The Impact of Landscape Composition and Configuration on Nitrogen Compound Concentrations in Small Polish Lowland Rivers During the Non-Vegetative Season

Abstract

Understanding how landscape structure affects nutrient pollution is essential for contemporary effective river basin management. This study examined the influence of landscape composition and configuration on concentrations of nitrate (NO₃⁻), nitrite (NO₂⁻), and ammonium (NH₄⁺) in 30 small lowland catchments of central–eastern Poland during the cold period. Water samples were collected monthly from September 2021 to April 2022, and land-use patterns were quantified using landscape metrics derived from high-resolution spatial data at the catchment scale and within riparian buffer zones. The results showed that the impact of land use on nitrogen concentrations was strongly dependent on both landscape type and spatial scale. Forests, meadows, wetlands, and water bodies generally acted as sink landscapes, reducing nitrate and nitrite levels. The effect was more pronounced in catchments where forest patches (mainly coniferous) covered a larger area, had greater total Edge Length, and were more complex in shape. It was advantageous when meadow patches were large, cohesive, and weakly fragmented. In contrast, arable land and built-up areas consistently functioned as source landscapes, contributing to higher nitrogen concentrations when characterized by a larger share, size (both), and aggregation degree of patches (arable land). Higher landscape diversity at the catchment scale was associated with lower nitrate and nitrite concentrations. Overall, land-use effects were best explained at larger spatial extents, especially the entire catchment and the 500 m buffer zone. These findings emphasize the need to integrate landscape structure and appropriate spatial scale into nutrient management strategies for lowland agricultural catchments.

1. Introduction

Land use and land cover (LULC) changes are among the most important drivers of disturbances in the hydrological cycle and of water-quality deterioration on a global scale [1]. Research on the impact of land use on water quality has been continuously conducted for several decades, with the first papers on this topic appearing before the advent of GIS and its widespread adoption [2]. Ref. [3] distinguished three main stages of research development in this field. The first “wave” of investigations in this research trend began in the 1960s with the exploration of the impact of a catchment’s morphological features on turbidity, dissolved oxygen concentration, and water temperature. The second stage developed in the 1970s, when scientists creating this trend focused on analyses at the catchment scale. Contributors to the third, clearly distinct, and ongoing trend employ remote sensing, GIS, and multivariate statistical analysis to elucidate the impact of LULC on suspended matter, nutrient concentrations, and the ecological coherence of rivers [3]. The development of GIS tools and the increased availability of more accurate, high-resolution spatial data allow consideration of the impact of landscape structure on water quality at various spatial scales—from catchment [4] to various variants of buffer zones along streams [5,6] and rings around springs [7] or measurement points [8].

In landscape ecology, landscape structure can be characterized by two groups of indicators: composition, which describes the number and diversity of LULC, and configuration, which describes the spatial arrangement of patches of different land cover types [9]. Landscape metrics provide an objective description of these characteristics. These measurable units enable the determination of landscape structure and the tracking of its changes, thereby describing and quantifying spatial patterns and ecological processes across time and space [10]. So far, many studies on water quality have used landscape metrics, including space metrics such as PLAND (percentage of land cover of a given class) [11,12], landscape configuration metrics [13,14], and metrics that account for distance from the stream (inverse distance-weighted method) [15,16].

Recent scientific work on the relationships between LULC and water quality, based on landscape metrics, has spanned nearly all continents [17,18], resulting in a rich body of literature across diverse environmental contexts. However, in Poland, this trend in landscape ecology is essentially nonexistent, and only individual studies address it within a very limited scope [18,19]. According to [20], based on a review of studies conducted between 2017 and 2022, lowland catchments remain relatively underrepresented in research examining relationships between landscape structure and water quality: more than 70% of reviewed studies were based in mountainous, upland areas with diverse morphologies, e.g., [21,22,23]. It should also be noted that a significant portion of studies in this field were based on 20 or fewer measurement points/catchments, e.g., [7,24,25,26,27]. Simultaneously, it must be emphasized that the issue of nutrient pollution in Poland is extremely important, as the Baltic Sea is listed alongside the North American Great Lakes, Chesapeake Bay, and the Gulf of Mexico as eutrophic aquatic environments experiencing low oxygenation (hypoxia) or lack of oxygen (anoxia) caused by the death of algae and their decomposition [28,29]. The analyzed catchments belong to the Vistula River basin, which ultimately drains into the Baltic Sea, and therefore contribute indirectly to nutrient loads entering this marine ecosystem. According to the predictive model system developed by [30], Poland represents one of the hotspots of dissolved inorganic nitrogen and phosphorus introduced to coastal ecosystems.

This paper presents a comprehensive study of how landscape composition and configuration influence the concentrations of nitrate (NO₃⁻), nitrite (NO₂⁻), and ammonium (NH₄⁺) ions in 30 lowland catchments of the Bug and Narew tributaries during winter. The choice of the cold season was aimed at reducing the influence of macrophytes on nitrogen concentrations, given their intensive development in the studied streams during the warm period, and at enhancing the detectability of landscape-related effects. The specific research objectives were (1) identification of the spatial pattern of concentrations of selected nitrogen compounds during the non-vegetative season; (2) assessment of relationships of landscape composition and configuration on the concentrations of selected nitrogen compounds; (3) determination of the spatial scale (the entire catchment, buffer zones of 50, 250, and 500 m on both sides of the stream) at which relationships between landscape metrics and nitrogen compound concentrations in lowland streams perform best.

2. Study Area

The research covers lowland streams and their catchments belonging to the Bug and Narew rivers, major tributaries of the Vistula River (Figure 1, Appendix A). The investigated catchments are distributed across the Central Masovia Lowland and the Southern Podlasie Lowland within the Central Poland Lowlands subprovince [31]. Due to the Warthanian glaciation of the area, catchments formed in glacial rocks, mainly glaciofluvial sands and gravels, except in the southeast, where they formed in glacial clays. The area’s morphology is determined by postglacial plateaus in the south, divided by the Liwiec valley, and the extensive Bug valley in the north [32]. The area is characterized by a mosaic of Podzols and Luvisols (according to the WRB soil classification). In the western catchments, muck soils constitute a significant proportion of the soil cover [33]. According to the Köppen–Geiger climate classification, modified by [34], the study area is classified as a warm-summer humid continental climate (Dfb). The studied streams exhibit a nival regime, with the highest streamflow rates occurring in spring due to snowmelt and the lowest during summer and autumn [35]. The area’s landscapes are dominated by agriculture, with exceptions (Figure 2). The crop structure is dominated by corn, oats, triticale, rye, and potatoes [36]. Nitrogen fertilizer consumption (expressed as pure nutrient) in the 2019–2020 period in the Masovian Voivodeship, where the studied catchments are located, amounted to 60 kg N·ha⁻¹, which was slightly lower than the national average of 69 kg N·ha⁻¹ [37]. It should be emphasized that, in addition to mineral fertilizers, organic fertilizers are widely applied in the region. During field investigations, the authors observed manure being transported and temporarily stored on agricultural fields in the vicinity of the stream. Dominant tree species include Scots pine (Pinus sylvestris L.), silver birch (Betula pendula R.), common oak (Quercus robur L.), and, in some river valleys, white willow (Salix alba L.), common aspen (Populus tremula L.), and black alder (Alnus glutinosa (L.) Gaertn.). The surveyed area is predominantly rural, with the small town of Łochów present in P8. Some of the streams are presented below (Figure 3).

3. Materials and Methods

3.1. Field and Laboratory Investigations

Field investigations were conducted in 30 catchments, with areas ranging from 9.77 to 30.96 km². The selection of catchments and sampling sites was guided by the following criteria: catchments differed in the share of main land-use categories (the surveyed area included forest, meadow, arable, and mixed catchments), and sampling points were located above dams and wastewater inflows, factors previously shown to significantly modify water quality [38,39,40,41,42], which could weaken the influence of landscape composition and configuration on water quality. However, agricultural drainage effluents originating from tile drainage systems may still contribute nitrate to the streams as a diffuse source of pollution; such drainage effluent was observed below site P26 and was sampled twice. Sampling was conducted monthly from September 2021 to April 2022, covering the non-vegetative period and its marginal parts, which significantly limit the impact of instream vegetation on nutrient ion accumulation [43]. Water samples were collected from the main current of the streams into polyethylene bottles. After immediate transport to the laboratory, samples were subjected to membrane filtration, and NH₄⁺, NO₃⁻, and NO₂⁻ concentrations were measured photometrically with an LF300 spectrophotometer. Calibration was performed using standard solutions according to the manufacturer’s procedures. The limit of detection (LOD) values for the applied methods were 0.1 mg·dm⁻¹ for NO₃⁻, 0.02 mg·dm⁻¹ for NO₂⁻, and 0.05 mg·dm⁻¹ for NH₄⁺. The photometric accuracy of the instrument is ±3% according to the manufacturer’s specifications (Slandi Ltd., Michałowice, Poland).

3.2. Landscape Metrics

Given the broad use of high-resolution LULC data in landscape ecology, e.g., [8,21,44], the current study is based on 10 m or higher-resolution raster data from the Copernicus program [45] and the Sentinel Global Land Cover program [46] (Appendix B). Copernicus High-Resolution Layers data were used to compute class-level landscape metrics, while S2GLC data were used for landscape-level diversity metrics and metrics for arable land, due to the lack of such data in the Copernicus. Copernicus data consist of several thematic products focusing on specific land cover types, such as impervious surfaces, forests, grasslands, wetlands, and small woody features. Due to dedicated classification procedures and the integration of multiple data sources, these products are generally considered more accurate for selected thematic layers and therefore constitute the basis of the analyses. In contrast, S2GLC represents a comprehensive land cover map derived primarily from automated classification of multi-temporal Sentinel-2 imagery. Consequently, some differences in land cover classification were observed between the datasets. In the study area, built-up areas are underestimated in S2GLC compared with the Copernicus imperviousness layer, often being misclassified as arable land. Wetlands (i.e., water bodies and areas that are permanently or temporarily wet) represent the least consistent class between the datasets. In S2GLC, they appear to occupy relatively large areas within the catchments, whereas the Copernicus Water and Wetness layer frequently overlaps with meadow areas identified in both datasets. These discrepancies result from differences in classification approaches: in S2GLC, wetlands are treated as a land cover class, whereas in Copernicus HRL, they represent a wetness-related category that may overlap with other land cover types. Therefore, the Water and Wetness layer was evaluated only in terms of its proportional area within the catchment, and no additional landscape metrics were calculated for this class. In contrast, grasslands and forests show relatively high agreement between the datasets.

Landscape metrics were computed at the catchment scale and for 50, 250, and 500 m wide buffer zones along rivers from the sampling site to the springs, using the QGIS (version 3.22; QGIS Development Team, Open Source Geospatial Foundation, Beaverton, OR, USA) plugin LecoS—Landscape Ecology Statistics (version 3.0.0). The choice of landscape metrics was guided by the following conditions: (1) the metrics are suitable for describing the landscape diversity of the study area and for understanding relationships between land use and nitrogen concentrations in rivers; (2) the metrics represent different categories—area and edge metrics, shape metrics, dispersion and interspersion metrics, partition metrics, isolation metrics, and diversity metrics—and are commonly used in landscape ecology studies; (3) the metric set is not redundant, so, on the basis of Principal Component Analysis (PCA) (Figure 4), the most independent metrics were chosen. Before PCA, variables were tested to establish whether they meet the assumptions of factor analysis. Bartlett’s test of sphericity was significant (p < 0.001), which indicates that the variables were statistically significantly related to each other and could be reduced to a smaller number of components, whereas the Kaiser–Meyer–Olkin test result was 0.677, which suggested that the sampling was adequate on the average level. The first principal component (PC1), explaining 53.76% of the total variance, represented a gradient of landscape fragmentation. Positive loadings were associated with metrics describing large and cohesive patches (e.g., Largest Patch Index, Patch Cohesion Index, Effective Mesh Size, Mean Patch Area), whereas negative loadings were related to metrics indicating high landscape fragmentation, such as Patch Density, Edge Density, Splitting Index, and Landscape Division. The second principal component (PC2), explaining 15.31% of the variance, reflected differences in patch configuration and edge complexity. High negative loadings were associated with metrics describing a large number of patches and extensive edge structures (e.g., Number of Patches and Edge Length), whereas positive loadings corresponded to metrics related to patch adjacency and patch size. Ultimately, seven landscape metrics were selected: Shannon Diversity, Landscape Proportion, Edge Length, Mean Patch Area, Fractal Dimension Index, Patch Cohesion Index, and Landscape Division (

Table 1. Landscape metrics used in the study.

Metric’s Name	Description	Abbreviation	Formula		Unit	Value Range	Value Explanation
Landscape proportion	share of pixels of a given land cover class in all pixels of the catchment	PLAND	${\displaystyle \frac{{\sum }_{j=1}^n{a}_{ij}}{A}}$	where aij is the area of all patches of a given class, and A is the area of the entire landscape	-	0 < x < 1	equals 0 when there are no pixels of a given class, reaches the value 1 when the entire area is occupied by one patch of a given class
Edge length	length of edges of all patches of a given land cover class	TE	${\displaystyle \sum _{k=1}^m}{e}_{ik}$	where eik is the edge length	meters	x >= 0	increases when a given land cover class is more fragmented
Mean patch area	average patch area of a given land cover class	AREA_MN	mean(AREA[patchij])	where AREA[patchij] is the area of each patch	square meters	x > 0	increases as the size of patches of a given class increases
Fractal dimension index	mean fractal dimension index of the patches; the fractal dimension index is based on the perimeter and surface of the patches, and describes the complexity of the patches	FRAC_MN	mean(FRAC[patchij])	where FRAC[patchij] is the fractal dimension of each patch	-	0 <= x <= 2	equals 0 if all the patches are circles, reaches 2 when all the patches are maximally irregular
		FRAC	${\displaystyle \frac{2\times ln\times \left(0.25\times p_{ij}\right)}{\mathit{ln}{a}_{ij}}}$	where pij is the perimeter of the patch in meters, and aij is the patch area
Patch cohesion index	a measure of the coherence of a given land cover class	COHESION	$1-\left({\displaystyle \frac{{\sum }_{j=1}^n{p}_{ij}}{{\sum }_{j=1}^n{p}_{ij}\sqrt{a_{ij}}}}\right)\times {\left(1-{\displaystyle \frac{1}{\sqrt{z}}}\right)}^{-1}\times 100$	where pij is the perimeter, aij is the area [m2], and Z is the number of pixels	percent	0 < x < 100	equals 0 if the patches are strongly isolated, increases with increasing degree of aggregation
Landscape division	determines the probability that two randomly selected cells (pixels) will not be within the same land cover class	DIVISION	$1-{\displaystyle \sum _{j=1}^n}{\left({\displaystyle \frac{a_{ij}}{A}}\right)}^2$	where aij is the area [m2], and A is the area of the entire catchment [m2]	-	0 <= x <= 1	equals 0 if a given area is completely occupied by one patch of a given land cover class, reaches 1 when each patch is a separate pixel
Shannon Index	the metric takes into account the number of land cover classes and the share of each class	SHDI	$-{\displaystyle \sum _{i=1}^m}\left(P_i\times \mathit{ln}{P}_i\right)$	where Pi is the share of a given class	-	x >= 0	equals 0 if the entire area is completely occupied by one patch, increases without limit as the number of land cover classes increases, while the proportions between classes are equal

). Multicollinearity among chosen landscape metrics was evaluated using the variance inflation factor (VIF). All variables showed VIF values below the commonly accepted threshold of 5, indicating no strong multicollinearity among the selected landscape metrics.

3.3. Statistical Analysis

The influence of landscape composition and configuration metrics on nitrogen compound concentrations in selected streams was assessed using correlation analysis. Because the data were not normally distributed, as confirmed by the Shapiro–Wilk test (<0.05), the analysis used the nonparametric Spearman rank correlation coefficient. Given the relatively small number of catchments and the large number of landscape metrics, the statistical analysis was designed as an exploratory approach based on Spearman rank correlations rather than multivariate regression models, which could lead to unstable parameter estimates and model overfitting. To account for multiple testing, p-values obtained from the correlation analysis were adjusted using the Benjamini–Hochberg false discovery rate (FDR) procedure. All statistical analyses, including Principal Component Analysis (PCA) and graphs showing average and extreme values for nitrate, nitrite, and ammonium ion concentrations, were performed using Statistica 13.3 (TIBCO Software Inc., Palo Alto, CA, USA). Descriptive statistics, including mean, maximum, minimum, and standard deviation for each measurement point, were also calculated. Climatological characteristics for the study period, i.e., average monthly air temperature and monthly precipitation totals, were obtained from the Siedlce meteorological station (Institute of Meteorology and Water Management—National Research Institute). These values were compared with corresponding values for the multiyear period 1990–2020. Daily data were obtained from the NOAA Global Surface Summary of the Day service [47]. Moreover, hydrological background of the study period was provided by streamflow data (Q), acquired from the only available gauging station located in the downstream part of the Liwiec River (Łochów). A summary of the applied methods is presented in Figure 5.

4. Results

4.1. Hydrometeorological Background

Meteorological and hydrological characteristics of the study period, compared with long-term data, are presented below (

Table 2. Mean monthly temperature (T), precipitation (P), and streamflow (Q) values from September 2021 to April 2022 in comparison to analog data from the years 1990–2020.

Year	Month	T [°C]	T 1990–2020 [°C]	P [mm]	P 1990–2020 [mm]	Q [m3·s−1]	Q 1990–2020 [m3·s−1]
2021	IX	12.3	13.2	69.6	54.4	9.1	6.0
	X	8.7	8.1	6.9	38.5	7.5	6.9
	XI	4.8	3.4	29.2	33.0	6.6	9.8
	XII	−1.7	−0.8	16.7	34.0	8.5	10.5
2022	I	0.2	−2.1	49.5	29.7	13.6	11.6
	II	2.7	−1.0	39.4	27.6	25.4	14.8
	III	2.3	2.6	3.4	29.5	9.2	18.0
	IV	6.2	8.5	48.0	37.6	13.9	15.9
Study period	IX–IV	4.4	4.0	262.6	284.2	11.6	12.0

). Overall, the period from September 2021 to April 2022 can be considered as generally similar to long-term average values, as mean air temperature was only 0.4 °C higher than mean value from 1990 to 2020, whereas precipitation sum in the period in question was slightly lower, constituting nearly 93% of long-term sum. Runoff conditions directly reflected the prevailing meteorological conditions; mean streamflow values for the Liwiec River at Łochów gauging station reached 11.6 m³·s⁻¹, which was similar to the long-term mean value. However, from October to December and in March and April, mean streamflow was lower than the long-term mean, while in January and February peak flow values were observed, mainly as a result of mixed rain–snow precipitation (Table 2).

4.2. Spatial Differentiation of Nitrogen Compound Concentration

The mean NO₃⁻ concentration during the cold period did not exceed 1 mg·dm⁻³ at most sampling sites, with the lowest mean of 0.11 mg·dm⁻³ at site P21 and the highest of 28.61 mg·dm⁻³ at site P34 (Figure 6). In this last stream, the maximum concentration reached 54.76 mg·dm⁻³. The standard deviation of this ion’s concentration in the stream was 17.84 mg·dm⁻³. According to government standards, the mean value allows the classification of the waters of this river as below good quality status. In addition, in the context of the Nitrates Directive (91/676/EEC), concentrations above 50 mg·dm⁻³ indicate strong nitrate pressure. Thus these high values were verified during data screening and were not treated as statistical outliers. Most contaminated rivers were located in the central and eastern parts of the study area. For NO₂⁻, less variability was noted, but the spatial pattern of its concentration was closely related to NO₃⁻ (Figure 7). In most sites, the concentration of NO₂⁻ ions did not exceed 0.02 mg·dm⁻³, with the lowest mean of 0.004 mg·dm⁻³ (P6). Again, site P34 exhibited the highest mean concentration (0.130 mg·dm⁻³), maximum concentration (0.37 mg·dm⁻³), and the highest standard deviation (0.144 mg·dm⁻³). The spatial distribution of NH₄⁺ concentration differed from that of NO₃⁻ and NO₂⁻ (Figure 8). The mean concentration of this ion ranged from 0.18 to 0.68 mg·dm⁻³. Surprisingly, these edge values were computed for neighboring catchments, P12 and P10. In the latter, the maximum concentration was documented, reaching 3.088 mg·dm⁻³. The standard deviation of this ion’s concentration in the stream was 0.992 mg·dm⁻³. It is worth noting that drainage effluents may significantly contribute to contamination of streams in agricultural catchments, based on the concentration of nitrogen compounds documented in such effluents below site P26 reached in January (NO₃⁻ 62.82 mg·dm⁻³, NO₂⁻ 0.039 mg·dm⁻³, NH₄⁺ 0.318 mg·dm⁻³) and in April (NO₃⁻ 73.82 mg·dm⁻³, NO₂⁻ 0.059 mg·dm⁻³, NH₄⁺ 0.245 mg·dm⁻³). Spatial autocorrelation of nitrogen concentrations among the studied catchments was evaluated using Moran’s I statistic. Significant positive spatial autocorrelation was detected for NO₃⁻ (Moran’s I = 0.334, p < 0.001) and NO₂⁻ (Moran’s I = 0.165, p = 0.003), indicating spatial clustering of similar concentration values. In contrast, NH₄⁺ concentrations showed no significant spatial autocorrelation (Moran’s I = −0.080, p = 0.463).

4.3. Landscape Metrics’ Effects on Nitrogen Compound Concentration

According to Spearman’s rank correlation coefficient, forests, especially coniferous forests, may reduce the concentrations of NO₃⁻ and NO₂⁻ ions and minimize variation in these concentrations, as measured by standard deviation (

Table 3. Spearman’s rank correlation coefficients describing relationships between landscape metrics of (A) forests and Shannon Diversity Index (SHDI), (B) small woody features, meadows and wetlands, (C) arable land and developed areas, and the mean concentrations and standard deviations (SDs) of NO₃⁻, NO₂⁻, and NH₄⁺. Bold values indicate correlations that remained statistically significant after Benjamini–Hochberg FDR correction (p < 0.05).

(A)
LULC	Spatial Scale	Metric		Mean NO3−	Mean NO2−	Mean NH4+	SD NO3−	SD NO2−	SD NH4+
Landscape	catchment	SHDI		−0.43	−0.50	0.14	−0.51	−0.45	−0.03
	500 m buffer	SHDI		−0.30	−0.36	0.25	−0.39	−0.34	0.06
	250 m buffer	SHDI		−0.10	−0.14	0.19	−0.19	−0.14	0.07
	50 m buffer	SHDI		0.29	0.23	0.05	0.30	0.21	0.09
Forests	Catchment	LanPro	overall	−0.44	−0.47	0.18	−0.47	−0.39	0.02
			deciduous	−0.26	−0.20	0.33	−0.16	−0.04	0.25
			coniferous	−0.43	−0.53	0.07	−0.50	−0.52	−0.05
		EdgLen		−0.53	−0.58	0.15	−0.55	−0.61	−0.17
		MeaPatAre		−0.06	−0.12	0.01	−0.04	−0.09	0.00
		FraDimInd		−0.21	−0.25	0.12	−0.28	−0.24	−0.02
		PatCohInd		−0.27	−0.35	0.05	−0.29	−0.36	0.00
		LanDiv		0.16	0.23	0.08	0.19	0.20	0.05
	500 m buffer zone	LanPro	overall	−0.19	−0.25	0.23	−0.21	−0.19	0.02
			deciduous	−0.36	−0.32	0.37	−0.27	−0.16	0.32
			coniferous	−0.12	−0.23	−0.11	−0.17	−0.25	−0.17
		EdgLen		−0.39	−0.43	0.27	−0.46	−0.49	−0.06
		MeaPatAre		0.15	0.07	−0.16	0.18	0.08	−0.07
		FraDimInd		−0.21	−0.10	0.41	−0.29	−0.06	0.20
		PatCohInd		−0.07	0.02	0.27	−0.03	−0.04	0.38
		LanDiv		0.07	−0.08	−0.34	0.04	−0.06	−0.23
	250 m buffer zone	LanPro	overall	−0.01	−0.11	0.16	−0.11	−0.06	0.09
			deciduous	−0.23	−0.17	0.41	−0.23	−0.04	0.35
			coniferous	−0.06	−0.13	−0.07	−0.10	−0.16	−0.15
		EdgLen		−0.24	−0.30	−0.31	−0.35	−0.35	0.03
		MeaPatAre		0.20	0.16	−0.10	0.24	0.19	0.01
		FraDimInd		−0.47	−0.43	0.01	−0.52	−0.36	−0.18
		PatCohInd		0.04	−0.03	−0.07	0.01	−0.01	−0.03
		LanDiv		−0.18	−0.17	0.27	−0.20	−0.24	0.05
	50 m buffer zone	LanPro	overall	−0.07	−0.04	0.19	−0.08	0.04	0.02
			deciduous	−0.20	−0.16	0.28	−0.22	−0.05	0.16
			coniferous	0.08	0.02	−0.03	0.01	0.00	−0.05
		EdgLen		−0.09	−0.15	0.15	−0.19	−0.20	−0.05
		MeaPatAre		0.40	0.35	−0.03	0.45	0.38	0.10
		FraDimInd		0.15	0.09	−0.21	0.25	0.08	−0.16
		PatCohInd		0.23	0.18	0.00	0.24	0.20	0.02
		LanDiv		−0.35	−0.34	0.21	−0.39	−0.40	0.10
(B)
LULC	Spatial scale	Metric		mean NO3−	mean NO2−	mean NH4+	SD NO3−	SD NO2−	SD NH4+
Small Woody Features	Catchment	LanPro		−0.16	−0.03	0.12	−0.11	0.02	−0.11
		EdgLen		−0.18	−0.12	0.21	−0.17	−0.18	−0.17
		MeaPatAre		0.45	0.50	−0.11	0.49	0.50	0.00
		FraDimInd		0.54	0.58	−0.17	0.58	0.60	0.05
		PatCohInd		0.28	0.32	0.16	0.42	0.32	0.05
		LanDiv		−0.20	−0.16	0.22	−0.22	−0.22	−0.05
	500 m buffer zone	LanPro		−0.35	−0.26	0.13	−0.28	−0.23	−0.18
		EdgLen		−0.08	−0.05	0.28	−0.13	−0.16	−0.12
		MeaPatAre		0.50	0.50	−0.20	0.55	0.50	0.01
		FraDimInd		0.57	0.56	−0.31	0.59	0.56	−0.01
		PatCohInd		0.12	0.12	−0.02	0.16	0.12	−0.09
		LanDiv		−0.16	−0.14	0.30	−0.24	−0.24	−0.03
	250 m buffer zone	LanPro		−0.23	−0.19	0.08	−0.21	−0.19	−0.24
		EdgLen		−0.07	−0.09	0.25	−0.15	−0.18	−0.13
		MeaPatAre		0.42	0.39	−0.27	0.51	0.35	−0.05
		FraDimInd		0.46	0.42	−0.36	0.52	0.40	−0.09
		PatCohInd		0.08	0.12	−0.06	0.11	0.06	−0.15
		LanDiv		−0.06	−0.08	0.28	−0.19	−0.17	−0.05
	50 m buffer zone	LanPro		−0.14	−0.04	0.19	−0.14	−0.06	−0.13
		EdgLen		−0.04	−0.03	0.22	−0.12	−0.13	−0.14
		MeaPatAre		0.27	0.30	0.06	0.35	0.26	0.08
		FraDimInd		0.27	0.30	−0.07	0.33	0.29	0.06
		PatCohInd		−0.05	0.02	0.26	−0.04	−0.02	−0.04
		LanDiv		0.05	0.05	0.23	−0.03	−0.05	−0.11
Meadows	Catchment	LanPro		−0.51	−0.46	−0.04	−0.51	−0.41	−0.20
		EdgLen		−0.12	−0.17	0.15	−0.24	−0.32	−0.07
		MeaPatAre		−0.52	−0.50	−0.20	−0.49	−0.46	−0.33
		FraDimInd		0.47	0.49	0.28	0.38	0.41	0.20
		PatCohInd		−0.72	−0.69	−0.17	−0.69	−0.63	−0.28
		LanDiv		0.63	0.58	0.14	0.60	0.47	0.21
	500 m buffer zone	LanPro		−0.72	−0.63	0.10	−0.67	−0.56	−0.09
		EdgLen		−0.02	−0.02	0.32	−0.10	−0.16	0.17
		MeaPatAre		−0.62	−0.58	−0.20	−0.52	−0.53	−0.31
		FraDimInd		0.46	0.50	0.19	0.41	0.51	0.20
		PatCohInd		−0.65	0.60	−0.01	−0.62	−0.57	−0.14
		LanDiv		0.59	0.53	0.13	0.57	0.43	0.17
	250 m buffer zone	LanPro		−0.73	−0.67	0.07	−0.68	−0.62	−0.12
		EdgLen		0.08	0.03	0.24	−0.02	−0.12	0.13
		MeaPatAre		−0.63	−0.60	−0.23	−0.52	−0.53	−0.26
		FraDimInd		0.43	0.43	0.09	0.47	0.54	0.20
		PatCohInd		−0.48	−0.41	0.12	−0.44	−0.40	0.06
		LanDiv		0.37	0.24	−0.01	0.31	0.13	0.01
	50 m buffer zone	LanPro		−0.36	−0.32	−0.10	−0.32	−0.35	−0.09
		EdgLen		0.08	0.02	0.22	−0.01	−0.12	0.14
		MeaPatAre		−0.21	−0.17	−0.18	−0.18	−0.20	−0.11
		FraDimInd		0.08	0.10	0.00	0.07	0.12	−0.03
		PatCohInd		−0.07	−0.12	−0.05	−0.04	−0.16	0.08
		LanDiv		0.13	0.13	0.14	0.08	0.04	0.01
Wetlands	catchment	LanPro		−0.49	−0.38	0.14	−0.44	−0.21	−0.10
(C)
LULC	Spatial scale	Metric		mean NO3−	mean NO2−	mean NH4+	SD NO3−	SD NO2−	SD NH4+
Arable land	Catchment	LanPro		0.66	0.68	−0.17	0.62	0.61	0.04
		EdgLen		0.07	0.04	0.09	−0.02	−0.14	−0.03
		MeaPatAre		0.66	0.67	−0.22	0.62	0.58	−0.01
		FraDimInd		0.20	0.23	0.18	0.27	0.21	0.01
		PatCohInd		0.62	0.64	−0.15	0.58	0.51	−0.03
		LanDiv		−0.64	−0.69	0.16	−0.62	−0.61	0.00
	500 m buffer zone	LanPro		0.68	0.70	−0.11	0.62	0.58	0.11
		EdgLen		0.38	0.33	0.17	0.22	0.12	0.07
		MeaPatAre		0.68	0.69	−0.18	0.64	0.60	0.07
		FraDimInd		−0.50	−0.45	0.14	−0.48	−0.35	−0.05
		PatCohInd		0.70	0.70	−0.14	0.67	0.60	0.12
		LanDiv		−0.65	−0.68	0.16	−0.62	−0.59	−0.10
	250 m buffer zone	LanPro		0.68	0.69	−0.10	0.62	0.56	0.11
		EdgLen		0.41	0.35	0.14	0.28	0.16	0.08
		MeaPatAre		0.68	0.70	−0.15	0.66	0.62	0.12
		FraDimInd		−0.17	−0.11	0.16	−0.26	−0.19	−0.04
		PatCohInd		0.66	0.65	−0.22	0.70	0.59	0.11
		LanDiv		−0.60	−0.65	0.19	−0.63	−0.58	−0.08
	50 m buffer zone	LanPro		0.73	0.71	−0.02	0.70	0.60	0.25
		EdgLen		0.61	0.55	0.11	0.50	0.37	0.21
		MeaPatAre		0.70	0.67	−0.11	0.73	0.60	0.19
		FraDimInd		0.41	0.46	0.19	0.38	0.45	0.29
		PatCohInd		0.66	0.61	−0.10	0.68	0.46	0.18
		LanDiv		−0.65	−0.63	0.12	−0.68	−0.54	−0.17
Developed areas	Catchment	LanPro		0.28	0.33	0.25	0.36	0.32	0.22
		EdgLen		0.05	0.04	0.24	0.10	−0.06	0.13
		MeaPatAre		0.44	0.46	0.10	0.58	0.48	0.11
		FraDimInd		0.20	0.17	−0.08	0.24	0.16	−0.07
		PatCohInd		0.28	0.32	0.16	0.42	0.32	0.05
		LanDiv		−0.35	−0.40	−0.21	−0.46	−0.42	−0.21
	500 m buffer zone	LanPro		0.40	0.40	−0.16	0.47	0.45	0.18
		EdgLen		0.23	0.20	0.31	0.15	0.07	0.26
		MeaPatAre		0.18	0.20	−0.08	0.34	0.37	0.22
		FraDimInd		0.03	−0.01	0.05	0.07	−0.01	0.16
		PatCohInd		0.37	0.41	0.09	0.35	0.46	0.02
		LanDiv		0.25	0.25	−0.13	0.35	0.35	0.20
	250 m buffer zone	LanPro		0.34	0.32	0.16	0.36	0.27	0.18
		EdgLen		0.29	0.22	0.19	0.22	0.13	0.14
		MeaPatAre		0.34	0.41	0.27	0.44	0.41	0.30
		FraDimInd		−0.07	−0.06	0.20	−0.01	0.02	0.25
		PatCohInd		0.38	0.43	0.28	0.45	0.36	0.29
		LanDiv		−0.33	−0.36	−0.19	−0.40	−0.32	−0.21

A). This effect was more pronounced in catchments with longer forest edges (also observed in the 500 m buffer zone). In buffer zones, fewer statistically significant correlations were observed. The negative correlations of NO₃⁻ concentrations and standard deviation with the Fractal Dimension Index of forest patches in the 250 m buffer zone indicate that irregular, complex forest patches improve water quality. Notably, relatively high positive correlation coefficients (R) were found between the proportion of deciduous forests in the 250 m and 500 m buffer zones and mean NH₄⁺ concentrations; however, these relationships did not reach statistical significance. In the case of small woody features, two metrics seem to be important for the analyzed parameters: Mean Patch Area and Fractal Dimension Index (Table 3B). Such objects could be considered source landscapes, as an increase in area and shape complexity of small woody features contributed to a rise in NO₃⁻ and NO₂⁻ concentrations and in the variability of these concentrations. Such relationships were noted at catchment and wider buffer scales (250 and 500 m).

Greater potential for self-purification of the catchment from NO₃⁻ and NO₂⁻ ions was observed in catchments with a higher share of meadows, characterized by large Mean Patch Area, low spatial isolation (Patch Cohesion Index), and fragmented meadow patches (Landscape Division) (Table 3B). Higher mean concentrations and standard deviations of NO₃⁻ were observed in catchments with more irregularly shaped meadow patches, as measured by the Fractal Dimension Index. Meadows within the 50 m buffer zone did not affect the concentrations of NO₃⁻ and NO₂⁻ ions. The strength of correlations increased in wider buffer zones. The only metric for which no statistically significant correlations were observed at the 250 m and 500 m spatial scales was Edge Length. Wetlands and water areas also play a sink-landscape role—a negative correlation was found between the share of such areas in the entire catchment and the average concentrations of NO₃⁻ (Table 3B).

Arable land played a vital role in shaping water quality, often degrading it (Table 3C). Across spatial scales, the share of arable land and its average area were positively correlated with the average concentrations of NO₃⁻ and NO₂⁻, as well as with their variability. Furthermore, the spatial isolation (Patch Cohesion Index) and fragmentation (Landscape Division) of arable land were found to be important for water quality. Analyses in buffer zones reveal new relationships: for example, increasing the length of arable land edges negatively affects water quality (50 and 500 m buffer zones), as measured by nitrogen parameters, whereas shape complexity (Fractal Dimension Index) is positively correlated in the 500 m zone. Notably, the strongest correlations were observed for arable land compared to other LULC types. To assess the influence of high nitrate concentrations associated with drainage inputs, correlations were recalculated after excluding catchments with mean NO₃⁻ concentrations above 15 mg·dm⁻³ and 20 mg·dm⁻³ (likely covered by a subsurface drainage network). The results showed that correlation coefficients for arable land decreased from 0.39 to 0.31 (without four catchments with mean concentration above 20 mg·dm⁻³) and 0.26 (without six catchments with mean concentrations above 15 mg·dm⁻³).

The significance of developed areas was definitely lower than that of the LULC forms discussed above. Deterioration of water quality, measured by standard deviation of NO₃⁻ and NO₂⁻ ion concentrations, may be greater in catchments with larger patches (Table 3C). Only a single statistically significant relationship was observed for the buffer zones. An increase in the proportion of built-up areas within the 500 m buffer was associated with increased variability in NO₃⁻ concentrations. Although correlations with configuration metrics were relatively strong, they did not reach statistical significance.

The greater landscape diversity of the catchment (Shannon Diversity Index) favored a reduction in the seasonal variability of NO₃⁻ and NO₂⁻ ion concentrations in rivers (Table 3A). Landscape diversity may therefore contribute to the reduction in these ion concentrations.

4.4. Spatial Scale of Analysis and Strength of Correlation Relations

Numerous landscape metrics were effective predictors of nitrogen concentrations at larger spatial scales, such as the catchment scale or larger buffer zones (Table 3). This was documented for the diversity metric (stronger correlations in the catchment, meadows (at every scale except 50 m), small woody features (increased correlation strength in wider buffer zones), and forests). For the two forest metrics, Landscape Proportion and Edge Length, correlations decreased as the spatial range narrowed. Regardless of spatial scale, arable land metrics are crucial for understanding nitrogen compound concentrations in streams. The spatial scale at which developed areas affect water quality remains undetermined. Based on the mean absolute correlation coefficients (R) calculated for each spatial scale, the strength of the relationships increased with increasing spatial extent. The mean R values were 0.29 for the catchment and 500 m buffer, 0.26 for the 250 m buffer, and 0.21 for the 50 m buffer.

5. Discussion

5.1. Spatial Differentiation of Nitrogen Compound Concentration

The current study shows that the mean concentration of NO₃⁻ ions in most of the streams analyzed is consistent with that observed in other lowland catchments in Poland. For instance, similar concentrations were found in 14 tributaries of the Świder River, ranging from 0.35 to 4.72 mg·dm⁻³, with a peak in winter, though not exceeding 8 mg·dm⁻³ [19]. In this context, sites P24, P25, P26, P27, P28, P31, and P34 stand out negatively—these rivers are characterized by NO₃⁻ concentrations exceeding 10 and 20 mg·dm⁻³. The drainage network is undoubtedly a significant factor shaping concentrations in these streams; drainage effluents weaken the watershed’s resistance to contaminants mobilized and transported within it [48]. Drainage effluents were observed below the P26 site. In one drainage effluent located directly next to the measurement point, the NO₃⁻ concentration exceeded 60 mg·dm⁻³. It is likely that the catchments with the highest mean NO₃⁻ concentrations have subsurface drainage systems, similarly to the P26 catchment. After excluding catchments with mean concentrations above 15 and 20 mg·dm⁻³ from the correlation analysis, a noticeable decrease in the average R value was observed. These results suggest that elevated nitrate inputs may have contributed to the strength of relationships between LULC and nitrogen compound concentrations. However, it should be emphasized that despite the presence of subsurface drainage networks, no clear weakening of the correlations for cropland configuration metrics relative to their composition was observed. Mean NO₂⁻ levels in these streams were similar to those reported by [49]. NH₄⁺ concentrations generally oscillated around 0.2 mg·dm⁻³ and were similar, for example, to the results reported by [49]. It is worth noting that, for NH₄⁺ ions, no stream in the investigated area exceeded the class limits set by government standards. It must also be emphasized that during the study period, significantly higher concentrations were documented, exceeding 1 mg·dm⁻³ and 3 mg·dm⁻³; however, higher maximum values were recorded by [50] in the Orla River and its tributaries, exceeding even 10 mg·dm⁻³. Moran’s I analysis revealed significant positive spatial autocorrelation for NO₃⁻ and NO₂⁻ concentrations, indicating spatial clustering of similar values among the studied catchments. This suggests that their distribution is influenced by broader landscape-scale factors, such as land-use patterns and agricultural activities. In contrast, NH₄⁺ concentrations showed no significant spatial autocorrelation, implying that this form of nitrogen is primarily controlled by more localized processes within individual catchments and local point sources of nitrogen.

It must be emphasized that the spatial variability of nitrogen compounds is also dependent on season, which could be related to hydrometeorological conditions, controlling the mobilization and mobility of nitrogen [51,52]. Numerous studies have indicated that the influence of LULC on water quality is season-dependent [53,54], and in many cases, relationships between landscape structure and water quality are most pronounced during the wet season [6,24,55]. Hydrometeorological conditions during the study period were close to the long-term average. The period included snowfall events followed by snowmelt. Moisture conditions likely played an important role in the transport of nitrogen compounds, facilitating the detection of relationships between their concentrations and landscape composition and configuration. However, the investigated streams are not included in the national monitoring network, and no discharge measurements were conducted during the study period, which limits the ability to assess the impact of hydrological controls on nitrogen dynamics and to quantitatively assess changes in LULC–water quality relationships across different hydroclimatological conditions. In the absence of discharge data, it is not possible to quantify nutrient loads or to distinguish between dilution effects, flushing processes associated with runoff events, and sustained export of nitrogen from the catchment.

The spatial variability of nitrogen compound concentrations is also influenced by soil properties within the catchment. Podzols, due to their coarse texture and high permeability, favor nitrate leaching, whereas Luvisols, with higher clay content, enhance nitrogen retention [56]. Organic-rich muck soils may both retain and release nitrogen (especially organic forms), contributing to variability in nitrogen concentrations [57]. However, this factor, similarly to hydrological conditions, was not explicitly included in the analysis.

5.2. Landscape Effects on Nitrogen Compound Concentrations

5.2.1. Forest Patches

The study highlighted the positive impact of forests on water quality, as documented across various geographical regions [7,58,59,60]. Forests are geoecosystems characterized by multiple mechanisms for retaining nutrients, including nitrogen compounds—from bioaccumulation by trees and understory plants [61] to retention in soil organic matter [62,63]. Regarding forest composition in the catchment, quantified using the Landscape Proportion metric, coniferous forests appear to play a dominant role in river water purification during the cold period. This might be linked to soil chemical differences; coniferous forests tend to have lower NO_3- concentrations due to a higher C:N ratio, which reduces NO₃⁻ leaching potential compared to deciduous forests [64]. Finally, the cold period is the time of the lowest retention of inorganic nitrogen compounds in the catchment [65], as reflected in the relatively weak (compared with expectations) impact of forests on water quality in the study area.

Forest patch Edge Length influences mean and standard deviation of NO₃⁻ and NO₂⁻ concentrations, with longer edges enhancing a catchment’s self-purification capacity [66,67]. This effect results from increased surface friction and extensive root systems that intercept nutrient loads in flows [51]. Significant links between the Normalized Shape Index, Fractal Dimension Index (in this study), and nitrogen concentrations suggest that irregularly shaped forest areas near streams (250 m) have better self-purification potential, a finding confirmed by [68]. An analogy can be drawn to water bodies—lakes with complex shorelines are more resistant to water quality degradation [69], and forests with complex shapes have greater potential to reduce migratory nitrogen load.

5.2.2. Meadow Patches

For meadows, relationships were clearly unidirectional across all spatial scales, and more features of meadow patch configuration had the potential to shape NO₃⁻ and NO₂⁻ concentrations. Catchments with more meadows and a larger share of them are characterized by lower concentrations, especially NO₃⁻. This aligns with previous studies [19,70,71,72] and highlights the benefits of minimal management, like avoiding fertilization. Permanent plant cover and organic-rich soil help filter pollutants, with microbial processes facilitating nitrogen retention [73,74,75]. Retention of nitrogen compounds (and most pesticides) is mediated by microbial processes, including denitrification, degradation, and decomposition [69]. Significant relationships between meadow configuration and water quality show that larger, compact patches with low isolation degree and fragmentation enhance a catchment’s self-purification capacity. This is a classic set of relationships characterizing biogeochemical barriers, so-called “sink landscapes”. Ref. [70] used indices like ENN, IJI, and PLAND to demonstrate this influence, while [76] noted that fragmentation can worsen water quality; at the same time, they suggested that both meadows and forests can transform from “sink landscapes” into “source landscapes” if the considerations move from the time scales of a season or a year to the scale of single episodes [71].

5.2.3. Wetland Patches

Spearman’s rank correlation coefficient analysis confirms that wetlands and aquatic areas are denitrification “hot spots,” as negative correlations were observed between the share of wetlands in the entire catchment and the mean concentrations of NO₃⁻. Previous studies [77,78] highlighted the positive impact of wetlands on mineral nitrogen concentrations, noting that they can shift from being a “sink landscape” to a “source landscape” based on water quality parameters [79]. The presence of wetlands and water bodies is generally positively correlated with other biogeochemical barriers—a greater proportion of wetlands is found where meadows and forests occupy larger areas. The common elements of the ecosystems discussed above are (usually) soils rich in organic matter and the presence of permanent plant cover. Restoring the good chemical condition of water by vegetation is achieved through mechanical processes—adsorption, filtration, interception, and sedimentation—as well as biochemical processes: biofiltration and denitrification [75,80].

5.2.4. Arable Land Patches

The negative impact of arable land on water quality, as demonstrated in this study, was also observed in numerous investigations [81,82,83,84,85]. This influence may be related to the high susceptibility of chemical components in inorganic fertilizers to leaching and gaseous losses, even under appropriate fertilization practices. Furthermore, intensively used agroecosystems are typically characterized by low soil organic matter, which in turn limits the microbiological processes that contribute to the storage and reduction of nitrogen compounds in the soil [86]. Seasonally, arable land is particularly vulnerable during the cold months, when soil is exposed and plant cover is lacking, leading to chemical losses and erosion. Cover crops can mitigate nitrogen losses by absorbing nitrate ions during the off-season; the authors of [87] found that they can reduce NO₃⁻ losses by 40% while also enriching soil microbiology by supplying high-quality carbon, confirmed by [88].

Surface water enrichment with NO₃⁻ and NO₂⁻ ions is most pronounced in catchments with larger, less fragmented cultivated areas and low spatial isolation, aligning with previous studies [89,90,91]. Correlation analysis indicates that the shape of arable land patches affects water quality, with [91] finding a negative relationship between Fractal Dimension Index and nitrogen export, while [92] saw higher NO₃⁻ concentrations linked to greater Fractal Dimension Index. Ref. [93] reported no impact of land-cover fractal dimension on water quality in Iran. Research remains inconclusive on how arable land shape affects water quality. Extending field edges near streams may raise nitrate and nitrite concentrations, with stronger correlations closer to watercourses. Borders of arable land with forests may not effectively reduce pollution because of the narrowness of the biogeochemical barrier and the increased likelihood of contact with watercourses.

5.2.5. Developed Areas

The negative impact of developed areas has been demonstrated in both urban [94,95] and rural landscapes [96]. According to [96], even small settlements significantly affected the chemical status of water (in this case, small reservoirs), typically causing deterioration. In the current study, only two landscape metrics were significant, Landscape Proportion and Mean Patch Area, with effects similar to those of arable land. The similarity in the effects of development and arable land, as measured by landscape metrics, has also been documented in other studies [79,91,97]. The relationship between water quality degradation and increases in built-up areas [67,79], their aggregation [98,99], and reduced fragmentation [91] has been demonstrated repeatedly using various landscape metrics. In this study, relationships between nitrogen parameters and configuration metrics, such as aggregation and fragmentation, almost approached statistical significance, whereas the shape of developed patches showed no significant effect on water quality.

Built-up areas adversely affect the water cycle by increasing surface impermeability, thereby accelerating runoff [97]. Rural regions contribute potential nitrogen sources, such as septic tanks and silage piles [100,101], which can act as point sources of pollution. However, leaking septic tanks that are not connected to drainage systems should be treated as diffuse pollution sources. Some villages in the study area lack sewer systems, and poor domestic sewage management has been observed. This can lead to domestic sewage entering streams in catchments with high NH₄⁺ concentrations, such as P10 and P32, where point sources may minimize the impact of land use on NH₄⁺ levels.

5.2.6. Small Woody Features

Small woody features are typically considered to support the self-purification of catchments from non-point source pollutants. The role of both mid-field and waterside woodlots, which create ecotones, has been widely emphasized [102]. Small tree patches are also regarded as an important component of sustainable agricultural landscapes, helping to mitigate the negative effects of landscape homogenization [103]. However, in the present study, these features were associated with higher concentrations of NO₃⁻ and NO₂⁻ and were therefore classified as “source landscapes”, similarly to arable land and urban areas. In particular, larger and more complex shelterbelts were linked to increased nitrogen concentrations.

This pattern may be related to enhanced nitrogen cycling within these features. Organic matter inputs from woody vegetation can stimulate mineralization processes, increasing nitrogen availability and potentially promoting nitrate leaching [104]. The process is intensified if woodlots are combinations of species associated with arbuscular mycorrhiza (AM), such as Acer and Populus [104], as well as species that, with the involvement of symbiotic bacteria, fix nitrogen, such as Alnus glutinosa [105], which is common in the study area. Additionally, the effect of small woody features may depend on their age [106,107]. As indicated by [107], former land use plays a key role in nitrogen leaching from forest soils—significant nitrogen losses are observed in middle-aged stands (around 50 years for deciduous forests) established on former agricultural land. Ref. [108] notes that at the beginning of the 21st century, abandoned arable lands occupied almost one-tenth of arable land in Poland. This is a phenomenon commonly observed in the study area, particularly on soils with low agricultural suitability.

These findings suggest that the role of small woody features in nutrient dynamics is not unidirectional and may vary depending on local environmental conditions and landscape context. At the same time, direct comparison with previous studies is limited, as small woody features are rarely analyzed as a separate land-use class in studies linking landscape metrics with water quality.

5.2.7. Landscape Diversity

In the context of the study catchments’ resilience to increased nitrogen migration during the cold season, high landscape diversity proves beneficial. A wide variety of LULC forms helps limit the dominance of any single type, particularly those called “source landscapes”. Recent studies using the Shannon Diversity Index (SHDI) yielded mixed results on its impact on water quality. For instance, ref. [72] found that landscape diversity’s effects are seasonally dependent—promoting lower nitrate concentrations in the dry season but potentially increasing them in the wet season. Refs. [67,109] suggested that higher SHDI might intensify eutrophication, indicating more anthropogenic areas and nitrogen sources. Similarly, studies have linked higher SHDI to water-quality deterioration [13,68,110] and to landscape fragmentation driven by anthropogenic influences [111]. However, ref. [21] reported no correlation between SHDI and water quality.

5.3. The Influence of Landscape Metrics on Nitrogen Compounds and the Spatial Scale of Analysis

Identifying the spatial scale at which land use impacts nitrogen compound concentrations is challenging. Furthermore, it is difficult to identify such a scale for a single land-use type or for a specific metric or ion. Generally, the smallest spatial range (50 m buffer zone) is the least effective, while stronger correlations are seen at larger scales, particularly in catchments and 500 m buffer zones. Studies, such as [112,113,114], show increasing relationships between land use and water quality at larger spatial scales. However, some research [115,116] contradicts these findings. Recommendations in the literature suggest prioritizing specific spatial scales for water quality management, with [109] finding correlations ranking reach > buffer > catchment, and buffer sizes of 50 m > 100 m > 150 m, underscoring the significance of buffer zones in ecohydrology.

Studies indicate the importance of considering land use both adjacent to streams and across entire catchments [117]. Discrepancies in results may arise from the scale and accuracy of spatial data [5]. The impact of land cover on aquatic ecosystems is influenced by factors like forest ecotone quality, catchment size, and other pressures [118]. In mountainous regions, significant disturbances occur near streams, while areas farther away may remain more natural [119]. Ref. [59] noted that the limited impact of land use within a 50 m buffer might be due to its small contribution to the overall catchment. The effectiveness of spatial analysis can also depend on data characteristics, with a 15 m resolution potentially being insufficient [5]. Nevertheless, the selection of the best spatial extent is determined by (1) land cover type [72,120], (2) metric used for analysis [54,120], (3) water quality parameter [54], and (4) season [53,109]. Ultimately, understanding river ecosystems requires a multifaceted approach that accounts for spatial and temporal dimensions.

6. Conclusions

The present research identified several significant findings regarding the influence of landscape metrics on nitrogen compound concentrations in lowland catchments.

• Most streams exhibited low mean concentrations of NO₃⁻, NO₂⁻, and NH₄⁺ ions. High NO₃⁻ levels in some catchments may stem from drainage effluents, whereas elevated NH₄⁺ concentrations are linked to the application of organic fertilizers during the cold season and poor domestic sewage management. The latter may explain the small number of statistically significant relationships between NH₄⁺ concentrations and LULC.
• According to the obtained results, the group of “sink landscapes”, i.e., biogeochemical barriers, could include forests, meadows, wetlands, and water bodies. The purification effect is facilitated by a greater share of forests (mainly coniferous), a greater total length of edges (considered at the catchment and 500 m buffer zone scale), and complex forest patch shapes (in the 250 m buffer zone). The greater the share of meadows, the larger the patches, the more compact they are, and the less spatial isolation and fragmentation they exhibit, the lower the concentrations of nitrate and nitrite ions in the analyzed streams. Wetlands and water bodies in the catchment are denitrification “hot spots”, i.e., areas characterized by anaerobic conditions that favor the reduction of nitrates to ammonium ions and ultimately to gaseous forms.
• The function of “source landscapes” of mineral nitrogen compounds is performed by arable land, developed areas, and small woody features. The negative impact of arable land increases with its share in the catchment area and buffer zones, when patches are characterized by a low degree of fragmentation and spatial isolation. In turn, the negative impact of developed areas on water quality is intensified when built-up areas are larger and their share in the catchment increases. Contrary to expectations, small woody features may contribute to water quality degradation.
• The analyses carried out show that landscape diversity at the catchment scale, measured using the Shannon Diversity Index, favors the reduction and seasonal changes in nitrate and nitrite ion concentrations, which is not confirmed in most publications.
• It is impossible to clearly identify the area of the analysis where land use has the greatest impact on nitrogen compound concentration. To generalize, the smallest spatial range, i.e., the 50 m buffer zone, is the least useful, while the strongest correlations were obtained for larger spatial ranges, the catchment, and the 500 m buffer zone.
• The results should be interpreted with caution, as the absence of discharge data prevents the quantification of nutrient loads and limits the ability to distinguish between dilution, flushing, and export processes. Consequently, the identified relationships reflect concentration-based patterns rather than actual nitrogen fluxes, and the role of hydrological controls could not be directly assessed. Future studies should incorporate discharge measurements to better quantify nutrient transport processes.