Predicting water quality trends resulting from forest cover change in an agriculturally dominated river basin in Eastern Ontario, Canada

At over 348 million ha, Canada's forested lands account for around 9% of the world's forest cover. In 2013, 45,900 ha or under 0.02% of forests were lost to deforestation nationally (Natural Resources Canada 2016). That rate of deforestation represents a decrease from around 60,000 ha/yr (ca. 1990), and this rate is expected to continue to decrease in the future (Masek et al. 2011); but in many agro-ecosystems experiencing agricultural expansion, forest/tree/hedgerow cover may be decreasing at much higher rates (e.g., SNC 2016). Burpee et al. (2018) reports that between 1990 and 2015, the nationwide conversion of land to agriculture (clearing for pasture or cropping) accounted for 42% of deforestation, followed by natural resource development (24%), urban/transport/recreation development (16%), hydroelectricity development (13%), and forestry roads (6%). The South Nation watershed, a ∼3,750 km² watershed located in Eastern Ontario, Canada, has lost 4.1% of forest between 2008 and 2014, while some agriculturally intensive areas in the watershed experienced over 11% forest cover loss during this period as a result of, primarily, agricultural expansion. The watershed now has less than the 30% of forested area recommended for environmental health purposes (Bryan & Henshaw 2013).

Forests, treed areas, and hedgerows can provide an array of ecological services including, but not limited to, sequestration of carbon, water filtration, habitat for wildlife and beneficial insects, and temperature, wind, and flood regulation; they also provide recreational, aesthetic, and spiritual services (Foley et al. 2005; Bonan 2008; Bryan & Henshaw 2013; Tallis et al. 2013; Mitchell et al. 2014).

Forested and treed areas in watersheds can critically impact water quality/quantity in downstream hydrological receptors. Zhang et al. (2017) reviewed global studies conducted on 312 watersheds with the intent to examine the hydrological impacts (primarily runoff) of forest cover change and to define the respective contributions of spatial scale, climate, forest type, and hydrological regime to these impacts. A majority of the watersheds reviewed showed that loss of forest cover could increase annual runoff across multiple spatial scales, although the effect of forest cover gain was more complicated and inconsistent. Allan (2004) indicated that landscape-scale deforestation could critically degrade habitats and reduce biodiversity and pollutant buffering capacity. Several studies (e.g., Richards et al. 1996; Roth et al. 1996; Johnson et al. 1997; Wang et al. 1997; Sponseller et al. 2001; Ding et al. 2015; Liu et al. 2015; Meneses et al. 2015) have tied declining habitat, water quality, and reduced biological assemblages to the spread of agricultural land use in watersheds. Other studies that have considered the influence of forest cover on the health of stream systems have found that forest cover and riparian areas generally reduce nutrient and contaminant concentrations and loads and increase the diversity of invertebrates (Connolly et al. 2016; Kändler et al. 2017; Paula et al. 2018). Woodlands, riparian buffers (e.g., Schultz et al. 2004; Dosskey et al. 2010; Sunohara et al. 2012; Vigiak et al. 2016), hedgerows (Burel 1996; Benhamou et al. 2013), and/or combinations of these features (Herzog 2000) can trap sediments and reduce loadings of agro-chemicals and nutrients. In agricultural areas in Eastern Canada, Clément et al. (2017) found that streams with a forest cover ≥47% of the watershed had relatively ‘good’ water quality; below that threshold, however, eutrophic and meso-eutrophic conditions were more frequent. Conversely, due to the increased precipitation interception and evapotranspiration of forests, the removal of forest cover has been tied to increase in the baseflow and the streamflow (Maes et al. 2009; Nagy et al. 2011; Price 2011). Still, examinations of the watershed-scale influence of forest change on water quality in agricultural basins, particularly those located in cold temperate climates, are generally scarce.

Understanding water quality trends resulting from land-use changes, such as deforestation or conversion of treed lands to agriculture in agro-ecosystems, is paramount in determining where, when, and how to deploy practices that address water quality burdens. Furthermore, projections and simulations of potential water quality changes resulting from forest/tree land uses in a watershed can be used in nutrient-trading initiatives that use fixed infrastructure designed to reduce nutrient loading in targeted river systems (O'Grady 2008). Forests and treed areas can be considered fixed ‘natural’ infrastructure/features for such purposes. The impacts of deforestation and reforestation are also intertwined with climate variability effects on hydrology, and great care should be taken not to confuse the influence of one factor with the influence of another. While multiple approaches can be used to quantify the relationship between forest cover and hydrology (Wei et al. 2013), hydrological modeling is the most practical approach for watersheds. Modeling of large complex ecosystems may be the only viable way to forecast and project ecological indicators resulting from anthropogenic activities that interact variably with known and unknown transient biophysical processes (Krysanova et al. 1989; Coughenour 1992; Grizzetti et al. 2003; Doren et al. 2009; Vigerstol & Aukema 2011; Piroddi et al. 2015). Hydrological modeling tools, in particular, can be very useful for quantifying the response of hydrological systems to changes in land use (e.g., El-Khoury et al. 2015; Fan & Shibata 2015; Hlásny et al. 2015; Álvarez-Cabria et al. 2016; Fan & Shibata 2016) and land management practices (e.g., Kaini et al. 2012; Bosch et al. 2013; Motsinger et al. 2016; Haas et al. 2017; Hanief & Laursen 2019). Studies such as these inform planning of land-use management practices in order to help avoid water quality issues.

In this study, the Soil and Water Assessment Tool (SWAT) was used to predict the effect of forest cover change on water quantity and quality in a river basin in Eastern Ontario, Canada experiencing acute forest/woody vegetation loss over the past decade. Additionally, the influences of forest cover change on water quality are contrasted with water quality influences imposed by several greencover Best Management Practices (BMPs) (vegetated filter strips (VFS) and grassed water ways).

The South Nation watershed covers ∼3,750 km² in Eastern Ontario, Canada (Figure 1). Land use in the South Nation watershed is predominantly tile-drained cropland (corn–soybean–forage livestock production) and forest/tree cover with the remainder classified as primarily wetlands or urban/light rural areas. Around 101,040 ha (27% of the watershed) was forested/treed land in 2014, down by 5,124 ha (4.8% of the forested regions) from 2008 (Ontario Ministry of Natural Resources and Forestry 2008, 2014; South Nation Conservation 2016). Total forest/tree cover in 2014 was below the ‘high-risk’ threshold of 30% forest cover recommended by Environment and Climate Change Canada (ECCC). Below that threshold, less than half of all potential species richness and only marginally healthy aquatic systems are expected to be supported (ECCC 2013). Additionally, total P in the surface water exceeds the provincial standard (0.03 mg-P/L) throughout the river basin (South Nation Conservation 2014).

The SWAT is a deterministic, semi-distributed, continuous-time model to assist with assessments of the impact of land management on water supply and quality in watersheds (Arnold et al. 1998). The model allows for the assessment of hydrological response in complex watersheds by dividing the system into subwatersheds or sub-basins, which are generally delineated such that the entire area drains to a single outlet. Each sub-basin contains one or more hydrological response units (HRUs), which define a fraction of the sub-basin with consistent land-use, management, and soil characteristics. HRUs do not have spatial properties (and may not be contiguous); each is defined as a fraction of 1 for each sub-basin. Loadings (runoff with sediment, nutrients, etc.) are calculated for each HRU and then summed to define total sub-basin loadings.

The watershed limits and 31 sub-basins (Figure 2(a)) were delineated using Shuttle Radar Topography Mission elevation data (Jarvis et al. 2008). Sub-basins were further divided into HRUs based on common land-use and soil properties. Soil properties (Figure 2(b)) were derived from the Soil Landscapes of Canada version 3.2 (Soil Landscapes of Canada Working Group 2010). The land-use distribution was developed by combining data from the Agriculture and Agri-food Canada (AAFC) Crop Inventory for 2015 (selected for its superior resolution, Figure 2(c)) and forest/tree coverage from Digital Raster Acquisition Project Eastern Ontario 2008 (DRAPE, Ontario Ministry of Natural Resources and Forestry 2008, 2014). Land uses across the watershed have been generalized as water bodies, wetlands, urban/light urban areas, forest/treed, and cropland. Table 1 outlines the fractions of each land use across the watershed for year 2008 (a year relevant for assessing forest cover change in this study).

Cropland was ascribed a standardized 8-year rotation of corn–corn–corn–soy–forage–forage–forage–forage, with the starting point in the cycle randomized for each cropland HRU. This rotation was based on rotation examples from producers in the study region. Forages are often composed of grass, alfalfa, and clover mixtures, yet SWAT requires consistent crop specification for each HRU with the random distribution producing a mix of crop plantings across each sub-basin. The modeled crop and land management cycle is presented in the Supplementary Material (Table A1). Timing and fertilizer specifications were based on Sunohara et al. (2015) and Que et al. (2015) and personal communication with representative agricultural producers in the region. For parsimonious modeling purposes, we used the simplest option for fertilizer application available in the SWAT, essentially applying nitrogen and phosphorus on specific dates. The crop management practices used in the modeling were not defined to hindcast water quantity or quality trends but to provide a communized basis for scenario testing forest/tree cover losses on hydrological endpoints.

Tile drainage was defined within the SWAT using the following assumptions: (i) all tiles are at 1 m depth; (ii) tile drains are spaced 15 m apart (Sunohara et al. 2015); (iii) drain radius is 40 mm; (iv) lateral saturated hydraulic conductivity (k_sat) is equal to k_sat as defined by soil data; (v) time to drain to field capacity and drain lag time calibrated to 39.9 and 69.7 h, respectively.

From the suite of BMPs already packaged in the SWAT software, terracing, contour planting, and strip cropping were not considered relevant for the generally low relief South Nation watershed, while residue management was considered outside the scope of this study. However, VFSs and grassed waterways (GWWs) have been examined in SWAT for their potential to reduce sediment and nutrient loading: furthermore, these BMPs are germane to the region of study (see Sunohara et al. 2012; Wilkes et al. 2013).

Both VFSs and GWWs can help reduce agricultural and urban pollutants before they reach downstream receptors. VFSs are maintained strips of planted and/or indigenous vegetation located downslope from non-point pollution sources such as agricultural fields, while GWWs are artificial or natural vegetated channels that reduce and regulate flow (Kaini et al. 2012).

Meteorological inputs required by SWAT simulations include daily precipitation, maximum and minimum air temperatures, solar radiation, wind speed, and humidity. Precipitation, temperature, and wind speed data were obtained from data archives for four stations in or near the watershed at Russell, St Albert, South Mountain, and Morrisburg (ECCC 2018), while solar radiation and humidity data were extracted from the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis archive (Kalnay et al. 1996).

Streamflow data for calibration and validation of the SWAT model were sourced from ECCC (2018), while monthly water quality data were sourced from the Ontario Provincial (Stream) Water Quality Monitoring Network (Ontario Ministry of the Environment, Conservation and Parks 2020) for stations near the South Nation river basin outlet at Plantagenet, Ontario. Calibration and validation were limited to a whole-basin approach using the Plantagenet station due to poor monitoring coverage (across space and/or time) elsewhere in the basin. Additionally, due to issues with completeness of data records for stations within the basin, weather data for validation were also drawn from Kemptville, Ontario which is just several km west of the South Nation watershed boundary.

Calibration and validation were conducted using a monthly time scale. Metrics were calculated for months where observations were available and evaluated based on the recommendations of Moriasi et al. (2007).

To investigate the impact of forest/tree cover changes on water quality properties in the South Nation River basin, simulations were repeated for a series of forest/tree cover scenarios:

• The initial land-use scenario was based on land uses outlined in the model structure and data sources (section 2.2.1) and Table 1 for 2008, constructed from the land-use and soil-type sources cited previously.

• The second land-use scenario was developed by comparing forest/tree cover data from the 2008 and 2014 DRAPE surveys (DRAPE, Ontario Ministry of Natural Resources and Forestry 2008, 2014). For this study, reductions in forest cover from 2008 to 2014 were assumed to be converted to tile-drained cropland (a primary land conversion in the region), with other covers (urban, wetland and water) unchanged. Soil properties associated with forests were provisioned for those land uses; however, following the conversion of forested/treed areas to agricultural land, soil properties were assigned soil attributes reflecting the dominant soil types for that area with the inclusion of surface cultivation layers (Ap horizons) associated with cropland management practices.

• Further scenarios were generated to represent land uses in 2020, 2026, and 2030, assuming that the rate of deforestation was the same as between 2008 and 2014, i.e., an annual loss of 0.8% of forest cover (about 0.2% of the watershed area). The land-use changes were generated by adjusting the fraction in each forest cover and cropland HRU from the 2008 land use. This resulted in scenarios with a relatively narrow range of overall forest/tree cover (22.2–28.3% of the watershed).

• Scenarios 6–10 (Table 3) were developed with the watershed total forest/tree cover at 10, 20, 30, 40, and 50% to provide an extrapolated indicator of ‘watershed’ response to enhanced and denuded forest cover pressures. Land cover profiles for these scenarios were calculated by scaling the 2008 forest and crop cover HRUs for each sub-basin by the value required to adjust the total 2008 forest cover to each scenario total. In this way, the variation of forest/tree cover between sub-basins was maintained. The last three scenarios also correspond to the forest/tree cover thresholds considered high, medium, and low risk, respectively, for overall ecosystem health metrics provided by ECCC (2013). Again, it was assumed that all cropland is tile-drained, and in the case of reforestation, remnant tiles on reforested agricultural land were not considered operational.

The 11 scenarios are outlined in Table 3. The SWAT model was then used to simulate the quality and quantity of water in the outlet of each of the 31 subwatersheds and the basin outlet for each of these 11 scenarios with 31 years (1981–2011) of meteorological observations.

VFSs were simulated using default SWAT parameters. Of critical importance for the VFS simulations is the ratio of field area to filter area. Based on an average field size of approximately 15 ha in the South Nation basin, a VFS of width 5 m along two sides of a square field yields a ratio very close to the SWAT default of 40 which was used here and consistent with protected stream riparian zones evaluated by Wilkes et al. (2013) and Sunohara et al. (2012) in the South Nation watershed. For the simulation of GWWs, the average width of the waterway was set to 5 m and depth to 0.5 m based on the ratio used by Hanief & Laursen (2019) on the Grand River, Ontario and, again consistent with Wilkes et al. (2013) and Sunohara et al. (2012). These BMPs were simulated separately in all cropland HRUs and across all forest cover scenarios.

Calibration and validation performance statistics are given in Table 4. Based on the statistical measures alone, calibration can be considered at least satisfactory for the key parameters of interest for this study, while validation was considered acceptable.

Generally, the model appears to underestimate the highest spring flows and loads slightly and also slightly overestimate the summer flow and loads. However, for this exercise, the most important parameter is flow, given that our nutrient and agronomic parameterization is a generalization of land-use and agricultural management activities in the basin, and not specified and optimized to precisely hindcast water quantity and quality properties. All the parameters mentioned in the paper (flow, sediment, total phosphorus, and total nitrogen) had a satisfactory performance in calibration. Total phosphorus missed the satisfactory threshold by very little (0.47 instead of 0.5) in calibration. The performance is lower for nitrate (0.27), but in line with other work including El-Khoury et al. (2015), Shrestha et al. (2021), and Zango (2021). Hence, calibration and validation results can be considered, in the context of this study, as acceptable.

For a study such as this, it is useful to consider the typical quality, quantity, and route of water exported from HRUs, with the caveat that flows and loadings cannot be calibrated at this scale.

Figure 3 shows average annual flows to the reach by HRUs for the six cover types included in this study as well as the average of all crops (since cropped areas rotated through all five crops). Surface runoff from forested HRUs is approximately 5% lower than the average of cropped areas, while the absence of tile drainage in the forested areas allows for significantly higher soil water contents and higher interflow and groundwater flow. Figure 4 shows the nitrogen applied to each crop and the nitrate exports to the South Nation River for each cover simulated, as transported by surface runoff, interflow, and tile drainage.

It can be seen in Figure 4 that nitrate flushing by lateral flow (interflow) is generally insignificant compared with surface runoff and tile flow. Results (not shown) also indicate, unsurprisingly, that nitrate exports are largely proportional to the application of nitrogen applied to each cover type, except for red clover. The nutrient inputs for the forage crops are assumed to be provisioned by manure-based fertilizers; manure (solid and liquid) are commonly applied to forage soils in the region.

SWAT partitions P into organic P, soluble P, and P sorbed to the sediment. The annual breakdown of the export of these species, by cover type, is shown in Figure 5. Since the rate of phosphorus fertilizer application in the simulation was largely consistent between the crop types (30 kg-P/ha/yr for corn and 20 kg-P/ha/yr for other crops), most of the variation can be attributed to variations in the growth properties of the different plant cover types. Notably, however, the unfertilized forest HRUs generated significantly less soluble and sediment-bound P while releasing more organic P than cropped areas.

Again, when considering the flows and loadings into the reach, it is important to note that calibration was limited to the conditions at the basin outlet, so results for individual reaches may be less reliable. Nonetheless, it is immediately apparent from Figure 6 that the influence of changing land-use scenarios on runoff is minor, while the other flows are much more sensitive. We previously showed that runoff from forested HRUs is slightly lower than for cropped HRUs, although the result at the subwatershed scale is less consistent. For the sub-basins in the model, runoff increased with increasing forest cover for 13 sub-basins, while for the other 18, runoff decreased with increased forest cover. The net change averaged across the basin is small, however. These simulations suggest that a further reduction to 20% forest cover across the watershed, from ∼28% forest cover in 2008, is likely to increase the maximum monthly loadings of nitrate and total phosphorus by over 10% and sediment by almost 7%. Early winter peaks in tile flow and nitrate coincide with early winter snowmelt after the harvest of crops. Table 5 shows the peak monthly and annual figures for selected land-use scenarios and the relative variations in percent compared with the 2008 calibration land-use scenario. Here too, there is little variation in surface runoff, even at peak flow, while all other parameters vary significantly, with nutrient loads increasing with decreasing forest cover. At least part of the increase in tile flow and nitrate in tile flow is due to the increase in the area that is fertilized and drained by tiles as forest cover is converted to tile-drained agriculture in this study.

It can be seen in figures and tables of the average annual flows and nutrient/sediment loads into the South Nation River (Table 5) that groundwater, lateral, and tile flows are far more sensitive to land-use change than surface runoff. Paired t-tests of the annual totals compared with the 2008 scenario found all results to be statistically significant (p < 0.05). Similar tests of monthly flows and loadings also found all results to be statistically significant with the exception of monthly runoff in the 50% forest scenario. Paired t-tests of the annual totals compared with the 2008 scenario found all results to be statistically significant (p < 0.05). Similar tests of monthly flows and loadings also found all results to be statistically significant with the exception of monthly runoff in the 50% forest scenario. This is expected from the previous observations that runoff varies relatively little between crops and forest, while the other components are significantly different. The decrease in tile flow with increasing forest cover overrides the increasing trend from the other three components, leading to a decrease in water yield to the reach with increasing forest cover.

Variations in surface runoff loadings, however, are quite significant – reducing forest cover from the 2008 starting condition to 20% forest cover will likely increase annual contributions of sediment, nitrogen, and total phosphorus from runoff by 5–10%. Total tile flow and nitrate from tiles again appear to be very sensitive to the land-use conversion since tile flow and nitrate therein is significant from cropland, but absent from forests where N-fertilizing is absent.

As shown in Figure 7, similarly to runoff, streamflow at the basin outlet is mostly insensitive to forest cover variations in this study. Also, in keeping with the previous observations, nutrient loadings are significantly impacted by forest cover, particularly during the spring melt period. Of particular interest is the sediment loading. Despite our previous finding that forest cover has a significant influence on the sediment carried to the reach with runoff, sediment loading at the Plantagenet river basin outlet shows relatively small variations between land-use scenarios (deforestation to 20% cover increases the sediment in runoff by ∼7% and the sediment at Plantagenet by less than 1%).

Figure 7 shows only small variations in streamflow and sediment loads across the simulated scenarios, while there was a more significant change in nutrient loads. Total nitrogen loads behave differently, however. Around the spring peak, Total N appears less sensitive to lower forest cover, and in lower flow periods through the summer, deforestation scenarios actually produced more total nitrogen (TN) leaving the basin, while the other nutrients were all reduced by increased forest cover.

Overall, flows and nutrient/sediment loads are generally inversely related to forest cover (Supplementary Material, Table A3). Changes in flow and nutrient loading related to recent forest loss (and near-term projected scenarios) are relatively small. More extreme, but not improbable, scenarios such as reducing forest cover to 20% or even 10% could be expected to trigger significant increases in nutrient outputs, with associated risks to river ecosystem (e.g., eutrophication) and public health (e.g., cyanobacteria and hydrologically entrained agriculturally engendered pathogens (Wilkes et al. 2014)).

In addition to examining the impact of converting forested land to cropland (or vice versa), the same land-use scenarios were simulated with the addition of GWW and VFS that are intrinsically designed to help reduce runoff and nutrient loads. These BMPs also have important co-benefits such as the provision of wildlife habitat and related refugia for beneficial insects (Bryan & Best 1994; Kleijn et al. 1998; Boutin et al. 2003; Cole et al. 2015).

The small change in the average sediment loads in reaches and the river basin outlet, despite significant local changes at the HRU scale, suggests that stream or reach processes dominate the balance of sediment in the river. This effect is observed again here, where sediment carried in runoff is significantly reduced by VFS and GWW, but sediment export from the basin is almost unaffected. However, both BMP options have a notable influence on nutrients transported to the reach and exported from the basin.

Figure 8 shows monthly average loads of the sediment and nutrients to the reach and at the basin outlet, respectively, for the 2008 land-use scenario. The installation of VFSs has the potential to significantly reduce the sediment and nutrient loadings in runoff (by 50% or more for these simulations), particularly during the flood period when GWWs are less effective, particularly for nitrate. Similarly, GWWs have a negligible impact on nitrate and total nitrogen exported from the basin, while VFSs have little influence for most of the year except for May–June (when fertilizer is applied in these simulations) and October–December which is typically after crop harvest. Mineral and total phosphorus loads were reduced consistently year-round, with VFSs once again leading to greater load reductions than GWW. Although not shown here, simulations of both BMPs together showed negligible changes from the simulations associated with VFSs only.

Detailed results for all land-use scenarios as simulated with and without VFSs and GWWs, for average annual total nutrient and sediment loads to the reach and at the basin outlet, are provided in the Supplementary Material (Table A4). Again, the impact of these BMPs on runoff loads is much more significant than on the basin loads. The impact on phosphorus export is greater than on nitrogen export. Comparison of Supplementary Material, Tables A3 and A4 shows the capacity of these BMPs to (partially) compensate for the changes in sediment and nutrient loads due to forest cover change. At the HRU/reach level, even the most optimistic reforestation scenario cannot match the reductions indicated by the BMP simulations. While at the basin outlet, reach processes reduce the advantage of the BMP simulations and the loading reductions simulated by VFS are comparable to more modest reforestation scenarios.

The simulations show that forest/tree cover losses in the basin impact the quantity and quality of water throughout most parts of the South Nation watershed, but the water quality changes vary significantly between the runoff entering the reach and the streamflow leaving the basin. The following characteristics were identified at the HRU/subbasin scale:

• Forested HRUs return lower surface runoff and total water yield to the reach and increase soil water content and subsurface flows.

• Although forested HRUs yield more sediment than HRUs under forage crops, sediment yield is much lower than corn, soybeans, or the average of all crops. It is notable here, again, that crop management parameterization was highly generalized for the purpose of scenario testing.

• Forested HRUs release much less nitrate since no nitrogen fertilizer is applied and there is no tile drainage, and

• Although there is a slight increase in organic phosphorus export from forested HRUs, phosphorus in soluble form and sorbed to the sediment is significantly higher from cropped HRUs.

The main findings of the study on the simulated effects of forest/tree losses at the sub-basin and watershed scale are listed below:

• The response of runoff to the reach to changes in forest cover is very inconsistent; some subwatersheds indicate increased runoff with forest gain while others indicate a decrease; the net result is a small change at the basin scale.

• Streamflow at the basin outlet is largely insensitive to forest/tree cover changes.

• Sediment loads to the reach are significantly increased by forest/tree cover loss, while sediment loading at the basin outlet is much less sensitive.

• Nutrient loads to the reach are more sensitive to forest/tree cover change than loadings at the basin outlet.

• Increasing deforestation drives increases in nutrient and sediment loads at the reach and watershed scale, and

• Stream or reach processes appear to reduce the load changes observed at the basin outlet compared with that noted in the flows into the reach from sub-basins.

The cause of the inconsistencies in runoff at the subwatershed scale and apparent insensitivity in runoff and streamflow at the basin scale are not entirely clear – there appears to be no relationship to sub-basin size, forest or crop fraction, or actual area of forest and/or crop. This inconsistency may be related to soil types, but assessing these kinds of cause and effect relationships was beyond the scope of this study. While data are not yet available about forest/tree cover changes in the river basin in 2020, observations in the region suggest ongoing and amplified forest/tree losses since 2014. Further forest clearing to 20% forest cover would lead to a negligible change in water yield to the river and annual average streamflow but will increase annual runoff-induced nitrate and total phosphorus loads to the reach by 8.1 and 9.4%, respectively with increases at the basin outlet, however, 7.7% (NO₃) and 10.3% (total phosphorus, TP). Maximum monthly concentrations of total nitrogen and total phosphorus at the basin outlet are set to increase by 2.9% to 5.2 mg-N/L (TN) and by 5.1% to 0.24 mg-P/L (TP), further above current CCME benchmarks of 3 mg-N/L nitrate and 0.035 mg-P/L total P. Although some studies have found forest loss leads to increases in monthly and peak flows (e.g., Cheng 1989), the influence of forest loss in South Nation appears to be only significant for large forest cover changes, perhaps due to the highly scattered distribution of most wooded features in the basin and the extremely low relief across the region. Yet, many low relief agricultural watersheds in Canada have a similar disposition, where wooded features are not confined to large blocks, but to fence lines, hedgerows, and along water courses.

In this equation, c_sp and spexp are calibrated terms (SPCON and SPEXP, respectively, in Table 2), and v_ch,pk is the peak channel velocity. Various values for SPCON and SPEXP were tested outside the calibration process with a negligible effect on the sediment exports from the basin. Examination of the reach processes driving this limitation was outside the scope of this study, but may be worthy of further study.

At the other end of the scale, even the most ambitious forest/tree cover scenario considered here is insufficient to reduce maximum monthly averages to CCME standards, notwithstanding that in the modeling, tile drainage was removed from land where reforestation of agricultural land occurred. The 50% forest cover scenario represents a gain of over 860 km² forest cover from the 2014 DRAPE forest cover profile. Although this scenario sees an annual average nitrate concentration of 1.1 mg-N/L, the maximum monthly average nitrate remains around 3 mg-N/L, while both annual and maximum monthly average TP concentrations are significantly above the 0.035 mg-P/L threshold defined as eutrophic by CCME. It is clear from these results that forest expansion and management alone is not sufficient to meet water quality standards and other BMPs are necessary to meet water quality targets.

It is apparent as well that streamflow out of the basin is only marginally influenced by changes to forest/tree cover, while nitrogen and phosphorus loadings are significantly affected. These observations are in line with the results published by El-Khoury et al. (2015) for the South Nation River and Tu (2009) for eastern Massachusetts.

The following key conclusions can be drawn from the analysis of simulations executed with the inclusion of VFSs and GWWs:

• VFSs are more effective than GWWs at reducing the sediment and nutrient flows to reach and at the basin outlet.

• The wide-scale adoption of VFSs is more effective at lowering nutrients than most reforestation scenarios.

• GWWs are more effective at reducing loads of sediment and phosphorus than at reducing loads of nitrate or total N; likely due to improved trapping of sediment and attached phosphorus.

• Again, reach processes appear to neutralize gains made to runoff loadings, particularly sediment.

• Combining VFSs and GWWs makes negligible gain over VFSs only, and

• These BMPs are not sufficient to reach nutrient concentration standards with the current land use balance.

In terms of the sediment and nutrient loadings carried by runoff to the reach, VFS can achieve loading reductions under all forest cover scenarios that are greater than even the ambitious 50% forest cover scenario considered here, both at peak flows and in annual totals. Nonetheless, the lowest nutrient loads are delivered by scenarios with VFS and 50% forest cover.

As with the deforestation simulations, the influence of the BMPs is attenuated at the basin outlet, such that VFSs only partly offset the effects of deforestation. While 50% forest cover with VFSs remains the best-case scenario for nutrient loadings, partial deforestation (to 20% cover) combined with VFSs achieves total P loads similar to the 45–50% forest cover scenario, and nitrate loads being similar to the 30% forest scenario. VFSs are consistently more effective than GWWs in this context, and, therefore, only quite significant reforestation has any potential to match the nutrient export reductions from the installation of VFS. It is important to consider, particularly for VFS which would replace in some cases cropland, the cost-benefit balance of adopting alternative management practices – the simulated VFS here represents approximately 2.5% of a 15 ha field.

The influence of deforestation/tree losses on streamflow and nutrients in the South Nation Watershed, a predominantly agricultural watershed undergoing deforestation, was examined using hypothetical land-use changes and hydrological modeling. The impact of VFSs and GWWs on water quality was also simulated.

Results showed a consistent inverse relationship between forest cover and all nutrient loads assessed, although simulated forest changes only caused minor changes in streamflow and sediment export. Further loss of forest/tree cover is likely to drive significant increases in nutrient loads while reversing the recent land conversion trend would decrease nutrient loading. Furthermore, this study indicates that, considered in isolation, significant reforestation (500 km²) would be required to reduce total nitrogen concentrations below water quality guidelines, and even the most optimistic scenario tested returns total phosphorus levels well above provincial standards.

The introduction of VFS to cropland HRUs reduced sediment and nutrient loads, particularly phosphorus. However, even under the most optimistic reforestation scenarios combined with VFS, concentrations of nitrate and total phosphorus would remain above benchmarks recommended by the Canadian Council of Ministers of the Environment (2012a, 2012b).

It is well established that forests/treed areas sequester carbon and release oxygen, influence air temperature through evapotranspiration and shading, reduce erosion by wind and water, cycle soil nutrients, and provide habitat for native wildlife, including some that provide pollination and pest control benefits to farmers (Foley et al. 2005; Bonan 2008; Bryan & Henshaw 2013; Tallis et al. 2013). The findings contributed here, considered with the array of other climate, ecological and cultural services provided by forests/treed areas in the landscape, enhance the credibility of reforestation programs and regulations limiting the amplified tree/forest cutting in agri-ecosystems across Canada.

This study was supported by Dr Seidou's NSERC discovery grant and funds from Agriculture and Agri-Food Canada's Environmental Change & One Health Observatory (ECO²).

All relevant data are included in the paper or its Supplementary Information.