge. Results suggest that ~15 km2 of the PVGB may be highly suitable for 758 MAR projects, as delineated by having a suitability index in the upper quartile of the quantitative 759 range. Using a numerical model to simulate MAR projects, we show that project sites on high 760 MAR suitability areas could reduce seawater intrusion to a greater extent than if MAR projects 761 were located on low suitability areas. Modeling suggests that reducing seawater intrusion is 762 achieved most efficiently with MAR projects distributed throughout the PVGB in highly suitable 763 locations, rather than focusing only along the coast. 764 Groundwater development is expected to continue increasing around the world, providing 765 significant economic benefits and maintaining food security for the growing population. 766 Unfortunately, the financial returns from increased agriculture and industrial development are 767 rarely used towards water management, resulting in declining water tables, reduction in 768 groundwater storage, and water quality issues. MAR projects may contribute towards sustainable 769 groundwater use as a low-cost, low-maintenance, and potentially distributed method. This paper 770 illustrates a nuanced approach for identifying suitable locations for MAR projects, and for 771 determining the relative impacts of various recharge project scenarios using numerical modeling.

28 772 Careful field studies and assessment of implemented projects will be required to test and refine 773 these conclusions. 774

775 Acknowledgements 776 We thank Mike Cloud, Michael Cahn, and Marc Los Huertos for their thoughtful advice on land 777 use and Pajaro Valley geology and hydrogeology. This work was supported by the National 778 Science Foundation Graduate Research Fellowship Program (ID# 2009083666), the National 779 Institute for Water Resources (Grant 08HQGR0054), and The Recharge Initiative 780 (rechargeinitiative.org).

781 References

782 Abarca E, Vázquez-Suñé E, Carrera J, et al. (2006) Optimal design of measures to correct 783 seawater intrusion. Water Resources Research 42:1–14. doi: 10.1029/2005WR004524 784 Adham M, Jahan C, Mazumder Q, et al. (2010) Study on groundwater recharge potentiality of 785 Barind Tract, Rajshahi District, Bangladesh using GIS and Remote Sensing technique. 786 Journal of the Geological Society of India 75:432–438. 787 Aller L, Lehr J, Petty R (1987) DRASTIC: A Standardized System to Evaluate Ground Water 788 Pollution Potential using Hydrogeologic Settings. Journal of the Geological 38–57. 789 Bouwer H (2002) Artificial recharge of groundwater: hydrogeology and engineering. 790 Hydrogeology Journal 10:121–142. doi: 10.1007/s10040-001-0182-4 791 Brabb EE, Graham S, Wentworth C, et al. (1997) Geologic map of Santa Cruz County, 792 California. US Geological Survey, 97–489. 793 Chenini I, Mammou A Ben (2010) Groundwater recharge study in arid region : An approach 794 using GIS techniques and numerical modeling. Computers and Geosciences 36:801–817. 795 doi: 10.1016/j.cageo.2009.06.014 796 Chenini I, Mammou A, El May M (2010) Groundwater recharge zone mapping using GIS-based 797 multi-criteria analysis: a case study in Central Tunisia (Maknassy Basin). Water Resources 798 Management 24:921–939. doi: 10.1007/s11269-009-9479-1 799 Chitsazan M, Akhtari Y (2009) A GIS-based DRASTIC model for assessing aquifer 800 vulnerability in Kherran plain, Khuzestan, Iran. Water Resources Management 23:1137– 801 1155. doi: 10.1007/s11269-008-9319-8 802 Chow V (1959) Open-Channel Hydraulics. McGraw-Hill, New York 803 Chowdhury A, Jha M, Chowdary V (2010) Delineation of groundwater recharge zones and 804 identification of artificial recharge sites in West Medinipur district, West Bengal, using RS, 805 GIS and MCDM techniques. Environmental Earth Sciences 59:1209–1222. doi: 806 10.1007/s12665-009-0110-9

29 807 Clark J, Dupre W, Rosenberg L (1997) Geologic map of Monterey County, California. US 808 Geological Survey, 97–30. 809 Dettinger M (2005) From climate-change spaghetti to climate-change distributions for 21st 810 Century California. San Francisco Estuary and Watershed Science 3: 811 Dillon P, Pavelic P, Page D, et al. (2009) Managed aquifer recharge: An Introduction. 1–65. 812 Dupre W (1990) Quaternary geology of the Monterey Bay region, California. In: Garrison R, 813 Greene H, Hicks K, et al. (eds) Geology and tectonics of the Central California Coast 814 region, San Francisco to Monterey. American Association of Petroleum Geologists, 815 Bakersfield, CA, p 314 816 Foster SSD, Chilton PJ (2003) Groundwater: the processes and global significance of aquifer 817 degradation. Philosophical transactions of the Royal Society of London Series B, Biological 818 sciences 358:1957–72. doi: 10.1098/rstb.2003.1380 819 Fryar AE, Macko S a., Mullican III WF, et al. (2000) Nitrate reduction during ground-water 820 recharge, Southern High Plains, Texas. Journal of Contaminant Hydrology 40:335–363. doi: 821 10.1016/S0169-7722(99)00059-5 822 Giordano M (2009) Global Groundwater? Issues and Solutions. Annual Review of Environment 823 and Resources 34:153–178. doi: 10.1146/annurev.environ.030308.100251 824 Gleeson T, VanderSteen J, Sophocleous M a., et al. (2010) Groundwater sustainability strategies. 825 Nature Geoscience 3:378–379. doi: 10.1038/ngeo881 826 Hanson R (2003) Geohydrologic framework of recharge and seawater intrusion in the Pajaro 827 Valley, Santa Cruz and Monterey Counties, California. US Geological Survey, Sacramento, 828 CA 829 Hanson R, Schmid W, Faunt C, et al. (2013) Hydrologic Model of Pajaro Valley, Santa Cruz and 830 Monterey Counties, California. US Geological Survey, In Review 831 Hanson R, Schmid W, Lear J, Faunt C (2008) Simulation of an Aquifer-Storage-and-Recovery 832 (ASR) System using the Farm Process in MODFLOW for the Pajaro Valley, Monterey Bay, 833 California. Modflow and More - Ground Water and Public Policy. Golden, CO, pp 501–505 834 Hanson RT, Schmid W, Faunt C, Lockwood B (2010) Simulation and analysis of conjunctive 835 use with MODFLOW’s farm process. Ground Water 58:674–689. 836 Harbaugh AW (2005) MODFLOW-2005, The U.S. Geological Survey modular ground-water 837 model—the Ground-Water Flow Process: U.S. Geological Survey Techniques and Methods 838 6- A16. 839 Hatch CE, Fisher AT, Ruehl CR, Stemler G (2010) Spatial and temporal variations in streambed 840 hydraulic conductivity quantified with time-series thermal methods. Journal of Hydrology 841 389:276–288. doi: 10.1016/j.jhydrol.2010.05.046 842 Jasrotia A, Kumar R, Saraf A (2007) Delineation of groundwater recharge sites using integrated 843 remote sensing and GIS in Jammu district, India. International Journal of Remote Sensing 844 28:5019–5036. doi: 10.1080/01431160701264276 845 Jenks G (1967) The Data Model Concept in Statistical Mapping. International Yearbook of 846 Cartography 7:186–190.

30 847 Jha M, Chowdhury A, Chowdary V, Peiffer S (2007) Groundwater management and 848 development by integrated remote sensing and geographic information systems: prospects 849 and constraints. Water Resources Management 21:427–467. doi: 10.1007/s11269-006- 850 9024-4 851 Konikow LF, Kendy E (2005) Groundwater depletion: A global problem. Hydrogeology Journal 852 13:317–320. doi: 10.1007/s10040-004-0411-8 853 Ma L, Spalding R (1997) Effects of artificial recharge on ground water quality and aquifer 854 storage recovery. Journal of the American Water Resources Association 33:561–572. 855 Maliva R, Missimer T (2012) Managed Aquifer Recharge. Arid Lands Water Evaluation and 856 Management. pp 559 – 630 857 Missimer TM, Drewes JE, Amy G, et al. (2011) Restoration of wadi aquifers by artificial 858 recharge with treated waste water. Ground water 50:514–27. doi: 10.1111/j.1745- 859 6584.2012.00941.x 860 Muir K (1972) Geology and ground water of the Pajaro Valley area, Santa Cruz and Monterey 861 Counties, California. US Geological Survey, Menlo Park, CA 862 Munevar A, Marino M (1999) Modeling analysis of ground water recharge potential on alluvial 863 fans using limited data. Ground Water 37:649–659. 864 Murray J, Mcdaniel PA (2003) Development of a GIS database for ground-water recharge 865 assessment of the Palouse Basin. Soil Science 168:759–768. doi: 866 10.1097/01.ss.0000100474.96182.5f 867 NRCS (2010a) National Resources Conservation Service, Soil Data Mart. In: Santa Cruz 868 County, CA. http://soildatamart.nrcs.usda.gov/Survey.aspx?County=CA087. Accessed 7 869 Jan 2012 870 NRCS (2010b) National Resources Conservation Service, Soil Data Mart. In: Monterey County, 871 CA. http://soildatamart.nrcs.usda.gov/Survey.aspx?County=CA0853. Accessed 7 Jan 2012 872 Piscopo G (2001) Groundwater vulnerability map explanatory notes: Castlereagh Catchment. 873 PVWMA (2013) 2012 Basin Management Plan Update, Draft: January 2013. 80. 874 Racz AJ, Fisher AT, Schmidt CM, et al. (2011) Spatial and Temporal Infiltration Dynamics 875 During Managed Aquifer Recharge. Ground water 1–9. doi: 10.1111/j.1745- 876 6584.2011.00875.x 877 Rahman MA, Rusteberg B, Gogu RC, et al. (2012) A new spatial multi-criteria decision support 878 tool for site selection for implementation of managed aquifer recharge. Journal of 879 Environmental Management 99:61–75. doi: 10.1016/j.jenvman.2012.01.003 880 Rauch-Williams T, Hoppe-Jones C, Drewes J (2010) The role of organic matter in the removal 881 of emerging trace organic chemicals during managed aquifer recharge. Water Research 882 44:449–460. 883 Rosegrant MW, Cai X (2009) Global Water Demand and Supply Projections Global Water 884 Demand and Supply Projections Part 2 . Results and Prospects to 2025. 37–41. 885 Ruehl C, Fisher a. T, Hatch C, et al. (2006) Differential gauging and tracer tests resolve seepage 886 fluxes in a strongly-losing stream. Journal of Hydrology 330:235–248. doi: 887 10.1016/j.jhydrol.2006.03.025

31 888 Russo TA, Fisher AT, Winslow DM (2013) Regional and local increases in storm intensity in the 889 San Francisco Bay Area, USA, between 1890 and 2010. Journal of Geophysical Research: 890 Atmospheres 118:1–10. doi: 10.1002/jgrd.50225 891 Saraf A, Choudhury P (1998) Integrated remote sensing and GIS for groundwater exploration 892 and identification of artificial recharge sites. International Journal of Remote Sensing 893 19:1825–1841. 894 Schmid W, Hanson RT (2009) The Farm Process Version 2 (FMP2) for MODFLOW-2005— 895 Modifications and Upgrades to FMP1: U.S. Geological Survey Techniques and Methods 6- 896 A-32, 102 p. 897 Schmidt C, Fisher A, Racz A (2011) Rapid nutrient load reduction during infiltration of managed 898 aquifer recharge in an agricultural groundwater basin: Pajaro Valley, California. 899 Hydrological Processes. doi: 10.1002/hyp 900 Shammas MI (2007) The effectiveness of artificial recharge in combating seawater intrusion in 901 Salalah coastal aquifer, Oman. Environmental Geology 55:191–204. doi: 10.1007/s00254- 902 007-0975-4 903 Shankar MNR, Mohan G (1998) A GIS based hydrogeomorphic approach for identification of 904 site-specific artificial-recharge techniques in the Deccan Volcanic Province. October 905 Smith AJ, Pollock DW (2012) Assessment of Managed Aquifer Recharge Potential Using 906 Ensembles of Local Models. Ground Water 50:133–143. doi: 10.1111/j.1745- 907 6584.2011.00808.x 908 Taylor RG, Scanlon B, Döll P, et al. (2013) Ground water and climate change. Nature Climate 909 Change 3:322–329. doi: 10.1038/NCLIMATE1744 910 Wallace M, Lockwood BS (2009) Pajaro Valley Water Management Agency Annual Report. 911 Yeh H-F, Lee C-H, Hsu K-C, Chang P-H (2009) GIS for the assessment of the groundwater 912 recharge potential zone. Environmental geology 58:185–195. doi: 10.1007/s00254-008- 913 1504-9 914 915

32 916 917 Figure 1 Location of the Pajaro Valley, CA, with extent of seawater intrusion measured in 918 2001(Hanson 2003), elevation and major streams. Area shown is the local water management’s 919 (Pajaro Valley Water Management Agency) boundary of operation. The Harkins Slough MAR 920 project is indicated with a square, and the mid-basin measurement point used in the modeling 921 section is indicated with a white circle. Add slough shapefile.

922 923 924

33 925

926 Figure 2. Comparison of dataset weights used in other studies to map groundwater recharge with 927 a GIS. The normalized weights used in this study are shown in orange. Values shown for land 928 use and slope are calculated means of values used, because these data sets were used as 929 modifiers for other data sets, as discussed in the text. Replace with Andy’s new B&W version.

34 930

931 Figure 3 Example calculated effective infiltration (IE) values for a given infiltration capacity (IC)

932 value of 5, roughness coefficients 14 to 100, and three slope values. The IE curve will move

933 down for larger slopes and smaller IC values. 934

935 936

35 937 938 Figure 4 Pajaro Valley Hydrologic Model (PVHM), (A) map view of model domain showing 939 grid cells, (B) cross section showing model layers along transect A-A’. Modified from (Hanson 940 et al. 2013). 941

36 942 943 Figure 5 Classified surface and subsurface properties used to determine relative MAR 944 suitability. (A) surficial geology, (B) effective infiltration, (C) effective transmissivity, (D) 945 storage availability, (E) change in groundwater elevation (2010-1998), (F) measured streambed 946 infiltration.

37 947 948 Figure 6 Map of relative MAR suitability determined by GIS-based integration. The location of 949 the existing Harkin Slough MAR project is indicated with a circle.

950 951

38 952 953 Figure 7 Histogram of the MAR suitability index values for the PVGB. The suitability index 954 value of the Harkin Slough project site is 78, which represents field tested managed recharge of 955 approximately 106 m3/yr. Four percent (8.7 km2, 2160 ac) of the PVGB has similar or higher 956 suitability index values.

39 958 959 Figure 8 MAR scenario location groups shown on the MAR suitability index map. Figure 4 960 shows the spatial extent of the model domain. Ten site locations are shown for each of the four 961 groups: Coastal, Back-basin, Good, and Poor. Head levels were compared to the Basecase 962 simulation at a location in Watsonville (Figure 10), indicated with a filled black circle. Add mid- 963 basin monitoring location.

40 964

965 966 Figure 9 Increase in head levels in Layer A2 at model yr-34 relative to the Basecase due to MAR 967 projects simulated in Good Run-22 (A) and Coastal Run-8 (B) locations. Both scenarios have 10 968 MAR projects applying 4.6 x 105 m3/yr each indicated by black open circles.

41 970 971 Figure 10 Increase in head levels near the center of the PVGB (location shown in Figure 1) 972 relative to the Basecase due to MAR projects simulated in four regions of the basin, (A) in Layer 973 A1, (B) in Layer A2, and (C) in the Layer A4.

42 974 975 Figure 11 Benefits relative to the Basecase due to MAR projects simulated in four regions of the 976 basin, respectively, for (A) Reduction of seawater intrusion shown versus time, and (B) Increase 977 in flow to offshore zone shown versus time. Each scenario has 5 MAR projects, applying 9.8x105 978 m3/yr each, and operating 12-mo/yr. 979 980

43 981 982 Figure 12 seawater intrusion reduction relative to the Basecase due to simulated MAR projects 983 with varying rates of total applied water. Each scenario uses 10 MAR projects operating 12- 984 mo/yr, located at (A) Coastal sites, and (B) Good sites. The efficiency, with respect to seawater 985 intrusion, is shown for MAR projects located at (C) Coastal sites, and (D) Good sites.