A rainfall simulation experiment was conducted in a typical radish (Raphanus sativus L.) planted vegetable field at Yixing, China (31°14′ N, 119°53′ E) where has a subtropical monsoon climate with average annual temperature 15.6°C and precipitation 1180 mm). At the beginning of the experiment, the primary fertility properties of the 0–20 cm topsoil were as follows, pH 6.25 (soil and water mixed in a 1:2.5 ratio, w/v), nitrate N 120 mg/kg, total N 1.56 g/kg, Olsen-phosphorus (P) 56 mg/kg, available potassium 83 mg/kg, and organic matter 26.4 g/kg.

Rainfall simulation is a versatile and widely used research tool [17], according to which we conducted the current field experiment using an artificial rainfall simulation system and a runoff collection device. The device mainly comprises rainfall nozzles, centrifugal pumps, filters, and water storage tanks. The device (with a height of 2 m) covered a rainfall area of 1 m² (1 m × 1 m). As there were no blind spots in the rainfall area, the rainfall uniformity in our experiment was >90%. Rainfall intensity could be adjusted by controlling the pressure of the centrifugal pump, with an adjustable range of 50–200 mm/h. The runoff device is mainly composed of poly vinyl chloride (PVC) partitions and runoff collection barrels, surrounded by PVC partitions and ridges. The PVC partitions were placed 10 cm above the soil surface and buried in the soil at 30 cm depth. One of the PVC partitions had a hole to connect the runoff overflow pipe to ensure that all surface runoff could flow into the catchment bucket after the runoff occurred.

In this experiment, we assessed three N fertilization levels, i.e., no N (Control), low N (LN, 291.9 kg/ha, scientifically recommended N input), and high N (HN, 417 kg/ha, farmers’ traditional N input). According to the general rainfall records in the Tai Lake area [1,5,6], the rainfall intensities were set at 55 and 75 mm/h for each N input level, respectively. Therefore, there were six treatments with three replicates each for the experiment. The simulated rainfall experiment was conducted from 15 September to 25 October 2022, with a total of 18 rainfall tests. For vegetable growth, chemical N fertilizer in form of urea was applied as basal and supplementary fertilizers on 16 August and 13 September 2022, respectively. At the same time of basal N fertilization 2160 kg/ha of organic fertilizer, 540 kg/ha of calcium and magnesium phosphate, and 120 kg/ha of potash were amended into the vegetable soils. In addition, the radish was harvested on 17 October 2022.

Data were statistically analyzed using Excel 2016 and by ANOVA using the SPSS 21.0 statistics software. They were analyzed for the significance of differences using Duncan’s test (p < 0.05). We plotted the figures using Origin 8.0 software and Excel 2016.

Either LN or HN had a short-term effect on the soil NH₄⁺-N and available P contents, while the differences were not statistically significant. However, HN treatment significantly (p < 0.05) increased the soil NO₃⁻-N content by 111.1% (Table 1). Excessive fertilization can elevate the soil NO₃⁻-N which leading to an increased risk of runoff and leaching N losses [18,19]. Therefore, emphasis should be placed on soil NO₃⁻-N content to control non-point source pollution from intensive vegetable production. Meanwhile, how N fertilization whether with a low or high rates impact soil available N and P contents should be studied in a long-term.

The two N amended treatments produced 103.6%–146.4% more radish production than the control treatment (Table 2), as N is the essential nutrient for vegetable plant growth and the final yield [20]. As shown in Table 2, the vegetation coverage were increased by 75.0% on average following urea N fertilization. However, no difference in vegetation coverage between the HN and LN treatments. Moreover, the high N input treatment produced a 21.0% higher radish yield than the low N input treatment. Within a reasonable application N application range, crop/vegetable production always increased with N nutrient input rates [21,22]. What is more, with high N input root vegetables (like radish in the current study) accumulate a significant amount of N in their roots, increasing the dry matter biomass in the storage roots [23], i.e., the production of radish in our work. However, farmers should avoid applying excessive N fertilizer that exceed the demand of vegetable growth to guarantee vegetable production and reduce N pollution meanwhile.

Under the both rainfall intensities, higher runoff volumes were recorded under the two N-applied treatments, compared with no N treatments (Fig. 1). For the 55 mm/h rainfall intensity, the cumulative runoff volumes in an hour under HN, LN, and Control were 9.4, 7.9, and 5.1 L, respectively (Fig. 1A), and there were 14.1, 9.6, and 6.5 L respectively for the 75 mm/h rainfall intensity (Fig. 1B). Overall, N fertilizer application led to more runoff from vegetable field soil. Moreover, we found that HN treatment resulted in significantly higher runoff rates than the LN treatment. Generally, soil with more vegetation would generate lower runoff [24,25]. And N fertilizer increased the vegetable biomass. However, the data shown in Table 2 summarized the vegetation coverage did not explain the difference in runoff rate difference between the control and N-added treatments. Intensive crop cultivation under intensive crop rotation always results in higher soil bulk density and lower water infiltration rates, and fertilizer application could exert negative impacts on soil porosity and deteriorate the structure as reported in previous works [26,27]. Generally, inorganic fertilizer application increased the macro-pores but not the intro-aggregate pores in soil cores, compared to no-fertilization [28]. Yield and nutrient concentration of root vegetables were elevated with increasing fertilizer [29,30]. In addition to these factors determining the runoff volume, we hypothesized that roots and rhizomes of radish gradually occupy the soil, and runoff plots with high N application follow the tendency to produce large amounts of runoff at the same rainfall intensity. This leads to a tendency for runoff water to accumulate on the soil surface, and thus runoff is generated. Of course, more long-term and full-scale experiments should be done to verify this phenomenon.

The dynamics of NH₄⁺-N, NO₃⁻-N, DON, and STN concentrations in runoff water from treatments with different N application rates under 55 mm/h rainfall intensity are shown in Fig. 2. The dynamic pattern of runoff NH₄⁺-N under HN and LN were similar, and the average NH₄⁺-N concentration only increased from 1.25 to 1.32 mg/L when N application increased from 291.9 to 471 kg/ha (Fig. 2A). Under two rainfalls intensities, the N-amended treatments had significantly higher NH₄⁺-N concentrations than the control treatment, by 10.9 times and 10.3 times on average.

Similar changing trends were observed for NO₃⁻-N, DON, and STN concentrations of runoff water under the 55 mm/h rainfall intensity (Fig. 2B–D). Overall, all of them in the N-applied treatments were higher than in the control treatment. The application of N fertilizers increased the soil mineralization rate immediately. Half of the soil mineralized N originated from applied N fertilizers, readily increasing the soil available N content of the soil at these high mineralization rates, which further led to a higher concentration of all forms of nitrogen in the runoff that than in the non-N added treatment [31,32]. Therefore, different fertilization practices suitable for precipitation conditions and vegetable species should be developed for vegetable cultivation, which can reduce runoff N losses from vegetable fields under heavy rainfall cases.

Under 55 mm/h rainfall, the initial NO₃⁻-N concentrations of runoff water from N-amended treatments were from 1.34 to 1.73 mg/L, which was close to that from the control treatments (Fig. 2B). Thereafter, the average NO₃^–-N concentrations under LN and HN were up to significantly 140% and 184%, significantly higher than that from the control, respectively. With the rainfall continued, the NO₃⁻-N concentrations were 0.81–1.22 mg/L during the first 35 min, which thereafter decreased to only 0.06–0.39 mg/L until the end of the rainfall experiment. Comprehensively considering the previous and current research results, there is an optimum time to control the runoff from the intensive vegetable field [33]. Our data suggested that the first 5–10 min after runoff generation produce higher runoff volume and sediment quantity. And till 20 min, the runoff sediment gradually decreases and tends to be stabilized. Therefore, controlling the runoff generated in the primary 20 min after flow production help to more effectively in inhibiting runoff N losses from vegetable fields.

At the initiation of the runoff case, the differences in initial DON and STN concentrations in runoff water under 55 mm/h rainfall intensity between LN and HN were minor (8.35 vs. 6.83 mg/L and 13.5 vs. 11.1 mg/L, respectively). By contrast, we recorded significantly lower DON (0.11–0.25 mg/L) and STN (2.5–3.8 mg/L) concentrations in the control treatment. Regarding their temporal dynamics, both DON and STN concentrations under LN were higher than that under HN (average 6.2 vs. 4.0 mg/L) during the first 30 min. However, after 30 min, HN treatment had higher DON and STN concentrations than LN (average 8.3 vs. 5.3 mg/L). At the beginning of the runoff, the surface soil was not eroded. Along with the runoff event, the N contained in the water-soil interface is also high, and the soil N easily migrates to the runoff at a fast rate, resulting in a sharp increase in runoff N concentrations [34]. As rainfall progresses, the topsoil is continually compacted by the raindrops, less soil lost with the runoff. Meanwhile, surface soil N is carried away by runoff or leaching into the sub-layer soil; therefore, the water-soil N concentration gradient decreases as well as for the runoff N concentration decreases [35].

The dynamics of runoff NH₄⁺-N and NO₃⁻-N concentrations were similar at different N application rates under 75 mm/h rainfall intensity (Fig. 3A,B). The mean NH₄⁺-N concentrations under HN and LN (1.57 and 1.31 mg/L) were almost similar (1.6 vs. 1.3 mg/L) and were approximately 13 and 11 times higher than under Control, respectively. The runoff NH₄⁺-N nitrogen concentrations under HN and LN decreased by 1.22–1.27 mg/L during the first 10 min after runoff flow production. During the first 25 min we found a significant decrease in NO₃⁻-N concentrations 0.65–1.60 mg/L under two N amended treatments. Previous research reported that the remarkable decreasing trend of NH4⁺-N at the beginning of rainfall (approx. first 10 min) of runoff was consistent with the turbulence processes that brings N nutrients to the runoff [36]. This phenomenon was confirmed by our investigation, i.e., the dynamics of decrease and stabilization of the runoff NH4⁺-N concentrations [36].

Data in Fig. 3 also showed a similar changing pattern for runoff DON (Fig. 3C) and STN (Fig. 3D) concentrations through the rainfall case. Runoff exhibited a downward trend. During the initial 25 min, the DON and STN of runoff water from two N applied treatments reduced by 5.1–8.1 mg/L and 7.9–11.0 mg/L in the LN and HN treatments, respectively. Compared to the non-fertilizer control, runoff N concentration was higher in all fertilizer N added which was easily understood. However, the control treatments had no changes in runoff DON and STN concentrations. A significant correlation has been observed between soluble N concentration and rainfall intensity on sediment flux, with nutrients from runoff sediments gradually diffusing into runoff water as rainfall intensity increased [37]. At the early stage of runoff, the soil surface water and eroded soil interacted sufficiently, allowing continuous infiltration of moisture into the soil. As the rainfall continued, the soil surface was waterlogged and the stripping of the nutrients from the soil surface was reducing, and the impact effect of the raindrops on the soil was reduced, which might be the reasons for the stepwise decreasing trend in the concentration of the soluble nitrogen fractions in the runoff from vegetable lands [38].

Under 55 mm/h rainfall intensity, the dynamics of runoff and TN concentrations at different N application levels is shown in Fig. 4A,B. Generally, with the increasing N application rate, the average PN and TN concentration of runoff water decreased from 37.4 to 23.9 mg/L (Fig. 4A), and from 38.0 to 30.1 mg/L (Fig. 4B). A similar trend was found under 75 mm/h, particularly during the first 25 min after runoff was generated (Fig. 4C,D).

There was a dilution effect during one runoff cases [39], which explained the significant decrease in PN and TN concentrations under HN treatment during the period 30 through 60 min. Under both 55 and 70 mm/h rainfall intensities, the soluble N concentrations of runoff water from the control and N-application treatments were consistent in the following order: DON > NO₃⁻-N > NH₄⁺-N (Figs. 2 and 3). This result suggested that DON was the main form in runoff N losses from the vegetable production system. By contrast, one previous study have found that in the N lost from agricultural runoff was dominated as the PN, then the DON and, which were mostly derived from the applications of organic and inorganic/chemical fertilizers [40]. To mitigate the non-point N pollutants, previous work has suggested that ecological ditched and wetland rice paddies can effectively reduce 50%–60% of vegetable runoff N losses ecological ditches and wetland rice paddies can effectively reduce 50% to 60% of the nitrogen N losses [41]. Meanwhile, considering the uncertainty the forms of N in runoff from agricultural lands during each runoff in a short term, optimizing the amount of N fertilizer is still the most effective approach to decrease runoff N from insensitive vegetable production in China.

The cumulative N losses from vegetable system through simulated rainfall experiment under 55 and 75 mm/h rainfall intensities were 18.5–26.0 kg/ha and 27.4–44.4 kg/ha, respectively (Fig. 5). It is clear that heavier rainfall leads to more nutrient losses from agricultural soils. As the intensity of rainfall increases, the ability of raindrops to erode the soil surface intensifies, resulting an increase in runoff N losses from agricultural land [42]. Compared with the non-N control urea N application through fertilizers increased total N losses via runoff from vegetable field soil by 22.3%–40.7% (Fig. 5A) and 28.2%–62.2% (Fig. 5C), under 55 and 75 mm/h of rainfall intensities, respectively (Fig. 5A,C). Consistent with previous studies work conducted in paddy field, runoff N losses from the fertilizer application N applied treatments could be near 4.0 times higher than that from the no-N treatment [43]. They also found that there was a significant positive correlation between runoff N losses and N applied rates [43,44], which was also confirmed by our results of STN (Fig. 5A). However, there was no significant difference in cumulative N losses between LN and HN in our study. Interestingly, our data demonstrated that PN is the main constituent of the cumulative N losses via runoff, and its percentage reached 80%–99% and 84%–96% (Fig. 5D) under 55 and 75 mm/h, respectively (Fig. 5B,D). Some previous studies that also conducted under field condition found that soluble N dominated the N lost via runoff from agricultural soils [45,46]. For intensive vegetable soils, we revealed that PN is the primary form in runoff N. Therefore, implementing measures to reduce sediment runoff should first be considered when controlling N pollutants from intensive vegetable production systems. Only mitigating the soluble N is not effective enough.

Many factors determine the N losses via runoff from farmland, such as amounts of irrigation water, and N fertilizer, plant uptake capacity as well as the mulch managements [47,48]. As shown in Fig. 6, we here conducted a principal component analysis (PCA) to clarify the factors determining the runoff N losses from vegetable plantations. Overall, 89.3% of the total variation could be explained by PCA plot. Of which, PC1 explained the highest proportion of the variance (57.1%). Moreover, PN, STN, and TN runoff losses were positively correlated with the runoff volume, indicating that runoff volume mainly determined the N losses from vegetable fields. These results suggested that reducing the runoff rate from vegetable soil is of great importance to mitigate non-point source N pollutants from vegetable-producing systems.

In addition to chemical N fertilizer, N retained in the soil is also lost via runoff during rainfall events. Nitrogen fertilizers applied into the farmland and organic N in the soil increase the runoff N losses; therefore, appropriate fertilization can reduce fertilizer application rate and soil N losses [49]. In terms of runoff N fertilizer losses that from fertilizer applied in current season, there was minor difference in runoff loss rates between the two rainfall intensities (Table 3).

With the N fertilizer applied rate raise, the runoff losses of TN, STN, and PN from fertilizers generally increased. For example, under 55 mm/h rainfall intensity, runoff N lost via PN from fertilizers in the HN treatment was 4.5-fold higher than that in the LN treatment. Furthermore, under 75 mm/h, the runoff N losses in forms of both TN and PN from fertilizers that applied into HN treatment were significantly greater than LN treatment. These results indicated that more N would lost via runoff when the chemical N fertilizers are applied at high rates, which was overlooked in previous studies [50,51]

As shown in Table 3, the percentages of different forms of runoff N losses from fertilizer applied were STN (79.7%–95.5%), TN (17.8%–37.5%), and PN (2.8%–28.6%), respectively. Therefore, we suggested that most runoff N losses via soluble and particle N were contributed by the N that provided from fertilizers and that retained in retained in topsoil, respectively.