The study area is located in Guangrong Village (47°34^′–47°38^′ N, 126°81^′–126°88^′ E), Hailun City, Heilongjiang Province, within the central MRNC (Fig. 1A). The region experiences a continental monsoon climate, with annual precipitation of 300–900 mm during 2000–2022, of which approximately 80.7 % falls between June and October, coinciding with the peak period of soil erosion. The mean annual temperature is ∼ 1.5 °C (-25.6 to 26.6 °C), with crop sowing typically commencing in mid-April. The terrain comprises gently rolling hills, and the soils are classified as Mollisols (Chernozem) with a silty clay loam texture, 45 %–60 % silt content, and an organic matter content of >3 % in the ploughed layer. These conditions support intensive maize and soybean cultivation, but sustained anthropogenic disturbance has caused a ∼ 20 % decline in soil fertility. In particular, gully erosion on sloping farmland leads to an annual arable land loss of ∼ 0.097 %, with gully density reaching 1.5 km km⁻² (Chen et al., 2025c).

To assess the morphological characteristics and activity status of the gullies in the region, a comprehensive gully survey was conducted in May 2021 prior to hydrological monitoring. The results revealed that over 90 % of farmland gullies were highly active, with average widths and depths of 13.3 and 3.4 m, respectively. On this basis, two permanent gullies in farmland catchments (F1 and F2) were selected (Fig. 1B–E), as they exhibited similar catchment areas, land use proportions, and typical morphological and topographic features. Each catchment contained only one permanent gully, and both gullies exhibited clear signs of active development. The gully heads were highly susceptible to headward erosion under rainfall-driven runoff. In addition, vegetation cover on the gully slopes was relatively sparse, particularly in the upstream sections of the gullies (Fig. 1F–G). The characteristics of the two catchments and their gullies are described as follows. The F1 and F2 catchments cover 4.3 and 3.4 ha, respectively. Farmland is the dominant land use, comprising 83.4 % of F1 and 85.5 % of F2. The area directly occupied by the gully accounts for 9.6 % and 15.2 % of the total catchment area in F1 and F2, respectively, with a mean value of 12.4 %. In contrast, the upslope drainage area of the gully head (UDGH) accounts for 64.8 % and 43.9 % of the catchment area in F1 and F2, respectively (mean: 54.3 %), and is entirely covered by farmland. Moreover, gully dimensions were consistent with the survey averages: the gully in F1 measured 0.38 ha in area, 242.3 m in length, 17.7 m in width, and 3.8 m in depth, and the gully in F2 measured 0.54 ha, 293.7, 18.4, and 4.8 m, respectively. Gully slope gradients (F1: 36.2°; F2: 39.5°) were significantly steeper than those of the adjacent farmland slopes (F1: 4.3°; F2: 3.4°). In addition, within the catchments, basal fertilizer was applied at the end of April during ridge formation and sowing using a fertilizer seeder, such that fertilization and sowing were completed simultaneously. The remaining fertilizer was then top-dressed in mid- to late June at the maize jointing stage. Meanwhile, during the rainy season, crop cover on the agricultural upslope areas exceeded 90 %, while vegetation cover within the gullies exceeded 70 %. It should also be noted that a 2 m-wide unplanted buffer along the gully bank, maintained for machinery access, was colonized by natural grass cover (Fig. 1F–G). Field monitoring during intense rainfall indicated that these grass strips, together with wheel ruts, effectively diverted lateral runoff downslope along their margins, reducing direct flow into the gullies (Chen et al., 2025b). Therefore, this minor component was excluded when estimating the contribution of the gully to runoff and dissolved NH4+, NO3-, and P transport fluxes.

From June to October in both 2022 and 2023, tipping-bucket rain gauges (Jian Darenke Electronic Technology Co., Ltd., Jinan, China) with a resolution of 0.2 mm were installed in each catchment to characterize rainfall conditions at the catchment scale (Fig. 1B, C, and H). A rainfall event was defined as a continuous precipitation event separated from the next rainfall event by at least 6 h without rainfall; shorter dry intervals were considered part of the same event, whereas longer intervals were treated as separate events. Effective rainfall events were defined as those that generated observable surface runoff at the catchment outlet. Accordingly, only rainfall events associated with observable runoff were included in the subsequent analysis (Chen et al., 2024b). Because the monitored catchments are hillslope systems without baseflow under dry conditions, runoff occurred only in response to rainfall and ceased shortly after rainfall ended. Therefore, the beginning and end of each runoff event were determined by combining field observations with automatic monitoring records at the measuring weirs, with runoff initiation defined as the time when flow was first detected and runoff termination defined as the time when flow ceased. To evaluate the impacts of different rainfall types on dissolved NH4+, NO3-, and P transport fluxes, five parameters were selected for cluster analysis: rainfall depth, duration, average intensity, maximum 30 min intensity, and rainfall erosivity. The calculation of rainfall erosivity (RE, MJ mm ha⁻¹ h⁻¹) is shown in Eqs. (1)–(3): 1RE=Ke⋅I302Ke=∑(Pr⋅Er)3Er=0.291-0.72e-0.082irt In Eq. (1), Ke represents the total kinetic energy of a rainfall event (MJ ha⁻¹), and I30 denotes the maximum rainfall intensity in 30 min (mm h⁻¹); In Eq. (2), Pr represents the rainfall amount of segment r (mm); and in Eq. (3), Er denotes the unit kinetic energy during rainfall segments r (MJ ha⁻¹ mm⁻¹). Here, r=1, 2, …, n refers to the consecutive rainfall segments within a single rainfall event, which were defined according to the temporal variation in recorded rainfall intensity; and ir is the rainfall intensity during rainfall segment r (mm h⁻¹).

To capture runoff variations during rainfall events, measuring weirs were installed at both the gully head (upslope drainage area of the gully head; UDGH) and the catchment outlet (Fig. 1I–K). HOBO Water Level Probes (Onset Computer Corporation, Bourne, MA, USA) recorded runoff dynamics at 10 min intervals by measuring pressure differences relative to identical probes placed in the air (Cheng et al., 2023; Chen et al., 2025b). Runoff samples were manually collected during rainfall events at the rising, peak, and recession stages of runoff using 1000 mL polyethylene bottles. Depending on runoff duration and flow variability, 3–23 runoff samples were collected for each event, with an average of 6 samples per event. After the rising and peak stages had been adequately characterized, sampling intervals were gradually extended during the late runoff stage to ensure full event coverage. All collected samples were immediately delivered to the laboratory for dissolved NH4+, NO3-, and P analysis. Notably, no baseflow was observed in either gully during non-rainfall periods; therefore, its potential influence on the runoff process was excluded from consideration.

A subsample was filtered through a 0.45 µm Millipore membrane to obtain the filtrate for nutrient analysis. Concentrations of dissolved NH4+, NO3-, and P were determined using standard spectrophotometric methods: Nessler's reagent spectrophotometry for NH4+, ultraviolet spectrophotometry for NO3-, and ammonium molybdate spectrophotometry for dissolved P. The runoff volume for each rainfall event was calculated using the calibrated weir depth-discharge curve and an empirical formula. By integrating high-frequency runoff sampling and dissolved nutrient concentrations, the dissolved nutrient transport flux for each rainfall event was determined (Eq. 4). Specifically, nutrient concentrations measured from discrete runoff samples were assigned to their corresponding sampling intervals, and the event-scale dissolved nutrient transport flux was calculated by summing the products of runoff volume and nutrient concentration across the entire runoff process (Bender et al., 2018). A detailed description of the calculation process can be found in our previous study (Chen et al., 2025b). In addition, after rainfall events were classified, dissolved nutrient transport fluxes were further aggregated within each rainfall type to compare differences in cumulative nutrient transport fluxes among rainfall types. 4F=∫t1t2Qt⋅C1000dt Where F is the transport flux of dissolved NH4+, NO3-, and P (kg). Qt refers to the runoff discharge at time t (m³ h⁻¹). t1 and t2 correspond to the times when runoff begins and ends, respectively (h). C represents the concentrations of dissolved NH4+, NO3-, and P (mg L⁻¹).

Rainfall types were classified using K-means clustering analysis via the R package “cluster” (v.2.1.3). To compare event-scale dissolved nutrient transport fluxes among different rainfall types, data normality and variance homogeneity were first assessed using Shapiro's test and Levene's test, respectively. If these assumptions were met, one-way ANOVA followed by Tukey's HSD test was used to compare dissolved NH4+, NO3-, and P transport fluxes across rainfall types; otherwise, the Kruskal-Wallis nonparametric test was applied. A statistically significant difference (p<0.05) was interpreted as evidence that rainfall type significantly influenced dissolved nutrient export dynamics. To quantify changes in dissolved NH4+, NO3-, and P concentrations during transport through the gully, a dilution ratio was calculated for each event as the outlet concentration divided by the corresponding concentration at the gully head. Values lower than 1 indicate dilution during transport through the gully, whereas values greater than 1 indicate enrichment. Correlation analysis was used to examine the relationships between dissolved NH4+, NO3-, and P transport fluxes and rainfall characteristics. Redundancy analysis (RDA) was employed to explore the individual effects of rainfall, runoff, and dissolved nutrient concentrations on dissolved NH4+, NO3-, and P transport fluxes. Model significance was assessed using a Monte Carlo permutation test with 999 permutations, and the relative importance of each explanatory variable was then determined through hierarchical partitioning. In addition, to assess the effects of gullies and rainfall types on dissolved NH4+, NO3-, and P transport fluxes, the relationships between nutrient transport fluxes and rainfall depth were fitted using either power or linear functions. A significant power function relationship (F=aRb) was observed between rainfall depth and the transport fluxes of dissolved NH4+, NO3-, and P, where the coefficient a indicates the sensitivity of nutrient transport fluxes to rainfall (higher values reflect greater mobilization potential) and the exponent b represents the efficiency with which transport fluxes respond to changes in rainfall depth. In the linear function (F=aR+b), parameter a likewise reflects the sensitivity of dissolved NH4+, NO3-, and P transport fluxes to rainfall depth. All statistical analyses were performed in R (v.4.5.0; R Core Team, Vienna, Austria).

From June to October of 2022 and 2023, 30 and 29 rainfall events were recorded in F1 and F2, respectively. K-means clustering classified these events into three distinct rainfall types (Table 1). Type A was characterized by low rainfall depth (20.8 mm), moderate duration (8.2 h), moderate intensity (3.9 mm h⁻¹), and low erosivity (71.9 MJ mm ha⁻¹ h⁻¹). Type B featured moderate rainfall depth (23.8 mm), short duration (0.9 h), high intensity (28.3 mm h⁻¹), and moderate erosivity (267.4 MJ mm ha⁻¹ h⁻¹). Type C exhibited high rainfall depth (79.6 mm), long duration (48.7 h), low intensity (1.7 mm h⁻¹), and the highest erosivity (333.7 MJ mm ha⁻¹ h⁻¹). Among these rainfall types, Type A was dominant, occurring 23 times in each catchment, while Types B and C were less frequent (F1: 4 and 3; F2: 3 and 3, respectively). However, the erosivity of Types B and C was 3.6 and 4.3 times higher than that of Type A, respectively (Table 1).

During Type C rainfall, cumulative runoff volume in the UDGH was 3.9 and 21.0 times higher than that under Types A and B, respectively, based on the mean values of the F1 and F2 catchments (Fig. 2A–B). At the outlet, cumulative runoff under Type C was 3.3 times higher than that under Type A and 19.0 times higher than that under Type B. On average, Type C rainfall generated significantly more runoff than Types A and B at both locations (P <0.05) (Fig. 2A–B). Specifically, the average runoff volume in the UDGH during Type C was 29.8 and 24.5 times greater than that under Types A and B, respectively, while at the outlet, it was 25.6 and 22.1 times higher than that under Types A and B, respectively (Fig. 2A–B). Although Type B produced more runoff than Type A, the difference was not statistically significant (P >0.05) (Fig. 2A–B).

Gullies accounted for only 12.4 % of the catchment area but contributed an average of 36.1 % of total runoff (based on the mean value of the F1 and F2 catchments) (Fig. 2C–D). This contribution varied with rainfall type, with the highest value observed under Type A (43.2 %), followed by Type B (40.1 %), and the lowest value under Type C (33.8 %) (Fig. 2C–D).

At both the gully head and the outlet, rainfall depth was the factor most strongly correlated with runoff volume and dissolved NH4+, NO3-, and P transport fluxes, followed by rainfall erosivity (Fig. S1). Moreover, redundancy analysis indicated that dissolved NH4+, NO3-, and P transport fluxes were primarily influenced by runoff volume, rainfall depth, and rainfall type, which ranked as the three most important factors, while their correlations with the corresponding concentrations were not significant. This indicates that dissolved NH4+, NO3-, and P transport fluxes were influenced more strongly by runoff and rainfall than by concentration (Fig. S2).

The power-law relationships between dissolved NH4+, NO3-, and P transport fluxes and rainfall depth revealed the regulatory role of gullies in modulating transport sensitivity (Fig. 8). The results showed that the gully significantly increased the sensitivity coefficient (a) for dissolved NH4+, NO3-, and P transport fluxes, with increases of 3.04, 1.71, and 2.84 times, respectively. However, gullies also reduced the overall transport efficiency (parameter b), by 4.8 %, 2.0 %, and 19.5 % for dissolved NH4+, NO3-, and P, respectively (Fig. 8). Among the different rainfall types, Type C rainfall markedly enhanced the sensitivity of dissolved NH4+, NO3-, and P fluxes compared to Types A and B (Fig. 9). Furthermore, within the same rainfall category, comparison of the slope values of the linear relationships between the gully head and the outlet showed that the presence of the gully amplified the rainfall sensitivity of dissolved NH4+, NO3-, and P transport at the catchment outlet (Fig. 9).

Our findings indicated that, although gullies occupied only 12.4 % of the catchment area, they contributed 36.1 % of the total runoff, highlighting their promotive effect on runoff generation (Fig. 2). Compared with the gently sloping farmland covered by dense crops during the rainy season (Sect. 2.1), the steep topography and relatively sparse vegetation of the gullies provided favorable conditions for runoff generation and concentration, which is consistent with previous studies (Chen et al., 2025a; Zhang and Zhang, 2025). These studies showed that, compared with bare land, crop-covered slopes can reduce runoff by 55.8 %–92.2 % (Chen et al., 2025a), and that the runoff coefficient under vegetation cover is only about 10 % of that on bare land (Zhang and Zhang, 2025). This effect is mainly related to rainfall interception by vegetation (Zhang et al., 2025), while steeper gully slopes further promote flow concentration (Zhang and Zhang, 2025; Xu et al., 2026). Together, these results suggest that gullies act as efficient hydrological connectors, rapidly transferring water from upslope farmland to the catchment outlet, which agrees with previous studies (Hou et al., 2022; Chen et al., 2024b; Chen et al., 2025b). Notably, gullies also showed a clear dilution effect on dissolved nutrients, especially NO3-, for which the average concentration ratio between the outlet and the gully head was 0.65 (Fig. 3). This pronounced reduction in runoff NO3- concentration may have resulted from the formation of ponded, anaerobic, or reducing microenvironments in locally flat sections of the gully bed, where denitrifying microorganisms could convert NO3- into gaseous nitrogen, thereby significantly lowering its concentration (He et al., 2026). Furthermore, runoff NO3-, as a highly mobile anion, is not readily adsorbed by sediments and tends to remain dissolved in gully water. As runoff accumulated, NO3- was more prone to dilution than retention (Wang et al., 2024; Zhao et al., 2025). In contrast, NH4+, as a positively charged ion, is more likely than NO3- to be adsorbed onto soil colloids, retained through cation exchange, and temporarily stored in sediments on the gully bed (Zhao et al., 2025). Therefore, the observed reduction in NH4+ concentrations may have been more the result of physical retention than dilution (Wang et al., 2024). Unlike DN, DP does not undergo gaseous transformation and is primarily governed by adsorption processes (Liu et al., 2020; Yang et al., 2024). Its response of DP was therefore more variable, likely because DP concentrations reflected not only runoff transport, but also interactions at the sediment-water interface within the gully (Bender et al., 2018). Our previous results showed that phosphorus concentrations in gully soils and sediments were significantly lower than those on adjacent farmland slopes (Chen et al., 2024c, 2025b; Wang et al., 2026). This pattern suggests that the equilibrium phosphorus concentration in gully sediments was lower than that in runoff water, which may favor phosphate release from sediments, especially under anoxic conditions (Bender et al., 2018). Interestingly, while gullies generally reduced the concentrations of dissolved NH4+, NO3-, and P from upslope runoff, they also amplified the sensitivity of transport fluxes to runoff (Fig. 8). This indicates that runoff volume, rather than concentration, primarily governed dissolved NH4+, NO3-, and P transport (Fig. S2). Once runoff connectivity was established, the increase in water volume outweighed the decrease in concentration. Therefore, managing runoff pathways within catchments may be important for reducing dissolved nutrient transport fluxes.

This study classified rainfall events using five rainfall parameters to examine how rainfall type affects runoff and dissolved N and P transport fluxes in gully-dominated catchments. This raises an important question: how do the hillslope and gully differ in their hydrological responses under different rainfall types? Our results showed that the runoff contribution of gullies was highest under Type A rainfall (43.2 %), followed by Type B (40.1 %), and lowest under Type C (33.8 %) (Fig. 2C–D). This pattern indicates clear differences in flow-path connectivity between hillslopes and gullies among rainfall types. Previous studies have shown that intense rainfall can activate surface hydrological connectivity through saturation-excess runoff and near-surface lateral flow, thereby connecting more distant potential runoff pathways and allowing runoff from remote hillslopes to participate in the hydrological process (Winter et al., 2022; Bian et al., 2026; Lei et al., 2026). This may explain why hillslope runoff contributed more, whereas gully runoff contributed less, under Type B rainfall with higher intensity and Type C rainfall with greater erosivity (i.e., extreme rainfall events). Different hydrological response patterns are likely to create different conditions for nutrient transport (Winter et al., 2022). In this study, concentrations of dissolved NH4+, NO3-, and P were significantly higher during the relatively light rainfall events of Types A and B than during the extreme rainfall events of Type C (Fig. 4). A similar pattern has been observed in Southwest China, where nutrient concentrations in runoff decreased with increasing rainfall following straw return practices on sloping farmland (Zhang et al., 2024; Feng et al., 2025). In contrast, monitoring in micro-catchments comprising paddy fields and drylands found that peak concentrations of dissolved NH4+ and NO3- followed the order of heavy rainstorm > rainstorm > moderate rain (Zhang et al., 2011). In the Jinglin River watershed of the Three Gorges Reservoir area, rainfall intensity was also found to enhance DP concentrations (Chen et al., 2024a), which differs from our results, where DP concentrations were lowest during extreme rainfall events (Fig. 4). As discussed in Sect. 4.1, the mechanisms driving DN and DP transport differ. Under varying rainfall conditions, the heterogeneity in gully soil, topography, and vegetation may intensify these differences (Wenng et al., 2020; Winter et al., 2022), leading to inconsistent patterns of nutrient concentrations across rainfall types (Feng et al., 2025). In contrast to concentration, nutrient transport fluxes showed a more consistent pattern (Zhao et al., 2026). Extreme storms produced much greater dissolved nutrient fluxes than Types A and B (Figs. 6; 7), indicating that hydrological forcing, rather than by concentration alone, mainly drove nutrient export during these events (Zhang and Zhang, 2025). At the plot scale in a potato-maize-sweet potato rotation system, dissolved NH4+, NO3-, and P transport fluxes during intense storm events were 613.8 %, 220.5 %, and 268.0 % higher, respectively, than those during moderate rainfall events (Feng et al., 2025). At the catchment scale, monitoring of 11 agricultural catchments in Canada showed that only three extreme storm events per year contributed 14 %–44 % of annual dissolved P flux (Ross et al., 2022). These findings help highlight why a small proportion of rainfall events accounted for most dissolved nutrient transport fluxes at the catchment scale (Chen et al., 2018). In addition, the sensitivity of dissolved nutrient fluxes clearly differed among rainfall types. The power-law coefficient a and the slope of the linear relationship with rainfall depth reflected the vulnerability of dissolved nutrient transport to rainfall forcing (Fig. 8). Our results showed that this sensitivity increased markedly under extreme rainfall and was further amplified by the presence of gullies (Figs. 8; 9). From a practical perspective, nutrient management in gully-dominated catchments should pay particular attention to low-frequency but high-impact storm events, because a single extreme storm may generate dissolved nutrient transport fluxes comparable to those from several ordinary rainfall events combined (Bian et al., 2026).

Compared with artificial drainage ditches, natural gullies are more dynamic because they have active headcut erosion, irregular morphology, steeper slopes, and stronger sediment interactions (Kumar Bhattacharya et al., 2024; Su et al., 2024; Chen et al., 2025b). This highlights the need to treat gullies as distinct geomorphic units rather than simply as natural drainage ditches. This study demonstrates that under natural rainfall conditions, gullies in agricultural catchments play a dual role. They can reduce dissolved nutrient concentrations through dilution or retention, but they can also enhance dissolved nutrient export by increasing runoff connectivity and transport efficiency. These findings provide important guidance for developing best management practices (BMPs) in gully-dominated catchments. The sharp increase in dissolved N and P fluxes during extreme rainfall events was mainly associated with enhanced surface hydrological connectivity. Therefore, a key management priority is to disrupt rapid flow connectivity during heavy storms and improve water and nutrient retention within the catchment. On agricultural upslopes, conservation tillage has been shown to increase water storage, improve surface roughness, and lengthen runoff pathways (Bayad et al., 2022; Chen et al., 2025a; Cui et al., 2025; Feng et al., 2025). In addition, microtopographic modifications such as terracing can effectively intercept runoff and slow surface flow connectivity on hillslopes (Wang et al., 2023; Wu et al., 2025). Because fertilization replenishes nutrient stocks in surface soils, low fertilizer-use efficiency may further aggravate water-quality deterioration, especially in intensively cultivated catchments exposed to frequent storms. Synchronizing fertilizer application with forecast rainfall patterns and using organic fertilizers and slow-release fertilizers may help improve crop nutrient uptake and reduce storm-driven non-point source pollution (Liu et al., 2020; Wenng et al., 2020). Within gullies, increasing vegetation cover on gully slopes and establishing buffer strips along gully margins may be especially important for slowing and weakening rapid surface runoff connectivity during major storms (Krzeminska et al., 2023). In addition, planting nutrient-intercepting vegetation or constructing small wetlands in the middle and lower parts of gullies can reduce pollutant loads and improve surface water quality (Krzeminska et al., 2023). Although these measures provide a practical basis for promoting sustainable agriculture and protecting water quality in the Mollisols region, their implementation still needs to be adjusted to local topography, land use, and rainfall conditions (Wenng et al., 2020).

This study also has several limitations. First, although the catchments were small, rainfall in each catchment was characterized using only one rain gauge, and thus spatial heterogeneity in rainfall within the catchment could not be resolved. Second, although extreme rainfall events were captured during the two-year monitoring period, climate change is expected to alter the frequency and intensity of such events. Longer-term monitoring is therefore needed to test whether the observed patterns remain valid across broader temporal scales and under future climate conditions. Third, a small amount of runoff from the gully banks could not be directly quantified. Although field observations suggested that this component was minor because grass cover and wheel ruts along the gully margins reduced direct flow into the gully, its contribution may still have led to a slight overestimation of the gully effect. These limitations should be considered when interpreting the results and planning future studies. Future work should further examine how gully morphology, vegetation recovery, and sediment deposition interact with rainfall extremes to regulate dissolved nutrient export. Long-term monitoring across more catchments is also needed to determine whether the patterns observed here are consistent across different gully sizes, developmental stages, and land-use settings. In addition, combining field monitoring with tracer techniques or process-based modeling could help disentangle the relative contributions of hydrological dilution, sediment retention, and in-channel biogeochemical transformation to dissolved nutrient dynamics.