Introduction

Rivers play a dual role in our connection to nature. Their linear structure and hierarchical geometry enable widespread access to freshwater for extensive industrial, municipal, and agricultural use, but they also serve as major pollution sinks, disposing of large volumes of industrial waste and municipal effluents (Mishra et al. 2021; Patel et al. 2021). Due to their continuous throughflow structure (Furtado 1979), rivers also serve as the end recipient of the materials coming from the surrounding and upper stream areas such as debris and agricultural runoffs, exposing them to various pollutants such as microplastics (MPs). The term MPs, introduced by Thompson et al. (2004), defines small water-insoluble synthetic solid particles or polymer matrices ranging from 1 µm to 5 mm (Campanale et al. 2020). These emerging environmental concerns are associated with diverse anthropogenic sources in river environments (Yano et al. 2021). While the biological effects of MPs are still a subject of debate (Zhang et al. 2019), their high production rate has attracted significant scholarly attention, such that over six billion tons of MPs are being produced globally, with nearly 80% left unrecycled (Law 2017).

The occurrence and abundance of riverine MPs can be a function of upstream land use (Mani et al. 2015), runoff (Werbowski et al. 2021), relative volume of effluent discharge (Ziajahromi et al. 2016), river hydrology and morphology (Yan et al. 2021) and the population density, urbanization and industrialization near the river (Aslam et al. 2022). However, a critical knowledge gap persists concerning the importance and relative contribution of environmental behavior, contamination and transfer mechanisms that affect the fate of MPs in these ecosystems (Skalska et al. 2020). To bridge this gap, an increasing number of studies aim to pinpoint sources and migration drivers of MPs. They revealed that decreasing river flow (He et al. 2020), decreasing river slope (Mani et al. 2015), and increased morphological complexities like mid-stream channel bars (Horton and Dixon 2018) significantly amplify MP deposition in river sediment. Physio-chemical attributes of river sediments, such as organic carbon content and texture, also significantly influence MP mobility (Liu et al. 2021). In terms of MP source identification, certain studies have delved into the composition, diversity, and physical traits of MPs. The concept of "MP community" was introduced to discern connections among different MP groups, geographic regions, or biological communities (Li et al. 2021; Yuan et al. 2022). However, despite these valuable findings, visually identifying physical characteristics for source determination may be misleading due to color fading and structural fragmentation of MPs (Song et al. 2020). To address this, a diversity of linear (Liu et al. 2022a, b; Mora-Teddy and Matthaei 2020) and nonlinear (La Daana et al. 2017; Liu et al. 2022a, b) models have been used successfully to determine the contribution of each possible source. The results of these studies are highly incongruent, highlighting the importance of local conditions and the scale of analysis in MP source identification.

In Iran, MP pollution remains largely unnoticed. Human activities in the central arid regions of the country heavily rely on river ecosystems both as a freshwater source (Madani and Mariño, 2009) and as a major sink for wastewater disposal (Karimian et al. 2020). The Zayandeh-rood River, situated in Central Iran, serves as a stark example of this situation. It flows through the third largest Iranian urbanized landscape, surrounded by numerous mid-sized industrial parks and two of Iran’s largest steel production centers. The river is predominantly covered by active farmlands (Asgarian et al. 2016) and is a tourist hub for visiting historical bridges and river-side public parks. Due to the intense frequency of these activities, the Zayandeh-rood River has already been acknowledged as a severely polluted aquatic ecosystem (Chamani 2020). Given the rapid population growth and the ensuing expansion of urban and industrial zones around the river, it is essential to monitor river contamination, particularly in relation to emerging pollutants like MPs.

This study is the first attempt to analyze the Zayandeh-rood River MP contamination focusing on the impact of nearby land use, sediment physicochemical properties, and river geometry using spatial modeling. Sediment samples were collected at the mouths of river sub-catchments and analyzed for MPs in terms of size, color, and polymer type. We assumed that the abundance of MPs at the sampling stations is related to the distribution and extent of the human settlements and agricultural areas whose spatial connection to the mouth of sub-catchments is not blocked. The extent of these activities was identified through the utilization of user-defined pseudo-catchments. Moreover, we measured some river geometry parameters and sediment physio-chemical characteristics to be used, alongside settlement and agricultural patterns, as predictors of the sedimentary MPs. The links were studied using the multiple linear regression (MLR) model as a linear and Artificial Neural Network (ANN) as a nonlinear model. Ultimately, a comparison was made regarding the prediction performance and importance of the selected variables.

Material and methods

Study area

This study was performed on the Zayandeh-rood River in the Gavkhouni Basin, Central Iran (Fig. 1). With a length of more than 420 km, the river originates from the western summits of a sub-range in the Central Zagros Range, known as Zard-Kooh, and flows into the Gavkhouni International wetland in the east. The altitudinal range of the river varies between 1450 and 3800 m above sea level. Topographically and according to the distribution of human activities, the river can be classified into upstream, midstream and downstream reaches. In the upstream zone, the river runs through mountains and valleys with sharp decreases in the channel slope. The river flow rate in this area is controlled by the Zayandeh-rood Dam and several ponds and bridges to approximately 38 m3/sec on average (Poustizadeh and Najafi 2011). Due to topographical limitations, the human presence and activities (especially agriculture) are significantly low in this zone. The long-term mean annual precipitation and air temperature of the upstream area are 316.9 mm and 12.9 °C, respectively. After the covering of approximately 140 km, the river enters the midstream zone due to the significant decrease in topographical complexities as well as in mean altitude to below 2000 m above sea level. In this zone, the river runs through hills, and basins and then into plains and opening flat terrains, which allowed extensive development of agricultural activities. The downstream part of the river basin starts from Isfahan City, the third most populous city in Iran (2.24 million residents reported in 2021 (Madiseh et al. 2021)) and ends at the Gavkhouni International Wetland. The midstream and downstream reaches are characterized by an arid and semi-arid climate, according to De Marton’s classification system, and receive less than 100 mm annual precipitation on average (Asgarian et al. 2016).

Fig. 1
figure 1

Location of the Gavkhouni basin in Central Iran, Zayandeh-rood River and sediment sampling stations along different river reaches

Full size image

To identify suitable locations for sampling sedimentary microplastics (MPs) in the study area, we utilized the 30-m spacing ASTER global digital elevation model (DEM) to generate sub-catchments within the basin. These sub-catchments were required to meet two specific criteria: first, the streams within them should flow directly into the river, and second, they should have a minimum spatial coverage of 1000 hectares. These selected sub-catchments were considered potential sampling stations for studying sedimentary MPs. However, some sub-catchments are highly modified by various structures and were excluded from the research. The most important excluding structures that severely block suspended materials were regulatory dams, mainstem reservoirs and bridges as well as multiple flood defense structures constructed on high-flow, steep-sloped streams. The selected sub-catchments were then significantly trimmed to only include parts of the river and stream network that do not carry suspended materials from the previous sub-catchment. Specifically, the process of extracting pseudo-catchments is as follows.

  1. (A)

    Classification of the river basin into 1000 ha < catchments using both automatic and visual methods,

  2. (B)

    Exclusion of the catchment in which the river water (and its associated materials such as MPs) is blocked severely and frequently,

  3. (C)

    Visual inspection of the stream network in each selected catchment to ensure the continuity of the water flow from each stream to the river.

  4. (D)

    The selected catchments were trimmed to exclude those parts of the stream network whose stream-river connection was completely blocked and were free of any human presence/ activity including built-up, road and agricultural lands.

In total, a total of 21 pseudo-catchments (Table 1 and Fig. 2) including 9 in the upstream, 6 in the midstream and 6 in the downstream zones were selected as sampling stations.

Table 1 Topographical characteristics of the sub-catchments and their location from the river origin
Full size table
Fig. 2
figure 2

Spatial distribution of the pseudo-catchments along the Zayandeh-rood River

Full size image

Sample collection, processing and analysis

The sediment sampling was carried out during the summer of 2021, 3 months after the last rainfall. Sediments were collected at the mouth of the pseudo-catchments from straight courses of the river to ensure that the samples were not gathered near debris-flow-prone areas and from deposition traps along the river bank. At each sampling station, three kg of surface sediments in an area of 30 × 30 cm to a depth of 10 cm were picked in three replications using a stainless steel shovel. We followed the guidelines of the National Oceanic and Atmospheric Administration (NOAA) with some minor modifications for the sample pre-treatment which includes wet peroxide oxidation and density separation (Masura et al. 2015). To do so, the sediment samples were oven-dried (60 °C, 48 h) and sieved. The digestion process was carried out using regular addition of hydrogen peroxide (H2O2—30% w/v) for 24 to 240 h depending on the type and quantity of organic matter. According to NOAA guidelines, hydrogen peroxide should be added regularly to the solution until the natural organic matter is destroyed. Hence, the more organic matter, the higher the amount of consumption and the duration of digestion. The density of the solution was then increased using the cost-effective and nontoxic method of salinity-based density separation to enhance the floatation of low-density MPs. To enhance the aqueous solution’s density, we conducted two separate density separation processes, each lasting 24 h (Besley et al. 2017). After the first 24 h, the upper aqueous phase was removed for the next treatment and the remaining precipitated solids were subjected to the second-density separation using the saturated NaCl solution. The upper liquid layer obtained from the second round of density separation was combined with the layer obtained from the first density separation. The combined solution was filtered using Whatman filter paper grade 42, and then, the filtrate was air-dried in the shade for further analysis (Eerkes-Medrano et al. 2015). Then, all the particles remaining on the filters were inspected visually under a standard light microscope (Nikon CFI LU Plan Fluor) with magnifications of × 40, × 100 and × 400. The identified MPs were classified into 6 types (fiber, fragment, film, foam, pellet and line) and 10 colors. To identify the polymer type, MPs were analyzed using the Raman spectroscopy (785 nm wavelength) to collect spectra from 50 to 3500 cm−1 with a resolution of approximately 4 cm−1. Using a matching threshold of 0.7, the results were compared with the reference spectra available in the Openspecy database (https://openanalysis.org/openspecy/).

In order to avoid sample contamination, the containers were pre-cleaned with distilled water and placed in aluminum foils. Cotton laboratory clothes and nitrile gloves were used during the sample preparation and MP identification. The laboratory environment was kept in isolation, and the experimental supplies were carefully rinsed in running water and dried. To assess the recovery rate, we selected a subset of sediment samples from which MPs had been previously extracted and washed it multiple times to ensure it was thoroughly cleansed of MPs. The samples were mixed with equal amounts (25 pieces) of polystyrene (PS), polyamide (PA), polypropylene (PP), and polyethylene (PE) and treated similarly. Finally, we counted the number of each type of MP and divided it by 25 to determine the recovery rate, which was found to be 100%, 98%, 96%, and 96% for PS, PA, PP, and PE, respectively.

Data preparation

Drawing from the literature review and the availability and importance of MP sources, the explanatory variables were collected into three groups: sediment physio-chemical, riverine and land-use/land-cover (LULC) variables. The sediment physio-chemical variables were pH, electrical conductivity (salinity-dS m−1), total organic carbon (TOC %), organic matter (OM %) (Walkley and Black 1934) and the sediment particle size distribution (Bouyoucos 1962). The riverine characteristics included four geometric variables: river length (RL in km), mean river slope (RS in %), river curvature (RC) index, and mean river width (RW in m). To produce the RC index, we divided the river length by its smoothed length measured using the Polynomial Approximation with Exponential Kernel (PAEK) smoothing algorithm (Shen et al. 2020) with a smoothing tolerance of 5000 km. In order to measure the river width, we utilized the mean length of the river cross section measured using the high-resolution Google Earth at 50 m intervals. The river flow rate was omitted from the analysis because it is highly controlled approximately between 35 and 38 m3/sec depending on the river section and distribution of LULCs. The LULC variables were related to six far-river characteristics: the area of settlements (Set in ha) and agricultural lands (Agr in km2), the mean distance of the settlements to the river mouth (MRD-Set in km), mean distance (km) and mean slope (%) of the streams connecting agricultural areas to the river mouth (MSD-Agr and MSS, respectively) and the estimated mean annual number of visitors including tourists and local people (VP in million people) which came to visit historical bridges and use public/private gardens along the river. Agricultural and settlements were extracted from the sentinel 2 images in spring 2021 using the maximum likelihood classifier (Ali et al. 2018) with an acceptable Kappa coefficient of 0.89. The river stream network data were obtained from the regional water company of Isfahan at a scale of 1:5000. The stream slope was estimated using a high-resolution DEM layer (10 m resolution) available for the region.

Statistical analysis

MLR and ANN models were used to explain the variability of MP abundance in relation to the selected explanatory variables. The MLR model assesses the simultaneous linear relationship between a single continuous dependent variable and a vector of independent predictor variables. The four main assumptions of the MLR model are data normality, multicollinearity, autocorrelation, and heteroscedasticity (Alita et al. 2021). To use the MLR model, all data were tested for normal distribution using the Kolmogorov–Smirnov test. The non-normal data were normalized using the Box–Cox transformation technique. Collinearity among the variables was assessed using the variance inflation factor (VIF) diagnosis. We used an entry criterion of p ≤ 0.05, an exclusion criterion of p ≥ 0.05 and the step-by-step procedure to execute the regression model as recommended in previous studies (Tsai et al. 2012). The performance of the resulting MLR models was evaluated based on the F-test of overall significance against a null model.

The ANN model was also used due to the probable nonlinear associations between the dependent and independent variables. In many studies, deep learning models such as ANN offered superior prediction capabilities in the simulation and forecasting tasks (Aggarwal 2018). This model seeks the nonlinearity complex relationship between the underlying variables and consists of many processing (input, hidden and output) layers and neurons whose functionality is determined by the weight factor. Iterative parameter adjustments and overtraining are, however, the key constraints of the ANN model which can lead to large prediction errors (Mijwel 2018). The back-propagation (BP)-ANN algorithm was employed to build an MP prediction model. It consists of the feed-forward and BP calculation phases (epochs) using which the input data enter into the input layer and reaches the output layer. In this research, the Batch training type was used to update the synaptic weights. We used the initial λ of 5e-7, initial σ of 5e-5, interval center of 0, and interval offset of ± 0.5 in this section. The minimum relative change in training error and error ratio was set to 1.0e4 and 1.0e3. The memory storage was set to a maximum value of 1000 cases. The best-performing hidden layer was selected between 1 and 2 with minimum and maximum units of 1–50. Finally, the Akaike information criterion (AIC) estimator (Cavanaugh and Neath 2019) was used to select the best-performing model. We also used the coefficient of determination (R2) between the observed and predicted cases to determine the performance of the MLR and ANN models in relation to the included and ranked variables. The sensitivity analysis was also performed to determine the relative importance of each variable in both models.

Results

Abundance, type, and size of MPs

MPs were observed in all stations even in the S21 which is not surrounded by any human activity. Upstream stations had relatively low average MP abundances ranging from 26 (S1, S6 and S7) to 142 (S9) items/kg d.w (mean abundance of 56 items/kg d.w). The average MP abundance in the downstream zone was as high as 1701 items/kg d.w, and the highest average MP abundance (5040 items/kg d.w) was recorded in S17 in this zone. The minimum and maximum average MP abundances in the mid-stream zone were 92 and 680 items/kg d.w, respectively. Overall, the average abundance of MPs followed the order: upstream (57.2 items/kg d.w) < midstream (269.2 items/kg d.w) < downstream (1701 items/kg d.w) and averagely was 588 items/kg d.w across all stations (Table 2). The correlation between the average MP abundance and distance from the river origin was also found to be marginally significant (r = 0.46, p = 0.036, n = 21).

Table 2 Average abundance, size and percentage of different MP types identified in the study stations
Full size table

The polymer type of MPs was identified using a Raman device, and their spectra were compared with the reference spectra provided in the Openspecy database, showing that the sediment MPs were of three different polymers including polypropylene (57.1%), polyethylene (28.6%) and polyamide (14.3%). By size, MPs were categorized into 4 groups: < 500 μm, 500–1000 μm, 1000–3000 μm, and 3000–4750 μm, with corresponding abundance percentages of 64.5%, 18.3%, 13%, and 4.2%, which demonstrates a reduction in the frequency of MPs with increasing size. Fiber was the dominant form of MPs in the upstream and midstream zones, accounting for more than 47 and 58% of the MPs. In the downstream zone, fragment (40.8%), followed by line (30.5%) and fiber (29.4%) were the most frequently observed MPs. Considering all stations, the percentage of the MP types had the order: fiber (45.3%) > fragment (28.8%) > line (9.9%) > pellet (7.8%) > film (6.8%) > foam (2.1%) (Table 2). By color (Fig. 3), fiber, fragment and line types exhibited a variety of colors such as black, blue and brown. Nearly 65% of the films were transparent, and foams were only colored white (90.9%) and blue (9.1%).

Fig. 3
figure 3

Percentage color distribution of MPs by type in the study stations

Full size image

The size of sediment particles affects the distribution of organic and inorganic pollutants (Zhao et al. 2019). Therefore, sediment texture (clay (less than 4 µm), silt (4–63 µm) and sand (63–6000 µm)) and the concentration of TOC, OM%, pH and salinity were measured to investigate their relationship with the abundance of MPs. The correlation results (at p value < 0.05; two-tailed) showed that clay (r = 0.15), silt (r = 0.39), TOC (r = 0.41), %OM (r = 0.41) and salinity (r = 0.59) were positively correlated with the MP abundance, while sand (r = − 0.27) and pH (r = − 0.48) had a negative correlation with the MP abundance. The average values of the three explanatory variables by river zones are given in Table 3 (the station-level data are provided in Appendix 2). A significant increase was observed in the mean salinity, TOC and OM with distance traveled downstream. The mean salinity increased from 0.32 dS m−1 in the upstream to 28.95 dS m−1 in the downstream zone. The mean OM and TOC increased by 44 and 244%, respectively. The sediments of the river upstream zone were highly alkaline with pH values of greater than 8 and then, decreased to 7.95 and 7.63 in the river midstream and downstream zones, respectively. Sand was the dominant substrate of the river bed in the upstream and midstream zones with respective percentages of 53.62 and 76.44. In the downstream zone, however, sediments had a clay-loam texture according to the USDA soil classification system. A negative significant correlation was found between pH and salinity (r (21) = − 0.70, p < 0.01) and pH and TOC (r (21) = − 0.50, p < 0.05). Moreover, significant positive relationships were observed between TOC and OM (r (21) = 0.75, p < 0.01).

Table 3 Mean values of the explanatory variables measured at three categories of physio-chemical, riverine and LULC variables in upstream, midstream and downstream stations; RL: river length, RS: mean river slope, RC: river curvature index, RW: mean river width, Set: the area of settlements Agr: area of agricultural lands, MRD-Set: mean distance of settlements to the river mouth, MSD-Agr: mean distance of the streams connecting agricultural areas to the river mouth, MSS: mean slope of the streams connecting agricultural areas to the river mouth and VP: the estimated mean annual number of visitors
Full size table

The highest mean river slope (9.33%) and lowest river curvature index (0.71) were observed in the upstream zone due to its mountainous nature and sharp altitudinal gradient. As expected, the river length of large pseudo-catchments in the downstream zone was 30.3 and 64.9% longer than that of the midstream and upstream areas. Due to the wide breadth of the river reservoir and dams in the upstream zone, the mean river width was found to be more than 145 m. The majority of the basin agricultural areas are distributed in the midstream zone while the upstream and downstream areas are, respectively, limited by topographical and soil constraints. Moreover, the water quality of the river downstream zone is significantly deteriorated by various anthropogenic activities like the discharge of industrial sewage and domestic wastewater. From the upstream down, settlements increased both by mean area and distance from the river course due mostly to the dense far-river distribution of industries and industrial parks in the downstream zone. Historical bridges and near-river private and public tree-covered parks are the most important tourist attractions of the region which are mostly distributed at the beginning of the downstream zone near and within the Isfahan City, especially in S16.

The best-performing MLR and ANN models with the lowest AIC are represented in Tables 4 and 5, respectively. The MLR model [F(2,18) = 36.68, p = 0.00] with an R2 of 0.803 was constructed using the Set (t = 8.002, p = 0.00) and VP (t = 3.059, p = 0.01) variables. Both variables had a positive effect on the dependent variable such that increasing area of settlements and the number of visitors led to increasing abundance of MPs (Table 4). Moreover, none of the other riverine and sedimentary variables, as well as the distribution of agricultural lands, were linearly significant (p < 0.05) in explaining the variability of the MP abundance. VIF values indicated a lack of colinearity between the selected variables. The simple incremental sensitivity analysis also showed that Set is the most important variable in the MLR model.

Table 4 Characteristics of the best-fit MLR model developed to predict the abundance of MPs
Full size table
Table 5 Characteristics of the best-fit ANN model developed to predict the abundance of MPs
Full size table

In the ANN model, all variables were included in developing the neural network. The best BP-ANN model was constructed with 1 hidden layer containing 6 units (the weights attached to each neuron are given in Appendix 3). The cross-entropy errors during the training and testing phases were 1.93 and 0.02, respectively. The training phase was stopped after reaching one consecutive step with no decrease in error. The incorrect prediction of the training and testing phases was also 0.24 and 0.28, respectively. According to the ANN sensitivity analysis (Table 5), the most important variable in building the ANN network was set (0.136), followed by TOC (0.082), river length (0.082) and MRD-Set (0.076). The least important variable was pH (0.013). The BP-ANN outperformed the MLR model with an R2 value of 0.995 (Fig. 4).

Fig. 4
figure 4

Observed versus predicted scatterplot of MPs in a logarithmic scale

Full size image

Discussion

Abundance and forms of MPs

Rivers are capable of accumulating significant quantities of MPs. In this study, we found that the mean abundance of sedimentary MPs trapped at the mouth of the modified sub-catchments was 588 items/kg dry weight. Additionally, the abundance of MPs exceeded 1000 items/kg dry weight in the midstream and the first two stations of the downstream zone, where the density of the human population and their industrial and agricultural activities is notably high. This indicates a substantial contribution of these activities as non-point sources of MPs in the region. Similar findings have been reported in other studies, such as the Shanghai River in China (802 items/kg d.w by Peng et al. (2018)) and the Brisbane River in Australia (up to 520 items/kg d.w by He et al. (2020)), where rivers in highly developed landscapes also accumulate significant quantities of MPs.

Despite receiving highly treated wastewater from the Isfahan WWTP, which is situated far from the next closest station (S17), and experiencing multiple within-river retentions, the number of MPs remained considerably high at this station where it receives urban, agricultural, and industrial runoff. This emphasizes that the pollution of river MP originates not only from non-point sources but also from point sources. This conclusion aligns with previous findings (He et al. 2023, Park et al. 2023), suggesting that both point and non-point sources contribute significantly to riverine MP pollution. Stations S1, S4, S6, S7, and S8 with fewer than 2 MPs per 50 g dry sediment had the lowest contamination in this study which could be attributed to the low population density and the presence of agricultural and orchard lands around them.

PP, PA and PE polymers we found in sediments are often used in fabrics, agricultural mulch, packaging bags, safety belts, parachutes, thread ropes and fishing nets (Claessens et al. 2011; Deopura et al. 2008). In the study river, however, they might be related to fishing nets and ropes, along with the discharge of domestic (laundry) and industrial wastewater. By form, the amount of fiber was highest in the sediment samples and the dominant type in 57% of the stations from different river zones. The inflow of domestic sewage into the Zayandeh-Rood River is likely one of the main causes for the high abundance of fibrous microplastics in this river, as studies indicate that laundry machine wastewater contains a significant amount of fibers (Hernandez et al. 2017).

Throwing garbage such as water bottles, plastics, and bags by tourists into the water and on the banks of the Zayandehrud River can be considered as the source of secondary MPs. Fragments emerged as the second most abundant category among MPs, suggesting that the majority of MP sources in the region may encompass clothing, carrying a substantial load of fibers, as well as the weathering and breakdown of larger plastics that generate secondary fragments (Bošković et al. 2022; Marrone et al. 2021). Additionally, road traffic near the river can potentially facilitate the transfer of small rubber and plastic particles resulting from tire abrasion runoff into the river (Yeganeh Far et al. 2020).

Considering land use variables and environmental factors, the morphology of MPs and their shape, color, and size can vary extensively (Eerkes-Medrano et al. 2015; Horton and Dixon 2018). For example, foam, being the least prevalent type of MPs, was only detected in specific stations such as S2 and S10, predominantly in white color (over 90%).

While pellets and foam were absent in the study stations, they were found abundantly in the Red River of Canada (Anderson et al. 2017), St. Lawrence River in North America (Castaneda et al. 2014), and the Antoã River in Portugal (Rodrigues et al. 2018). Similarly, however, fibers were detected in Pearl River sediments (Lin et al. 2018), Poyang Lake in China (Liu et al. 2019), and Venice lagoon sediments in Italy (Vianello et al. 2013). Along the shores of Lake Ontario (Ballent et al. 2016), the Manojo River (Wu et al. 2020), and the Dongting Lake River in China (Jiang et al. 2018), fine particulates were also predominant in riverbed sediments.

Various colors of MPs were observed which can easily be mistaken as food by aquatic organisms (Arias-Andres et al. 2018; Avio et al. 2015; Cole and Galloway 2015). Moreover, it is expected that color could indicate their primary sources (Sheikhi and Mirzaei 2023). For instance, colored fragments might originate from packaging materials (Andrady 2017), and colored beads could come from cosmetic and personal care products (Fendall and Sewell 2009). The size of all MPs ranged from 50 to 4750 μm. The results indicated that the category with sizes smaller than 500 μm accounted for the highest abundance at 64.5%. Such small MPs pose serious threats to aquatic organisms and ecosystems (He et al. 2020) since they can be ingested by a wide range of organisms (Moore et al. 2011). Furthermore, MPs serve as carriers for pollutants like organic compounds and heavy metals (Munier and Bendell 2018) due to their smaller size and higher specific surface area, enhancing adsorption capacity (Lahijanzadeh et al. 2020). Given the prevalence of MPs smaller than 500 μm in the study area, there is potential for pollutant absorption and increased bioavailability. Furthermore, MPs with lower density and smaller size are more prone to displacement by wind and water, potentially aiding biological invasion by facilitating the entry of invasive species into aquatic ecosystems (Zetteler et al. 2013).

Irrespective of the form, color and size of MPs, we made no conclusion about the mean size of MPs as they showed no reasonable trend and distribution pattern. Furthermore, we must interpret the significant correlation between the abundance of MPs and the distance from the river origin (Appendix 1) with caution because the observed increase in MP abundance toward downstream areas could potentially be a result of the escalating human activities in those regions, rather than a direct consequence of the gradual accumulation of MPs along the course of the river. The MP abundance correlated positively with clay and silt particles which is in agreement with previous studies such as Park et al. (2023) who showed that clay particles are strongly correlated with the concentration of MPs in sediments. It seems that MPs are more physically capable of binding to these fine particles through mechanisms such as electrostatic interactions. Moreover, clay and silt particles in sediment provide large surface areas and are known for their ability to adsorb substances.

Relation type and importance of predicting variables

The modeling approaches employed in this research offered insights into the influence of general variables on the abundance of MPs in Zayandeh-rood sediments. To provide more meaningful variables, we classified the influencing population into two groups: visitors who lack spatial relevance to the pseudo-catchments, and local residents whose daily activities generate plastic waste, which may ultimately find its way into the river environment. This perspective yielded more practical results with significant managerial implications for the region, differing from previous studies such as Kataoka et al. (2019) who employed the basin’s entire population as a non-point pollution source. The area of settlements, followed by the number of visitors, emerged as the two essential variables in the MLR model. Given that the settlement area serves as a proxy for the local population, it is evident that both residents and tourists collectively determine the abundance of sedimentary MPs, following a linear relationship. Using the nonlinear BP-ANN model, the area of nearby settlements was also identified as the most sensitive variable with regard to changes in MP abundance. These results underscore that the utilization of plastic products, coupled with their introduction into the environment through direct disposal or the discharge of domestic drainage, constitutes fundamental sources of MP pollution in the Zayandeh-rood River. Similar findings have been reported by previous studies, such as Kataoka et al. (2019), Nel et al. (2017) and Song et al. (2020), demonstrating that urbanization and population density play pivotal roles in MP pollution within densely populated basins.

By employing the BP-ANN model, we gained a more comprehensive understanding of the nonlinear impact of the selected variables. TOC and river length emerged as the second and third most influential predictors of MP abundance. The role of fundamental sediment chemical components, such as TOC, has been extensively emphasized in prior research. Consistent with our findings, studies by Maes et al. (2017), Liu et al. (2021) and Falahudin et al. (2020) found a significant association between the sediment TOC and MP abundance. Consequently, regions with high proportions of TOC are anticipated to exhibit high MP abundances, owing to similarities in their sources, such as WWTPs, making them valuable hotspots for further MP investigations. Despite the significance of human population density and certain sediment attributes, the contribution of river geometric characteristics and agricultural activities was relatively modest, which contrasts with the findings of some other authors (Mani et al. 2015; Piehl et al. 2018). One possible explanation for this discrepancy could stem from the significant differences in study design strategies and the selection of sampling sites (Skalska et al. 2020). Additionally, there exist notable distinctions in certain river flow characteristics, such as temporal variation and river geometry (e.g., bedform), among different rivers. For example, in the Yangtze River, river flow characteristics were identified as crucial determinants of the mobility and sedimentary deposition of MPs, a river with a considerably larger flow rate compared to the river studied in this research (Xiong et al. 2018). In summary, the abundance of MPs in the sediments of the Zayandeh-rood River was found to be closely tied to human (local and visiting) population and their municipal activities, particularly the release of wastewater into the river. The contribution of human distribution is further accentuated due to the cumulative increase in MP abundance and the escalating population density as we move toward the end of the river.

Study limitations

One of the most crucial limitations inherent in such studies revolves around the general categorization of input variables, particularly those related to non-point sources. While the significance of factors like population density and urbanization has been demonstrated in numerous studies, their contributions stem from a diverse array of sources and pathways. In the case of the Zayandeh-rood River, a more detailed categorization of population density, differentiating between occasional visitors and local residents, coupled with an examination of their daily habits, can substantially mitigate this limitation. The influence of river flow regulation, achieved through regulatory dams, mainstem reservoirs, and bridges, is another significant factor that impacts the results. Although we have partially addressed this limitation by defining and utilizing pseudo-catchments, further studies are necessary to precisely quantify the influence of these structures on MP transfer. Additionally, all variables were investigated at the same spatial scale, without considering the temporal variability of flow conditions.

Conclusion

This study evaluated the linear and nonlinear relationships between sedimentary, riverine and LULC characteristics and the abundance of MPs in Zayandeh-rood River, Central Iran. Our results showed that: (1) fiber and fragments emerged as the most abundant forms of MPs, distributed widely across all stations. This emphasizes the predominant contribution of non-point sources to river MP pollution, (2) the high MP abundance observed near the Isfahan WWTP underscores the significance of wastewater as a significant non-point source of MPs in river ecosystems, as demonstrated by multiple studies, (2) human density-related variables, including settlement area and the number of spatially independent visitors, were identified as the most crucial predictors of MP abundance. These factors contribute to sediment MP pollution through the disposal, weathering, and breakdown of larger plastics such as water bottles, plastics, and bags, (3) regions rich in TOC are likely to exhibit relatively high MP abundances, making them valuable hotspots for further MP investigations. Although the concept of pseudo-catchments facilitated the determination of cumulative impacts from surrounding non-point sources, the substantial intra-annual variations experienced by the river underscore the importance of conducting a comprehensive temporal investigation to enhance our understanding of MP transfer within the river. In conclusion, our study suggests that improving the culture of plastic use and disposal by local people as visitors, along with effective treatment of municipal and industrial wastewater released into the river, are essential steps to mitigate river MP pollution.