Watersheds located within the Long Branch Watershed in Macon and Adair Counties (39° 50’ N to 40° 05’ N and 90° 32’ W to 92° 20’ W), Missouri, USA, were studied during 1997-1999 (Figure 1). Three land management types were included: at one extreme, forest catchments (n = 2), and at the other extreme, row-crop (n = 3) catchments with virtually no riparian vegetation, and pasture/Conservation Reserve Program (CRP) catchments (n = 3) as the intermediate condition. The two forest catchments were within the state-owned Atlanta Conservation Area.

Study watersheds ranged in area from 21 to 259 ha, and landuse within each watershed is summarized in table 1. Areal land use patterns were measured from 1:24,000 areal photographs (provided by Adair and Macon County, USDA Service Stations). Landuse classification was verified by observations during the study. Watersheds were identified based on major land management type: forest (1xx), pasture (2xx), and row-crop (3xx). Within each land management category, the northern-most watershed was assigned the smallest number and vice versa (Figure 1). One row-crop watershed (400) received irrigation water from the La Plata Water Treatment Plant. Watershed 400 does not have a forested riparian buffer zone but has a 3 to 4 m wide dense grass filter strip along the stream. Therefore, this watershed was classified as a row-crop watershed with a grass filter strip (RC-GFS). These watersheds had low human population with no sewage outfalls. Cattle were present during certain periods in summer in watershed 202.

Pre-settlement vegetation was dominated by forests and tall, native grass prairies well-adapted to climate and soils of the area (Ferguson, 1995; Steyermark, 1996). The dominant trees in forest tracts and riparian areas were white oak (Quercus alba L.), black oak (Q. velutina Lam.), post oak (Q. Stellata Wangenh.), pin oak (Q. Palustris Muenchh.), hickory (Carya spp.), elm (Ulmus spp.), osage orange (Maclora pomifera) and cedar (Juniperus virginiana).

Agricultural watersheds were in a corn (Zea mays L.)‑soybean (Glycine max (L.) Merr.) rotation under minimum- or no‑till cultivation. Wheat and sorghum represented little land cover in row-crop watersheds. Pasture/CRP watersheds were mainly under CRP. Fescue (Festuca spp.), golden rod (Solidago spp.) and cool season grasses were common in pasture watersheds. Row-crop and pasture watersheds contained varying degrees of forest riparian buffer (Table 1).

Soils in the region are developed from glacial till, loess, and pedisediment (Ferguson, 1995). The surface was created by glaciation. Some soils are formed from loess and underlying pedisediment (Unklesbay and Vineyard, 1992). The landscape is characterized by gently to strongly sloping uplands.

The average snowfall in the study area is about 61 cm and generally lasts several weeks. The average winter temperature is ‑2.2°C and average daily minimum is ‑7.7°C. In summer, the average temperature is 23.9°C and average daily maximum is 30°C. The 30‑yr precipitation averages 967 mm, about 65% of which falls in April through September (Owenby and Ezell, 1992).

ISCO samplers and flow measuring devices (Lincoln, NE, USA) were installed annually in late February or during the first week of March (1997, 1998, and 1999) to record flow rate, water level, sampling time and collect water samples. Thus, the sample collection period extended from March to December each year. Generally, streams were frozen during winter, and little or no stream flow occurred.

All eight streams were surveyed and stream characteristics were fitted to Manning’s equation to estimate water flow from rating curves of flow versus depth. Flow measuring devices controlled the sampler to collect water samples. Samples were collected at 30‑minute intervals from the rising limb of the hydrograph and at 90‑minute intervals from the falling limb of the hydrograph. During 1999, samples were collected based on pacing volume, such that after a set amount of water flow, the flow meter would signal sample collection. Because runoff varies with watershed size, landuse patterns and season, pacing volumes were larger for large row-crop watersheds (300:1,400 m³ and 301: 4,000 m³) compared to the small row-crop (400: 200 m³), pasture (200: 300 m³ and 202: 200 m³), and forest (100: 600 m³) watersheds. Samples were delivered to collection bottles through vinyl suction tubes that were first rinsed with stream water. Sample inlets were positioned in the middle of the stream.

During some weeks, flow was either too low to generate an automatic sample or the sampler failed. If a sampler malfunction was detected, the sampler was replaced as soon as possible. Samplers on watersheds 100 (22‑6‑97 and 12‑3‑98) and 201 (18‑5‑97) failed due to sand blocking the suction line. The sampler and flow meter on watershed 301 was damaged twice (18‑9‑97 and 29‑6‑98) by lightning. The sampler housing for watershed 300 was placed on a four‑foot shelf, but stream flow inundated the stream bank area several times during the study period. After a runoff event, flow, level and sample intake time data were downloaded from recorders. Water samples were transferred from the field to the laboratory and individual samples were analyzed for all the parameters of interest during the first two years of the study. During the third year (1999), composite samples from each runoff event were analyzed.

A known volume of well mixed sample was filtered through a pre-weighed glass microfiber filter (934‑AH) using a vacuum pump (maximum vacuum 7 lbs in ‑2 above ambient) to estimate sediment weight. These filters were dried at 105°C to a constant weight and dry weight was recorded. Differences between the tare weight and sample volume were used to estimate the weight of total suspended sediment. Samples were refrigerated at 4°C until analysis.

Statistical analysis of data was performed using SAS (SAS Institute, 1999). Normal distributions of runoff and sediment losses were tested using Shapiro-Wilk and Kolmogorov-Smirnov tests and log transformed data were used for further analysis. Random variables were analyzed as a split-plot intime using the procmixed procedure in SAS. The main plot consisted of land management (treatments) and the subplot consisted of year and interaction of land management*year. Watersheds within a treatment were used as the denominator of F for the main plot (land management) effect. Watersheds within a land management and year was used as the denominator of F for the sub-plot (year and interaction of land management*year) effects. The residual mean square represented multiple observations within a year watershed combination. The fixed effects are land management, year and the interaction of land management*year. Mean differences were determined using Fisher Least Significance (LSD) and were calculated using Lsmeans statement within the Procmixed procedure. The variance covariance matrix was investigated using AIC coefficient to determine the most suitable mean separation procedure.

Long Branch (LB) and Kirksville Radio Station (KRS) weather stations received 21% and 13% less precipitation in 1999 (March to August) compared to the 30‑yr average, respectively (Figure 2). These two stations also received 18.5% and 10% less in 1997. Whereas during 1998, the region received 36% (LB: 43 and KRS: 30%) more rain compared to the 30‑yr average.

The study area received rain on 62 dates ranging from a trace to 43 mm in 1999. No measurable runoff events occurred after 17 May, 1999. During the 1997 sampling period, rain occurred on 136 dates, ranging from a trace to a maximum of 54 mm. Rainfall events generating runoff ranged from 2.5 to 54 mm in study watersheds. Rain ranging from a trace to 132 mm was recorded in 1998 and 22 storm events on average produced more than 13 mm of precipitation. Rainfall generating runoff ranged from 9 to 128 in the watersheds.

Measurable runoff events from watersheds generally followed precipitation (Table 2). During sampling in 1999, forest watershed 100, the pasture and the row-crop watersheds produced 8, 7 and 9 runoff events, respectively. In 1997, the two forest watersheds, on average, produced four measurable flows (100: 6 and 101: 2). Pasture and row-crop catchments recorded 9 and 17 runoff events in 1997, respectively. Sampling year 1998 resulted in 14, 17 and 20 measurable runoff events from forest, pasture, and row-crop land uses, respectively.

The 3‑yr mean annual discharge of water per area differed greatly among watersheds, ranging from 563 m³ ha^‑1 yr^‑1 (200) to 3771 m³ ha^‑1  yr^‑1 (301; Table 3). The 3‑yr mean runoff volumes from forest, pasture, row-crop and RC-GFS land management types were 939, 1,560, 3,434 and 1,175 m³ ha^‑1 yr^‑1, respectively (Figure 3). The 3‑yr mean annual runoff for the row-crop land management was 3.7, 2.2 and 2.9 times larger than that of forest, pasture and RC-GFS types, respectively. However, runoff volumes between the row-crop and RC-GFS watersheds were not significantly different (p ≤ 0.08). Pasture and RC-GFS watersheds produced 1.7 and 1.3 times more runoff than forest watersheds, respectively.

Discharge varied with rainfall, and maximum runoff from the study watersheds occurred on different dates. Watersheds 100 (375 m³ ha^‑1), 200 (237 m³ ha^‑1), 201 (361 m³ ha^‑1) and 400 (394 m³ ha^‑1) had maximum runoff on 4 July, 1998 (Figure 4). Runoff events on 30 March, 1998 caused maximum runoff on watersheds 300 (1,121 m³ ha^‑1) and 202 (333 m³ ha^‑1), respectively. Watersheds 101 (462 m³ ha^‑1) and 301 (1,111 m³ ha^‑1) had maximum runoff on 14 March, 1998 and 14 April, 1999, respectively. Due to lightning damage, runoff data were not available for watershed 301 on 4 July, 1998 which was probably the largest event during the study.

The largest single runoff event in each land management category accounted for 20%, 12%, 25% and 11% of the total runoff in forest, pasture, row-crop and RC-GFS watersheds, respectively (Table 4). The proportion of total runoff contributed by the largest runoff event varied among watersheds. Largest events from forest watersheds 100 and 101 accounted for 13% and 27% of the total discharge, respectively. Differences among watersheds within a given land use type were larger between the two forest watersheds than the other categories. The largest runoff events from the pasture watersheds accounted for 10‑14% of the total, and in row-crop, the value was 24‑26%.

Three-yr mean soil erosion values from the eight watersheds ranged between 738 (202) and 3,777 (300) kg ha^‑1 yr^‑1 (Table  5). The 3‑yr mean soil erosion on forest, pasture, row-crop and RC-GFS were 1,017, 1,241, 3,679 and 2,129 kg ha^‑1 yr^‑1, respectively, (Figure 5). Row‑crop watersheds lost 3.6 and 2.9 times more soil compared to forest and pasture watersheds, respectively. Row‑crop watershed with grass filter strip (400) and pasture watersheds lost only 58 and 34% soil, respectively, compared with row-crop watersheds (Figure 4).

Row-crop watersheds lost 2,901 kg TSS ha^‑1 yr^‑1 in 1997 (Table 5). Soil loss from row-crop watersheds in 1997 was 5.5, 4.2 and 6.8 times greater than forest, pasture and RC‑GFS land managements, respectively. Soil erosion in 1997 and 1999 on the RC-GFS was only15% and 41% of that on two row-crop watersheds (Table 5). Most storm events occurred early in 1999 and caused severe erosion when the ground was bare and before the soil lost its recharge. The two row-crop watersheds had more erosion in 1999 than in 1997 or 1998.

The largest single runoff event accounted for 21% to 28% of total soil loss among the various land management categories during the 3‑yr study (Table 6). The percentage of soil loss caused by the largest single event among watersheds varied between 15% and 33% in this study. Averaged over eight watersheds, the single biggest runoff event accounted for 23% of the total soil loss measured during the study period. The largest three runoff events accounted for over half of the total erosion from forest and row-crop watersheds. Pasture and RC‑GFS watersheds lost 45% and 42% of the total loss during the largest three events. The largest two and three episodes caused 39% and 49% of the total soil loss from all eight watersheds during the 3‑yr study period.

The 2‑yr mean erosion values for the forest and pasture watersheds were 947 and 1,427 kg ha^‑1 yr^‑1, respectively (Table 7). Using the 2‑yr mean erosion amounts on forest and pasture land managements as base values, we compared erosion on the row-crop watersheds (300 and 301) and the RC‑GFS (400) watershed to quantify the proportion of events on row-crop and RC-GFS watersheds that generated erosion equal to or less than the 2-yr mean erosion from forest and pasture watersheds.

None of the events in 1999 had erosion amounts similar to that from forest or pasture watersheds (Table 7). The above normal precipitation in the previous year and early season runoff in 1999 caused severe erosion on all row-crop watersheds even though the study area received below average precipitation. In 1997, 14% and 25% of erosion events on row-crop and RC‑GFS watersheds had erosion amounts similar to forest watersheds. In 1998, relatively few erosion events on row-crop watersheds generated erosion amounts that were similar to or less than those of forest and pasture watersheds. During the 3‑yr study, watersheds 300, 301 and 400, had 10, 12 and 11% of erosion events that were similar or less than the erosion on forest watersheds, respectively (Table 7). Approximately 90% of erosion events from row-crop watersheds exceeded the mean erosion on forest and pasture watersheds.

Annual, seasonal and within growing season variation in the frequency and intensity of precipitation influenced runoff. The study region received 16% less precipitation than the 30‑yr average in 1997 and 1999, and runoff from all watersheds was lower compared to runoff on 1998 regardless of land cover type (Table 3). Annual rainfall during 1998 at each weather station was about 36% greater than the 30‑yr average. Consequently, runoff volumes were also above long‑term averages for the area and most watersheds had the largest runoff. This observation is supported by a 10‑yr study with three adjacent corn‑soybean rotational watersheds in northeast Missouri, where 27 or more runoff events occurred when precipitation was above 40% and fewer events when precipitation was normal or below normal (Udawattaet al., 2002).

We also assumed that watershed boundaries coincided with topographically defined catchment of the surface runoff. All of our agricultural watersheds varied by less than 2 m elevation, and therefore, boundary demarcations and the true watershed area may be inexact. Surface runoff and ground water can cross topographically defined boundaries, so the collection area of the stream becomes uncertain (Garrisonet al., 1987). ISCO submerged probes measure and collection strainers collect when flow was greater than 2.5 cm. Runoff occurred due to low level flow events could also account for the slightly larger annual total discharges from these watersheds. Hydrologic variability among watersheds also may have resulted in a wide range of runoff volumes within a land use category (Edwards and Owens, 1991; Lowranceet al., 1985; Peterjohn and Correll, 1984; Quintonet al., 2001; Udawattaet al., 2002; 2004; Zhenget al., 2004). Differences in discharge can also be due to erosional stage of the watershed and vegetation condition.

In spite of differences in precipitation and soils among watersheds, the general effects of landuse on runoff and sediment loss were evident. Overall, forest watersheds did not have as many runoff events compared to pasture or row-crop watersheds. Forest watersheds had 18% and 72% less sediment loss than pasture and row-crop landuse. Literature supports that forest watersheds in general have less runoff and sediment loss (Lowranceet al., 1985). The two watersheds are located in the Atlanta Conservation Area and are > 85% forest. Forest vegetation reduces runoff, soil erosion and nutrient loss from watersheds and improves infiltration and water holding capacity of soils (Lowranceet al., 1985; Gilliam, 1994). In contrast, row-crop watersheds had the largest discharges of runoff (3,434 m³ ha^‑1 yr^‑1) and sediment (3,679 kg ha^‑1 yr^‑1). These two watersheds had degraded riparian buffers with sparsely distributed understory growth. The riparian corridor on watershed 300 has poorly drained soils and most of the time soils remained wet. The other row-crop watershed (301) has a small buffer strip along the stream and its stream incision is deep. Watersheds with less homogeneous soils, surface slopes, management and non-uniform buffers may result in short circuiting of surface runoff where runoff flows preferentially through only parts of the buffer strip.

Research shows that grass strips reduce runoff and sediment losses ranging from 30‑90% and 60‑90%, respectively (Pattyet al., 1997; Robinsonet al., 1996; Schmittet al., 1999). Discharge from the RC‑GFS watershed was 34%, 64% and 83% less than that from row-crop watersheds in 1997, 1998 and 1999, respectively. The grass filter strip along the stream was 3‑4 m wide and may have utilized water in spring and early summer, thus runoff and erosion losses were less than that of row-crop watersheds. During 1997 and 1999, the study area received 16% less precipitation than the 30‑yr mean and the grass filter strip on the RC‑GFS watershed reduced soil erosion by 85 and 60% relative to the other two row-crop watersheds, respectively. Most of the sediment reduction occurs within the first 7.5 m width of the strip (Schmittet al., 1999). In 1998, the RC‑GFS watershed lost 15% more soil than row-crop watersheds. During above normal precipitation in 1998, the width of the filter strip may have been insufficient to control erosion because transpiration loss of the permanent vegetation and infiltration capacity of the soil may have been exceeded by runoff and antecedent water content. The results showed that grass filter strips provide a way to improve surface water quality in agricultural watersheds, and they are more effective in years with normal precipitation distributions and under low flow conditions.

CRP/Pasture watersheds had 65%, 50% and 83% less runoff in 1997, 1998 and 1999 than row-crop watersheds and reduced sediment loss by 55% relative to row-crop land management. These watersheds, on average, contained 30% and 55% forest and pasture covers, respectively (Table 1). CRP/pasture has been shown to increase soil structure due to the presence of live and dead roots, fungal hypae and other organic material than do annual vegetation and thereby reduce runoff and sediment loss (Wienhold and Tanaka, 2000; Zhenget al., 2004). Darvie and Lant (1994) stated that conversion of row-crop areas in Illinois to CRP would reduce erosion by 30%. Row-crop watersheds with CRP/pasture in this study showed greater reductions in sediment loss than the CRP lands in Illinois. Within a watershed, CRP/pasture lands were located between the crop and forest areas and this could have improved sediment reduction in runoff. Many authors have shown the merits of a naturally vegetated riparian zone in improving water quality (Gilliam, 1994; Lowranceet al., 1985).

Control of fallow period and bare ground soil loss was important to reduce NPSP from row-crop watersheds (Ghidey and Alberts, 1998; Udawattaet al., 2004). Until crops establish a good ground cover, soils are susceptible to raindrop impact and erosion compared to the pasture areas. Studying soil erosion on corn, wheat and meadow watersheds, Edwards and Owens (1991) observed that 92% of the soil loss occurred in corn fields, 8% in wheat fields and no erosion in grass-meadow fields. It is unlikely that row-crop watersheds reduce soil and nutrients losses in runoff during the fallow period compared to the cropping period (Udawattaet al., 2004). A comparison of soil losses in 1999 from RC‑GFS and CRP/pasture versus row-crop watersheds proves that cool season grasses that cover the soil reduce soil loss early in the season and plant material and roots enhance settling of sediment. Results indicate that CRP/pasture land within a row-crop watershed improves water quality compared to a watershed with no CRP/pasture. The expected reduction in sediment loads through the conversion depends on time, watershed size, landuse, channel morphology and precipitation characteristics.

Large runoff events were responsible for removing substantial amounts of sediment and nutrients from row‑crop watersheds (Lal, 1976; Morganet al., 1986; Quintonet al., 2001; Robinsonet al., 1996; Udawattaet al., 2004). In this study, the three largest runoff events accounted for approximately half of the total erosion. In a 28‑yr study in Coshocton, Ohio, with more than 4,000 rainfall events, more than 66% of the soil loss occurred in large five events (Edwards and Owens, 1991). However, erosion producing large events accounted for differing amounts of the total erosion among watersheds within a given land use. For example, during the largest two and three erosional events, watersheds 301 and 300 lost 39% to 53% and 47% to 63%, respectively of the total soil eroded. Differences in erosion among watersheds within the same management demonstrate the importance of large storms on total erosion and long-term evaluation to determine best management practices. On the other hand, smaller events that occur more frequently may account for a greater proportion of total nutrient loss than infrequent large events (Quintonet al., 2001; Udawatta et al., 2004). These require less energy to detach soil particles (Quintonet al., 2001). In contrast, large events have more energy to remove nutrient-rich surface as well as nutrient-poor subsurface soils. In order to control NPSP in runoff, nutrient and sediment losses from all events need to be reduced.