Exploring the Effects of Landscape Structure at Multiple Scales on River Water Characteristics in the Khorramabad and Chalus River basins

Abstract


Introduction
Water pollution sources are divided into two categories: point source and non-point source. Pollution from point sources can be practically controlled by various methods, but due to the uncertainty of the type, amount, location, and how pollutants enter surface and groundwater streams, it is di cult to control nonpoint sources of pollution. Water pollution through non-point sources is the result of the use of a wide range of human activities. In many countries, all agricultural and livestock activities are considered as non-point pollutants and are recognized as an important factor in determining water quality and play an important role in freshwater nutrition (Mehaffey et al., 2005) because different land covers affect the type of nutrients that enter the river through runoff (Shen et al., 2015).
It is well known that rivers receive pollution from their surrounding landscape and the amount and intensity of this pollution are affected by the composition and con guration of the landscape features around the river or within a watershed (Clément et (Li et al., 2008;Liu et al., 2012). Therefore, factors such as the quality and ow of rivers, not only re ect the health of the river itself but also provides information about the watershed and the landscape through which they pass (Xie and Ng, 2013). Different studies con rm that the structural features of a watershed such as a slope and the number of land cover types in a watershed such as urban, rangeland, forest, and agriculture have signi cant effects on river water quality. for instance, the presence of forest cover improves river water quality in a watershed (Clément et al., 2017;Tong and Chen, 2002). Lee et al. (2009) also reported that the quality of river water decreases as the size of forest patches decreases. Therefore, altering landscape structure by clearing the forest for human activities such as agriculture has been recognized as important factors that affect the water quality of the river (Clément et al., 2017) Natural and semi-natural habitat loss and fragmentation are two important processes that drastically change the structural pattern of land covers (Xie and Ng, 2013). Habitat loss reduces the amount of the original habitat and fragmentation increases the isolation of remnants patches (Parker and Mac Nally, 2002). Several studies have found that changes in the landscape structure patterns resulted from habitat loss and fragmentation have signi cant effects on the quantity and quality of river water in a watershed (Amiri and Nakane, 2009 Effects of landscape patterns on river water quality occur at different spatial scales such as watershed and riparian zones (Zhou et al., 2012). Several studies have shown that land-use patterns close to the river have greater effects on the variability of river quality parameters than those farther away (Dodds and Oakes, 2008;JOHNSON et al., 1997;Shen et al., 2014). For example, at the riparian scale, the amount and spatial arrangement of different vegetation covers such as forest cover can have a signi cant impact on nutrient concentration, physical properties, and energy balance in a river (Casatti et al., 2012;Jackson et al., 2015). On the other hand, some studies have stated that measuring land use patterns at the watershed scale will yield more reliable results (Sliva and Williams, 2001;Zhou et al., 2012). From an ecological point of view, estimating the relationship between landscape patterns and water quality parameters at multiple scales provides valuable information about these scales and determines the spatial scale that has the most impact on rivers (Xie et al., 2018;Zhou et al., 2012). Due to human disturbances and the elements that make up a watershed, it is con rmed that there is no permanent and special scale for these effects (Alberti et al., 2007;Li et al., 2013;Margriter et al., 2014;Shen et al., 2015). Therefore, in each watershed, estimating the effects of the surrounding landscape and human disturbances on the river water quality at different scales is necessary to make suitable management decisions (Tudesque et al., 2014).
Landscape ecology is the study of the composition and con guration of ecosystems at the landscape level (Mitchell et al., 2013). Landscape con guration indicates the spatial distribution of patches within a landscape, but composition refers to non-spatial aspects of landscape features such as the area of  2007)) found that the landscape composition and con guration account for 21 to 86% of changes in river water quality. Some studies have claimed that the effects of landscape composition on river water quality are more determinative than con guration, for example (Alberti et 2018)) found that the landscape composition compared to its con guration has a greater impact on predicting changes in water quality. Clément et al. (2017) suggested that the in uence of landscape con guration appears at a certain threshold and in areas with intensive crop farming (>50%), and an improvement of water quality will come with more forests and wetlands, and the con guration of those patches is less relevant.
In the present study, the landscape and class metrics effects on the water quality of highly degraded watersheds have been explored to make such kinds of study results more applicable to urban landscape planning and water management at an operational level. This study aims to compare the effects of land use patterns in two watersheds that have a large difference in the presence of human disturbances. Both watersheds have agricultural, rangeland, forest, and urban covers, but the type of forest cover in these basins is different and also the number of human activities in these basins is different. We estimated the relationship between land use patterns and river water quality at both riparian and watershed scales using multivariate statistical analyses.
Our main questions were: (1) What are the landscape con guration and composition relative in uences on river water quality?
(2) Which landscape metrics and which land use cover is more related to water quality?
(3) How Riparian corridor and watershed-scale affect water quality characteristics?

Study area and data
The Chalus River basin, located in the west part of Mazandaran province, has an area of 110398 ha, and the length of the river in our study area is 100 km (Fig. 1). The River's branches drainage 19 catchments, mostly covered with forest, but different covers such as agriculture, grassland, bare soils, and urban areas are present. The average area of the catchments is 5810 ha (Table 1) and the average slope of the basin is 48%. The average percentage of urban areas is about 0.58%. we selected the study area consciously for especially examining the effects of landscape pattern on water quality in an area with the lowest human activities. The average percentage of forest cover is about 37% in the Chalus River basin, ranging from 1.3-77.3%.
The Khorramabad River watershed is located in the middle part of Lorestan province. It has an area of 244072 hectares and its longest river length (Khorramabad River) is 100 km (Fig. 2). The Khorramabad River's branches drainage 21 catchments, covered with different covers like agriculture, forest, grassland, bare soils, and urban. The Kakasharaf River ows to the southern and southeastern parts of the study area. These two rivers joined in the watershed outlet and ow into the Kashkan River. In general, the Khorramabad watershed most of the time has been accumulated from the alluvial deposits of permanent rivers and other seasonal rivers. The average slope and elevation of the watershed are 20 % and 1603 m, respectively ( Table 2).
Using a digital elevation model (DEM) at 30m×30m resolution, watershed boundaries, digital river network, and sampling points were delineated using ArcGIS Arc Hydro extension. For this study, 42 water quality sampling points were selected based on each output of the basins. The sampling was conducted one week after raining in the early days of the rainy season. Seven representative variables were selected for testing. These variables were electrical conductivity (EC), nitrite (NO2), nitrate (NO3), dissolved oxygen (DO), phosphate (PO4), phosphor (P), and temperature (TEMP). The sampling was conducted within 6 h of collection following standard methods (Shen et al., 2014).  Quanti cation of Landscape Pattern Sentinel 1A images with the least cloud cover, taken in august 2018, were used for the classi cation of land use and land cover (LULC) data using a maximum likelihood method. All the images were projected into the World Geodetic System (WGS) 1984 UTM 39N coordinates. Five land use and land cover types were classi ed: urban, forest, agriculture, grassland, and bare soil. Five spatial scales within the regional watershed, consisting of 100 m, 300 m, 500 m, and 1000 m buffer zones, and basins were created by buffering along the streams using ArcGIS 10

Statistical analysis
First, all the water quality variables and landscape metrics that did not follow a normal distribution were logarithmically transformed. The Kolmogorov-Smirnov (K-S) test was used to detect the normality of distribution of the variables for water quality, landscape composition, and habitat fragmentation (Olea and Pawlowsky-Glahn, 2009). Stepwise multiple regression analyses were performed to determine the direction and magnitude of the interaction between the landscape metrics and the individual water quality indicators. The stepwise regression has long been used to select descriptive variables for relating water quality to landscape descriptors ( In our study, we use the stepwise method, which starts at the forward selection, but at each stage, the possibility of deleting a predictor, as backward elimination, is considered (Chong and Jun, 2005). The probability value to enter variables into the stepwise models was set at 0.05 and the probability to remove was set at 0.1.
Redundancy analysis was used for the gradient analysis of the water quality/landscape relationship at multiple spatial scales. Before the RDA, the water quality data, including EC, NO2, NO3, DO, PO4, P, and temperature were imported into Canoco 4.5 software to test if the DCA gradient shaft length was less than 3. The result showed that the DCA gradient shaft length was less than 3. Therefore, six water quality variables, as well as all selected landscape metrics including landscape composition and con guration, were considered. RDA is a constrained linear ordination technique that describes the variation between two multivariate data sets (Ou and Wang, 2011). The forward selection method was used to identify the signi cant variables at multiple scales based on the Monte Carlo Permutation method (n = 499) in the process of the RDA analysis, avoiding the in uence of redundant variables. Based on the RDA analysis, the in uence of the landscape metrics for the catchment and riparian zones on all water quality variables was examined.

Results
The relationship between landscape structure and water quality variables The results of stepwise multiple regression models for Chalus watershed showed ( Almost all classes had a considerable effect on the NO3 variable. This variable was negatively correlated with the number of patches of agriculture class and had a positive correlation to the AREA_MN of forest cover. PO4 was negatively correlated to the PAFRAC of agriculture class, while P had a negative correlation to patch density of forest cover and a positive correlation to IJI of urban class. On the whole, both landscape composition and con guration metrics had a signi cant in uence on water quality in the Chalus watershed, but landscape con guration metrics were more related to the water quality variables (Table 3).  According to regression results presented in Table 4, landscape composition metrics had no signi cant in uence on water quality, and landscape con guration metrics were more associated with the water quality parameters. Almost all of the landscape metrics that appeared in the stepwise regression model had a positive effect on the water quality parameters. The relationship between the landscape structure and the water quality at multiple scales The results of the redundancy analysis of the Chalus watershed (Table 5) showed that the proportions used to quantify the landscape explained more than 75% of the variation in water quality. The rst two RDA ordination axes also explained 70 to 79% of the total correlation re ected by all axes. At the 100 m buffer zone scale, the landscape pattern metrics could account for 80% of the water quality variation.
Thus, the 100 m riparian buffer was identi ed as the main riparian zone that had the greatest impact on the water quality. When the scale increased to a 300 m buffer, the explanatory power decreased to 78%. However, there is a considerable decrease of explanatory power at 500 m buffer (71%), but the value increased again at 1000 m and basin scales. Table 6 shows the dominant landscape variable groups with the maximum explanatory power at each spatial scale selected based on the test of signi cance and importance and inspection of variance in ation factors (VIF < 10). Consistent with the results of the multiple stepwise regression, the metrics re ecting the landscape con guration had a greater impact on water quality. AI of the forest (0.33%), ED of urban areas (0.14%), and IJI of the agricultural lands (0.19%) were identi ed as the most important variables at the basin scale. AI of forest (0.35%), ED of urban lands (0.18%), and SPLIT of Agriculture class (0.17%) appeared as dominant metrics at the 100 m buffer zone. Similar to basin scales, the aggregation (AI) metric of forest cover also was the most signi cant metric for explaining water quality at a 100 m buffer zone. ED of urban class and SPLIT of agriculture class was dominant metrics at 300 m buffer zone with explaining powers of 0.28%, 0.29%, respectively. At the 500 m and 1000 m buffer zone scales, urban metrics were the most effective metrics in determining water quality.
The results of redundancy analysis for the Khorramabad watershed showed that the metrics that were used to quantify the landscape, explained more than 80% of the variation in water quality. The rst two RDA ordination axes also explained more than 80% of the total re ected correlation by all axes. The landscape pattern metrics could account for 89% of the water quality variation at the 100 m buffer zone scale. Thus, the 100 m riparian buffer was the main riparian zone that had the greatest effect on the water quality. When the scale increased to a 300 m buffer, the explanatory power decreased to the rate of 83%. However, as the buffer scale increased from 500 m to1000 m, the explanatory power decreased from 84-79%. Table 7 presents the dominant landscape variable groups with the maximum explanatory power at each spatial scale. Consistent with the results of the stepwise multiple linear regression, the metrics re ecting the landscape con guration had a greater effect on water quality. SPLIT and MESH of the urban land and for explaining water quality at a 300 m buffer zone similar to previous scales. These metrics were MESH and CLUMPY of urban areas and NP of agricultural lands with explaining powers of 0.32%, 0.24%, and 0.16%, respectively. Landscape con guration metrics of urban lands were the factors affecting water quality at all riparian buffer scales, as MESH, CLUMPY, and PLAND of this class were identi ed to be more related to water quality variance at the 500 m and 1000 m buffer zone scales.
Figures 3 and 4 display ordination diagrams derived from the RDA using the water quality variables and selected landscape metrics representing landscape composition and con guration. The plots can be interpreted quantitatively using the landscape factor arrow length to indicate how much is the water quality variance was explained by that factor. The water quality variable arrows pointing in the same direction as the landscape factor arrows indicate a positive correlation (the smaller the angle between the arrows, the stronger the relationship).

Discussion
Landscape con guration and water quality at landscape and class levels Our study indicated that the metrics re ecting landscape spatial con guration had a stronger relationship to the water quality variables than composition metrics for both basins under study (Tables 2 and 4).
Stepwise regression and redundancy analyses showed that the water quality variables were more explained by the con guration metrics (e.g., PD, IJI, MESH, and SPLIT) of land use classes in comparison with landscape composition metrics, as re ected by the high frequency of occurrences in the best models (Tables 2 and 4). Note that composition metrics also were important in the analysis, as they were more effective for some of the water variables than con guration metrics. Our results are inconsistent with studies that claimed landscape con guration indices are more important than composition indices in predicting stream water quality in watersheds ( a higher IJI indicates a fragmented landscape, and subsequently presents degraded water quality. In the present study, DO was negatively correlated to IJI, and P was positively correlated with SPLIT of agriculture areas in the Chalus basin (Table 2), and in the Khorranabad basin DO was positively correlated to patch density of agriculture (Table 4 grasslands are considered to act as a buffer, play a retention function, decrease the amount of surface runoff, absorbing part of runoff, and thus reducing non-point source pollution (Ouyang et al., 2010). In our study, MESH of grassland was negatively correlated with DO in the Chalus basin, implying that more fragmented grasslands lead to decreased water quality. In the Khorramabad basin, LPI and PLAND of grassland were negatively correlated with NO3, contrary to the study done by Lee et al. (2009) that showed the LPI of grassland was positively correlated with the water quality parameters. Lee et al. (2009) also showed that the LPI of grasslands was positively correlated with the water quality variable.
Streams and rivers that pass through urbanized landscapes often have higher levels of water pollutants and more nutrient loads, thus reduced biodiversity (Meyer et al., 2005). our study showed that distant urban patches in the Chalus basin lead to increase electrical conductivity (EC) because EC was mostly affected by urban metrics such as ED, ENN_MN, and PAFRAC and had a positive relationship with ENN_MN of urban patches. In the Khorramabad basin, NO3 was positively correlated with PD of urban lands, indicating that dispersed urban areas lead to increasing degraded water quality. This result is also true for TDS that had a positive correlation with SPLIT of urban areas ( Table 4).
Effects of forest cover on water quality in the Chalus basin showed that NO3 had a positive relationship with the mean area of forest patches and P had a negative relationship with the density of forest patches.
In the Khorramabad basin, only NO2 showed a negative relationship with the SPLIT of the forest, implying that fragmented forest cover leads to degraded water quality in the basin. Our results are consistent with the studies that claim in forested catchments, water quality will start to deteriorate as the landscape becomes more fragmented, and resulting in higher landscape diversity and lower landscape contagion Clément et al. (2017). In general, forest cover in the present study did not affect the water quality variable considerably in both basins and our results only showed that the more fragmented forest, the more deteriorated water quality. However, inconsistent with our results, Lee et al. (2009) reported a negative relationship between the LPI of forest and water quality parameters. meaning that the widespread distribution of forests could improve the water quality of the watershed to a certain extent.
On the whole, PD, IJI, and SPLIT of different land uses were the most important metrics which had signi cant relationships with the water quality indicators (Tables 2 and 4). Similar to Shen et al. (2015) our results also showed that metrics at the class level are a better predictor than metrics at the landscape level for the water quality variables.
Impacts of landscape metrics on water quality at multiple scales The in uence of the landscape on the water quality is scale-dependent, and our study also con rms this fact, as reported by others (Sliva and Williams, 2001;Zhou et al., 2012). The impact of landscape patterns on river water quality is different at riparian buffer zones and basin scales, thus a controversial issue. Some studies have shown that near covers to the river have greater effects on river quality than distant covers or the entire catchment. For example, JOHNSON et al. (1997) found that TP and TSS were much better explained by the land use within the stream riparian buffer than the entire catchment. While, Sliva and Williams (2001) showed that at the catchment, landscape metrics had a slightly greater in uence on the water quality than the 100 m buffer in their study area and Alberti et al. (2007) found that it was not possible to determine which scale was more correlated to river quality changes.
It is acceptable that the landscape patterns of riparian buffer zones have important effects on aquatic ecosystems (Alberti et al., 2007;Shen et al., 2015). Consistent with our results, Shen et al. (2015) showed that the water quality changes were better correlated with the buffer metrics than the basin scale. In both basins, the selected landscape metrics at the 100 m buffer zone explained more than 80% of water quality variation, which has proven the importance of the landscape pattern for the 100 m riparian buffer zone (Tables 5 and 7 (2016) also showed that the largest patch index of urban (LPI) and aggregation index of forest (AI) were the most important predictors for NH3-N, NO3-N, and TP. Clément et al. (2017) proposed that a forest edge is bene cial for water quality with a density of higher than 36 m/ha. They also compared two catchments with similar characteristics, and different forest edge densities, and water quality. Their results indicated that in the catchment with the higher edge density, water quality is better, implying complex woodlands following river corridors and gullies, improves the ltration capacity.
Although in Chalus basin human activities were insigni cant, they had considerable effects on Chalus river quality, and metrics of urban and agriculture were recognized as a dominant variable at all scales, implying that a large amount of forest cover cannot impede the effects of human activities in a basin because this kind of activities have been created near rivers and have more impacts on the river than distant natural covers.

Conclusion
In summary, our results showed that: 1. Water quality variables were better explained with the buffer riparian zones than basin scale, and 100 m buffer is the most effective buffer zone for affecting the water quality.
2. Both landscape composition and con guration had signi cant impacts on water quality, but in our study, landscape con guration indices were more effective than composition metrics in explaining river water quality.
3. Among the dominant landscape metrics representing both the landscape composition and the spatial con guration, the AI of the forest was recognized as the most signi cant variable in uencing the water quality at the basins. 4. Landscape metrics at the class level can predict river water quality in the study of watersheds more effectively in comparison with the indices at the landscape level. On behalf of all authors, the corresponding author states that there is no con ict of interest.
Availability of data and material Data are available on request from the authors only based on logical requests.

Code availability
Code available on request from the authors only based on logical requests.

Funding
There are no nancial con icts of interest to disclose.

Figure 1
Location of Chalus river basin in Iran.

Figure 2
Location of Khorramabad river basin in Iran.  Redundancy analysis biplots showing the correlation between the water chemistry variables and landscape variables in the Khorramabad watershed.