The Chenqi watershed is situated east of Chenqibao Village, Puding County, Guizhou Province (26°15'48"N, 105°46'06"E). It represents the headwater region of the Houzhai River and encompasses an area of approximately 1.29 km². The watershed includes several small rural communities with a combined population of roughly 500–800 residents; the nearest township center is Puding County seat, located approximately 15 km to the southwest. The terrain within the watershed is highly varied, with elevations ranging from 1316 to 1524 m above sea level, and it is surrounded by mountains on three sides. The area exhibits a typical karst peak cluster-depression landform of the Guizhou Plateau. According to the Köppen climate classification, the region falls under the humid sub-tropical climate with hot summers. The multi-year average temperature is 14.3°C, and the average annual precipitation is 1338 mm. Precipitation exhibits significant spatio-temporal heterogeneity, with over 80% of the annual rainfall occurring during the rainy season from May to October. Geologically, the watershed is underlain by the Middle Triassic Guanling Formation (T₂g¹–T₂g³), characterized by thick-bedded limestone and dolomite intercalated with minor thin-bedded marlstone and shale [32]. The strata exhibit a gentle dip (angle < 8°). The first member of the Guanling Formation (T₂g¹), composed of interbedded marlstone and shale, forms the aquitard basement of the watershed. Land use types include cropland, grassland, shrubland, and woodland. Cropland, predominantly distributed in the valley bottom flatlands, accounts for approximately 35% of the watershed area. The dominant soils in the region are classified as yellow soils and limestone-derived rendzina soils, featuring moderate fertility but high susceptibility to erosion in sloped terrain [33].

Due to the marked topographic variation within the watershed, agricultural land use exhibits a clear spatial pattern: paddy rice is primarily cultivated at higher elevations, whereas tomatoes are grown in lower-lying valley areas. This elevation-based crop zoning reflects both agronomic adaptation and terrain-driven water management needs. Importantly, such a configuration can be viewed as a form of non-structural best management practice (BMP), potentially influencing nutrient transport and hydrological connectivity. This topography-based division also provides a practical rationale for distinguishing sampling sites between upstream (rice-dominated) and downstream (tomato-dominated) zones.

Five hydrological monitoring points were established across distinct crop zones: the primary surface water outlet (SO), the primary well water outlet (WO), and three groundwater wells (W1–W3). W1 and W2 were situated in paddy rice fields at higher elevations, while W3 was positioned in a downstream tomato cultivation zone. With the exception of the SO, which samples surface water, all other points sample groundwater. This deployment strategy was guided by topography- based crop zoning, which reflects how karst terrain naturally segments agricultural land use into distinct cultivation zones. Such zoning constitutes a form of non-structural best management practice (BMP), facilitating the evaluation of land-use-related water quality from a landscape hydrology perspective [34]. The spatial distribution of these points is illustrated in Fig. 1, and their specific information is detailed in Table 1.

Water sample collection adhered to the recommended protocols of the Chinese Ecosystem Research Network (CERN) and the National Field Research Stations. The specific procedures implemented were as follows: 1) Sampling frequency: Samples were collected monthly in the middle of each month. Sampling in 2016 was limited to March, June, September, and December due to personnel adjustments and other logistical factors. These irregular data were used cautiously in seasonal averaging and excluded from PCA or interannual comparisons to avoid analytical bias. 2) Sample container preparation and collection: Pre-cleaned polyethylene bottles were rinsed three times with site water before sample collection. Detailed information, including sampling time, location identifier, and water temperature, was recorded directly onto the sample bottles using waterproof markers. 3) In-situ physicochemical parameter measurement: During sample collection, field personnel used a portable multi-parameter meter to measure and manually record water temperature, pH, and electrical conductivity (EC) on site. 4) Sampling depth protocol: For samples collected from the three groundwater monitoring wells (W1, W2, W3), the water sampler was positioned approximately 0.5 meters below the water surface. In contrast, samples from WO and SO were collected directly at the point of discharge using the prepared sampling bottles. The consistent sampling depth ensured comparability of groundwater data across crop zones. 5) Sample handling and preservation: Samples were kept refrigerated in coolers with ice packs and later stored at 4°C in the laboratory until analysis.

To reflect topographic segmentation in subsequent analysis, the three well points were classified into two crop-related zones for subsequent analysis: an upstream paddy rice zone (W1–W2) and a downstream tomato zone (W3).

The upper-left panels show the location of the study area at the continental (Asia), provincial (Guizhou), and municipal (Anshun) scales. The top-right panel illustrates the elevation distribution of Puding County, characterized by a typical karst terrain. The middle panel presents a satellite view of the observation area, indicating the positions of the surface water outlet (SO), well outlet (WO), and three groundwater wells (W1–W3), which align with the local agricultural topography. The bottom panel provides a conceptual cross-section showing the elevation gradient, crop types (rice vs. tomato), and the placement of groundwater wells along the slope, highlighting the hydrological connectivity across land-use zones.

Water quality analysis of the collected samples was conducted in accordance with standardized protocols, including those established by the CERN and the “National Field Scientific Observation and Research Station Observation Technical Specification Series—Technical Specifications for Karst Regional Ecosystem Network Observation”. These standards provide essential guidance for ensuring the scientific validity and reliability of the observational data. The physicochemical assessment of water quality in karst settings typically involves multiple indicators sensitive to nutrient enrichment and groundwater–surface water interactions, such as electrical conductivity (EC), nitrate (NO₃⁻), total phosphorus (TP), and dissolved oxygen (DO). These indicators are particularly responsive to fertilizer use, hydrological pathways, and lithological buffering capacity in karst systems [32, 35]. A total of 19 indicators were analyzed, including: water temperature (T), pH, EC, calcium (Ca2+), magnesium (Mg2+), potassium (K+), sodium (Na+), carbonate (CO32-), bicarbonate (HCO3), chloride (Cl-), sulfate (SO42-), phosphate (PO43-), nitrate (NO3-), total nitrogen (TN), total phosphorus (TP), dissolved organic carbon (DOC), dissolved oxygen (DO), chemical oxygen demand (CODcr), total dissolved solids (TDS). Cations were quantified using inductively coupled plasma optical emission spectrometry (ICP-OES), while anions were analyzed by ion chromatography (IC), with bicarbonate (HCO₃⁻) determined by acid-base titration; TN and TP were measured using alkaline potassium persulfate digestion-UV spectrophotometry and ammonium molybdate spectrophotometry, respectively following standard methods. EC was measured in situ using an electrode probe. These methods are widely accepted and validated in karst water quality research [32, 36].

All measured data were compiled to form the Water Environmental Characteristics Dataset of a Representative Karst Small Watershed in Guizhou (2016–2022) [31], a robust dataset that underpins the study’s long-term and multi-indicator investigation of water quality. This dataset also enables a comparative evaluation of non-point source pollution across different cropping zones under differentiated land-use regimes. Detailed analytical parameters, detection limits and instrument specifications are documented in Table A1.

CCME WQI was applied to evaluate the integrated status of outflow water quality at different conditions. It requires a minimum of four water quality indicators as model inputs. In this study, eight inputs (pH, Cl, SO42, NO3, TDS, DO, TN, TP) were selected to reflect both salinity and nutrient-related impacts on water quality in the karst environment. According to national environmental quality standards for groundwater and surface water in agricultural regions (GB/T 14848 − 2017; GB 3838 − 2002), the maximum allowable values for these parameters are: pH = 6.5–8.5, Cl⁻ = 250 mg·L⁻¹, SO₄²⁻ = 250 mg·L⁻¹, NO₃⁻ = 10 mg·L⁻¹, TDS = 1000 mg·L⁻¹, DO = 5 mg·L⁻¹, TN = 2 mg·L⁻¹, and TP = 0.02 mg·L⁻¹ [37, 38].

The CCME WQI integrates three dimensions—scope (F1), frequency (F2), and amplitude (F3)—to quantify the extent and severity of deviations from these objectives. The WQI is calculated using the following equations [39, 40]:

WQI = 100-[], (1)

=[]×100, (2)

=[]×100, (3)

=[], (4)

If a test value falls below the objective value, the excursion for that test value is calculated as follows:

nse=[], (5)

Conversely, if the test value exceeds the objective value, the excursion value is calculated as follows:

nse=[], (6)

This is calculated by an asymptotic function that scales the normalized sum of excursions (nse) of test values from objectives to yield a value between 0 and 100. The CCME model proposed five water quality classes: excellent (WQI = 95–100), good (WQI = 80–94), fair (WQI = 65–79), marginal (WQI = 45–64), and poor (WQI = 0–44) [39]. This classification provides a standardized framework for comparing groundwater and surface water quality across different zones, particularly useful for evaluating non-structural BMP effectiveness in upstream vs. downstream settings.

All the data were entered into Excel 2025 for preliminary collation. Prior to statistical analysis, the Shapiro-Wilk test and Levene’s test were conducted using IBM SPSS Statistics (version 27.0) to assess data normality and homogeneity of variance, respectively. One-way analysis of variance (ANOVA) was used to analyze the effects of location of wells on WQI.

In this study, ANOVA was further integrated with principal component analysis (PCA) and spatial zoning to explore the underlying structure of water quality variation and its association with land-use type. Similar multivariate statistical frameworks have been widely utilized in karst environments, including those leveraging hydrochemical and tracer variables to elucidate hydrodynamics [41].Pearson correlation analysis was used to evaluate relationships between WQI and individual physicochemical parameters, providing insight into how chemical species co-vary under different cropping systems [41]. The coefficient of determination (R², Eq. [8]) was used to assess linear model fitness during exploratory correlation analysis. All figures were generated using Origin 2021 (OriginLab, Northampton, MA, USA).

Figures 2A and 2B show the spatial distribution of Water Quality Index (WQI) values across different sampling points. Among the groundwater wells (W1–W3), W1 recorded the highest WQI (72.2), followed by W3, while W2 showed the lowest value (53.7). For the outlet sites, WO and SO exhibited lower WQI scores, averaging 57.2 and 52.8, respectively. These results suggest that groundwater samples generally fell within the “fair” to lower “good” quality category, while the outlet waters approached the “marginal” threshold. A comparison between influent (In) and effluent (Out) groundwater in Fig. 2B further highlights this pattern, with In values clustering around 62 and Out values around 54, indicating better overall water quality in the influent samples.

Figures 2C and 2D present the principal component analysis (PCA) results for influent (In) and outlet (Out) waters, respectively. In Fig. 2C, PC1 explains 24.5% of the total variance and is characterized by strong positive loadings for Mg²⁺, K⁺, and EC, and negative loadings for TN, NO₃⁻, and DO. PC2 (13.7%) shows a positive correlation with CO₃²⁻ and a negative correlation with HCO₃⁻. In contrast, Fig. 2D shows that PC1 for outlet water explains 32.6% of the variance and includes high loadings for TDS, EC, Mg²⁺, Ca²⁺, Na⁺, SO₄²⁻, and K⁺. These results indicate distinct controlling factors between groundwater and outlet water, with outlet samples exhibiting higher variance explained by the first principal component.

Figures 2E and 2F illustrate the Pearson correlation matrices between WQI and selected water quality parameters for In and Out samples, respectively. In the groundwater (In) samples (Fig. 2E), the correlations between WQI and measured variables are generally weak, suggesting a diffuse control. For outlet water (Fig. 2F), WQI shows a significant negative correlation with total phosphorus (TP) and a significant positive correlation with HCO₃⁻, indicating more pronounced relationships between WQI and specific chemical species in effluent samples.

Figure 3 presents the spatiotemporal variability of nitrate nitrogen (NO₃⁻-N), total nitrogen (TN), and total phosphorus (TP) across groundwater wells (W1–W3) and outlet sites (WO, SO) from 2016 to 2022. The left panels (A, C, E) display the time-series concentrations, while the right panels (B, D, F) show corresponding distributions across sites.

In Fig. 3A, NO₃⁻-N concentrations exhibited frequent fluctuations at W2 and WO, with multiple peaks observed throughout the monitoring period. W1 and W3 maintained relatively lower and more stable levels. The outlet site SO showed occasional surges but remained below WO for most of the period. Figure 3B supports these observations, with the widest interquartile ranges and highest maximum NO₃⁻-N concentrations found at W2 and WO.

Figure 3C illustrates the TN dynamics, where higher and more variable concentrations were observed at WO and SO, particularly during certain years. Among the wells, W2 recorded elevated TN values compared to W1 and W3. As shown in Fig. 3D, SO had the highest overall TN levels, while W3 displayed the narrowest range.

In Fig. 3E, TP concentrations were mostly low across groundwater wells but showed marked peaks at the surface outlet sites WO and SO. These variations were especially pronounced at SO during specific periods. The boxplots in Fig. 3F confirm that TP concentrations were substantially higher at WO and SO, while groundwater sites maintained consistently low levels throughout the study period.

Taken together, the figure highlights clear temporal variability and site-specific differences in nutrient concentrations, with surface outlets generally exhibiting higher fluctuations and elevated levels compared to groundwater wells.

Figure 4 illustrates the temporal and seasonal variation in Water Quality Index (WQI) across five monitoring sites. As shown in Fig. 4A, W1 and W3 exhibited the highest WQI values, indicating relatively better groundwater quality. In contrast, SO consistently recorded the lowest values, suggesting comparatively poorer surface water conditions. Overall, groundwater sites maintained higher and more stable WQI values than surface outlets.

Figures 4B–4F present WQI boxplots classified by dry and wet seasons for each site. Seasonal differences in WQI are evident, particularly at SO and W2. In Fig. 4B, WQI values at W1 are highly consistent between seasons, with minimal spread. Figure 4C shows a slightly wider distribution at W2, especially during the wet season. Figure 4D indicates that WQI at W3 is moderately affected by seasonal changes, but remains more stable than surface outlets. Figure 4E demonstrates that SO has the greatest seasonal variability, with lower WQI values during the wet season and a larger interquartile range. Figure 4F shows a moderate seasonal contrast at WO, but with smaller fluctuations than SO.

Across all sites, the coefficient of variation (CV) follows the descending order: SO > W2 > W3 > WO > W1. The highest CV (22.7%) is observed at SO, while the lowest (11.1%) occurs at W1. These patterns indicate that surface outlets are more influenced by seasonal variation, whereas groundwater sites maintain relatively stable water quality.

The relationship between daily rainfall and Water Quality Index (WQI) is shown in Fig. 5. A segmented regression model identified a breakpoint at approximately 32 mm/day. Below this threshold, WQI values showed minimal response to rainfall variation, with a very low coefficient of determination (R² = 0.002). In contrast, when daily rainfall exceeded 32 mm/day, WQI values increased with rainfall amount, as indicated by a stronger positive relationship (R² = 0.61). However, rainfall events above this breakpoint were infrequent during the study period, leading to limited data points in the high-rainfall range.

The spatial variation in WQI (Fig. 2A) reveals a notable inflection: groundwater quality at W2 (median WQI = 53.7) is significantly lower than at both upstream W3 (median = 62.0) and downstream W1 (median = 72.2). This pattern, contrary to a simple upstream-downstream degradation gradient, highlights the role of localized factors in pollutant accumulation and attenuation. These spatial differences in WQI (Fig. 2A) highlight the contrasting influences of natural hydrogeological processes and anthropogenic inputs within the Chenqi watershed. This mirrors patterns reported in other subtropical karst settings, where subsurface flow through epikarst and soil layers attenuates nutrient and sediment loads more effectively than open channels [15, 18]. In particular, W1, located in cropland near the outlet, showed the highest mean WQI (72.2), suggesting that recharge waters may benefit from sufficient residence time in soil and fractured bedrock. This allows for sorption and biotic uptake of nitrogen and phosphorus prior to aquifer recharge [13, 42]. In contrast, W2 (upper valley) and W3 (mid-valley) exhibited moderately lower WQI values (53.7 and 62.0, respectively), implying that shallower soils and steeper hydraulic gradients at those sites may permit more direct pollutant percolation [15].

W2’s poor water quality is likely linked to its position in a topographic depression with complex cropping systems. As a convergence zone for runoff from surrounding fields, W2 tends to accumulate nutrients from mixed agricultural inputs. Its soil, characterized by higher clay content and poor drainage, limits infiltration and promotes surface ponding, which increases nutrient leaching into groundwater. This corresponds with the frequent spikes in NO₃⁻-N and TN observed at W2 (Fig. 3), reflecting intermittent nutrient surges from fertilization and infiltration events.

In contrast, W1’s relatively high-water quality may stem from a combination of topographic filtration and tomato cultivation practices. Located in a downstream tomato zone with sandy loam soil, W1 benefits from enhanced infiltration and pollutant sorption, acting as a natural buffer. Tomato cultivation here, with lower planting density and more regulated irrigation compared to W2’s mixed cropping, may reduce nutrient accumulation in soil. Additionally, W1’s relative distance from rural settlements minimizes exposure to domestic wastewater, reinforcing its role as a groundwater recovery zone. This "inflection and recovery" pattern supports the hypothesis that crop zoning—combined with topography—modulates pollutant transport in karst systems.

This spatially heterogeneous pattern is consistent with studies in karst watersheds showing that agricultural land use, coupled with catchment hydrology, drives chronically elevated nitrate and strong seasonality in export [43].As a broader pattern, a China-wide meta-analysis of 221 datasets reported that nitrate leaching in vegetable systems averaged ~ 79 kg N ha⁻¹ per season, primarily driven by very high N fertilizer (~ 423 kg N ha⁻¹) and water inputs, with greenhouse systems > open-field [44]. Similarly, slope position modulates nutrient retention; toe-slope/riparian zones often show lower nitrate leaching owing to longer, slower, and more saturated subsurface flowpaths that enhance denitrification (e.g., [45]).

PCA further delineates the hydrochemical drivers across spatial gradients (Fig. 2C–D). At influent wells (Fig. 2C), PC1 (24.5% variance) correlated positively with Mg²⁺, EC, and TN, and negatively with K⁺ and DO, indicating that baseline groundwater chemistry is governed by carbonate dissolution and natural nutrient cycling [10, 12]. The strong positive loadings of Mg²⁺ and EC reflect the pervasive influence of carbonate mineral weathering (CaCO₃, MgCO₃) in the Triassic Guanling Formation (Fig. 1). PC2 (13.7%) loaded on CO₃²⁻ and HCO₃⁻, further supporting a geology-dominated signature in recharge waters.

In contrast, PCA of effluent waters (WO, SO; Fig. 2D) revealed PC1 explaining 32.6% of variance, with strong positive loadings on TDS, EC, Mg²⁺, Ca²⁺, Na⁺, SO₄²⁻, and K⁺. These ionic are commonly associated with both anthropogenic inputs, such as agricultural fertilizers (e.g., KNO₃, (NH₄)₂SO₄), soil leachate, rural domestic wastewater, and natural geogenic sources, including the weathering of carbonate rocks (e.g., CaCO₃, MgCO₃) [16, 17]. The elevated TDS and EC at the outlets, along with increased SO₄²⁻ and K⁺, underscore the role of surface runoff in mobilizing soluble salts and nutrients [46].

The negative correlation between WQI and TP (Fig. 2F) confirms phosphorus as a primary contaminant at discharge points, with intensification during fertilizer application (April–June) and subsequent rainy-season flushing [16]. Meanwhile, the positive WQI–DO relationship indicates that organic loading (e.g., manure runoff) can reduce dissolved oxygen levels, thereby exacerbating water-quality deterioration [47].

The observed water quality trends validate that the watershed’s crop zoning—rice at higher elevations (W3) and tomatoes downstream (W1)—functions as an effective Best Management Practice (BMP), termed the Karst Rice-Tomato Cascade System (KRTCS). This configuration leverages crop-specific traits to mitigate nutrient fluxes. Upstream rice paddies (W3), characterized by prolonged flooding, act as nutrient sinks. The anoxic conditions promote denitrification, reducing nitrate leaching, while the dense root systems facilitate phosphorus uptake, contributing to W3’s relative stable WQI. Downstream tomato fields (W1) enhance nutrient interception and soil filtration. With high nutrient demand and deeper root architecture, tomatoes efficiently absorb residual nitrogen and phosphorus from upslope runoff, while their well-drained soil matrix supports the sorption and immobilization dissolved pollutants, thereby reducing their migration into groundwater.

Seasonal WQI patterns (Fig. 3) further reflect the buffering role of this crop zoning system. Among all monitoring points, the surface water outlet (SO) exhibited the highest seasonal variability (CV = 22.7%), while W1 groundwater showed the lowest (CV = 11.1%), indicating greater stability in subsurface flows. The marked decline SO WQI during the wet season (May–October) coincides with both peak monsoonal rainfall and the primary agricultural fertilization window (April–June). This overlap likely increases nutrient mobilization, especially of nitrogen and phosphorus, from croplands into surface water bodies, thereby reducing water quality metrics [15, 18]. However, further quantification of runoff intensity and lag responses is necessary to confirm this linkage. During high-precipitation events, overland flow may rapidly carry agrochemicals into channels and karst conduits, overwhelming in-stream dilution capacity [42].

Conversely, in the dry season (November–April) reduced runoff and nutrient input, allowing for natural attenuation mechanisms, such as plant uptake, microbial transformation, and in-stream denitrification, to dominate, resulting in higher WQI values. These seasonal patterns underscore the capacity of KRTCS as a functional BMP to modulate contaminant load across hydrological cycles.

The biogeochemical buffering may also relate to subsurface soil processes. Soil enzyme activities, as indicators of microbial nutrient limitations, respond to seasonal vegetation dynamics and land-use changes. Farmland reversion and vegetation restoration have been shown to modify both biotic and abiotic soil conditions, potentially altering nutrient cycling efficiency and affecting the hydrochemical composition of adjacent aquatic systems [48]. Such buffering mechanisms have also been documented in monsoonal karst regions of Yunnan and Guangxi [18, 46].

Groundwater wells, particularly W1, exhibited muted seasonal fluctuations, reflecting the combined of slower groundwater velocities and epikarst matrix sorption[10, 12]. W2 and W3, located in irrigated paddy zones, showed moderate variability (CV = 15–17%), indicative of irrigation return flows and seasonally shifting water tables [13]. The temporal resilience of W1 suggests that protective measures around recharge zones, such as vegetative cover or conservation tillage [4], can further enhance its role as a BMP.

Vegetation type also affects soil respiration, microbial activity and overall soil physicochemical conditions in karst systems [49] Changes in land use (e.g., farmland abandonment or conversion to grassland) can impact soil aggregate stability, a sensitive indicator of erosion resistance [50]. When aggregate stability declines, surface runoff may mobilize more particles, increasing nutrient and pollutant loads in nearby water bodies [51]. Thus, maintaining stable vegetation – soil systems form another key BMP strategy for preserving groundwater and surface water quality.

The threshold behavior depicted in Fig. 4, where WQI improved significantly (R² = 0.61) only when daily rainfall exceeded 32 mm, underscores the nonlinear dynamics of monsoonal karst hydrology. Under light to moderate rainfall (< 32 mm/day, based on the local distribution), infiltration predominates, and pollutants may become more concentrated due to limited overland runoff, failing to produce effective dilution. Once rainfall intensity surpasses the threshold, surface runoff increases rapidly, flushing accumulated contaminants downstream and transiently enhancing water quality (WQI). This phenomenon parallels findings from southern China, where only intense precipitation events effectively transport pollutants from karst hillslopes to discharge points [42]. For water managers, recognizing such thresholds is vital: timing fertilizer application to avoid impending heavy rainfall can reduce nutrient leaching, while constructing vegetative buffer strips may intercept diffuse pollution during moderate rains [4, 15].

This study provides a seven-year perspective on groundwater quality in karst watershed, but several limitations should be acknowledged. First, land‐use dynamics (e.g., expansion of vegetable plots) were not quantified; integrating remote‐sensing data could clarify how spatial and temporal land cover changes affects hydrochemistry. Second, key biogeochemical processes—such as denitrification and carbonate precipitation — were inferred from ion patterns; future studies should employ isotope or tracer analyses (e.g., δ¹⁵N‐NO₃⁻, δ¹³C‐DIC) to more precisely distinguish natural attenuation from anthropogenic sources. Third, although a 32 mm/day rainfall threshold was identified, further research should couple hydrological models with downscaled climate projections to anticipate water‐quality trends under future precipitation regimes. Fourth, we did not continuously monitor groundwater level fluctuations or surface/groundwater discharge volumes, both of which are critical for understanding hydrological and transport mechanisms. Such data would enable quantification of exchange rates and residence times that govern pollutant transfer between surface and subsurface domains, currently limiting our ability to directly link spatiotemporal water quality variations to surface-groundwater interaction processes. Finally, our analysis focused on major ions and nutrients, future monitoring should incorporate trace metals and emerging organic contaminants to provide a more holistic assessment of karst water quality.

The findings confirm that crop zoning based on elevation and hydrology connectivity—as embodied in the Karst Rice-Tomato Cascade System (KRTCS)—functions as an effective BMP to mitigate diffuse agricultural pollution in karst landscapes. To enhance the performance of this BMP, we recommended several targeted interventions: (1) Improving drainage at W2 to reduce surface ponding and nutrient accumulation; (2) Formalizing fertilization schedule based on rainfall threshold to avoid nutrient losses during extreme events; (3) Expanding downstream tomato buffer to reinforce the recovery function of W1 as a natural filtration site.

Together, these strategies contribute to an adaptive management frameworks tailored to the characteristics of monsoonal karst agroecosystems, aiming to balance agricultural productivity and groundwater protection.