Introduction

Water pollution has become a major concern as it affects the environment and all living organisms. The hypoxic zone (HZ) in the Gulf of Mexico (GoM) has become an environmental challenge and it is monitored on a yearly basis (United States Environmental Protection Agency 2019). The size of the HZ fluctuates, and efforts have been made to reduce the hypoxic zone to an area of 5000 km2. The HZ size can be influenced by climatic and anthropogenic factors such as rainfall, hurricanes, and agricultural practices in the Mississippi River basin ((National Oceanic and Atmospheric Administration 2020). Nutrients coming from agricultural areas are the main contributor of the increase of the HZ (Rebich et al. 2011). Literature concurs that runoff is the main pathway for P losses from agricultural fields, while N losses are usually higher in subsurface flow (Melland et al. 2008; Norberg et al. 2022). Another study reported significantly higher TP (2.3–123 times) and total dissolved nitrogen (TDN) (2.6–5 times) loads in runoff compared to losses in subsurface flow (Ibrahim et al. 2013). A The Mississippi and Atchafalaya Rivers combined deliver approximately 90% of the nitrogen (N) exported to the GoM (Lu et al. 2020). The elevated concentration of nutrients getting into the GoM reflects the water quality of inland water bodies. The Mississippi River basin drains the corn belt, and agriculturally dominant areas, including the Midwest (Natural Resources Conservation Service 2013). For instance, 46% of the rivers and streams of the USA present a degraded biological status, and 21% of the lakes of the country have eutrophication problems (United States Environmental Protection Agency 2017). The report mentioned also that the leading causes of the poor condition of water bodies was nutrient pollution, sediment, and pathogens. Agricultural practices are closely associated with water pollution due to discharge of water enriched in sediments, N, P, and herbicides to water bodies (Jaynes et al. 1999; United States Geological Survey 2000; Denver et al. 2018; United States Environmental Protection Agency 2018; Mittelstet et al. 2019).

Although eradicating fertilizers and manure use in agriculture is not viable, the implementation of conservation practices like agroforestry can significantly reduce runoff, sediment, and nutrient pollution from agricultural watersheds (Udawatta et al. 2002, 2011). Agroforestry buffers are known to reduce runoff by enhancing infiltration and reducing soil moisture (Anderson et al. 2009). Field studies have found runoff reductions due to agroforestry buffer or vegetative filter strips. For instance, a study of small agricultural watersheds in the northeast and central Missouri by Udawatta et al. (2011) found that the agroforestry buffers within watersheds under corn-soybean rotation and grazed watersheds reduced runoff by 19% and 49%, respectively. Another study by Veum et al. (2009) found that the same grass buffers of the current study decreased runoff by 8.4%, while the agroforestry buffers did not have a significant effect. A review mentioned that agroforestry systems had 57% lower runoff compared to conventional crop systems (Fahad et al. 2022). Also, a plot study found that an 8-m fescue filter strip reduced runoff by 13% compared to plots without filter strips (Blanco-Canqui et al. 2004).

Additionally, agroforestry practices increase soil microbial biomass and improve the denitrification potential of soils (Zhu et al. 2020). Overall, agroforestry increases the retention time of water in agricultural fields and enhances mechanisms that reduce nutrient concentration and sediment loss in the discharge of watersheds.

Agroforestry buffers reduce sediment losses by enhancing soil aggregation, protecting soils from erosive agents, and decreasing runoff. For instance, agroforestry buffers improve soil hydraulic properties by enhancing macro porosity, which results in increased infiltration capacity of soils and less runoff (Schoeneberger et al. 2017; Alagele et al. 2019b). Also, tree buffers in agricultural fields have been shown to enhance soil organic carbon (SOC) retention that boosts soil water holding capacity, soil bonding, and nutrient retention capacity (Alagele et al. 2019b; Salceda et al. 2022). Intense precipitation events produce the most sediment losses in agricultural fields (Rankl 2004). Vegetative buffers reduce erosion by retaining the soil with their root systems (Noordwijk et al. 2015). Also, vegetative covers reduce the susceptibility of soil to erosion by acting as a barrier between erosive agents, and soil particles (Basche and Edelson 2017). Therefore, by reducing runoff, agroforestry buffers reduce sediment losses in agricultural fields. By using three paired watersheds, Udawatta et al. (2011) showed that agroforestry buffers reduced sediment loss in runoff by 30% and 28%, respectively. Also, by modeling the effects of agroforestry in combination with other conservation practices; (Natural Resources Conservation Service 2013) reported of possible sediment loss reductions from 20 to 99% in the Lower Mississippi River Basin.

The implementation of agroforestry buffers on agricultural fields also decreases N pollution by enhancing physical and chemical mechanisms of nutrient retention and reduction (Udawatta et al. 2006, 2011; Natural Resources Conservation Service 2013; Salceda et al. 2023).

Specifically, vegetative buffers in agroforestry decrease loads of N in runoff from agricultural fields to bodies of water by favoring sedimentation, nutrient adsorption, and denitrification (Muscutt et al. 1993; Lowrance et al. 1997; Mayer et al. 2005). Particulate N can be retained in the soil during infiltration, which favors reactions of N products with other compounds. Also, agroforestry buffers absorb N for their physiological processes, thus decreasing the amount of N available to export (Hickey and Doran 2004). Another important benefit of tree buffers is that they enhance conditions for denitrification by increasing soil microbial biomass and carbon retention; Thus, improving the denitrification potential of soils (Mayer et al. 2005; Zhu et al. 2020). According to (Udawatta et al. 2011), agroforestry and grass buffer treatments reduced total N losses in the runoff by 75% and 68%, respectively, in comparison with losses without the vegetative buffers. Also, Dillaha et al. (1989) reported that vegetative filter strips of 4.6 m and 9.1 m decreased N concentration by 54–73%, respectively. Hickey and Doran (2004) indicate that N removal from agricultural fields is highly variable, and the soil’s capacity for N removal can reach up to 97% in organic soils, while it is significantly low for mineral soil.

Agroforestry buffers reduce P concentration in the runoff by consuming it and enhancing its infiltration (Hickey and Doran 2004; Parkyn 2004; Mayer et al. 2005). Tree buffers can intercept runoff, and boost sedimentation, which reduces particulate P transport, and favors P uptake by plants. For example, Udawatta et al. (2004) found a strong relationship between sediment and P loss; thus, reducing sediment loss and P losses. Vellidis et al. (2003) reported P removal up to 66% in a wetland riparian environment. Moreover, Dillaha et al. (1989) found that vegetative filter strips of 4.6 m and 9.1 m width reduced P loss by 61–79% of the P load in the runoff. (Parkyn 2004) points out that vegetative buffers 9.1 m width, reduced P concentration in runoff by 79%, in a riparian environment. The establishment of agroforestry and grass buffers in small agricultural watersheds reduced total P loss by 26% and 22%, respectively (Udawatta et al. 2011).

Previous research on the same watersheds demonstrated that agroforestry conservation practices substantially reduced runoff, sediment, and nutrients losses (Udawatta et al. 2002, 2011). However, studies of the long-term effect of agroforestry buffers on water quality are limited. The buffers of the current study have had an increased performance at improving water quality as the system matures. From 2012 to 2019, crop areas of the agroforestry and grass buffer watersheds were managed with biomass crops (Switchgrass Panicum virgatum, Roundtree big blue stem Andropogon Gerardii, Rumsey Indiangrass Sorghastrum nutans (L.), Illinois bundle flower Desmanthus illinoensis, and Partridge pea Chamaecrista fasciculata). This perennial vegetation may have changed the hydrology and soil properties, thus affecting non-point source pollution (NPSP). The purpose of this study was to determine the long-term effects of agroforestry practices on NPSP. The specific objectives were to (i) quantify the runoff, sediments, and nutrients exported from the watersheds, and (ii) compare long-term effects of the agroforestry buffers and land managements on water quality.

Materials and methods

Site description

The study watersheds are located at the Greenley Memorial Research Center of the University of Missouri-Columbia, Knox County, Missouri USA (40°01′N, 92°11′W). The study site consists of three adjacent north-facing watersheds instrumented with water monitoring and sampling devices in 1991. The east watershed (control) has an area of 1.65 ha, in the center is located the agroforestry watershed (4.44 ha), and on the west is the grass buffer watershed (3.16 ha) (Fig. 1). For more information about the study site refer to Salceda et al. (2022).

Fig. 1
figure 1

A The inset map shows the location of the Missouri State in USA, Knox County in Missouri, and the three paired watersheds in Knox County. B Three watersheds of the paired watershed study at Greenley Research Center, Novelty, Knox County, MO. Wide cyan bands represent grass buffers on the contour grass strip watershed, while wide brown bands represent agroforestry buffers (tree + grass) on the agroforestry watershed. Wide green bands represent grass waterways on all three watersheds, and narrow lines represent 0.5-m interval contour lines

Full size image

The soils in the watersheds are mapped as Kilwinning silt loam (fine, smectitic, mesic Vertic Epiaqualfs), Putnam silt loam (fine, smectitic, mesic Vertic Albaqualfs), and a small portion of Armstrong loam (fine, smectitic, mesic Aquertic Hapludalfs) (SoilWeb, 2019). The southern part of three watersheds (1–3% slopes) has Putnam silt loam soil, the Kilwinning silt loam occupies the central portion of the three watersheds (1–5% slopes), and the northern part of the watersheds (5–9% slopes) consists of Armstrong loam soil (SoilWeb, 2019).

The soil profiles have Bt horizons ranging from 20 to 36 cm deep, with low saturated hydraulic conductivity. The shallow argillic horizons with low hydraulic conductivity enhance runoff during rain events. According to WebSoil (2019), the pH for Putnam and Kilwinning silt loam soils ranges 5.7–7.1, and for the Armstrong series ranges 6–7. (Udawatta et al. 2014) reported mean SOC on the surface of the control and agroforestry watersheds of 1.93% and 1.99%, respectively. The annual long-term precipitation (2000–2022) in the study area is 978 mm (http://agebb.missouri.edu). Approximately 67% of the precipitation falls in April through September. The mean temperature in January and July is − 1.79 °C and 24.4 °C, respectively.

Treatments and sampling procedure

The grass buffers (GB) consist of grass-legume strips 4.5 m (15 ft.) wide with Redtop (Agrostis gigantea Roth), brome grass (Bromus inermys Leyss.) and birdsfoot trefoil (Lotus corniculatus L.) spaced 36.5 m (120ft). The ABs and GBs had a spacing ranging from 22.8 m in the lower slope areas to 36.5 m in upper slope areas. The tree species in the AB were Pin Oak (Quercus palustris Muenchh.), Swamp White Oak (Q. bicolor Willd.), and Bur oak (Q. macrocarpa Michx.), located in the center of the grass-legume strips planted at 3 m apart. In 2012, crop areas of the agroforestry buffer and grass buffer watersheds were planted with a 9 kg ha−1 mix of Roundtree big blue stem (Andropogon gerardii Vitman) (38.3%), Rumsey Indiangrass (Sorghastrum nutans L. Nash) (28.3%), Kanlow switchgrass (Panicum virgatum L) (16.4%), Illinois bundleflower (Desmanthus illinoensis (Michx.)) (7.8%), and partridge pea (Chamaecrista fasciculata (Michx.)) (9.2%) until late 2019. Corn was planted only in the control watershed in the spring of 2019, soybean was planted on all three watersheds in 2020, corn was planted in 2021 and soybean in 2022 (Table 1).

Table 1 Summary of the agricultural activities on the three paired watersheds during the study
Full size table

The control watershed had the smallest drainage area and was instrumented with a 0.91 m (3 ft.) H-flume; the other two watersheds were instrumented with 1.37 m (4.5 ft.) H-flumes. H-flumes were placed at the discharge point of each of the three watersheds and equipped with bubbler flow measuring devices (ISCO 4230) (ISCO, Inc, Lincoln, Nebraska) and automatic samplers (ISCO 3700) to record flow rates and collect samples, respectively. The runoff on each watershed went through a grass waterway and was subsequently transported to a sampling structure and an H-flume. The instrumentation and calibration period started in 1991; during this period, sediment, total N, total P, nitrate, and ammonia data were collected; however, no treatment was applied. During the calibration period (1991–1997), the three watersheds were monitored for runoff, sediments, NN, TN, and TP to develop a baseline linear model to account for innate differences among the watersheds. The calibration linear models were used to estimate the response of the treated watersheds after the treatment was applied based on the response of the control watershed (Without disturbance). In 1997 treatments were established on the west and center watersheds, leaving the control watershed without treatment.

Sampling period was from November 2019 to December 2022. The bubbling lines and internal pump of the bubbler flow meters can be damaged as water freezes. Thus, the instruments were disconnected when freezing temperatures were expected. After measurable runoff events, samples were collected, flow-weighted, and analyzed. Some rain events did not produce enough runoff to activate the samplers.

Laboratory analysis

After rain events, samples were labeled and brought to the laboratory. The analysis of the water samples was performed in the laboratory of the Center of Agroforestry and School of Natural Resources of the University of Missouri-Columbia. First, samples were separated into two 65-mL bottles, unfiltered samples were used for total nitrogen (TN) and total phosphorus (TP) analysis, and filtered ones for nitrate (NO3−1), and dissolved P analysis. Unprocessed samples were preserved at 4 °C until their analysis. Filtration was performed by using previously washed, dried and weighted 934-AH glass microfiber filter paper (Whatman, Maistine, UK), placed into a vacuum pump system to obtain sediment weight per filtered volume. After filtration, filter papers were placed in an oven at 105 °C for 48 h, and weights of the dry filter papers were recorded. Loads of sediment were determined by analyzing the flow data retrieved from the flow meters placed on the fields, obtaining the total volume of runoff per rain event, and multiplying the volumes by sediment concentrations. Loads of sediment per unit of area was determined by dividing the total sediment load per event per catchment area of the watershed.

Total N, TP, nitrate, and dissolved P contents were determined by Lachat Quickhem Methods 10-107-04-4-B, 10-115-01-1-F, 10-107-04-1-O, and 10-115-01-1-A, respectively. The TN procedure requires the digestion of the samples in an autoclave, adding to the samples 5-mL of digestion solution conformed by established proportions of sodium hydroxide and potassium persulfate to a 10-mL of sample. The TP method also requires the digestion of the samples in an autoclave, adding 20 mg of ammonium peroxydisulfate to a 10-mL of sample. The detection limit for the three methods was ≤ 0.002 mg L−1.

Statistical analysis

A paired watershed approach was used to determine the effects of the AB and GB. A paired watershed approach assesses differences due to treatments by establishing a baseline between treated watersheds and a control watershed before the treatments are implemented (Calibration period) and comparing this baseline with the relationships between the same watersheds after the treatment is implemented (Clausen and Spooner 1993). The relationship used in the paired watershed approach is linear between a treated and a control watershed. The method indicates that the effects of the treatments are determined through comparing the treated versus control watershed line before and after the treatment establishment (Clausen and Spooner 1993). The method also considers that the relationship obtained for the parameters of interest during the calibration period represents the expected values in the treated watersheds if the treatment was not applied. The significance of the linear relationships during the calibration and treatment period were first explored with an ANOVA. Then, differences on slopes and intercepts were determined with an ANCOVA. Previous research established calibration relationships between the two treated watersheds and the control (Udawatta et al. 2002). These calibration equations showed highly significant relationships between watersheds for all measured constituents. A new linear regression analysis was performed by using the program R version 4.3.1. Relationships between the control and treatment watersheds for runoff, sediment loads, TN, TP, and Nitrate were determined. Based on the correlation coefficient and ρ values, the fit and significance of the data were evaluated. Once a linear model was developed, it was used to compare the effects of changes in the treatment with scenarios without treatment and previous treatments. The former calibration developed by Udawatta et al. (2002) was used to calculate the changes in measured variables in comparison with the no treatment scenario also reported as the expected values.

Results and discussion

Precipitation

The wettest month during the study was July 2021, where several floods were reported throughout Missouri. The driest month was February 2021, being only 18% (7.62 mm) of the long-term average for February (41 mm). Year 2022 was the driest during this study, representing 79% (775 mm) of the long-term mean annual precipitation (978 mm), followed by 2020 with 81% (796 mm) of the long-term mean annual precipitation. The wettest year was 2019 with 132% (1297 mm) of the long-term mean annual precipitation, followed by 2021 with 114% (1115 mm) of the long-term mean annual precipitation. Overall, the annual precipitation at the study site followed the pattern 2022 < 2020 < 2021 < 2019 (Fig. 2). The long-term precipitation in the study site shows a bimodal distribution with greater precipitation between May and June and September and October. Dry and wet years seemed to alternate during the study.

Fig. 2
figure 2

Monthly precipitation (bars) for 2019, 2020, 2021, 2022, and long-term mean (2000–2022) (thick black line) at the study site in Novelty, Knox County, Missouri

Full size image

Parameter relationships

Nine paired events were recorded for the AB watershed and six for the GB watershed during the study. The recorded events were used to develop linear relationships for runoff, TSS, NN, TN, OP, and TP. Various runoff events were recorded solely in one watershed but not in the respective coupled watershed; Thus, those events were not used for calibration purposes. Overall, runoff, NN, and TN had the strongest relationships, while sediment, DP and TP had a weakest correlation (Table 2). Overall, the AB watershed had stronger linear models than the GB watershed. The new linear relationships for runoff, NN and OP for the AB and GB watersheds were significant at the 95% confidence level. The new TN linear model was significant only on the AB watershed. Total suspended solids and TP loss relationships were not significant on both watersheds.

Table 2 Linear relationships for runoff, sediment, nitrate–nitrite (NN), total nitrogen (TN), ortho-phosphate (OP), and total phosphorus (TP) between the agroforestry and grass buffer watersheds and the control watershed
Full size table

The calibration equations from Udawatta et al. (2002) were used to calculate values for runoff, sediment, NN, TN, and TP losses (expected values) from the AB and GB watersheds for comparison with new observations (Table 3). Even though ortho-phosphates (OP) data were collected on the current study, Table 3 does not contain this parameter since there was no calibration equation for OP. The events recorded for the AB and GB watersheds were totaled for each of the five parameters in Table 3. Also, predictions calculated from observations in the control watershed were totaled and compared with observed values. Increases of runoff were observed in both watersheds compared to the predicted values. Total N losses in runoff increased in the GB watershed compared to predictions. All the other parameters reported reductions compared to the predictions.

Table 3 Agroforestry (AB) and grass buffer (GB) effect on changes in runoff, total suspended solids (TSS), nitrate–nitrite (NN), total nitrogen (TN), and total phosphorus (TP) at the studied watersheds
Full size table

Runoff

The grass buffer watershed had more recorded events (17) than the agroforestry buffer watershed (14) and the control watershed (12). However, paired events with the control watershed for the GB watershed were fewer than the events between the AB and control watersheds. Not all precipitation events produced runoff on the watersheds. Some events were not recorded or sampled due to instrument failures or animal damage. Therefore, the total events used for comparison between the AB, GB, and control watersheds were different (Fig. 3).

Fig. 3
figure 3

Change in runoff discharge (observed minus predicted) on the agroforestry buffer (AB) watershed and the grass buffer (GB) watershed during the treatment period between 2019 and 2022. The date format is (mm/dd/yy)

Full size image

The runoff linear relationships between the AB versus Control and GB versus Control were both significant (ρ < 0.05). After an analysis of covariance, it was found that the linear relationship between the AB-Control and GB-Control were not significantly different from each other. Also, the linear models during the calibration and treatment period were compared with an ANCOVA, finding that they were different (ρ < 0.05). The observed runoff in the two watersheds increased compared to the expected value (without the implementation of treatments) (Table 3). The slope of the runoff relationship between the grass buffer watershed and the control increased from 0.999 (Calibration) to 1.229 (Treatment after 25 years). On the other hand, the slope of the relationship in runoff between the agroforestry buffer watershed and the control watershed decreased from 0.770 (Calibration) to 0.572 (Treatment after 25 years). According to Table 3, the observed runoff in the AB watershed increased by 6.10% compared to the expected value. The runoff in the GB watershed was 30.46% greater than expected without the treatment. The greatest runoff reduction in the AB watershed corresponded to the runoff event collected in 04/11/2023, which had the greatest precipitation amount. The GB watershed only had runoff reductions in a single event, and it was not significant. The recorded events during the treatment period with their respective precipitation are shown in Table 4.

Table 4 Date of runoff collection with its respective total precipitation per event
Full size table

The runoff findings showed no reductions in runoff in either the AB or the GB watersheds compared to previous studies in the same watersheds. For instance, Udawatta et al. (2002) found runoff reductions in the same watersheds of 1% and 10% in the AB and GB watersheds, respectively, after three years of buffer establishment (Table 4). Another study on the same three paired watersheds from 2004 to 2008 showed runoff reductions by 15% and 23% for the AB and GB watersheds, respectively (Udawatta et al. 2011). A review by Zhu et al. (2020) on several agroforestry studies found that AB reduced runoff from 1 to 100%. The wide variation in runoff reductions in agroforestry systems indicates that site and management factors are important in runoff discharge. The increase in runoff found in the current study in the AB and GB watersheds might have been caused by runoff coming into the watersheds from surrounding areas in which the watershed berm divisions have deteriorated due to the traffic of heavy machinery. The control watershed is away from the main road and there is no evidence that the integrity of its surrounding berm has been compromised. A paired watershed study by Gassman et al. (2010), reported an 8% increase in watershed discharge of the treatment watershed compared to the control, like the current study.

Total suspended solids (TSS)

The new TSS relationships were not statistically significant at 95% confidence level for both watersheds. Overall, observed sediment discharge was lower than the discharge in the control watershed and the expected values on the treated watersheds (Table 3). The observed sediment discharge in the AB watershed was 62% lower than the expected discharge. The observed sediment discharge in the GB watershed was 71% lower than the expected discharge. The third and fifth event recorded the greatest sediment reductions (Fig. 4). Only a single event had increased sediment loss in the GB watershed and three in the AB watershed. Since the linear models during the treatment period were not significant, statistical differences in TSS between watersheds and monitoring periods were not reported. The greatest TSS reductions were observed on the events on 04/11/2021 and 06/19/2021. The first great TSS loss reduction event for the AB watershed occurred during the greatest precipitation event recorded on this study. The second and greatest runoff reduction event for both watersheds occurred in Summer 2021 when the soil was especially dry.

Fig. 4
figure 4

Change in sediment discharge (observed minus predicted) on the agroforestry buffer (AB) watershed and the grass buffer (GB) watershed during the treatment period between 2019 and 2022. The date format is (mm/dd/yy)

Full size image

A study in the same watersheds by Udawatta et al. (2002) found increases in sediment losses by 35% and 17% in the AB and GB watershed after three years on buffer implementation (Table 5). A more recent study (2004–2008) in the same watersheds indicated that sediment losses in the AB and GB watersheds were reduced by 30% and 28% (Udawatta et al. 2011). Zhu et al. (2020) indicated in a meta-analysis that agroforestry buffers reduced soil loss in runoff from 15 to 97%. Another paired watershed study by Gassman et al. (2010) found a 7% decrease in TSS in flow discharge in the treatment watershed (with a mix of conservation practices) compared to the control. The current study had TSS reductions within the reported range in the literature, even though the reductions were not significant at the 95% con. The decrease in sediment losses after 25 years compared to results of previous studies can be attributed to tree maturity and improvements in soil hydraulic conductivity, porosity, and biological soil properties (Seobi et al. 2005; Adhikari et al. 2014; Alagele et al. 2019a; Salceda et al. 2022). The tree height and diameter at the breast height (DBH) in the AB watershed were measured in 2021. The results showed that the pin oaks, swamp white oaks, and burr oaks increased their height from 8 to 10 times since 1999. Likewise, the mean diameter of the three species increased from 12 to 22 times since 1999. Water use by trees and perennials could have increased the soil’s water storage capacity, reducing runoff and sediment losses during rain events (Anderson et al. 2009; Udawatta et al. 2011; Sahin et al. 2016; Alagele et al. 2020). Well established conservation practices such as agroforestry buffers and vegetative filter strips can decrease erosion and TSS in runoff by acting as a barrier to slow down runoff, enhancing infiltration.

Table 5 Effect of the grass and agroforestry buffers on runoff, total suspended solids (TSS), nitrate–nitrite, total nitrogen, and total phosphorus
Full size table

Nitrate–nitrite (NN) losses

The nitrate–nitrite losses from the GB and AB watersheds were reduced compared to the expected values (Table 3). The AB-Control watershed linear model during the treatment period for NN losses was significant, while the relationship for the GB-Control relationship was not significant. Therefore, differences between the linear models between the AB and GB watersheds with the Control watershed were not reported. The AB watershed linear model during the treatment period was different to the model during calibration, indicating a significant reduction in NN. The AB watershed had 25% less NN discharge compared to the expected values. The reduction on the GB watershed was only 14%. The greatest reductions of NN in the AB watershed were observed in the third and fourth runoff events. The GB watershed had its greatest NN reductions during the second and third runoff events (Fig. 5). These events were associated with > 43 mm rainfall; thus, indicating effective control during large events. The slope of the linear relationship between the AB and the control watershed was also reduced compared to the calibration slope. The slope of the calibration NN relationship between the AB and the control watershed was reduced from 0.763 to 0.510, indicating that the AB effectively reduced NN losses. The relationship between the GB and control watersheds was also reduced, decreasing from 0.577 to 0.450, but it was not statistically significant. The greatest NN reduction occurred during the greatest rainfall event of the monitoring period. The GB watershed had NN reductions only in two events. Nevertheless, the two events with reductions in the GB watershed were significant enough to balance the NN change to an overall reduction of 14% in the GB watershed.

Fig. 5
figure 5

Change in nitrate–nitrite discharge (observed minus predicted) on the agroforestry buffer (AB) watershed and the grass buffer (GB) watershed during the treatment period between 2019 and 2022. The date format is (mm/dd/yy)

Full size image

Other studies in the same watersheds have reported NN reductions due to agroforestry buffers. For example, Udawatta et al. (2002) reported reductions of NN losses in the AB and GB watersheds by 26% and 39%, when the largest events were excluded, during a three-year treatment monitoring period (Table 5). The AB watershed had a greater NN loss reduction efficiency after 25 years compared to the efficiency after three years, suggesting that better established buffers increased their efficiency and retained applied fertilizers. Similarly, (Gassman et al. 2010) found that conservation practices in a paired watershed study decreased nitrate losses by 15% compared to the predictions. A review by Zhu et al. (2020) indicated a high variation in N loss reduction among different agroforestry systems and study sites, where N losses increased in some cases and decreased in others. Other studies found NN reductions in runoff of agroforestry systems from 24 to 97% (Schmitt et al. 1999; Schoonover et al. 2005).

The current study had NN reductions close to the lower extreme of the range reported in the literature. Agroforestry buffers and other conservation practices can reduce nitrogen and other nutrient non-point source pollution if enough time for their effect is allowed (Meals et al. 2010). The GB watershed had an increase in NN loss probably due to the increase in runoff coming from surrounding cropland areas result of damaged sections of the perimetric berm. The berms on the limits of the GB and AB watersheds should be evaluated and restored where needed for future studies. The AB watershed was less affected by incoming runoff than the GB watershed. Therefore, despite an increase in runoff, the AB watershed had recorded a decrease on NN losses compared to predictions and previous studies. The GB watershed had a lower reduction of NN losses compared to reductions of previous studies, but it was lower compared to the expected values. The reductions of NN losses despite the increase in runoff from surrounding areas indicate that the effect of the AB and GB on NN loss is greater than the numbers on this study indicate.

Total nitrogen (TN) losses

The TN linear models between the AB and GB watersheds with the control watershed were significant. The linear models of the AB-Control and GB-Control during the treatment period were not different from each other at the 95% confidence level, but at the 90% (ρ = 0.06). Additionally, both treatments were different (ρ < 0.05) when comparing their calibration and treatment period, indicating a significant reduction in the AB watershed and an increase in the GB watershed. The AB watershed had TN loss reductions of 64% compared to the expected value (Table 3). On the other hand, the GB watershed had an increase in TN losses. The increase of TN losses can be explained by runoff with TN coming from adjacent areas into the GB watershed. The third and fourth events had the greatest reductions on TN losses in the AB watershed (Fig. 6). The slope of the TN discharge relationship between AB and Control watershed was reduced from 0.636 to 0.180. Conversely, the slope of the TN discharge between the GB and Control watershed increased. As it was found with the NN losses, the greatest TN reduction in the AB watershed was found in the runoff event collected in 04/11/2021 that had the most rainfall during the monitoring period. This single event was responsible for the overall TN reductions found. The buffers effect had a stronger influence in water quality in heavy rainfall events compared to mild precipitation events.

Fig. 6
figure 6

Change in total nitrogen discharge (observed minus predicted) on the agroforestry buffer (AB) watershed and the grass buffer (GB) watershed during the treatment period between 2019 and 2022. The date format is (mm/dd/yy). The recorded runoff on 04/11/2021 in the AB watershed was truncated with a white diagonal rectangle for visualization purposes. This event had the largest reduction of TN loads corresponding to − 52 kg ha−1

Full size image

A study by Udawatta et al. (2002) reported TN reduction in the AB and GB watersheds by 21% and 20%, respectively (Table 5). Udawatta et al. 2011) found that the TN loss in the AB and GB watersheds between 2004 and 2008 was reduced by 11% and 13%, respectively. The AB watershed further reduced TN losses in runoff, going from 21 and 11% in previous studies to 64% in the current study. The reported TN reduction in the AB watershed (64%) is the greatest TN reduction found in the watersheds by this and previous studies. The GB watershed, conversely, had an increase in TN losses, which can be explained by the increase of runoff in the GB watershed from surrounding areas. Dillaha et al. (1989) found that vegetative buffers 4.6 and 9.1 m wide decreased TN in runoff in plots from 23 to 92% depending on slope and rainfall duration. The 30% runoff increase in the GB watershed from adjacent areas is believed to be the reason for the increased TN loss. A review reported highly variable results on AB effects on TN, with increases and reductions in TN, ranging from − 265 to 100% (Zhu et al. 2020). The current study concurs with the reductions reported in the review report.

Total phosphorus (TP)

Even though numerical TP reductions were found, the linear relationships between AB-Control and GB-Control during the treatment period were not significant. The AB and GB watersheds had numerical reductions in TP losses of 23% and 33%, respectively (Table 3). Total P reductions were reported despite the increase of runoff in both watersheds, indicating a possible benefit of the buffers on TP reduction. Only two runoff events reported increases in TP losses (Fig. 7). Since none of the TP relationships during the treatment period were significant, differences between watersheds and before and after treatments were not reported. The TP reduction in 04/11/2021 was the greatest on the AB watershed. Despite the increased in the first two events in the AB watershed, the subsequent events had a significant effect on the overall TP loss. Similarly, the GB watershed had an increase on TP losses in only one event and TP reductions in all the other events.

Fig. 7
figure 7

Change in total phosphorus discharge (observed minus predicted) on the agroforestry buffer (AB) watershed and the grass buffer (GB) watershed during the treatment period between 2019 and 2022. The date format is (mm/dd/yy)

Full size image

Studies on the same watersheds have reported TP reductions throughout the years ranging from 8 to 26%. Udawatta et al. (2002), reported TP loss reductions by 17% and 8% in the AB and GB watershed, respectively, after three years of buffer establishment (Table 5). Another study of the paired watersheds from 2004 to 2008 had TP loss reductions in the AB and GB watersheds of 26% and 22%, respectively (Udawatta et al. 2011). The GB watershed had the greatest TP loss reduction reported in the studied watersheds, and the AB had the third largest TP loss reduction. Blanco-Canqui et al. (2004) tested 8-m wide fescue filter strips, finding that plots with fescue strips reduced P (OP + TP) up to 84% compared to plots without strips. Total P loss reductions by agroforestry buffer vary widely due to site and management factors, with reported ranges from − 90 to 100% (Zhu et al. 2020). The current study falls within the reduction range in the literature.

Summary and conclusions

The current study reported significant and numerical reductions for some parameters of interest despite having some inflow from agricultural areas under the same management. The changes in sediment losses were insignificant at the 95% confidence level. However, the AB and GB watersheds reported 385 kg ha−1 and 355 kg ha−1 less sediment loss than the expected values without the buffers. The sediment loss reductions concur with the literature and current knowledge, indicating that the buffers helped reduce erosion in the treated watersheds. The effect of the buffers in NN losses was only significant in the AB watershed, reporting 6.13 kg ha−1 less than expected. The GB watershed had minimal reductions that were not significant. The GB watershed was the watershed that received more inflow from surrounding areas, which is assumed to have caused negligible reductions in NN. Similarly, TN reductions were found only in the AB watershed, and an increase was reported in the GB watershed. The AB watershed prevented 50 kg ha−1 of TN from being discharged during the treatment period (2019–2022). Reductions in TP were reported for both watersheds; however, these reductions were not significant. The AB and GB watersheds had 3.5 kg ha−1 and 3.17 kg ha−1 less TP losses due to the buffers effect. The results of this study indicate that despite the challenges related to having some inflow in mainly the GB watershed, the buffers influenced the water quality of the watersheds, favoring the reduction of water pollutants.

The performance of the buffers over time in TSS reduction was improved, even though the reductions were not significant at the 95% confidence level. The reductions reported in the current study were 1.8 to 4.2 times greater than the reductions reported in previous studies. The reductions of NN in the current study were lower than the reductions found in the 2002 study. No reduction in TN was found in the GB watershed. On the other hand, the AB watershed had a significant reduction in TN that was three to five times greater than reductions found in previous studies conducted in 1999 and 2008. As mentioned above, the inflow from surrounding areas in the AB and mainly the GB watershed is assumed to cause lower NN reductions in the current study compared to reductions in data collected in 1999. Also, differences between grass and tree N uptake with age can affect the dynamic of N loss reductions. The TP reductions found in the current study were not significant, but they were numerically more remarkable than most of the reported reductions, except for the reduction in the AB in the study with data from 2008. Despite the lack of significance of the TP data, the buffers had a greater performance in reducing TP than the reductions reported in previous studies.

The results of this study indicate that the age of the system can affect the performance of agroforestry practices on water quality. Overall, reductions in the current study (25 years after buffer establishment) in TSS, TN, and TP were greater than reported reductions in the studies with data from 1999 and 2008. Also, the ABs produced a greater reduction of TSS, NN, TN, and TP than the GBs. The AB and GB watersheds continued to reduce agricultural pollutants after 25 years, even increasing their performance in some cases. These results should be used to promote perennials in farmlands to allow them to mature to protect the environment and nutrient retention for economic purposes. The findings of this study emphasize that long-term studies should be promoted to demonstrate the benefits of conservation practices accurately. The results also evidence the importance of maintaining these practices for a longer term for numerous water and environmental benefits.