Groundwater Recharge in a Fire‐Adapted, Semi‐Arid Forest: A Watershed Water Balance Approach

Abstract

Climate change induced aridity and Euro‐American settlement have altered the historical disturbance and flow regimes of large portions of the ponderosa pine forests of northern Arizona. The increased occurrence of high‐severity wildfires due to these changes has led to the establishment of various forest restoration programs to protect the region's forests and their watersheds. In 2014, a paired‐watershed monitoring project was implemented to compare the impacts of differing levels of forest thinning to watershed hydrology in seven experimental watersheds nested within the Upper Lake Mary (ULM) watershed in Arizona. This study expands the calibration phase of the ULM paired‐watershed by synthesizing historic precipitation, surface runoff, groundwater recharge, soil moisture data, and evapotranspiration (ET) data to perform regression analyses and create a holistic water balance for each watershed. The magnitude and timing of seasonal groundwater recharge events were quantified for the first time in this region using a water table fluctuation method. The results showed that recharge did not occur every year and was heavily dependent (P < 0.05) on total winter season precipitation and snowpack duration. On average, recharge composed 9% of the total water budget when present. The results of this study lay the foundation for a greater understanding of how forest restoration alters northern Arizona's forest hydrology and will provide crucial information that should be used in water policy and water resource decision‐making as the region plans for future water availability.

Introduction

Global climate change continues to stress groundwater resources and public health as temperature and aridity increase (USGCRP ²⁰²³). Extreme events, including longer drought periods, dangerous heat waves, and more frequent catastrophic wildfires, are a few direct results (USGCRP ²⁰²³). Additionally, water‐limited arid and semi‐arid regions, such as the western United States, are more likely to experience water stress, including increased drought severity, higher evaporative demand, reduced river flow, and decreased groundwater recharge (Das et al. 2011; Niraula et al. ²⁰¹⁷; Milly and Dunne ²⁰²⁰; Overpeck and Udall ²⁰²⁰). The western United States is primarily dependent on mountain snowmelt from winter precipitation to replenish surface water, soil water, and groundwater resources. Studies conducted in the western U.S. have shown that snow‐water equivalent (SWE) in the region has decreased by ∼30% since the mid‐20th century (Mote et al. 2018) and predict up to 60% loss of total SWE within the next 30 years (Fyfe et al. 2017). While the region continues to struggle with declining renewable water availability due to climate change, it continues to experience the largest population growth in the country (DeJohn ²⁰²³), providing challenges for water resource planning.

Rising temperatures and concurrent dry periods, coupled with increased forest tree density from historical fire suppression and Euro‐American settlement, are also responsible for an increase in the severity, frequency, and duration of wildfires across the globe, particularly in semi‐arid forests (Covington and Moore 1994; Abatzoglou and Williams ²⁰¹⁶; Holden et al. ²⁰¹⁸; Mueller et al. ²⁰²⁰). High‐severity wildfires pose a significant threat to human health by reducing air quality through the release of pollutants within wildfire smoke (Reisen et al. 2015; Burke et al. ²⁰²⁰) and cause substantial economic losses by destroying human infrastructure and residential communities (Moritz et al. 2014; Kramer et al. ²⁰¹⁹). Nearly 49 million residential homes in the United States reside in the wildland–urban interface (WUI), an indicator of wildfire risk (Burke et al. 2020). High‐severity wildfires can also diminish short‐ and long‐term water quality within forested watersheds that serve as drinking water sources for nearby urban and rural communities by increasing nutrient loads and turbidity within water supplies due to increased post‐fire soil erosion (Smith et al. 2011; Hohner et al. ²⁰¹⁹). Post‐fire alterations to water quality present new challenges and uncertainties in protecting potable water supplies for resource managers across the region (Bladon et al. 2014). To respond to increasing extreme wildfire occurrences, forest restoration programs have been developed to restore the historical disturbance and flow regimes of these fire‐adapted semi‐arid forests. One such program is the Four Forest Restoration Initiative (4FRI) in Arizona, USA. Covington et al. (1997) described the combination of mechanical thinning and prescribed burning treatments that could reduce wildfire severity and suggested that it would also increase moisture availability by reducing tree competition in ponderosa pine forests in the region. In this article, the terms forest restoration and treatment are used interchangeably.

Forest restoration practices influence every component of the water budget (Wyatt et al. 2014; Moreno et al. ²⁰¹⁶; Saksa et al. ²⁰¹⁷; O'Donnell et al. ²⁰²¹; Jones et al. ²⁰²²). Dense mature second‐growth forests, such as those found in Arizona, exhibit high canopy interception, high evapotranspiration (ET), and moderate water storage in soils (Howe and Hendrix 1951; Jones et al. ²⁰²²). In the early stages of forest succession, such as those following restoration, there is decreased canopy interception, decreased ET, higher infiltration and soil water storage, as well as increased groundwater recharge and surface runoff (Baker 1986; Dore et al. ²⁰¹²; Wyatt et al. ²⁰¹⁴; O'Donnell et al. ²⁰¹⁸; O'Donnell et al. ²⁰²¹; Jones et al. ²⁰²²). Schlesinger and Jasechko (2014) suggested that transpiration (T) is 55 ± 15% of total ET in temperate coniferous forests; however, this ET amount is highly variable depending on local conditions, outlining the challenges associated with ET calculations. The contribution of T to ET has been directly linked to leaf area index (LAI; Wang et al. ²⁰¹⁴), with higher LAI resulting in larger contributions of T (Wei et al. 2017). Reducing total forest canopy density through forest restoration aids in the reduction of ET by reducing canopy interception and leaf area index, which increases water availability for surface runoff and soil moisture due to more precipitation reaching the forest floor (del Campo et al. ²⁰²²). Soil moisture has been shown to respond to burning and thinning practices in ponderosa pine forests, with marked increases in soil moisture measurable decades after posttreatment (O'Donnell et al. ²⁰²¹). Restoration efforts do not lead to permanent, long‐term changes to semi‐arid forest hydrology due to vegetative regrowth posttreatment. Continued restorative maintenance must be conducted to maintain any desired alterations to hydrologic function.

Groundwater is an essential water resource across the region and has historically been utilized by semi‐arid communities and Indigenous peoples as a resilient water resource (Famiglietti 2014; STACCWG ²⁰²¹). However, few studies have quantified how partial thinning and burning restoration treatments impact groundwater recharge in semi‐arid forests (Schenk et al. 2020). One reason is that recharge fluxes occupy a small proportion of the water budget compared with other variables such as evapotranspiration. Aldridge (2018) utilized the chloride mass balance approach of measuring recharge in meadows, ephemeral channels, and unthinned and thinned ponderosa pine forests. The results of the study provided reasonable estimations of recharge and concluded recharge is heavily controlled by available precipitation and vegetation density. Wyatt et al. (2014) employed a groundwater‐flow model to map predicted changes to groundwater recharge resulting from restoration treatments as a part of the 4FRI project in northern Arizona. Results suggested a small increase (∼3%) in recharge because of the forest restoration practices; however, supplemental climate models point to the overall decrease of recharge over the next several decades because of anticipated decreases in available precipitation for recharge and increases in water lost to ET (Maurer et al. 2007). Recharge increases only for a limited amount of time after forest restoration until vegetation succession following treatments reduces its effects. This finding further emphasizes the need for long‐term monitoring of hydrological responses to forest treatment (Wyatt et al. 2014; Aldridge ²⁰¹⁸).

In 2015, O'Donnell et al. (2015), along with the City of Flagstaff, Arizona (COF), designed the Upper Lake Mary (ULM) paired‐watershed study, which remains in the calibration phase, to analyze the effects of 4FRI forest burning and thinning treatments on water yields in seven subwatersheds that contribute directly to local surface and groundwater drinking water reservoirs. The paired watershed approach is an effective method for measuring how forest restoration modifies a watershed's hydrologic function (e.g., Stednick ¹⁹⁹⁶; Brown et al. ²⁰⁰⁵; Goeking and Tarboton ²⁰²⁰; Dymond et al. ²⁰²¹; del Campo et al. ²⁰²²). Baker (1986) conducted a long‐term paired‐watershed study in the Beaver Creek drainage of Arizona, USA, to compare how commercial forest management practices affected water yields (streamflow) in ponderosa pine forests. The study measured significant initial mean increases of water yields within the study watersheds around 15–45% following 30–100% overstory removal; however, watersheds with 77% overstory removal experienced the longest duration of significant increases to water yields. Water yield responses on watersheds with southern aspects were shorter lived (∼6–10 years) than northern aspects (>10 years). The results of Baker (1986) heavily informed the development of the ULM paired‐watershed study. The study presented here expands upon the calibration phase of the ULM paired‐watershed study by creating annual water balances that include the contributions of groundwater recharge and streamflow within the experimental watersheds. The objectives of the study were to (1) create annual water balances for the paired‐watersheds using meteorological and hydrological data from the 2015–2023 water years, (2) measure the magnitude, timing, and climatological conditions that lead to groundwater recharge within the study area; including the 2024 spring recharge season, and (3) determine how climatological conditions lead to increased or decreased water yield in these watersheds.

Methods and Materials

Site Description

The Upper Lake Mary (ULM) paired‐watershed study area, located ∼19–24 km (12–15 miles) southeast of Flagstaff, Arizona, USA (Figure 1), was established in 2014. The study area consists of seven separate first‐order subwatershed sites (LM‐1, LM‐2, LM‐2B, LM3L, LM3U, LM4, and LM5) that range in size from 1.84 to 4.53 km² (0.71–1.75 mi²) (Table 1). LM3L and LM3U are nested watersheds, meaning streamflow from the upper gradient LM3U watershed directly drains into the lower LM3L watershed. Each study site is proposed to have differing levels (low, moderate, or high) of mechanical thinning and prescribed burning forest restoration treatments, conducted after calibration of the paired watershed relationship; except for LM‐1, which acts as the reference site. LM3U and LM3L are nested watersheds, selected with the intent of comparing high and low treatments to watershed hydrology (Masek Lopez et al. 2013; Davis ²⁰²²).

Figure 1.

Study area map of (a) United States with Arizona highlighted, (b) Map of Arizona with Coconino County highlighted and general study area outlined, (c) ULM paired watersheds with marked points of interest and important faults. NOAA Flagstaff Pulliam Airport weather station is ∼5 miles northwest of the map extent. Upper Lake Mary is contained within the Lake Mary Graben between the two mapped faults. Huebi (https://commons.wikimedia.org/wiki/File:Map_of_USA_AZ.svg). “Map of USA AZ,” https://creativecommons.org/licenses/by‐sa/2.0/legalcode.

Table 1.

Summary Table of Geographical Information Measured Using DEM Rasters From the USGS Earth Explorer tool and processed using ArcGIS Pro. Proposed Treatment Levels for Each Subwatershed are Listed (DEM Entity IDS: SRTM1N35W112V2, SRTM3N34W112V2).

Watershed ID	Area (km2)	Max/Min Elevation (m)	Average Slope (%)	Treatment Level
LM1	2.4	2240/2150	1.32	Reference
LM2	4.5	2348/2194	2.43	Low
LM2B	1.9	2343/2164	1.52	Low
LM3L	3.9	2233/2157	0.98	High
LM3U	1.8	2233/2177	1.22	Low
LM4	2.3	2203/2115	1.47	High
LM5	3.3	2216/2139	2.35	Medium

Northern Arizona resides on the southwestern edge of the Colorado Plateau at an elevation of ∼2130 m (7000 ft) and hosts the largest continuous stand of ponderosa pine trees in the world. The region experiences cold winters and warm summers with average temperatures of −1 °C (30 °F) and 18 °C (64 °F) respectively. Northern Arizona is classified as a semi‐arid region, with average annual precipitation totaling ∼521 mm between 1991 and 2020 (Flagstaff Pulliam Airport; NOAA). Notably, over 85% of this water is lost as evapotranspiration (Dore et al. 2012). Precipitation in northern Arizona is partitioned as ∼40% monsoonal storms during the summer months, peaking in July and August, and 60% during the winter months peaking from December to February (Hereford et al. 2014). The hydrology of the area is highly seasonal and primarily snowmelt driven, with nearly all surface water discharge occurring between late December and April, and groundwater recharge typically occurring between February and May. On occasion, stream channels in the study area will flow during high‐intensity summer monsoon storms; however, these channels receive no consistent baseflow throughout the year.

The dominant bedrock within the ULM subwatersheds is young Quaternary Basalt from the local volcanic field and the Permian Kaibab Limestone (Moore et al. 1960). Alfisols of hydrologic soil group B occupy the largest surface area within the subwatersheds at 99%, with Inceptisols and Mollisols comprising the remaining 0.7% and 0.3%, respectively (Davis 2022). The overall topography of the subwatersheds is flat to moderately flat (Table 1). All seven subwatersheds drain in a northward direction into the 9 km long, 3.8 km² Upper Lake Mary, a surface drinking water reservoir used by the COF and adjacent rural communities. At full capacity, the lake can hold ∼2.01 × 10⁷ m³ (16,300 acre‐feet) of water. When water levels exceed storage capacity spillover flows into the adjacent Lower Lake Mary (Hornewer and Flynn 2008).

Groundwater is another significant source of drinking water in the region. A wellfield in the Lake Mary area provides potable groundwater extracted from the regional Coconino‐Supai aquifer, more commonly referred to as the C aquifer (Bills et al. 2000). The C aquifer in the study area consists of the karstic Kaibab Formation, Coconino Sandstone, and upper‐middle Supai Formation, and ranges in depth up to ∼450 m (1500 ft) below ground service at some localities (Bills et al. 2000). Multiple shallow and perched aquifers exist within the Kaibab Formation limestones in and around the Lake Mary Graben, overlain by fractured Quaternary Basalts and alluvium that range in thickness from ∼0 to 20 ft. The Lake Mary Graben is delineated by the larger, normal, Anderson Mesa Fault to the north and the Lake Mary Fault to the south, plus minor faults within the graben (Figure 1; Ben‐Horin and Pearthree ²⁰²¹, Bills et al. ²⁰⁰⁰). The minor faults act to divide the Lake Mary Graben into smaller graben blocks that create a stair‐stepping structure with the rock units inside the graben. These faults have a large influence on the flow paths of groundwater inside the Lake Mary area. In general, precipitation that percolates into the groundwater supply flows in a northward fashion towards the Lake Mary Graben, where it connects with the normal faults and is recharged into the deeper C aquifer (A. Springer, personal communication, 2024). Additionally, lateral groundwater flow into the study area is assumed to be minimal, as the subwatersheds are situated near the regional groundwater divide, limiting exchange with the adjacent southern watersheds. While quantification of groundwater recharge in the study area has yet to be estimated, previous studies indicate that recharge is primarily driven by snowmelt from winter precipitation (Meixner et al. 2016; Donovan ²⁰²¹). Throughout all the paired watersheds, the water table is not connected to the land surface, leaving a large vertical area for groundwater recharge to occur. This study aims to provide initial estimations of recharge in the paired watershed.

Water Balance Equation

Water balance calculations were used to understand how water enters and exits the watersheds at differing temporal resolutions. Quantifying the components of a water balance in the ULM paired‐watershed is complex, but vital in understanding the site's basic hydrology. Water balances were calculated at the annual scale to measure inflows and outflows of the watersheds for the 2015–2023 water years (Oct 1‐Sep 30). Precipitation is the only input to these watersheds. Equation 1 describes the ULM water budget equation, modified from Masek Lopez et al. (2013) and Killingtveit et al. (2003):

$$\epsilon =P-[Q+R+\mathit{ET}]$$ (1)

where, P is precipitation, Q is surface runoff, R is groundwater recharge, ET is evapotranspiration, and ε is error; all in mm. ε should approach zero if all components of the water balance are accounted for properly (Killingtveit et al. 2003); and we assume there is no change in soil water storage at the annual time scale. All terms were normalized by the total precipitation in a given water year to yield a percent for each, including the error term. Mean absolute error for a given watershed was calculated as the average of each individual year's absolute error percentage.

Precipitation

“P” in the water balance is the total annual precipitation (mm) received by each catchment in the study during the hydrologic year (Oct–Sep). The Flagstaff Pulliam Airport weather station, operated by the National Oceanic and Atmospheric Administration (NOAA), was used for field observations of precipitation for the duration of the study period. The weather station is located ∼22 km northwest of the study area; however, it sits at an elevation similar, but slightly lower, to that of the paired watersheds at 2140 m. Monthly precipitation data were obtained from the Western Regional Climate Center (WRCC ²⁰²²). Due to a lack of reliable precipitation data for each subwatershed individually, the calculated annual precipitation from the NOAA station was applied to each subwatershed's water balance. Snow data (SWE, snow depth, snowpack duration) were collected from the Mormon Mountain SNOTEL station (site 640) which sits at the top of Mormon Mountain south of the study area (NRCS ²⁰²⁴).

Runoff

The Q term of the water balance was directly calculated using on‐site surface runoff measurements from Salt River Project (SRP) Flowtography stations. Surface runoff data from 2014 to 2023 was retrieved from the City of Flagstaff's “My Watershed” online database. Gaging stations consist of OTT Orpheus Mini pressure transducers (accuracy ± 0.1 in.) and an in‐stream graduated staff gage (marked in increments) which relate water depth (pressure) to stage height in 15‐min intervals. Computation of stream discharge is indirectly measured using rating curves designed by SRP which relate stage height to discharge in cubic‐feet‐per‐second. During the early stages of the project, the stations also used a game camera to take photos of a reflective graduated vertical marker within the channel to provide an independent measurement of stream flow depth to corroborate transducer readings. However, because of challenges with photo interpretation and image archival, cameras were replaced with simpler and more robust measurement instruments. Hourly runoff measurements were converted from cfs to mm/h using watershed area, then summed daily to gather total mm of runoff for each day in the record. Annual runoff in mm for each year of the subwatersheds water balances was calculated by the summation of the daily runoff measurements. Data gaps within the runoff dataset were common during the winter months due to icing of the transducers, which affects the logger's ability to record data. During small data gaps (<12 h) the runoff hydrograph was extrapolated to gap‐fill missing data. During larger data gaps, flow was approximated by correlating rainfall intensities to expected storm runoff from storms during the periods of missing data, using rainfall–runoff models developed for the watersheds by Denver (2024).

Evapotranspiration

Evapotranspiration (ET) is the largest flux of the annual water balance, accounting for >85% of total precipitation in ponderosa pine forests (Dore et al. 2012). The largest challenge in understanding the magnitude of ET flux in water‐limited, forested environments is the difficulty of accurately measuring ET over large areas. Many studies have attempted to measure ET fluxes through a variety of methods, including remote sensing (Bhattarai and Wagle 2021; Tan et al. ²⁰²¹), meteorological models (Ha et al. 2014, 2017), and soil water balance models (Thornthwaite and Mather 1957). Ha et al. (2014) explored the efficacy of multiple meteorological models to predict evapotranspiration rates in ponderosa pine forests by comparing model predictions to field‐measured eddy covariance data. The McNaughton and Black (M‐B) model used in the study is targeted specifically for forest environments and proved to be highly accurate at predicting potential evapotranspiration (PET) rates in restored forest plots (McNaughton and Black ¹⁹⁷³; Federer et al. ¹⁹⁹⁶); however, it can underestimate PET during the summer in untreated forest plots (Ha et al. 2014). The M‐B model equation is:

$$\lambda \mathit{E}=\frac{\rho ·\mathit{C}\rho ·\left(e_s-e\right)}{\mathrm{Y}·r_s}$$ (2)

where λE is PET as evaporative flux (MJ s⁻¹ m⁻²), ρ is air density (1.23 kg m⁻³), C _ρ is the specific heat of air (∼1.01 × 10⁻³ MJ kg⁻¹ °C⁻¹), e _s is the saturation vapor pressure (kPa), e is the vapor pressure (kPa), Y is the psychometric constant (kPa °C⁻¹), and r _s is the bulk stomatal resistance (s m⁻¹, 416.67 from Ha et al. ²⁰¹⁴). The measurement of (e _s − e) is also known as the vapor pressure deficit (kPa). λE was multiplied by 86,400 (86,400 s = 1 day) to obtain daily flux measurements and converted to ET in mm using the conversion, ET (mm/day) = λE/2.45 (Allen 1998). Daily ET measurements were summed into monthly PET totals.

The M‐B model does not require previous knowledge of energy flux data or aerodynamic resistance above the canopy, which is required for alternative models such as the Penman–Monteith model (Penman 1948; Monteith ¹⁹⁶⁵). Due to a lack of available energy flux and supplemental aerodynamic data in the study area, the M‐B model was determined to be the best choice for calculating ET rates in the study area. The unknown variables in the model equation were calculated using off‐site meteorological data from the Flagstaff Pulliam Airport weather station (Data S1).

The M‐B model calculates potential evapotranspiration (PET), or the amount of water loss due to ET given there is enough water available within a given system. Northern Arizona is a water‐limited system, and actual evapotranspiration (AET) rates are typically much less than PET on an annual basis. PET was converted to AET by comparing monthly PET and total available moisture (TAM, monthly precipitation + monthly soil moisture). To measure soil moisture, a cosmic‐ray soil moisture observing system (COSMOS) was installed in the LM5 (Figure 1) watershed in December 2016. The COSMOS station provides readings of volumetric soil moisture at the hectometer horizontal scale and to depths of ∼10–70 cm (Zreda et al. 2012). The singular COSMOS station was the only source of soil moisture data throughout the study. Although the subwatersheds exhibit similar soil characteristics, topography, and are within proximity to each other, soil moisture has been shown to vary significantly at small spatial distances (<10 m, Famiglietti et al. ¹⁹⁹⁹; O'Donnell et al. ²⁰²¹) and there is expected error associated with taking soil moisture from only one location in the larger study area.

Average monthly volumetric soil moisture (%, VWC) was calculated for each month in the period of record. Kerhoulas et al. (2013) concluded that the functional rooting depth for water uptake by mature ponderosa pine trees in northern Arizona was at least 400 mm, which is also the mean depth at which the COSMOS station can measure soil moisture. We assumed water stored in the soil at and above 400 mm is readily available for ET; therefore, we can estimate soil moisture for AET conversions to be the average monthly soil moisture × 400 mm. AET conversions are as follows:

• If TAM > PET, then AET = PET (water is not limiting ET).

• If TAM < PET, then AET = TAM (water is limiting ET).

Groundwater Recharge

In August 2017, two exploratory, non‐pumping wells in the Lake Mary wellfield (LMEX wells) were instrumented with In‐Situ Rugged Troll 100 (30 m 43.5 psia) data loggers to measure hourly water table elevation through time (Figure 1). Drilling reports from the construction of these wells in 1992 noted non‐pumping water levels of 47 and 6 m for the two wells, revealing the presence of an unconfined shallow aquifer within the Kaibab Formation limestone layer. Pumping tests conducted on the wells concluded that the shallow aquifer is separate from the regional aquifer; however, the hydrologic communication between both aquifers is poorly understood. It was assumed that the shallow aquifer either leaks into the regional aquifer or discharges at spring locations by fracture flow along the many fractures and faults within the local geology. It is noted that the LMEX wells reside outside the study watersheds; however, both locations are contained within the same structural domain of the Lake Mary Graben. We assume both locations share similar enhanced permeability through structural faulting and fracture densities, which should create similar responses to recharge. Therefore, recharge responses in the LMEX wells can be used as a proxy for recharge within the watersheds. Additionally, the catchments have limited heterogeneity in snow cover and SWE to influence spatial variability in recharge.

The LMEX monitoring well data provided a high‐resolution water table elevation data set for estimating groundwater recharge through the water table fluctuation (WTF) method (Healy and Cook 2002; Nimmo et al. ²⁰¹⁴). The WTF method provides groundwater recharge estimations through the assumption that rises in the water table seen in observation wells are caused by recharge across the aquifer (USGS ²⁰¹⁷). The formula, as defined by USGS 2017, for the WTF method is as follows:

$$R\left(t_j\right)={S_y}^{*}\text{∆}H\left(t_j\right)$$ (3)

where R(t _j ) is recharge occurring between times t ₀ and t _j , S _y is specific yield (unitless), and ∆H is the water level rise during the recharge period.

There are multiple ways of implementing the WTF method to estimate recharge, one being the graphical method (Delin et al. 2007) which builds an extrapolated recession curve used to calculate ∆H. The graphical method of WTF works by measuring the total water level rise of a recharge event (∆H) by taking the difference between the peak water table elevation during recharge and the water table elevation on the extrapolated recession curve at the time of the peak (Figure 2). Creation of the extrapolated recession curves and calculations of ∆H required manual attention and were performed using R Studio (Posit Team 2023) and Excel. Because there are no documented specific yield values for the monitoring wells, or from any wells in the Lake Mary well field due to the high drilling costs for monitoring pumping tests, specific yield was estimated. Estimations of S _y values across the regional C‐aquifer range from 0.0002 to 0.14; however, the shallow aquifer in the study area is thought to be a separate system composed of limestone which has a S _y range of 0.005–0.05 (Bills et al. 2000). The shallow aquifer is thought to be controlled by fracture flow due to the heavy faulting from the Lake Mary fault zone in the study area and the presence of a fault contact spring in the watershed. Şen (2015) noted that fracture flow systems typically have lower specific yields due to their drainable porosity. Specific yield values at the lower ends of these ranges were assumed to be the most effective for predicting recharge. Water balance studies, in conjunction with the WTF method, have been used to estimate unknown specific yield values (Lv et al. 2021). The WTF method was applied for S _y values between 0.0002 and 0.05 to evaluate what S _y best fit the annual water balance. A specific yield of 0.01 was determined to be a reasonable S _y value for annual water balance calculations of recharge. Groundwater recharge was estimated for the 2017–2023 water years to create the water balances. Additionally, recharge estimates from the 2024 season were included since data was available; however, a 2024 water balance was not generated due to incomplete datasets for the other water balance terms.

Figure 2.

Well hydrograph data showing changes in water table elevation through time for (a) LMEX1 and (b) LMEX3 monitoring wells; data points recorded in 15‐min intervals. Water table fluctuations from recharge fluxes are well defined and the dates of recharge peaks are listed. Annotations depict how the graphical water table fluctuation method is used to calculate recharge using the extrapolated recession curve. ∆H measurements use in calculating recharge for each event are shown (R = ∆H * S _y ).

Results

Annual Water Balances

Precipitation data associated with each water year of the study period (2015–2023) were divided into winter (October–March) and summer (April–September) season precipitation amounts (Table 2). The 2024 winter season was also included. The 2023 water year was the wettest, with 681 mm of annual precipitation, and experienced the heaviest winter on record (536 mm) since the 1973 season. The second wettest water year during the study period was 2016, with 668 mm of total precipitation composed primarily of summertime monsoonal rains (55% of total P). The driest water year was 2018, with only 382 mm of precipitation, resulting in zero recharge or discharge in any of the watersheds (Figure 3). Mean annual precipitation across the study period was 558 mm; 58% fell in the winter season and 42% fell as summer season precipitation.

Table 2.

Summary Table of Precipitation and Snowpack Data From Water Years 2015 to 2023, Including the 2024 Winter Season.

Water Year (Oct–Sep)	P total (mm)	P winter (mm)	P summer (mm)	Snowpack Duration (days)	Snowpack Daily Average SWE (mm)
2015 1	602	292 (48)	310 (52)	—	—
2016	668	298 (45)	370 (55)	—	—
2017	637	400 (63)	237 (37)	—	—
2018	382	128 (33)	255 (67)	—	—
2019	589	442 (75)	146 (25)	123	83
2020	440	366 (83)	74 (17)	107	82
2021	518	210 (40)	308 (60)	71	78
2022	509	238 (47)	271 (53)	113	94
2023	681	536 (79)	145 (21)	176	249
2024	—	284	—	100	84
Average	558	323 (58)	235 (42)	115	112

Note: Numbers in parenthesis indicate seasonal percentage of P _total. Average values only incorporate complete datasets from 2015 to 2023.

¹ Data record begins November 2014.

Figure 3.

Normalized water balances for (a) LM1, (b) LM2, (c) LM2B, (d) LM3L, (e) LM3U, (f) LM4, and (g) LM5. Dashed red line represents 100% or 0% error. Water years on x‐axis are written in the shortened format, 2015 = “15.” Data tables used for the creation of each plot can be found in Data S1.

Water balances, normalized by total precipitation, for nine water years (2015–2023) were calculated for each watershed (Figure 3)—the LM2B subwatershed was instrumented later, thus the record starts in water year 2018. The LM2 and LM2B subwatersheds show the least amount of error with mean absolute error of 15 and 8%, respectively. LM3L and LM3U have the highest amount of error with mean absolute errors of 62 and 35%, respectively. LM1, LM4, and LM5 show moderate error on a yearly basis and have lower mean absolute error values (20, 18, 22%) when compared with LM3L and LM3U, but higher associated errors than LM2 and LM2B. Large, calculated contributions of runoff (Q) during the 2017, 2019, and 2023 water years are the main drivers of error within the high and moderate error‐producing subwatersheds. Surface runoff readings during these years range from reasonable (∼16–27% of total water balance) to highly erroneous (155–209%) within these subwatersheds. Freezing, or “icing,” of pressure transducers during the winter is thought to be the source of these large errors. LM2 and LM2B are devoid of significant icing error and provide the most reliable surface runoff data throughout all subwatersheds.

AET was the largest component of the water balance (Figure 3)—72% on average. Surface runoff was the second largest output of the water budget. The calculated water balances show a large, highly variable contribution of stream discharge (0–155% of total incoming precipitation) throughout all the subwatersheds and occurred almost exclusively during the winter and early spring months when snowmelt is occurring (Dec‐Apr, Figure 4). Groundwater recharge only occurred during 4 years of the study. Recharge events for these four seasons were observed in both the LMEX1 and LMEX3 wells; however, the magnitude and duration of water table rise (∆H) for a given event varied between the two sites (Figure 2, Table 3). Calculations of recharge, from the WTF method, were the average of the two monitoring wells. On average, groundwater recharge composed 9% of the annual water budget. Recharge of 49 mm (8%), 29 mm (7%), and 85 mm (12%) was calculated for the Spring 2019, Spring 2020, and Spring 2023 recharge seasons (Table 3)—percentages represent each recharge event's contribution to their respective water years' water balance by comparing the magnitude of recharge to the water year's total precipitation. The Spring 2024 season was included due to available data and recorded 11 mm of recharge (Table 3). Average recharge event durations in the LMEX1 well were around 78 days, while the LMEX3 well was longer at 95 days. Recharge event durations do not appear to have a strong relationship with the total magnitude of recharge; although more years of observation might improve this relationship.

Figure 4.

(a) Linear regression model of total wintertime precipitation versus annual stream discharge in the LM2 (R ² = 0.83, P = 0.0006) and LM2B (R ² = 0.97, P = 0.0004) subwatersheds. Equations of trendlines are LM2—y = 0.68x − 116.4, LM2B—y = 0.70x − 89.4. (b) Average monthly contributions (%) to annual streamflow from the 2015 to 2023 water years. Distribution of runoff was created using LM2 and LM2B data.

Table 3.

Summary Table of the Magnitude and Timing of Recharge Events Seen in the LMEX1 and LMEX3 Wells.

	Recharge Event	Start	End	Event Duration (days)	P winter (mm)	∆H (mm)	S y	Recharge (mm)
LMEX1	Spring 2019	2/12/19	4/12/19	60	442	3645	0.01	36
	Spring 2020	2/26/20	4/21/20	56	366	3074	0.01	31
	Spring 2023	2/10/23	5/31/23	111	536	5056	0.01	51
	Spring 2024	3/22/24	5/5/24	44	284	1420	0.01	14
LMEX3	Spring 2019	1/26/19	5/22/19	117	442	6103	0.01	61
	Spring 2020	3/5/20	6/7/20	95	366	2590	0.01	26
	Spring 2023	3/9/23	5/4/23	57	536	12,003	0.01	120
	Spring 2024	3/27/24	4/26/24	30	284	725	0.01	7

Note: Recharge values used for water balances were taken as the average between each well's calculated recharge for a given event.

Climatic Controls on Water Yields

Regression modeling was performed to determine the importance of climatic conditions such as winter precipitation and snowpack duration on the local hydrology. Watersheds with low overall error and little to no susceptibility to icing (LM2 and LM2B) were utilized to assess what parameters affect annual stream discharge in the study area. Wintertime precipitation (x‐variable) was plotted against observed annual streamflow (y‐variable) to determine the importance of winter season precipitation inputs (Figure 4a). Both LM2 (R ² = 0.83, P = 0.0006) and LM2B (R ² = 0.97, P = 0.0004) had a statistically significant relationship, suggesting that increased total winter season precipitation is correlated with proportional increases in total annual stream discharge in the same year (Figure 4a). Runoff primarily occurs during the winter season and into the early spring (88%, December–April)—with the most flow occurring during the month of March (36%) when snowmelt is at its highest (Figure 4b). Summer monsoon season (June–September) was only responsible for a small portion of annual runoff at 6%.

A similar winter precipitation trend is seen when setting groundwater recharge as the y‐variable. For years with available water table elevation data, recharge occurred during the years with the highest amount of winter precipitation (Figure 5). Both the LMEX1 and LMEX3 wells showed significant (P = 0.01, P = 0.02; respectively), highly correlated (R ² = 0.97, R ² = 0.96; respectively) relationships with wintertime precipitation. Donovan (2021) determined that ephemeral springs discharge in similar hydrostratigraphic units near the study area are highly correlated with seasonal duration of snowpack (periods of >60 days of continuous snow cover). Because springs are locations of aquifer discharge, there should be a similar relationship between groundwater recharge and snowpack duration. Regression analysis shows a statistically significant relationship (R ² = 0.81, P = 0.01) between seasonal snowpack duration (x‐variable) and the magnitude of springtime recharge (y‐variable) for years with available well data (Figure 6). Snowpack was defined as a period of snow cover >60 days, and the 2018 water year was excluded due to no continuous snowpack that season.

Figure 5.

Winter precipitation (mm) versus recharge (mm) for observed recharge events in the LMEX1 (R ² = 0.97, P = 0.01) and LMEX3 (R ² = 0.96, P = 0.02) wells. Equations of trendlines are LMEX1—y = 0.14x − 23.9, LMEX3—y = 0.45x − 130.6.

Figure 6.

Linear regression model of the relationship between seasonal duration of snowpack in days versus the total amount of groundwater recharge recorded in a given water year in mm. Annotations represent the water year each point correlates with. Data points are sized according to each water year's total wintertime precipitation (mm) provided in Table 2.

Discussion

Water balance analysis and regression modeling addressed the individual goals of this study and advanced understanding of the relationship between climate and the paired‐watersheds' hydrologic behavior. This study offers novel insight into groundwater recharge of this aquifer and provides the first published values of annual recharge for the Southern Colorado Plateau. Results of the study show that groundwater recharge does not occur every year and is highly correlated to the duration of snowpack and runoff (Figures 4b and 6). Interestingly, the 2022 water year had a longer snowpack duration than both the 2020 and 2024 years; yet the former produced no spring season groundwater recharge. Observations of total winter season precipitation illustrate that groundwater recharge is more likely controlled by a combination of snowpack duration and total snowfall, consistent with studies across the western United States (Meixner et al. 2016; Donovan ²⁰²¹) (Figure 4). Differences in the timing and magnitude of recharge seen in the LMEX wells (Figure 2, Table 3) are most likely due to differences in the surrounding structural geology between the wells, as well as proximity to karstic geomorphological features (sinkholes) and locations of high secondary porosity (fractures and faulting); however, these differences are not deeply explored in this article. The steeper slope of the LMEX3 trendline in Figure 5 suggests that recharge around this well has a stronger response to wintertime precipitation than LMEX1; but, due to the small sample size of available data, continued monitoring of groundwater recharge in the LMEX wells is needed for strengthening our understanding of this relationship. Specific aquifer characteristics (S _y ) for the shallow karst limestone aquifer in the study area are not well understood (Bills et al. 2000; Wood ²⁰¹⁹), and performing a pumping‐test derivation of S _y would greatly increase the precision and accuracy of recharge estimations from the WTF method (Remson and Lang 1955). Additionally, a further limitation of this study is that it does not consider how variables such as midwinter snowmelt, antecedent soil conditions (frozen soils), and freezeback events impact groundwater recharge (Hyman‐Rabeler and Loheide ²⁰²³). Future research that encompasses these parameters is required to understand how recharge will be affected by climate change and forest restoration in this region.

A wide range of climatic conditions were observed during the period of study, including the dry year of 2018 (382 mm) to the very wet year of 2023 (681 mm), which was crucial for understanding how the watersheds behaved under differing conditions. On average, 58% of the rainfall fell during the winter season and 42% fell during the summer season, which agrees with the 60% winter, 40% summer precipitation trend described in Hereford et al. (2014) for northern Arizona. Winter season storm fronts blanket the subwatersheds with relatively equal amounts of snowpack, reducing spatial variability during the season. High‐intensity convective monsoonal storms during the summer are much more localized, and using the NOAA Flagstaff Pulliam Airport weather station outside of the study area most likely led to some errors in the precipitation data. Monsoonal storms that only passed over the watersheds would be missed in the record; however, storms that only passed over the airport would be misinterpreted as inputs into the watersheds. We assume that these two scenarios should balance each other out and that the weather station precipitation data is sufficient for the water balance analysis. Evapotranspiration was the largest component of the water balance, as was found by prior eddy‐covariance and energy‐balance modeling studies (Dore et al. 2012; Ha et al. ²⁰¹⁴). Over the entire period of record, the M‐B model estimated ET at ∼72% of the total water budget, an underestimation from the typical regional rate of 85% (Dore et al. 2012) but agrees with the findings of Ha et al. (2014) who suggested the M‐B model underestimates ET in ponderosa pine forests. The 2019 and 2023 water years produced the least amount of AET (<60%). Both years represented the two wettest winters that were followed by below average dry summer monsoon seasons. Low AET values during these years are most likely due to colder, moist conditions reducing evaporative demand in the winter season, followed by increased water‐limiting conditions during the drier summers. The water balance in the dry 2018 water year was composed entirely of AET for all watersheds. Calculations of AET for 2018 were estimated as 110% of total P, an overestimation of 10%, which we assume to be error associated with the usage of the M‐B model. Error yearly with this method is not precisely known but is most likely due to our conversions of PET to AET. Moreover, the assumption of available soil moisture taken from one point source (LM5 COSMOS station) is not likely representative of the entire soil profile across all watersheds since soil moisture values can vary significantly at small distances (Famiglietti et al. 1999). O'Donnell et al. (2021) utilized specialized soil moisture probes to understand the spatial and temporal resolution of soil moisture values between control and treated forest plots in a nearby study. The collection of spatially representative soil moisture data across each subwatershed would improve the understanding of how soil moisture and ET rates vary spatially. An additional source of error may be the ability of vegetation to access stored rock moisture, particularly during periods of drought (Rempe and Dietrich 2018; Stocker et al. ²⁰²³). The role of groundwater or rock moisture contributions to ET is not known at this site but has been shown to increase with increasing aridity (e.g., Evaristo and McDonnell ²⁰¹⁷). Due to the proximity of the subwatersheds and the nearly homogenous vegetative cover and topographical characteristics of the study area, we assume similar ET rates across the watersheds. But, with modifications to vegetative cover post‐treatment, it is necessary to understand how the ET rates of the subwatersheds will differ with treatment.

Water balance analyses estimated a wide range of variation in total surface discharge amounts both spatially and temporally within the paired watersheds (Figure 3). Challenges in understanding these variations stem from errors associated with streamflow measurements. Overestimations of annual stream discharge were associated with heavier winter seasons that cause “icing” of the pressure transducers, which results in erroneously high water level readings due to logger malfunctions. Consequently, the water years and individual watersheds that experienced the highest degrees of error seem to have been the most susceptible to these overestimations in stream discharge (Figure 3d and 3e). Below freezing temperatures, common during the winter, are known to affect transducer measurements in this region, and proper protection from freezing is mandatory for ensuring the reliability of streamflow measurements. The use of sight board and camera equipment was attempted to mitigate periods of transducer malfunction; however, due to reduced visibility from snow cover and frozen channel conditions, this technique was unreliable in estimating discharge. Manual correction of data gaps due to icing could be performed when these gaps were small (<12 h); however, this was challenging during large periods of missing data (∼1–2 months). Rainfall–runoff relationships developed for each watershed were used to estimate discharge during storm events that occurred during these large data gaps (Denver 2024). Another potential source of error in the streamflow data is lateral flow entering the catchment from outside the study area, which could artificially raise discharge magnitudes. Given the thickness of permeable alluvium and fractured basalts overlying the fractured Kaibab Limestone bedrock, vertical infiltration is likely the dominant hydrological process in the study area, with lateral flow playing a minimal role except potentially during extreme saturation events (i.e., summer monsoons). The ephemeral character of the channels, coupled with the absence of base‐flow contributions throughout the year, indicates that streamflow in the catchments is generated exclusively from precipitation inputs within the watersheds.

The results of this study reinforce the region's reliance on winter season precipitation for water resources. Stream discharge is much higher during the winter months with very little streamflow occurring during summertime monsoonal storms, when ET is at its highest and the soil is at its driest (Figure 4b). Soil moisture during the winter months approaches or exceeds field capacity (∼22%) for much of the winter season, initiating surface runoff more easily. There is a significant linear relationship (high R ² values, low p‐values) between wintertime precipitation and stream discharge in the subwatersheds (Figure 4a). This is consistent with previous regression analyses of rainfall–runoff relationships in northern Arizona (Baker 1986; O'Donnell et al. ²⁰¹⁸). Similarly, all groundwater recharge occurs in the springtime following snowmelt from wintertime precipitation. Wintertime precipitation and snowpack presence are important to surface and groundwater yields within these subwatersheds (Figures 4, 5, 6). With current and future predicted losses of snowpack and SWE in the region due to climate change (Fyfe et al. 2017; Mote et al. ²⁰¹⁸), plus a transition from snow‐ to rain‐dominated winter precipitation (City of Flagstaff n.d.; Hammond et al. ²⁰¹⁹), surface and groundwater reservoirs are becoming threatened in this region. The results of this study suggest that groundwater reservoirs in semi‐arid northern Arizona are exceptionally at risk due to rain being less effective at producing recharge than snowpack/snowmelt, a trend seen across the semi‐arid American southwest (Szecsody et al. 1983; Winograd et al. ¹⁹⁹⁸; Earman et al. ²⁰⁰⁶).

The Four Forest Restoration Initiative's (4FRI) overall goals are to restore the fire‐adapted, ponderosa pine forests of Arizona to their range of historic variability in structure and health, and decrease the likelihood of high‐severity burns through mechanical thinning and prescribed burning techniques. Zomlot et al. (2015) note that the factors influencing groundwater recharge the most are precipitation, soil texture, and vegetative cover/land‐use type, with pine forest regions having a high positive effect on recharge. High‐severity wildfire would result in decreased groundwater recharge rates due to canopy loss and soil permeability reduction, which causes a shift towards increased surface runoff (Guzmán‐Rojo et al. ²⁰²⁴). Furthermore, large high‐severity burn areas are likely to remain without forest cover for decades to centuries if a nearby seed source for ponderosa regeneration is absent, leaving these areas vulnerable to nonnative species encroachment and shrubland formation, which would further limit successful pine regeneration (Haffey et al. 2018). 4FRI's restoration efforts can reduce the risk of high severity burns and, given a favorable climate, help restore and/or amplify flow regimes and groundwater yields within these forests through a variety of mechanisms (Wyatt et al. 2014; Moreno et al. ²⁰¹⁶; O'Donnell et al. ^{2018, 2021}; Jones et al. ²⁰²²). This is important since, as seen in Figure 2, water levels within the regional shallow limestone aquifer drop quickly following the spring and rely heavily on consistent seasonal recharge annually to keep water levels stable. A single high‐severity burn could be detrimental to the regional groundwater supply. While long‐term trends of water levels in the aquifer are primarily driven by climate, forest restoration projects, such as this one, create an opportunity for more fire‐resilient forests and can increase groundwater recharge (Wyatt et al. 2014). The continued monitoring conducted within the Upper Lake Mary paired‐watershed study is essential in understanding how these forest treatments will alter the hydrology in the study area posttreatment.

Conclusions

This study helps explain the current hydrologic conditions of the ULM paired‐watershed study subwatersheds and the mechanisms that control them. Using historical meteorologic and hydrologic data, seven annual water balances were created for each of the ULM paired‐watershed study subwatersheds from the 2015–2023 water years, which included novel data that quantified groundwater recharge. Groundwater recharge values presented in this study offer the first published values of annual recharge from the Southern Colorado Plateau measured using the water‐table fluctuation method. Groundwater recharge does not occur every year in the study area but is elevated by increased total winter seasonal precipitation and longer durations of seasonal snowpack presence. Surface runoff shows a similar relationship to wintertime precipitation, with 88% of runoff measured during the winter and early spring months when snowmelt is occurring in the channels. On average, only 8% of runoff occurs during the summer monsoon season. Water balances are limited by the reliability of runoff data due to data gaps and error caused by freezing of in‐stream pressure transducers. Mitigation of the error produced from these freezing events is essential for increasing the resolution of runoff data and decreasing overall error within the study. The LM2 and LM2B subwatersheds' transducers are less susceptible to freezing, allowing us to interpret climatic controls on runoff from these subwatersheds. As expected, evapotranspiration was the largest component of the water balance (∼72%), while groundwater recharge was the smallest (∼9%), when present. More work is needed to translate the findings of this study to regional‐scale water balance behavior. However, these findings highlight the importance of winter season precipitation on ground and surface water yields. As 4FRI treatments are not expected to occur for a couple more years, continued monitoring is essential for improving the understanding of climate on water yields, specifically groundwater, because only three recharge events were seen in the data record. Future work within the study area should explore aquifer properties, such as specific yield and groundwater flow paths, in more detail. Additionally, the study could benefit from the installation of more soil moisture probes and an eddy covariance station for ET measurements to help better constrain these data.

Supporting information

Data S1. Supporting Information.

Data Availability Statement

Research data are not shared.