Previous Work

Previous investigations of groundwater flow in the Mississippi Embayment as part of the USGS Gulf Coast Regional Aquifer System Analysis studies include Arthur and Taylor (1990, 1998), which focused on describing the hydrogeologic framework and groundwater flow in the Tertiary System, and Ackerman (1989, 1996), which described groundwater flow in the Mississippi River Valley alluvial aquifer. These models simulated groundwater flow in the Tertiary System from 1886 to 1987 and in the alluvial aquifer from 1906 to 1987 using a uniform grid with cells of 5 miles on each side. Arthur (2001) modeled transient groundwater flow in the alluvial aquifer in the Mississippi Delta area from January 1988 to December 1996, with a uniform grid with cells of 1 mile on each side, to help determine the capacity of the aquifer for continuing agricultural growth.

As part of the Mississippi Embayment Regional Aquifer Study (MERAS), Hart and others (2008) improved the hydrogeologic framework of the area. Clark and Hart (2009) and Clark and others (2011) modeled the Mississippi River Valley alluvial aquifer and the Tertiary System in a single groundwater flow model (referred to here as the “MERAS 1 model”) to quantify groundwater availability throughout the Mississippi Embayment. The MERAS 1 model simulated groundwater flow from 1870 to 2007 in 13 layers using a uniform grid with cells of 1 mile on each side. Clark and others (2013) improved MERAS 1 (MERAS 2.0 model) and used it to evaluate potential future water-level conditions. Additionally, Barlow and Clark (2011), Clark and others (2011), and Haugh (2012, 2016) all used the MERAS model to evaluate future water withdrawal scenarios. Haugh and others (2020a) extended the MERAS 2 model simulation through 2014 (MERAS 2.1) to evaluate water management scenarios in the Mississippi Delta; furthermore, Hunt and others (2021) updated the MERAS 2 model with the Newton-Raphson solver in MODFLOW–NWT (Niswonger and others, 2011) and an increased density of simulated streams for a study focused on the alluvial aquifer in the Delta (MERAS–NWT).

Each of these modeling efforts showed significant changes in the water budget and flow system because of increased groundwater pumping over time. In all studies, the sources of water to meet pumping demand included recharge, streamflow leakage, and storage, but the relative contribution of water from each of these sources varied considerably in the previous models. In a study comparing satellite-derived estimates of storage changes in major U.S. aquifers with monitoring and modeling results (Rateb and others, 2020), the MERAS model had the largest discrepancy in simulated storage change compared to the satellite-derived estimates, which may indicate oversimulation of the contribution of water from storage to meet pumping demand. Water budgets from Hunt and others (2021; MERAS–NWT) indicated larger contributions of water from stream leakage than previous MERAS models, likely because of the increased density of simulated streams. Accurate characterization of the water balance in the Mississippi Alluvial Plain—especially the inflows to the groundwater system and the distribution of terrestrial recharge compared to leakage from streams—remains a key topic of interest.

Limitations of Previous Models and Recent Advances

Use of the MERAS model for decision support in the Mississippi Delta is limited by a large number of computational cells, most of which are outside of the Delta. Additionally, most of the hydrostratigraphic units represented in the 13 layers of the MERAS model are not present everywhere in the model domain. The MODFLOW–2005 framework (Harbaugh, 2005) used for the MERAS model requires layer continuity. Discretization approaches that use layers to represent individual units are thus required to include thin “pinch out” cells where the units are absent, which can contribute to solution instability and greatly increase the number of active cells. In previous versions of the MERAS models, these issues resulted in long runtimes of as much as 16 hours or more for the model spin up and history matching periods.

The MERAS and Arthur (2001) models were also limited by their representation of surface water and recharge. The MERAS 2 models represented recharge in piecewise-constant zones based on soil type and surficial geology, with temporal rates based on fractions of precipitation. Arthur (2001) used a single recharge zone for the entire Mississippi Delta, with rates also based on fractions of precipitation. Both the Arthur (2001) and the MERAS 2 model included only the largest streams as boundary conditions. Arthur (2001) represented streams using the River (RIV) Package, with monthly stages specified from measurements collected by the U.S. Army Corps of Engineers. Although the MERAS 2 models used the Streamflow-Routing (SFR) Package, seasonal variability in streamflow and stages was simplified to a temporal discretization of biannual stress periods. Surface water runoff was based on assumed fractions of precipitation and was not parameterized in the history matching process. Simulated stream stages, a key driver of groundwater/surface water interactions, were also not considered (Clark and Hart, 2009). Since the construction of the MERAS model, there has also been increasing recognition of the importance of groundwater/surface water interactions and the larger integrated water cycle in groundwater flow simulations. Although the importance of surface water to the groundwater flow system in the Mississippi Delta has long been recognized (for example, Arthur, 2001), its inclusion in these models was historically limited by available software and computing resources and accompanying standards of practice.

Recent software advances have facilitated the incorporation of more realistic processes in groundwater models in a way that is tractable for decision support. An enhanced solver for MODFLOW (MODFLOW–NWT; Niswonger and others, 2011) provides superior handling of dry model cells (Hunt and Feinstein, 2012), which facilitates better representations of aquifer storage. The Soil-Water-Balance (SWB; Westenbroek and others, 2018) code provides a more representative simulation of groundwater recharge using an easy-to-use, physics-based approach that estimates net infiltration past the root zone, surface water runoff, and crop water demand on a cell-by-cell basis. SFRmaker (Leaf and others, 2021) automates the construction of realistic surface water input to the MODFLOW SFR Package from existing hydrography data such as NHDPlus (McKay and others, 2012). Machine learning techniques have allowed for monthly surface water inflows from other basins to be reliably estimated from climate and drainage basin characteristics (Dietsch and others, 2022). Finally, since the creation of the original MERAS model structure (Clark and Hart, 2009), there has been extensive new data collected in the Mississippi Embayment, including the largest airborne electromagnetic (AEM) survey flown in the United States to date (Minsley and others, 2021), waterborne electrical resistivity profiling of streambed sediments in major streams (Miller and others, 2016), improvements in water use metering and estimation (for example, Wilson, 2021), and detailed simulation of the terrestrial water balance (Nielsen and Westenbroek, 2023).

Although some of these updates were incorporated into the existing MERAS framework by Hunt and others (2021), the MERAS–NWT model had long runtimes (greater than 6 hours) because of simulating the entire embayment and an abundance of thin cells. The MODFLOW 6 groundwater modeling framework (Langevin and others, 2017) allows for discontinuous layering in the context of a regular grid via “vertical pass-through cells.” This capability, along with more efficient solution techniques, can produce faster model runtimes that are critical for effective decision support (for example, Doherty and Moore, 2020).

Modeling Goals and Approach

This study sought to produce updated groundwater models for quantifying current groundwater resources in the Mississippi Alluvial Plain and potential responses to future conditions, including expansion of agriculture, water conservation, climate change, and water supply mitigation. The work here focuses on the Mississippi Delta region. A key objective of the updated models was to leverage the previously mentioned advances in modeling techniques and software and synthesize the many new data and analyses produced by the larger USGS Mississippi Alluvial Plain Project. A secondary goal was to achieve faster runtimes to maximize model usefulness for decision support. Finally, automation was used in the model construction workflows to facilitate rapid refinement of local areas within the model domain and rapid model updating for new forecasts in response to new data or new questions or areas of interest.

Unlike previous studies that focused on a single model covering the entire Mississippi Embayment, the work here uses a simplified, faster-running model of the MERAS footprint (fig. 1) that can be used to produce specified flux boundary conditions for more detailed inset models of local areas. This updated MERAS model (MERAS 3.0 or MERAS 3; also referred to in this report as the “regional model”) uses a horizontal discretization of 1 kilometer (km; Clark and others, 2018a) and a vertical discretization of three layers. A more detailed inset model of the Mississippi Delta region (Mississippi Delta model or “Delta model”) with 500-m horizonal discretization and 21 layers was inset within the regional model solution using specified flux boundaries. Both models were created using MODFLOW 6 (version 6.3.0; Langevin and others, 2022). Detailed information on the discretization and construction of these models is included in appendix 1.

Python packages were developed (Leaf and Fienen, 2022; Leaf and others, 2021) to robustly automate the construction of the regional model and any inset model within the MERAS footprint. A stepwise approach (Haitjema, 1995) was taken starting with an easily achievable model configuration and adding features such as surface water runoff or denser parametrization as needed to better fit observed time series of groundwater levels, stream flows, and stream stages. Parametrization of model inputs was done using the PstFrom automated framework in pyEMU (White and others, 2021), which allows parametrization schemes at different scales (for example, adjusting hydraulic conductivity via pilot points or by zone) to be rapidly implemented and tested. History matching for parameter estimation was done using the iterative ensemble smoother (iES) algorithm implemented PEST++ version 5 (White and others, 2020). The iES method generates a stochastic ensemble of model input parameter configurations (conditioned on observations) that calculates probabilistic model uncertainty.

Purpose and Scope

This report documents the construction of the conceptual model and the construction and history matching of the MERAS 3 and Mississippi Delta models; formulation of the numerical models from preexisting data and data produced by other components of the USGS MAP project; history matching methods and results; simulated water budgets, groundwater levels and groundwater/surface water interactions; and the development and results of baseline forecast scenarios under varying potential future climates.

Climate

The region has a humid subtropical climate, with mean annual temperature ranging from about 15.5 degrees Celsius (°C) at Cairo, Illinois, in the north to 19.4 °C at Vicksburg, Miss., in the south (Alder and Hostetler, 2013) and mean annual precipitation ranging from about 122 centimeters per year (cm/yr) in the north to about 142 cm/yr or more in the south (for example, Clark and others, 2011; Reitz and Kress, 2019). Precipitation is generally highest in the winter and spring and lowest in later summer or early fall (Alder and Hostetler, 2013). Droughts are not uncommon during the summer and fall (Cushing and others, 1964).

Hydrostratigraphy

The Mississippi Embayment is a southward-plunging syncline with an axis primarily paralleling the present-day Mississippi River except where the axis curves around structural uplifts in southeastern Arkansas-northwestern Louisiana and southwestern Mississippi (Hart and others, 2008). Sedimentary deposits within the embayment generally dip and thicken towards the syncline axis and southward. Initial folding associated with the Ouachita orogeny during the Paleozoic Era was later deepened with additional downwarping and downfaulting associated with sedimentation during the Cretaceous period, creating the Mississippi Embayment (Hosman, 1996; Saucier, 1994). Subsequent cyclic advances and retreats of Cretaceous seas deposited mostly marine sediments throughout the embayment, creating the base for ensuing Tertiary and Quaternary deposits (Arthur and Taylor, 1998, Hosman, 1996; Saucier, 1994). Pleistocene glaciation resulted in incisement of the Mississippi River Valley by glacial meltwater flowing towards the Gulf of Mexico. As sea level rose, sediments filled the entrenched valley, forming the Mississippi River Valley alluvial aquifer.

The primary hydrogeologic units in the MERAS area include the 10 units described by Hart and others (2008). Table 1 in Clark and Hart (2009) shows a comprehensive depiction of the major units and their relation to local hydrostratigraphy and MERAS model layers. Together, the following units constitute the Mississippi Embayment aquifer system: the Quaternary Mississippi River Valley alluvial aquifer (alluvial aquifer), and underlying Tertiary-aged units: (the Vicksburg-Jackson confining unit, the upper Claiborne aquifer, the middle Claiborne confining unit, the middle Claiborne aquifer [also referred to as the Sparta aquifer in many places; for example Fitzpatrick and others (1990)], the lower Claiborne confining unit, the lower Claiborne aquifer, the middle Wilcox aquifer, the lower Wilcox aquifer, and the Midway confining unit). In this report, the term “Tertiary aquifer system” is used to differentiate the Tertiary units below the alluvial aquifer from the “Mississippi Embayment aquifer system,” which may include the alluvial aquifer. The alluvial aquifer consists primarily of gravel and sand deposits, and the middle Claiborne aquifer consists primarily of sand, silt, and clay deposits (Clark and others, 2011). The upper and middle Claiborne aquifers are bounded above and below by confining units in most places. The lower Claiborne-upper Wilcox aquifer and the middle and the lower Wilcox aquifers are not separated by confining units. The lower Claiborne confining unit and the lower Claiborne aquifer undergo a facies change and merge into the middle Claiborne aquifer in the northern part of the embayment model area (Hart and others, 2008).

Throughout much of Mississippi because of a facies change, the lower part of the lower Claiborne confining unit is the Winona-Tallahatta aquifer, which includes the greenish-gray, siliceous, sandy claystone of the Tallahatta Formation of Claiborne Group and the glauconitic, fossiliferous, medium- to coarse-grained sandstone of the Winona Sand of Claiborne Group (Spiers, 1977; Mancini and Tew, 1994; Clark and Hart, 2009). The sandy deposits of the three Wilcox aquifers are more heterogeneous than the other Tertiary units and consist of a highly variable arrangement of massive to thinly bedded sand and thin clay beds (Renken, 1998). The middle Wilcox aquifer differs substantially from the upper and lower Wilcox aquifers because its thin beds of sand and clay result in lower hydraulic conductivity than that of the massive, more permeable units above and below it (Renken, 1998). The middle and lower Wilcox aquifers, however, are undifferentiated throughout most of Arkansas and Louisiana (Clark and Hart, 2009). In parts of Tennessee and Mississippi, the lower Wilcox aquifer may be separated into two units, the lower Wilcox aquifer and the Old Breastworks confining unit (Clark and Hart, 2009), with the Old Breastworks Formation of Wilcox Group consisting of clay, silt, and lignite (Warwick and others, 1997). The lower Wilcox aquifer is the bottommost Tertiary aquifer within the embayment (Lloyd and Lyke, 1995).

Both the Vicksburg-Jackson and the Midway confining units consist of massive clay beds. In the southern part of the embayment, the Vicksburg-Jackson confining unit separates the embayment aquifer system from the overlying coastal lowlands aquifer system (Weiss, 1992). The top of the Midway confining unit forms the base of the embayment aquifer system, separating the aquifers in sediments of Tertiary age from underlying aquifers of Cretaceous age.

Terrestrial Water Balance

The terrestrial water balance describes the partitioning of precipitation input into evapotranspiration, near-surface runoff to streams, and net infiltration past the root zone that ultimately becomes groundwater recharge. Average annual precipitation ranges from about 122 cm/yr in the north to about 142 cm/yr or more in the south (for example, Clark and others, 2011; Reitz and Kress, 2019). Evapotranspiration (ET) accounts for about 65 percent of the terrestrial water balance (Reitz and Kress, 2019). Previous estimates of average annual groundwater recharge have ranged from near 0 to 14.5 cm/yr, with values of around 5 cm/yr or less being common (for example, Broom and Lyford, 1981; Arthur and Taylor, 1990; Arthur, 2001; Reed, 2003; Clark and Hart, 2009). More recent studies by Reitz and Kress (2019), Hunt and others (2021), and Westenbroek and others (2021) have estimated higher values of around 10–13 cm/yr (about 8 percent of the water balance). It should be noted that the estimate of Westenbroek and others (2021), produced by SWB, denotes net infiltration past the root zone and not actual recharge to the water table (see also Nielsen and Westenbroek, 2023). Some net infiltration may discharge laterally to streams through perched aquifer systems or as interflow, resulting in a lower amount of recharge to the regional aquifer.

The remainder of precipitation—27 percent by Reitz and Kress’s (2019) estimate—runs off to streams. A key question for groundwater sustainability in the Mississippi Embayment and especially the Mississippi Delta is the amount of water that ultimately reaches the water table as recharge. Given the small size of the recharge component relative to evapotranspiration and surface water runoff and that recharge is estimated via water balance closure, small errors in evapotranspiration or surface water runoff could result in large errors in estimated recharge (for example, Reitz and Kress, 2019).

Groundwater Balance and Water Use

The groundwater system is fed primarily by terrestrial recharge. Stream leakage during periods of high stage and along major rivers that originate outside of the Mississippi Embayment constitutes an important secondary source of water, especially in localized areas. Groundwater within the Mississippi Alluvial Plain discharges primarily to pumping wells. Along the margins of the Mississippi Embayment, a greater part of groundwater discharges to streams.

Agricultural water use in the Mississippi Embayment varies widely depending on the year, location and crop, but is heavily concentrated in the alluvial plain (for example, Wilson, 2021). Irrigation accounts for most groundwater use: 87 percent or more in the alluvial plain (Clark and others 2011) and 97 percent of water use in the Mississippi River Valley alluvial aquifer (Lovelace and others, 2020). Estimates of average annual irrigation in the alluvial plain range up to 100 cm, with areal average values (including nonirrigated lands) of around 23–25 cm (Bristow and Wilson, 2023; Westenbroek and others, 2021). On irrigated lands, average annual irrigation estimates range from about 28 to 43 cm (Westenbroek and others, 2021). Annual irrigation for a single year can vary from the mean by as much as 50 percent depending on precipitation (Wilson, 2021).

The large amount of pumping relative to recharge has resulted in a loss of groundwater from storage and a chronic decline in water levels in many parts of the alluvial plain since at least the 1940s (see, for example, Barlow and Clark, 2011; or the water level record at National Water Information System [NWIS] site 333959090435001 near Cleveland, Miss.).

Effects of Surficial Geology on Groundwater

Previous mapping efforts (for example, McGuire and others, 2021a) indicate a correspondence between drawdown of groundwater levels within the alluvial plain and the presence of fine-grained, low-permeability deposits near the land surface. The surficial geology of the alluvial plain is complex and highly variable, ranging from coarse sands and gravels deposited by stream channels to fine-grained materials deposited in backwater areas. Stream channel meandering through time has produced variability with depth in addition to the spatial variability apparent in the recent AEM survey (Minsley and others 2021; fig. 3). Saucier (1994) provides a more detailed description of the depositional history and plates of mapped surficial geology. McGuire and others (2021a) provide detailed maps of current groundwater levels throughout the alluvial plain and its subregions.

Figure 3 shows surficial connectivity classifications developed by Minsley and others (2021) based on the thickness of low electrical resistivity layers detected in the AEM survey within the upper 15 meters (m) of sediment. Yellow colors indicate high electrical resistivities near the surface, which are interpreted to indicate a lack of fine-grained, low-permeability materials and therefore a strong connection between the surface and regional water table that can more readily transmit recharge or stream leakage. Purple colors indicate a high thickness of low electrical resistivity (and presumably low permeability materials) and therefore a poor connection between the surface and regional water table. The patchy, linear patterns in the classification zones that parallel the course of the Mississippi River reflect shifting depositional environments through time in response to stream channel meandering.

In contrast, the underlying Tertiary sediments that subcrop along the margins of the embayment (orange and gray colors) are relatively continuous because of their marine origin. Higher permeability units such as the Sparta Sand (part of the Middle Claiborne aquifer) that subcrop at the land surface can provide important sources of recharge to the regional aquifer (for example, Arthur, 2001); for example, in figure 3B the parts of the Bluff Hills underlain by the Sparta Sand (outlined in red) are apparent in the AEM results as having high surficial connectivity.

Where present, fine-grained deposits at the land surface serve to limit terrestrial recharge and interactions between surface water and the regional groundwater system. Conversely, coarse-grained materials at the surface may allow for focused groundwater recharge and especially leakage from perennial streams. Inspection of the potentiometric surface (fig. 3B) indicates a correspondence between low groundwater levels and low surficial connectivity. Fine-grained deposits at intermediate depths between the rooting zone and water table may allow for perched aquifer or variably saturated interflow systems that intercept net infiltration past the root zone and convey it to nearby streams, effectively bypassing the regional groundwater system. In times of high stream stages, such systems may also accept bank storage (leakage) from streams and later return the water to the streams in times of low stage. Such perched systems are difficult to simulate numerically in a regional groundwater model and therefore present a key challenge to effective simulation of the flow system as they create conditions of apparently high recharge and base flow that are inconsistent with regional groundwater levels. Examples of such systems in the Mississippi Embayment may include the L’Anguille River in Arkansas, which flows perennially yet originates in an area where the regional water table is too deep for the alluvial aquifer to provide any discharge, and the central part of the Big Sunflower River in Mississippi, which supports local groundwater levels that are often 6 m above the regional water table, as indicated at NWIS site 333251090323801 (USGS, 2021).

Finally, thick low-permeability deposits near the surface may create locally confined conditions and therefore locally higher drawdown because of the smaller amounts of water released by confined storage (aquifer compression) compared to unconfined storage (aquifer drainage). As indicated by figure 3B, the presence or absence of low permeability deposits near the surface may have a controlling influence on the groundwater flow system.

Regional Groundwater Flow System

Although the MERAS area receives streamflow input from many regional watersheds outside of the model domain, groundwater flow across its perimeter is assumed to be negligible because of the presence of the Midway confining unit near the surface (Clark and Hart, 2009) or assumed to be limited to local areas such as the interface with the Ozark System (Clark and others, 2018b). Within the MERAS area, groundwater generally flows from topographically high (and relatively low water use) areas along the margins towards the lower-lying streams and pumping centers within the interior of the embayment including the alluvial plain and the Mississippi Delta. Regional groundwater flow from the Tertiary system may enter the alluvial aquifer laterally through shallow flow paths or vertically through deeper flow paths where the bottom of the alluvial aquifer overlies aquifer units such as the Sparta Sand and Cockfield Formation of Claiborne Group (part of the Middle Claiborne and Upper Claiborne aquifers, respectively; for example, Minsley and others, 2021; Arthur, 2001).

Water-level declines have altered the flow system by reducing groundwater discharge to streams and in many cases reversing the hydraulic gradient such that streams that previously drained the alluvial aquifer now provide recharge on average through stream leakage (for example, the Big Sunflower system in Mississippi; Clark and others, 2011; Arthur, 2001). Flow directions have also been altered. In large parts of the Mississippi Delta and alluvial plain regions of Arkansas, groundwater that previously flowed towards streams now flows inward towards regional cones of depression (McGuire and others, 2021a; Clark and others, 2011; Arthur, 2001). In theory, declining groundwater levels in these areas reduce the aquifer thickness available to transmit water, which can in turn increase the loss of water from storage in a nonlinear feedback loop. Observed water level declines thus far have been mostly linear, with the exception of recent wet years in the mid to late 2010s that have slowed declines or produced rebounds in some areas (for example, NWIS site 333742090303801 [L0027] in Sunflower County, Miss.; fig. 3), indicating that groundwater sustainability in the alluvial plain may depend at least in part on future precipitation patterns.

In addition to multi-decadal trends, groundwater flow and groundwater/surface water interactions in the Mississippi Embayment and especially the alluvial plain are highly transient on a seasonal basis (see for example, Arthur, 2001). Stream hydrographs are generally flashy because of the high ratio of surface runoff to recharge. As a result, even moderately sized streams such as the Big Sunflower River near Merigold, Miss. (NWIS site 07288280), can experience swings in stage of as much as 10 m or more over periods of days to months. At any given point in time, groundwater/surface water exchanges are controlled by the stream stage relative to groundwater levels in the underlying aquifer. Precipitation events can cause stream leakage as stage rises more rapidly than surrounding groundwater levels. Subsequently, groundwater discharge to streams can occur if groundwater levels are high and stage subsides more rapidly. Conversely, dry conditions that would otherwise be base-flow dominated in unaltered areas can produce stream leakage to groundwater where groundwater levels are chronically below the stream stage and ultimately cause streams to run dry at times (for example, Barlow and Clark, 2011). In other instances, streams perched above the water table may be supported during dry periods by agricultural return flows, bank storage and local perched aquifer systems, or groundwater-derived base flows from headwater areas.

Groundwater Flow in the Mississippi Delta

Similar to other parts of the alluvial plain, the groundwater system in the Mississippi Delta receives water from terrestrial recharge through surficial sediments, stream leakage, and regional groundwater flow. The Mississippi River forms a hydraulic boundary along the western side that fully penetrates the alluvial aquifer although some localized underflow through the Tertiary units may be possible. The Tallahatchie/Yazoo system forms a major hydraulic boundary near the eastern edge. All other streams within the Delta originate within its boundaries. The Tallahatchie/Yazoo system is fed by four flood control reservoirs (Arkabutla, Sardis, Enid, and Grenada Lakes; fig. 1) in the Bluff Hills, which are managed to store water from large events (thereby dampening discharge peaks downstream) and release water during lower flow periods. Together, outflows from the flood control reservoirs account for approximately one third of the flow in the Yazoo River at the Redwood, Miss., streamgage (NWIS site 07288800). Regional groundwater flow into the Delta mostly comes from the Bluff Hills and areas to the east. Although the Tallahatchie/Yazoo river system intercepts at least some of this flow, it also likely acts as a source of recharge in many areas, especially in the central part of the Delta where the rivers run over coarse surficial deposits (for example, Minsley and others, 2021).

The area of drawdown in the central part of the Mississippi Delta receives groundwater flow from all directions. In predevelopment times, this flow would have drained to the Big Sunflower and Bogue Phalia Rivers (Arthur, 2001). Current water levels, however, are mostly below the beds of these streams; most groundwater flow in this area therefore discharges to irrigation wells.

Previous work has suggested that the overall net flow of water between the alluvial aquifer and Mississippi River is small, with large inflows from the Mississippi River during periods of high stage approximately balanced by large outflows to the river during periods of low stage (Arthur, 2001).

Aquifer Thickness and Hydraulic Properties

Within the MERAS footprint, the Mississippi River alluvial sediments mostly range from 25 to 50 m thick, with an average thickness of about 39 meters (see Minsley and others, 2021 for a map of thickness). In the Mississippi Delta, the saturated thickness of the alluvial aquifer averages about 34 m. Literature values for hydraulic conductivity in the alluvial aquifer generally range from tens of meters per day up to about 200–300 meters per day, reflecting bulk averages at the scale of groundwater models (for example, 1 square mile [mi²] cells; Arthur, 2001; Reed, 2003; Clark and Hart, 2009; Clark and others, 2013) or testing in wells completed in the productive intervals of the alluvial sediments (for example, Pugh, 2022; James Hoffman, Mississippi Department of Environmental Quality, written commun., 2020). As noted previously, the Mississippi River alluvial sediments are highly heterogeneous, ranging from coarse gravel channel deposits to very fine backwater deposits. Local values of hydraulic conductivity could therefore be substantially higher or lower than the literature estimates. In general, hydraulic conductivities of the fine-grained backwater deposits are not well characterized.

The Tertiary deposits below and adjacent to the alluvial aquifer vary greatly in thickness throughout the MERAS footprint, from thin to absent along the margins to more than 1,000 meters thick in the central and southern areas of the domain. As noted previously, the Tertiary deposits are thought to be less heterogenous within mapped units, but widely varying in permeability between units. Literature values of hydraulic conductivity from previous modeling studies range from less than 1 meter per day in the confining units to about 40 meters per day in parts of the Middle Claiborne aquifer (Clark and Hart, 2009; Clark and others, 2013).

Model Domains and Discretization

An extensive description of the model construction is given in appendix 1; a simplified version is presented here. The spatial extents for the MERAS 3 and Mississippi Delta models are shown in figure 1. The MERAS 3 model covers the same extent as previous MERAS models (Clark and Hart, 2009; Haugh and others, 2020a; Hunt and others, 2021; fig. 1) at a 1-km uniform grid resolution that aligns with the USGS National Hydrogeologic Grid (Clark and others, 2018b). Vertically, the MERAS 3 model consists of three layers that distill the regional aquifer system into three parts: the Mississippi River Valley alluvial aquifer (layer 1); a lumped confining unit consisting of the Vicksburg-Jackson confining unit, the Upper Claiborne aquifer, and the Middle Claiborne confining unit (layer 2); and a lumped lower aquifer (layer 3) that includes the Middle Claiborne aquifer and all underlying units in the Tertiary sequence down to the top of the Midway confining unit (fig. 4).

The Mississippi Delta model includes the geographic extent of the Mississippi Delta (fig. 1) and the outlets of the flood control reservoirs in the Tallahatchie/Yazoo River system at a uniform horizontal resolution of 500 meters that also is aligned with the National Hydrogeologic Grid (Clark and others, 2018b). Layering in the Delta model consists of uniform 5-m-thick layers down to the base of the alluvial aquifer (bottom of layer 8); uniform layering of expanding thickness below the alluvial aquifer bottom to the deepest extent of AEM data; and layer surfaces based on the hydrostratigraphic surfaces of Hart and others (2008) between the deepest AEM data and the top of the Midway confining unit (see app. 1; fig. 5).

Time discretization for both models begins in 1900, with an initial predevelopment steady-state period (without pumping) followed by 6 multiyear stress periods extending to April 2007. The multiyear stress periods are structured to approximately align with step changes in the pumping history in previous MERAS models (Clark and Hart, 2009, fig. 10). Starting April 1, 2007, and extending to the history matching end time of January 1, 2019, the models have monthly stress periods of one timestep each (see app. 1, table 1.1).

Boundary Conditions

Boundary conditions in the MERAS 3 and Mississippi Delta models include terrestrial recharge originating from precipitation, groundwater/surface water interactions, and pumping fluxes representing water use. Similar to past MERAS models, MERAS 3 assumes a no-flow boundary around its perimeter. The Mississippi Delta model perimeter consists of transient fluxes extracted from the independent MERAS 3 model solution. Interior boundary conditions are summarized as follows (with more detail given in app. 1):

Subsurface Properties

Aquifer hydraulic conductivity and storage properties were represented structurally in the MERAS 3 and Mississippi Delta models at multiple scales using a blend of the conceptual models produced by the AEM survey by Minsley and others (2021) and the original MERAS framework by Hart and others (2008). In both models, aquifer properties were represented at the coarsest level using piecewise-constant zones based on the logarithmically binned electrical resistivity facies classes by Minsley and others (2021). In the MERAS 3 model, electrical resistivities were vertically averaged to develop zones for the single layer representing the alluvial aquifer. In both models, areas lacking AEM data were zoned based on the relation of their cell centers to the hydrostratigraphic surfaces of Hart and others (2008). Finer-scale variability in horizontal and vertical hydraulic conductivity and specific yield (in the AEM based zones only) was incorporated within zones using pilot points spaced every 5 km as described in appendix 2. Initial streambed leakance input to the SFR Package was estimated on a cell-by-cell basis using results from waterborne surveys of subsurface electrical resistivity (Leaf, 2023; Adams and others, 2019; Killian, 2018). Additional details describing the generation of the subsurface properties and how they were parameterized for both models can be found in appendix 2. Maps of the final aquifer properties for the models are also shown in appendix 2.

Parameter Estimation and Uncertainty Analysis

After constructing the models, parameter estimation was performed via history matching (for example, Anderson and others, 2015, chap. 9). The models were parameterized to allow for systematic adjustments to their inputs in response to model fit to field observations (history matching). A more detailed description of the parameter estimation methods and results is given in appendix 2. The overall approach to history matching is similar to that of Corson-Dosch and others (2022).

Parametrization of the models followed a multiscale approach as described in White and others (2021) and references therein. At the coarsest scale, piecewise-constant parameter values were estimated for zones covering large areas of the model. At successively finer scales, one or more multiplier parameters were then applied cumulatively on top of the coarse scale parameters. This approach apportions model input uncertainty at different scales, which can improve history matching, reduce overfitting and improve detection of model error phenomena such as parameter compensation (White and others, 2021). For example, the coarse scale parameters address large-scale biases in the model input, while the finer multipliers can address the effects of local-scale heterogeneity. Setup of the multiscale parametrization was enabled by the PstFrom routines in the pyEMU Python package (White and others, 2016; White and others, 2021).

Horizontal hydraulic conductivity, hydraulic conductivity vertical anisotropy, specific storage, specific yield, aquifer recharge, surface runoff and specified inflows to the SFR Package, streambed leakance, and water use (pumping) were all parametrized. Coarse-scale parameters included piecewise-constant zones representing spatial scales ranging up to the extents of the models. Intermediate scale parameters included multipliers on the parts of individual zones within a layer or within a geographic area. Fine-scale parameters included pilot points and stream-reach multipliers at approximately the kilometer scale, and temporal multipliers that applied to certain boundary conditions within a stress period. Table 2.2 provides a summary of the model parametrizations.

Field observations of aquifer heads and stream flows were processed into multiyear or monthly averages corresponding to model stress periods. An objective function consisting of the sum of squared weighted residuals between observed values and their simulated equivalents (phi) was assembled. Observations were initially weighted based on measurement uncertainty, and grouped by observation type (heads, flows, etc.), geographic area and other criteria (table 2.1). Subsequently, observation weights in each group were multiplied together to achieve a desired balance in the objective function that resulted in an acceptable model fit to the primary observations of interest.

Initial trial-and-error model runs focused on manual adjustment of coarse-scale parameters to improve history matching and gain insight about the key aspects of the groundwater flow system. The iES algorithm implemented in PEST++ (White, 2018; White and others, 2020) was then used to formally estimate both coarse- and fine-scale parameters. In the iES, an ensemble of parameter sets (realizations) is carried through the analysis. At each step, empirical correlations between the parameter ensemble and changes in observation values are used to iteratively improve model fit to observations while constraining the parameter values (reducing uncertainty). The iES results therefore provide estimates of parameter uncertainty in addition to a minimum error variance set of “base” parameter values that can be used as a “best” parameter set.

Successful history matching required numerous iterations following a stepwise approach in which the iES was run, the results were evaluated, and the model structure, parametrization, observations, or observation weighting were adjusted in response to the results. In particular, the history matching focused on a set of “priority wells” that were distributed around the Mississippi Delta and familiar to cooperators, as well as the overall fit to all head observations, especially in the central Delta region of greatest drawdown (fig. 2). History matching was considered complete when a satisfactory fit was achieved to head observations with a coherent mass balance, reasonable stream stages, and reasonable model input values that were consistent with the conceptual model(s) for the Mississippi Delta and larger Mississippi Embayment region.

History Matching Results

History matching results are described in detail in appendix 2; for brevity, only time series of observed and simulated heads and streamflows of interest in the Mississippi Delta (simulated by the Delta model) are shown here.

In general, heads at the priority wells of interest are well matched in terms of absolute water levels and the seasonal fluctuation between high water levels in April and low October water levels at the end of the growing season (fig. 6). The “base” member of the ensemble shows the results from the best-fit parameter set, and the remaining ensemble members show the possible spread (or uncertainty) in the simulation results. Differing degrees of influence from surface water can be seen in these hydrographs from sites scattered around the Delta geographically (fig. 2.1). The two Sunflower County wells (L0027 and B0003; figs. 6B, 6C) are respectively near the center and edge of the central Delta area of greatest drawdown (figs. 2 and 3) and are relatively distant from surface water influence. A lack of surface water connection because of a surficial confining unit (fig. 3) is thought to be a key driver of drawdown in this area. Water levels at these sites exhibit less seasonal variability than other wells such as O0037 (fig. 6G), and the spread of water levels simulated by the ensemble is wide, indicating differing degrees of surface water connection allowed by the different parameter sets. Identifying a suitable parametrization structure and parameter values to match observed drawdowns in this area posed a key challenge in past studies too as indicated in figure 6B and 6C where there is a larger difference between the green lines (from the MERAS 2.1 and MERAS–NWT models by Haugh and others [2020a] and Hunt and others [2021]) and the measured values.

In contrast, the Leflore County (O0037; fig. 6G) and Humphreys County (F0059; fig. 6D) sites illustrate strong connections with surface water. Drawdown at these two locations is minimal, with heads close to the average levels of nearby surface water. Experimental model runs indicated that the fit for the Leflore site was mostly controlled by the streambed leakance in nearby reaches of the Yazoo River, which is approximately 500 m away (and 100 m above the open interval for O0037). As a result, the hydrograph for this well exhibits more seasonal influence from the river but also is well fit by the model. Variability in the ensemble results is low at this site because simulated stages are similar across all members. The Humphreys site is similar; it is only 250 m away from the Yazoo River (with a shallow screen), but there is a greater mismatch between the simulated and observed time series, indicating a structural defect in the model at this location. In other words, an essential process driving the hydrograph at this site is apparently not captured in the model.

The Bolivar County site (F0020; fig. 2.1F), which is about 9 km from the Mississippi River and about 4 km from the Bogue Phalia River, represents an in-between condition that is relatively well matched by the model. A somewhat wider set of ensemble results here indicates variability in hydraulic conductivity and recharge parametrization across the ensemble members.

Figure 7 shows time series of monthly mean observed stream flows and their simulated equivalents for 2007 through 2018 for six priority streamflow gages in the Delta area. The “base” member of the ensemble shows the results from the best-fit parameter set, and the remaining ensemble members show the possible spread (or uncertainty) in the simulation results. Streamflow observations were not weighted highly in the history matching (table 2.1) because of incomplete representation of key streamflow generating processes in the model including the approximate nature of the runoff estimates from SWB, missing intermediate processes including perched groundwater, interflow and bank storage, and lack of information on agricultural return flows and their contributions to base flow during low-flow periods. Logarithmic y-axes are provided here to better illustrate match across the full range of stream flows.

The general agreement between simulated and observed values seen in all the plots indicates a realistic stream water balance in the model; however, the missing streamflow generating processes are evident in the bias towards undersimulation, especially in the low-flow periods within the three Sunflower River sites’ time series (figs. 7A, 7B, 7C). The Tallahatchie and Yazoo River time series (figs. 7E, 7F) have less undersimulation bias, because a substantial part of their flow (about 50 and 30 percent, respectively) comes from the Arkabutla, Enid, Sardis, and Grenada flood control reservoirs. Outflows from the flood control reservoirs are measured daily by the U.S. Army Corps of Engineers and were specified in the model.

Figure 8 shows time series of streamflow gain between three pairs of upstream and downstream locations and their simulated equivalents. In many streams, gains or losses of streamflow between sites can provide valuable integrated measurements of groundwater discharge to surface water and, ultimately, recharge into the groundwater system. In the Mississippi Delta, however, interpretation of streamflow gains or losses is less straightforward because of the confounding effects of poorly characterized intermediate runoff processes and an often small groundwater discharge component. In the case of the Sunflower River between sites 07288280 and 07288500 (fig. 8A), the river stage is mostly above the water table (indicating stream leakage to groundwater), but the difference in the monthly mean flows measured at the two sites generally indicates an increase in streamflow. Much of the increase is probably because of runoff from storm events that is included in the monthly means, but previously mentioned intermediate flow processes (perched groundwater and bank storage) and agricultural return flows may also be contributing to the gains. The measured record does show a few instances (for example, in early 2010 and 2011) of a net loss in flow between the two sites, but there are many more of these events simulated by the model, which is an artifact of the missing processes in the model. More frequent months of streamflow loss and the very low flows or drying simulated at 07288500 (fig. 7C) mean that any future flow conditions in the Sunflower River simulated by the model are probably conservative (less than what might be expected in reality for a given scenario).

The gain in flow across the Tallahatchie/Yazoo system downstream from the flood control reservoirs (fig. 8B) and the total outflow from Steele Bayou (fig. 8C; which includes the Sunflower and Bogue Phalia Rivers) show that overall, the stream water balance is well represented in the model. Periodic sharp reversals in the measured flow record are associated with backwater periods, when high water levels in the Mississippi River cause water to flow upstream.

Average parameter estimates for the two models are generally consistent with previous work (table 1). A 24-percent higher simulated mean annual recharge for the Mississippi Delta model compared to Arthur (2001) may reflect relatively wet conditions during the history matching period, especially after 2012. Substantially lower recharge simulated in the MERAS 2.2 model (Haugh and others, 2020a, b) may reflect parameter compensation to limit excessive rise of the water table between sparsely represented streams (Leaf, 2023). Different storage values estimated for the MERAS 3 model compared to the Mississippi Delta model and other models may reflect local heterogeneity in the various storage mechanisms (pore drainage, confining unit leakage, and aquifer compression) that is not well resolved in the three-layer representation of the system. Vertical hydraulic conductivity is difficult to assess because it is not easily measured in the field and, more importantly, is scale dependent.

Historical Mississippi Delta Model Results

Results for the historical version of the Mississippi Delta model (before 2019) are presented here; selected results for the MERAS 3 model are presented in appendix 3. Figure 9 shows simulated water table elevations and groundwater/surface water interactions in the Mississippi Delta. Under relatively dry conditions such as those at the end of September 2016 (figs. 9A, 9B), when stream stages are low, many streams, including the Mississippi River, receive groundwater discharge. Other streams are dry or leaking because the regional water table is below the streambed or stream stage. In the central Delta, streams generally always lose water to the alluvial aquifer, though they may intermittently receive groundwater discharge from perched water, including bank storage (which is not simulated in the model). Lighter shades of red or gray in this area indicate lower rates of stream leakage because of the intervening presence of the shallow confining unit. Under wet conditions such as those at the end of February 2018 (figs. 9C, 9D), when stream stages are high, most streams leak water to the groundwater system, except for some streams in the Bluff Hills. Figure 3.1 shows similar results for the larger alluvial plain simulated by the MERAS 3 model.

Figure 10 shows average annual net groundwater budgets for the Mississippi Delta. The years before 2007 in figure 10A represent averages for the multiyear spin-up stress periods (table 1.1). Historically, streams in the Mississippi Delta were mostly gaining from groundwater flow (see fig. 10, 1900 and 1950 budgets), but pumping since the mid-twentieth century has shifted the balance to be more variable on an annual basis between net leakage and net discharge (see also fig. 11). In dry years, such as 2007, 2012, and 2010, the loss of aquifer storage can constitute a significant source of water to meet increased pumping demand and groundwater discharge to streams at low stages. Conversely, in wet years such as 2009 and 2018, some aquifer storage is replenished by increased recharge, stream leakage to groundwater, and reduced pumping demand. Over the long term, however, aquifer storage has declined, resulting in the observed drawdowns in groundwater levels. In recent years (since about 2012), declines in water levels have leveled off or even rebounded at some sites (for example, L0027 in Sunflower County and F0020 in Bolivar County; figs. 6C, 6F), presumably because of wetter-than-normal conditions leading to increased recharge and stream leakage and less irrigation pumping. The hydrographs in figure 6 show the model generally capturing this trend.

During the history matching period of 2010 through 2018, terrestrial recharge constituted the largest inflow component at 56.8 percent (fig. 10B) of the overall groundwater budget or 3.2 annual inches (fig. 2.20; see app. 2 for additional discussion of estimated recharge). Groundwater flow from the Bluff Hills was second at 34.9 percent and seems to be relatively steady in absolute terms across years. A 7.6-percent net inflow from aquifer storage indicates a net loss of aquifer storage through the history matching period. Groundwater/surface water interactions with interior streams (excluding the Mississippi River) were nearly balanced between groundwater discharge to streams and stream leakage to groundwater. Similar to previous studies (for example, Arthur, 2001), the net groundwater flow to the Mississippi River is close to zero. Note that the Mississippi River component here also includes potential underflow to or from Arkansas, which cannot be distinguished from flows to or from the Mississippi in the current modeling framework, but regardless, net groundwater flow to the Mississippi is most likely small on an average net basis.

Budget results from the SFR Package (fig. 11) of the Mississippi Delta model illustrate the sources of streamflow in the Mississippi Delta model. Outflows from the four flood control reservoirs to the Tallahatchie and Yazoo Rivers (external inflows term in fig. 11) constitute approximately one-third of all streamflow. Groundwater discharge makes up approximately 20.9 percent of the remaining streamflow generated within the model area, with the remaining 79.1 percent coming from various other runoff processes. Inspection of the SFR Package budget for the Mississippi Delta footprint, which excludes the eastern tributaries to the Tallahatchie/Yazoo system, yields a similar proportion of groundwater-derived streamflow (24.9 percent; not shown).

Future Climate Scenarios and Forecast Hydrologic Effects

A key motivation for developing the Mississippi Delta model was to provide a tool to quantify groundwater responses to a range of potential future conditions. To demonstrate the model’s utility for this objective, a baseline scenario was developed to forecast groundwater levels out to 2056 under a range of potential future climate forcings. The baseline scenario assumes that current agricultural practices including the amount of irrigated acreage, crop types, and current levels of irrigation efficiency will continue but that the amount of irrigation pumping may change in response to changes in future climate. As such, the baseline scenario can be used to evaluate the potential effects of climate change on groundwater levels and as a basis for evaluating potential management scenarios such as changes in irrigated acreage, improvements in irrigation efficiency, changes in crop types, or water transfers. Input files and a reproducible workflow for setting up the baseline future climate scenario are provided in an associated data release (Leaf and others, 2023).

Future Model Forcings

Ten potential future climate forcings were applied to the Mississippi Delta model via the SWB model by Nielsen and Westenbroek (2023). Downscaled daily precipitation and temperature output (Localized Constructed Analogs method; Pierce and others, 2014; 2015) were obtained from the downscaled Coupled Model Intercomparison Project Phase 3 (CMIP3) and 5 (CMIP5) climate and hydrology projections (Bureau of Reclamation, 2022). Five General Circulation Models (GCMs) were selected for each of two future climate scenarios representing “middle” (representative concentration pathway [RCP] 4.5) and “high” (RCP 8.5) greenhouse gas emissions pathways (for example, Moss and others, 2010). The models were selected from the larger group of CMIP5 models included in the National Climate Change Viewer (Alder and Hostetler, 2013) based on the relative changes in mean precipitation and temperature simulated for the period of 2025–49 compared to 1981–2000. “Cool” (least temperature increase), “warm,” “wet,” and “dry” endmembers were selected from the RCP 4.5 and 8.5 scenarios, as well as the model that most closely matched the mean simulated temperature and precipitation changes for the state of Mississippi. Table 4.1 in appendix 4 summarizes the combinations of GCMs and scenarios included in the future climate ensemble used in this study.

Future climate forcings for Sunflower County, Miss., projected by all the CMIP5 models included in the National Climate Change Viewer (Alder and Hostetler, 2013) are summarized in figure 12. Although the magnitude varies, there is a clear consensus among the models that temperature increases observed since the late 20th century will continue, with the rate of increase diverging between the RCP 4.5 and 8.5 scenarios after 2050. Precipitation is more ambiguous, with substantial uncertainty and a flat central tendency out to 2100.

Gridded daily precipitation and minimum and maximum temperature output from the future climate ensemble of GCM/scenario combinations were used to drive 10 SWB simulations from 2020 to 2056 (Villers and Ladd, 2023). Results from the 10 future SWB runs were applied to the base ensemble member from the second iteration of the iES history matching run (containing the “best” parameter set). Only the base ensemble member was included in the future scenarios, as the effect of future climate forcings on model forecast uncertainty is typically much greater than the effect of model parameter uncertainty (for example, Hunt and others, 2013). This resulted in a future climate ensemble of 10 members (table 4.1). Model forcings not obtained from SWB (including perimeter groundwater fluxes, specified inflows to the SFR Package, and River Package stages) were developed from historical values, by repeating monthly averages for the period of 2010–15, which includes wet and dry years.

Daily net infiltration and surface runoff were averaged to the monthly stress periods and mapped to the nearest model grid cells (each 1-km SWB cell corresponds to exactly four 500-m groundwater model cells). All applicable time-invariant multiplier parameters (for example, those representing surficial connectivity zones or pilot points) were then applied to the mapped SWB output, using the same workflow as the historical model. Future pumping rates from SWB based on crop water demand were applied to the groundwater model in the same way, with layers for the pumping wells selected based on estimated production intervals (Torak, 2023; Knierim and others, 2019). To account for an apparent bias in historical SWB irrigation estimates relative to the AIWUM, a multiplier of 1.2 was applied to the SWB irrigation estimates used after 2019.

Potential Changes in Groundwater Levels, Streamflow, and Saturated Thickness

Figure 13 compares mean annual net water budgets simulated by the future climate ensemble (see table 4.1 in app. 4 for descriptions of each of the ensemble members) with the net water budget for the historical base period 2010–19. The ensemble members simulate varying amounts of potential future precipitation and temperature increase and, therefore (via SWB), differing amounts of net infiltration, runoff, and irrigation pumping. In general, the warm and dry climate scenarios (table 4.1) simulate greater amounts of pumping, lower amounts of recharge, and greater amounts of stream leakage and aquifer storage loss.

Figure 14 shows future groundwater levels simulated at the priority wells of focus (locations shown in fig. 2.1). The results vary depending on proximity to surface water. Wells that are well connected to surface water such as the Leflore County (O0037) and Humphreys County (F0059) sites (figs. 14C, 14D) show little drawdown into the future and little variability among the climate ensemble members. In contrast, the Sunflower County wells (L0027 and B0003; figs. 14E, 14G) that are less connected to surface water show a wide range of potential future drawdowns. This illustrates a relatively large uncertainty range (in drawdown) from future climate at many sites (about 10–15 m at B0003 and L0027) compared to the uncertainty range from model parameters (about 2–4 m at B0003 and L0027; illustrated by the ensemble results in fig. 6).

Figure 15 shows annual minimum monthly stream flows (the lowest monthly mean in each year) in the Sunflower River at USGS site 07288500, near Sunflower, Miss. (see fig. 2.2 for the location). The historical record (in black) shows substantial year-to-year variability but an overall declining trend consistent with the decline in groundwater levels. In recent years, the decline has resulted in periodic drying of some Sunflower River stream reaches during the summer months (Barlow and Clark, 2011). Future annual minimum monthly stream flows forecast by the climate ensemble show a continuation of the declining trend, with at least one month of contiguous dry conditions a year predicted in all the scenarios by mid-century. Although the stream flows simulated by the groundwater model may be biased low because of a lack of representation of some surface water processes (interflow, storage, and irrigation returns), they nevertheless indicate an increase in the frequency and duration of drying events.

Maintaining sufficient saturated thickness in the alluvial aquifer to allow for the production of irrigation water is a key concern of stakeholders. Figure 16 shows forecast changes in saturated thickness, determined as the difference between the simulated water table and the alluvial aquifer bottom surface (James and Minsley, 2021), between the end of the history matching period in 2019 and 2056. Sixty feet (ft) is often applied as a saturated thickness that provides sufficient capacity for typical alluvial plain irrigation pumping. Overall, across the 10 climate scenarios considered, the area fraction of the Mississippi Delta with at least 60 ft of saturated thickness could decline from near 100 percent currently to 80–90 percent by mid-century.

The largest declines in saturated thickness occur in the western part of the Delta between the Bogue Phalia and Mississippi Rivers. In this area, the alluvial aquifer is underlain by the Yazoo Clay of the Jackson Group and may also include lower permeability deposits or have a higher bottom elevation than reported by James and Minsley (2021); for example, the cross section in figure 5A indicates the presence of electrical resistivity facies class 3 (interpreted to represent lower hydraulic conductivities) at shallow depths approaching sea level near the Mississippi River (left side of the cross section). In plan view, figure 2.16 shows lower electrical resistivity facies classes and lower hydraulic conductivity estimates in this area (near the west side of the model domain) between approximate model rows 300 and 400. In addition to the increased drawdown that would be expected at lower permeabilities, a thin aquifer could promote accelerating drawdown rates with time, as a lowering of the water table results in a declining saturated thickness (transmissivity), which feeds back into larger decreases in the water table. Continued decline also occurs in the central Delta, and areas to the west between the Big Sunflower and Bogue Phalia Rivers. By mid-century, parts of this area may be at or below the saturated thickness threshold for sustaining irrigation pumping.