High water retention in urban garden soils leads to reduced leachate and elevated evapotranspiration

Among the ecosystem services provided by urban greenspace are the retention and inltration of stormwater, which decreases urban ooding, and enhanced evapotranspiration, which helps mitigate urban heat island effects. Some types of urban greenspace, such as rain gardens and green roofs, are intentionally designed to enhance these hydrologic functions. Urban gardens, while primarily designed for food production and aesthetic benets, may have similar hydrologic function, due to high levels of soil organic matter that promote inltration and water holding capacity. We quantied leachate and soil moisture from experimental urban garden plots receiving various soil amendments (high and low levels of manure and municipal compost, synthetic fertilizer, and no inputs) over three years. Soil moisture varied across treatments, with highest mean levels observed in plots receiving manure compost, and lowest in plots receiving synthetic fertilizer. Soil amendment treatments explained little of the variation in weekly leachate volume, but among treatments, high municipal compost and synthetic fertilizer had lowest leachate volumes, and high and low manure compost had slightly higher mean leachate volumes. We used these data to parameterize a simple mass balance hydrologic model, focusing on high input municipal compost and no compost garden plots, as well as reference turfgrass plots. We ran the model for three growing seasons under ambient precipitation and three elevated precipitation scenarios. Garden plots received 12–16% greater total water inputs compared to turfgrass plots because of irrigation, but leachate totals were 20–30% lower for garden plots across climate scenarios, due to elevated evapotranspiration, which was 50–60% higher in garden plots. Within each climate scenario, difference between garden plots which received high levels of municipal compost and garden plots which received no additional compost were small relative to differences between garden plots and turfgrass. Taken together, these results indicate that garden soil amendments can inuence water retention, and the high water retention, inltration, and evapotranspiration potential of garden soils relative to turfgrass indicates that hydrologic ecosystem services may be an underappreciated benet of urban gardens.


Introduction
Urban ecosystems are highly engineered environments that are often characterized by extensive impervious surface cover and highly altered hydrology. The built environment and impervious cover of urban ecosystems-roads, sidewalks, and roofs-increase the magnitude and speed of movement of people and goods, and also facilitate transport of water and nutrients, contributing to ooding and water pollution (Miller et al. 2014). Stormwater storage and in ltration is an important feature of urban green infrastructure design to mitigate ooding and reduce nutrient transport to water bodies, especially given increases in extreme precipitation events due to climate changes (Donat et al. 2016;Pathak et al. 2017).
Evaporation and transpiration (together, evapotranspiration or ET) are important factors in urban heat island reduction (Donat et al. 2016; Pathak et al. 2017). The extent of lawns and gardens and how these greenspaces are managed interacts with climate and in uence urban water demand for irrigation (Flörke et al. 2018).
Greenspaces play important roles in facilitating ecosystem services related to hydrology in urban ecosystems. Turfgrass is characterized by relatively high in ltration and ET, and low runoff rates (Monteiro 2017). Urban trees can decrease runoff through canopy interception, enhance percolation driven by tree roots, and subsequent transpiration (Kuehler et al. 2017; Berland et al. 2017; Rahman et al. 2019). Green stormwater infrastructure such as rain gardens and retention basins are intentionally designed for stormwater retention or in ltration (Karnatz et al. 2019). Green roofs, which are typically constructed in areas with high impervious cover, are designed to decrease runoff through retention and ET (Raimondi and Becciu 2020).
While the hydrologic properties of stormwater green infrastructure, turfgrass, and urban tree canopies have been relatively well-studied, less attention has focused on the hydrologic role of urban vegetable gardens. Urban food cultivation has increased in popularity in recent decades (Fox 2011), with primary goals such as increasing food security and diet diversity, particularly in low income areas (Warren et  function more similarly to engineered stormwater green infrastructure than to other types of urban greenspace. Urban gardens are characterized by high compost inputs (Small et al. 2019) and porous spoils, potentially resulting in high water storage capacity and ultimately higher ET from crops growing in nutrient-rich garden soil (Qiu et al. 2013). We previously documented seasonal ET rates in garden plots that were nearly twice as high as in reference turfgrass plots, with lower leachate uxes in garden plots despite additional water inputs from irrigation ).
The goal of this study was to quantify the effects of different soil management practices in urban gardens on the physical and biological processes affecting garden hydrology. We rst analyzed the results of three years of hydrologic data from experimental garden plots that received different types and amounts of compost and synthetic fertilizer. We then developed a simple mass balance hydrology model, parameterized using data from this study, and used this model to compare the fate of water in gardens or turfgrass across a range of precipitation scenarios. We tested the hypothesis that high compost garden soil is characterized by elevated water storage, decreased leachate, and ultimately higher ET.

Study area and design
We conducted a multi-year study using the Stewardship Garden at the University of St. Thomas in St.
Paul, MN. Established in 2011, the research garden contains 36 raised garden beds measuring 4 m 2 and 0.3 m deep ( Figure S1). At the start of the current study in 2017, soil from previous projects was replaced and homogenized. Each raised bed was divided into four subplots in which the following crops were planted and rotated annually: 1) carrots; 2) bush beans; 3) bell peppers; and 4) cabbage (2017) or collards (2018, 2019). We randomly assigned each of the experimental plots to one of 6 soil amendment treatments previously described by Shrestha et al. (2020). Brie y, soil treatments consisted of a: 1) control treatment in which no compost or fertilizer was added (nofert); 2) synthetic fertilizer to meet crop N demand and P (synthetic); 3) a higher application rate of manure compost targeted to meet crop N demand (high manure); 4) a lower application rate of manure compost targeted to meet crop P demand, with supplemental N fertilizer to meet crop N demand (low manure); 5) a higher application rate of municipal compost targeted to meet crop N demand (high municipal compost); and 6) a lower application rate of municipal compost targeted to meet crop P demand, along with supplemental N fertilizer to meet crop N demand (low municipal compost). Compost properties and application rates are described in Tables S1 and S2. Mean garden soil organic matter (loss on ignition method) ranged from 8% on the no fertilizer treatment to 12.6% in the high municipal compost treatment (Table S3). For more detailed information about the study area and the experimental design see Small et al. (2018) and Shrestha et al. (2020).
Total water inputs, soil moisture, and other meteorological data Soil moisture was measured 3-4 times per week at a depth of 5 cm, with three measurements recorded from each garden subplot, between June-August from 2017-2019, using a General DSMM 500 soil moisture meter. We used these direct measurements (data reported as %) for statistical analyses, but we converted values for the mass balance model (described below) in 2017 by collecting 16 40 mL soil cores from each soil amendment treatment and measuring water content as the difference between the initial (wet) mass and the mass after drying for 48 hours at 40°C (Fig. S2). Soil moisture readings are presented as % by volume ([mL water/mL soil] x 100).
Meteorological data was collected at hourly intervals in the research garden beginning in June 2017. Rainfall was measured using a ECRN-50 rain gauge (Part # 40655, METER); solar radiation was measured using a PAR sensor (Part # 40003, METER); temperature and relative humidity were measured using a VP-4 sensor (Part # 40023, METER); and wind speed was measured using a Davis Cup anemometer (Part #40030, METER). Data was recorded using an Em50 data logger (Part 40800, METER).
Throughout the growing season, soil moisture was maintained at > 15% in our study plots through a combination of rainfall and irrigation. When irrigation was required, we watered evenly over the 4 m 2 raised beds for a set time (30,45, or 60 seconds) and estimated the volume of water added by measuring the amount of time it took to ll an 11 L bucket at that ow rate. All garden study plots received equal irrigation inputs. Total water inputs for each weekly interval were calculated as the sum of ambient rainfall and irrigation inputs. Reference turfgrass plots located adjacent to the garden received ambient rainfall but did not receive supplemental irrigation.

Leachate collection
Prior to the beginning of the experiment in 2017, we installed lysimeters in the center of each of the 128 garden subplots, plus ve additional turfgrass reference plots. Lysimeter construction, installation, and data collection were previously described in Small et al. (2018) and Shrestha et al. (2020). Brie y, we constructed lysimeters by attaching plastic funnels with diameter of 11.8 cm to 1 L polyethylene bottles tted with Tygon tubing for sampling. We buried the lysimeters at a depth of 0.3 m. We collected and recorded leachate volume from the lysimeters weekly throughout the growing season by emptying the collection bottle with a 50 mL syringe.

Statistical analysis
We tested for differences in soil moisture and leachate volume among soil treatments using general linear models. For mean weekly soil moisture, our models included four predictor variables: soil treatment, weekly total water inputs (rainfall and irrigation), crop type, and year. Because the relationship between weekly water inputs and soil moisture was nonlinear above inputs of 5 cm/week, weeks exceeding this total were excluded from the statistical model (a total of 7 out of 41 weeks).
For weekly volume of leachate collected, our models included weekly total water inputs (rainfall and irrigation), crop type, year, and weekly mean soil moisture (on a volume basis). To identify the best t and most parsimonious models for both weekly soil moisture and weekly volume of leachate collected, we used multimodal inference and the Akaike Information Criterion (AIC) approach to model selection in R.
We tested assumptions of normal distribution using the diagnostic plots in R; we used residual vs. tted plots to test for equal variance and the Q -Q plot to assess normality. We also evaluated variance in ation factors and con rmed they were low (< 3), indicating insigni cant collinearity between our variables. Including four predictor variables in our models generated a total of 15 models. To select the best t model, we evaluated the 15 models from a R 2 , adjusted R 2 , AIC, DAIC, and model weight perspective.

Mass balance hydrology model
We created a simple mass-balance hydrology model to test assumptions about underlying processes by comparing model output with observed data. We modeled soil moisture (SM) as L of water within a 1m x 1m x 0.3m (300 L) experimental garden plot: dSM/dt = precipitation + supplemental irrigation -water leachate -evapotranspiration Daily precipitation and supplemental irrigation (mm/d, or L/m 2 /d) were inputs to the model as described above. Water leachate (mm/d, or L/m 2 /d) was modeled based on the difference between modeled soil moisture and soil water capacity. Water capacity was modeled as a function of soil % organic matter, based on the relationship between the mean %OM for each soil amendment treatment and the maximum observed soil moisture in that treatment (R 2 = 0.57). Water storage in excess of water capacity was assumed to be exported as leachate.
We calculated evapotranspiration (mm/d, or L/m 2 /d) based on the Penman-Monteith equation (Zotarelli et al. 2010), using mean daily solar radiation, maximum and minimum relative humidity, maximum and minimum temperature, and mean wind speed as inputs. The calculated reference evapotranspiration rate (representing turfgrass) was converted to potential crop evapotranspiration using seasonally varying crop coe cients ranging from 0.55-1.2, with maximum values in the middle of the growing season (based on values reported in Satler 2016). Potential crop evapotranspiration was multiplied by a correction factor, k s , that is a function of soil moisture (Zotarelli et al. 2010), adjusting ET downward in drier soil. Between soil moisture values of 6% and 21%, k s increases linearly from 0 to 1. During the parameterization process, we adjusted calculated ET using a correction factor of 2 to achieve a good correspondence between modeled and observed soil moisture and cumulative leachate values. ) achieved by multiplying any daily rainfall total greater than 5.08 cm (2 inches) by a factor of 1.25. Supplemental irrigation was maintained at ambient levels (no irrigation for turfgrass) in these scenarios. For each scenario, we also calculated cumulative leachate and ET uxes into water derived from precipitation and water derived from irrigation. To do this, we separately modeled stocks of soil water derived from precipitation and soil water derived from irrigation, with in ows being the known daily inputs from each source, and out ows were partitioned by multiplying the calculated total ux of leachate or ET by the relative composition of the total soil water stock.

Total weekly water inputs
The mean total weekly water inputs (the weekly sum of irrigation water + rainfall) were 3.65 (±SE, 0.59) cm in the 2017 growing season, 2.67 (±0.66) cm in 2018, and 3.52 (±0.85) cm in 2019. Total seasonal supplemental irrigation inputs mostly occurred during June and July, and totaled 14.9 cm in 2017, 11.0 cm in 2018, 7.03 cm in 2019. In 2017, 12 out of 14 growing season weeks had less than 5 cm of total water inputs, and in 2018 and 2019, total weekly water inputs were generally below 2.5 cm, though every year had at least one week over 7.5 cm (Fig 1.).
Soil moisture was positively related to weekly water inputs between 0-5 cm/week; soil moisture was not related to water inputs at higher levels than 5 cm/week (Fig. S3, S4). Excluding weeks with high (>5 cm) water inputs, the best-t model-as determined by R 2 , adjusted R 2 , and through AIC model selection criteria-for soil moisture included weekly total water inputs, soil treatment, crop type, and year (Table 1). This best t model and its four predictor variables explained 26.4% of the variation within weekly soil moisture and treatment explained 7.6% of that alone (Table 1). Weekly total water inputs (sum of irrigation plus precipitation) was a weak explanatory variable and was weakly correlated with weekly soil moisture (Fig. 1, Table 1). We report the coe cients of the logistical model in Table 2.
Mean observed soil moisture in garden plots varied across years (Table 1) (Table 2). Weekly soil moisture varied across treatments ( Table 2). The manure compost treatments (both high-and low-input levels) generally had the highest weekly soil moisture, whereas the low municipal compost and synthetic fertilizer treatments had the lowest values (Table 2). We found differences in weekly soil moisture among crop types (Table 1), with highest values observed in subplots growing beans (Table 2). Weekly total water inputs explained roughly one-third (R 2 ~ 0.32) of the variation in weekly leachate collected from 2017 to 2019 (Fig. 2). Total observed leachate was best explained by multiple predictor variables including treatment, crop, and year (Table 3, top row). This linear model explained 39.7% of the variation in weekly leachate collected and was the best t model from an R 2 , adjusted R 2 , and AIC approach. Leachate volume varied slightly across soil amendment treatments, with lowest values observed for the high municipal compost treatment and synthetic fertilizer treatments (Table 4). Leachate volume varied among crops as well, with lowest volume observed for collards, and highest volumes observed for peppers and cabbage (Table 4).

Climate scenarios
Garden plots received 12-16% greater total water inputs compared to turfgrass plots, but leachate totals were 20-30% lower for garden plots across climate scenarios, due to elevated evapotranspiration, which was 50-60% higher in garden plots (Fig. 3). Within each climate scenario, difference between garden plots which received high levels of municipal compost and garden plots which received no additional compost were small relative to differences between garden plots and turfgrass. High compost garden plots had ca. 7% higher evapotranspiration, and 10-15% lower leachate totals, compared to no compost garden plots.
Comparing across climate scenarios, elevated extreme rainfall events (Scenario 3) resulted in a 4% increase in total leachate without affecting evapotranspiration. Increased magnitude of smaller rain events (Scenarios 1 and 2) led to slight increases in evapotranspiration (due to modeled relationship between ET and soil moisture), in addition to larger increases in leachate. In the baseline scenario for the simulated high municipal compost treatment, 29% of rainfall was ultimately exported as leachate, compared to 17% of irrigation inputs. For the no fertilizer treatment garden plots in the baseline scenario, 34% of rainfall was ultimately exported as leachate, compared to 21% of irrigation inputs. Elevated rainfall in the climate scenarios increased both the fraction of rainfall and irrigation inputs that ultimately ended up as leachate. In Scenario 1, where total rainfall was highest, 40% of rainfall and 24% of irrigation inputs were exported as leachate in the high municipal compost treatment, whereas in the no fertilizer garden plots, 44% of rainfall and 29% of irrigation inputs were exported as leachate (Table 5).

Discussion
Our results show that garden management practices in uence the storage and fate of water, but these differences were small relative to differences between garden and reference turfgrass plots. Previous studies that have shown that the addition of organic material can increase water holding capacity (Young et al. 2014;Wadzuk et al. 2015), and our results show that the type of compost matters, with manure compost treatments maintaining highest mean soil moisture. We hypothesized that more soil water storage capacity should lead to a greater fraction of being water inputs ultimately exported as ET rather than leachate, as water should be retained longer in the soil, providing more opportunity for uptake of water by crop roots or physical evaporation. The comparison of empirical observations and model results between garden and turfgrass reference plots generally supports this hypothesis, as garden plots, which received higher water input due to irrigation, generally had lower leachate. Comparing across garden experimental treatments indicates more subtleties in the physical and biological processes controlling the fate of water. The lowest leachate values were observed in plots with municipal compost additions, and manure compost treatments were associated with higher leachate. It is possible that, by effectively retaining water in the soil, the manure compost treatments may have had a slightly lower capacity to store additional water inputs from rain events.
Some insight into the dynamics of water in garden soil can be inferred from the relationships between weekly total water inputs, soil moisture, and leachate volumes. The nonlinear relationship between weekly total water inputs and garden soil moisture (Fig. S3) shows that soils reached their maximum water holding capacity (between 0.25-0.30 mL water/mL soil) at input rates beyond 5 cm/week. The relationship between leachate volume and weekly total water inputs (Fig. 2) has an x-intercept of ~ 1, indicating that some leachate is expected with inputs above 1 cm/week. The slope is < 0.5, indicating that there is not a simple threshold beyond which all additional water becomes leachate, but rather, this garden soil has high water storage capacity, and more than half of additional water inputs are stored and ultimately exported through ET.
It is notable that total weekly water input variable alone explains relatively little variation in leachate volume (R 2 = 0.32; Table 3) and especially in soil moisture (R 2 = 0.133; Table 1). Experimental error accounts for some of this variation. High variation was commonly observed among replicate lysimeters. The relatively small areas of lysimeter compared to the study plots could lead to spatial heterogeneity in leachate uxes (e.g., due to aboveground vegetation and belowground root dynamics). The simple pan (zero-tension) lysimeters used in this study are known to work reasonably well under wet conditions, but in drier soil, divergence could lead to underestimating leachate uxes (Gee et al. 2004). Soil moisture measurements are also subject to variation due to spatial heterogeneity, error associated with conversion from instrument readings to volumetric soil water content (Fig. S1), and the differences in temporal resolution between instantaneous soil moisture measurements and cumulative weekly water inputs (e.g., a large rain event towards the end of a sampling interval would not affect soil moisture measurements collected earlier that week). More fundamentally, the lack of simple relationships between water inputs, soil moisture, and leachate underscores the complex relationships between storage and uxes that are better captured in dynamic mass-balance model. This model, based on simple assumptions (such as no differences between crops, and a simple relationship between soil organic matter and soil water capacity), performed reasonably well in capturing temporal dynamics and differences among treatments in observed soil moisture (Fig. S3) and cumulative leachate (Fig. S4), with a few exceptions, such as overestimating both leachate volume and soil moisture for the high municipal compost treatment in late 2017. Notably, we increased calculated ET (based on the Penman-Monteith equation, modi ed for common vegetable crops) by a factor of two (for both garden and turfgrass plots) to achieve good correspondence with our observed data, suggesting higher ET than would be expected from physical and biological conditions alone.
The different relative contributions of irrigation and rainwater inputs to leachate and ET (Table 5) illustrate that these different inputs have largely different fates, due to their timing, magnitude, and intensity. Irrigation occurred at moderate levels; the 60 irrigation events over three growing seasons had a median input level of 0.54 cm, with only one application exceeding 1 cm. By contrast, 62 days between 1 June 2017-31 October 2019 received rainfall in excess of 1 cm. Additionally, irrigation inputs occurred when there had not been recent rain (in contrast to the stochastic timing of rain events), so these inputs typically did not exceed the soil's capacity for storage. On the other hand, irrigation may keep soil moisture closer to the soil's water holding capacity, so that capacity to retain additional water from rain events is somewhat reduced.
Across the three growing seasons of this study, we observed ET uxes from turfgrass and experimental gardens on similar orders of magnitude compared to other urban greenspaces (Table 6). Previous studies in urban lawns in both subtropical China and southern California found ET uxes slightly higher than in our urban gardens (Litvak and Pataki 2016;Litvak et al. 2017;Qiu et al. 2017). These results are not particularly surprising, given the temperate climate in this study, though it may be a result of lower ET uxes from urban gardens compared to urban lawns. Compared to other urban greenspaces such as raingardens and hedges, our gardens had lower ET uxes (Table 6). Evapotranspiration uxes from urban hedges may be enhanced by greater leaf area and height (Saher et al. 2020) and rain gardens are explicitly designed to in ltrate water and increase ET (Hess et al. 2017).

Conclusion
Greenspaces including rain gardens, turfgrass, and tree canopies provide essential hydrological services by increasing stormwater in ltration and retention in urban ecosystems (Zhang et al. 2015;Hoover and Hopton 2019). Our results suggest that urban gardens may function in a similar manner to engineered green stormwater infrastructure, with a high capacity for water retention, in ltration, and ET. Through the generation of ET uxes, urban greenspace and gardens also provide some UHI mitigation bene ts ). These bene ts are notable because urban gardens are constructed and maintained for a variety of other purposes such as crop production, recreation, social and aesthetic goals (McDougall et al.

2019)
, and hydrologic ecosystem services provided by urban gardens are a secondary bene t achieved for free.
These soil management strategies can enhance hydrological services from urban greenspaces and the soil moisture pool to mitigate the urban heat island effect (Qiu et al. 2013;Zipper et al. 2017;Yang et al. 2018) or to increase garden productivity (and indirectly increase evapotranspiration, Taylor 2020). Together, these results suggest that urban garden soil management strategies may enhance urban greenspace hydrological ecosystem services through both biological and physical mechanisms.

Declarations
Ethics approval and consent to participate: This study did not use human or animal subjects Funding: This study was supported by a National Science Foundation CAREER award (award number 1651361) to GES.
Con icts of interest/Competing interests: The authors declare no con icts of interest Availability of data and material (data transparency): All data from this study are available from EJC upon request.
Code availability (software application or custom code): All R code from this study is available from EJC upon request.  Table 5. Leachate and ET fluxes originating from precipitation and from irrigation, for high municipal compost (HMC) and no fertilizer treatment garden plots, for the four climate scenarios. S0 = baseline precipitation, S1 = 30% increase in magnitude for all precipitation events, S2 = 30% increase in magnitude for spring and fall precipitation events, S3 = 25% increase in magnitude of precipitation events exceeding 5.08 cm. Values are reported in cumulative mm over the 880 day simulation period.