The Warszawicki Channel is located close to the boundaries of the largest peat bog in Mazovia – Bagno Całowanie (Całowanie Peatland, covering 35 000ha), located in the Mazowiecki Landscape Park, about 40 km south-east of Warsaw, Poland, in the Vistula River valley (Fig. ). In the past, large parts of peatland were reclaimed for agricultural purposes, and the Warszawicki Channel served as a water source for irrigation. The total catchment area is around 240 km2 and, in a hydrographic sense, it links the Wilga River system with the Vistula River to divert surface water reserves to the area of the Całowanie Peatland. The channel is also connected with several smaller watercourses to provide sufficient flood protection to the areas located between the Wilga and Vistula rivers. Indeed, those channels were designed to retain part of the floodwaters of the Vistula River to mitigate excess water hazards.

The experiments were conducted in a reach, about 500 m long, of the Warszawicki Channel. This channel was selected due to the varying cross-sectional vegetation patterns resulting from natural vegetation growth (see Fig. ). Typically, mechanical cutting and removal of bank and bottom vegetation are planned twice a year, with the local legislation requiring maintenance at least once per year. This fact might create variable conditions for the water flow or the solute transport, mostly due to different stages of plant development in the channel bed. In 2019, the channel vegetation was cleared only once at the beginning of October, using an excavator with a weed cutting bucket, and the channel bed was not dredged. These conditions were favourable for the present study, given that at the end of the summer, the channel vegetation was very dense, as shown in Fig. a.

We selected four sub-reaches (A between cross-sections P1 and P2, B between P2 and P3, C between P3 and P4, and D between P4 and P5) with varying vegetation coverage (see Fig.  for details). Their lengths differed as we attempted to delineate the sub-reaches so that a large range in the vegetation coverage could be obtained. We conducted investigations during fully vegetated conditions (Exp. 1, no maintenance, September 2019) and after complete cutting and removal of the channel and bank vegetation (Exp. 2, fully cut, October 2019). Figure  presents the channel view towards the downstream sub-reaches before (Fig. a) and after (Fig. b) the vegetation cutting.

Table  summarises the main properties of the four selected channel sub-reaches and the entire 467 m long reach, located between the P1 and P5 cross-sections (sub-reach ABCD), during both experiments. The flow discharge (Q) was roughly estimated based on flow velocity measurements performed before each tracer experiment in a few selected, well-accessible, cross-sections with reduced vegetation coverage. Flow velocity distributions were measured using an electromagnetic flow meter (Nautilus C 2000 OTT) to derive the flow discharge by integrating the point velocity measurements across the wetted cross-sectional area. There is increased uncertainty in the calculated flow rates due to lower water levels and the presence of vegetation in the channel affecting cross-sectional velocity measurements. No extreme events (e.g. heavy rainfalls and droughts) were recorded in the study period between the two experiments. However, during the field campaigns, controlling all environmental factors influencing the hydraulic conditions was not feasible, and possibly increasing uncertainties in the final estimations of the flow discharges should be considered. The channel slope was around 0.1 ‰. The slope was measured by multiple geodetic levelling of the water surface over 60–100 m. As is visible in Table , both experiments were performed with a comparable reach-averaged water depth. However, the water depth was slightly lower, particularly in the two most downstream sub-reaches in Exp. 2.

Unlike most available studies, the research proposed in the paper is not focused on individual plants or patches but on vegetation coverage at the reach scale in complex natural conditions.

Species that may be present in the Warszawicki Channel include the following: Phalaris arundinacea L., Phragmites australis, Glyceria maxima and Sparganium emerson, forming mixed, mostly emergent vegetation (Fig. ). Contrary to laboratory investigations where researchers can deal with controlled and well-described hydraulic and vegetation properties, in the field we are dealing with mixed vegetation (e.g. submerged and emerged, different species and densities). During our experiments, we did not collect detailed physical information on the particular plants growing in the channel. Instead, we aimed to investigate the influence of vegetation at the reach scale, and it is known that at the reach scale, the coverage is the factor mainly influencing the flow hydraulics . Thus, we hypothesised that the solute transport would also depend on coverage. Practical applications with “disorderly” natural vegetation motivated our work to investigate physically sound but easily measurable parameters like vegetation coverage.

The surficial vegetation coverage (V) of the studied reach was determined through UAV imagery using a DJI Phantom 4 drone equipped with an RGB camera (Fig. ). To ensure comparability of measurements during the experiments, the drone flights were performed in automatic mode with the same flight parameters and camera settings and similar weather conditions. In addition, the Pix4D application was utilised for programming and automatic implementation of the fully photogrammetric UAV missions. The flight took place at a speed of 4 m s-1 at a height of 35 m above ground with 70 % image overlap. The resolution of the obtained data was 1.5 cm. Three flight missions were carried out in a time interval of 40 min for the fully vegetated scenario and 10  min for the fully cut scenario, conditioned by the different velocities of the plume movement.

Based on the collected images, orthophoto maps were generated using the Agisoft PhotoScan software, applying the Structure-from-Motion (SfM) method . Those maps were analysed in the open source Quantum Geographic Information System (QGIS) (https://www.qgis.org, last access: 28 February 2023) to determine the surficial vegetation coverage in the channel in the case of fully vegetated conditions (light blue line in Fig. c, Exp. 1) as well as the precise location of the river bank line for the bare conditions (black line in Fig. c, Exp. 2). Similar water levels in the river channel during the two experiments (see Table  in Sect. 3) allowed the assumption that the bank line determined at the cut conditions was representative of the fully vegetated conditions.

Using map algebra, widely used in GIS studies , the percentages of vegetation coverage for the entire examined reach (between P1–P5 cross-sections) and for each individual sub-reach (see Fig. ) were calculated according to Eq. (): 1V=WC-WVWC, where V is the surficial vegetation coverage, WC is the surficial water area in the channel under bare conditions (polygon marked with a black line in Fig. c), and WV is the surficial water area in fully vegetated conditions (polygon marked with a blue line in Fig. c). In the case of Exp. 2 (fully cut conditions), V was assumed to be 0 %.

For the tracer experiments, we used Rhodamine WT which is a soluble, non-toxic fluorescent dye, conservative at the considered timescales . It is detectable in very low concentrations, and it has been used over many years in laboratory and field studies to estimate travel times, mean flow velocities, or dispersion coefficients in streams and rivers e.g.. In both Exp. 1 and Exp. 2, the Rhodamine WT was released instantaneously at P0, a non-vegetated area located 39 m upstream of P1 (see Fig. ). The dye concentration was measured at the cross-sections P1, P2, P3, P4, and P5 downstream of the injection point over a total distance of about 500 m. Distances between the sampling locations were 128, 34, 81, and 224 m for sub-reach A, B, C, and D, respectively (see Fig. ).

The water samples were manually collected from the central part of each cross-section using an aluminium sampling rod with the personnel standing outside the water without disturbing the flow. The samples were stored in black bottles to prevent rhodamine loss due to exposure to light. They were analysed in the laboratory under controlled temperature conditions with a 10-AU-005-CE fluorometer from Turner Designs. Furthermore, for Exp. 2, a handheld fluorometer (Turner Designs, AquaFluor handheld fluorometer) was used to check the concentration values in real time since the passage of the plume was very fast. This information was used to adjust the sampling frequency to ensure that the leading edge of the dye cloud and the concentration peak were captured correctly. We changed the sampling frequency based on expected/checked concentration values to optimise the usage of bottles for samples and the laboratory's sample measuring process.

During Exp. 1, we started sampling with 10 min intervals (except for the cross-section P1, when we started immediately with 5 min intervals). Then, the sampling frequency was increased to 5 min close to the expected peak (2–3 min for P1) and returned to 10 min (after the peak was captured). Finally, we measured from 10 to 60 min for the tailing edge as the concentration changed more and more slowly. In the case of non-vegetated conditions (Exp. 2), since the passage of the dye plume was quick, we sampled faster. Sampling frequency varied from 1 to 10 min. We sampled more frequently, close to the expected peak of concentration (from 30 s in P1 to 1–3 min in other cross-sections), and less frequently for the tailing edge from 5 to 10 min. The sampling period was adjusted to the actual cross-section concentration changing (using a handheld fluorometer on site).

Before starting both experiments, a few water samples were taken to establish the background concentration. Additional samples were taken during the experiments upstream of P0 to check that the background concentration was not changing. Background water samples have also been used for calibration and appropriate timing of the end of the sampling. For accuracy checking, Exp. 2 was repeated later on the same day under the same hydrological conditions after reaching the background values of the concentration (Exp. 2'). For Exp. 2', water samples were collected at selected cross-sections (P1, P2 and P4).

We derived parameters describing flow and mixing based on the obtained parameters during tracer test concentration data. They were derived separately for each sub-reach and the entire reach (P1–P5) based on the concentration curves at the corresponding upstream and downstream cross-sections (see Sect. 2.3).

The peak travel time (tp) and peak concentration (Cmax) were derived directly from the concentration distributions for each measured cross-section. Different methods may be applied to obtain the flow velocities and dispersion coefficients. The most commonly used are the method of moments and the routing procedure, described and compared, e.g. by . The second one required fixed time intervals in the concentration distribution. Taking into account our sampling procedure, we applied the method of moments , well-established and used for many years in tracer studies for details see e.g.. This method was initially proposed by , and nowadays, it is widely used in field and laboratory tracer studies, mainly for determining the longitudinal dispersion coefficient (DL).

The longitudinal dispersion coefficient value was determined based on the changes in the centroid and variance of the recorded temporal concentration distributions between two cross-sections. For each sub-reach j located between two sampling cross-sections (“1” – upstream and “2” – downstream cross-section), DLj was obtained from 2DLj=Uj2σt2x2-σt2x1tc2-tc1, where xi is the location of the ith cross-section, tci represents the time of passage of the centroid of the dye plume in ith cross-section, Uj indicates the mean velocity of the plume in the sub-reach j and σt2(xi) is the variance of temporal concentration distribution in the ith cross-section. The sub-reach mean velocity Uj is computed as 3Uj=x2-x1tc2-tc1. Based on the values of centroid travel times obtained at the upstream tc2 and downstream tc1 cross-sections of each sub-reach, the mean sub-reach centroid travel time was calculated as 4Tc=tc2-tc1.

The weakness of the method of moments is that the distribution variance is sensitive to concentration fluctuations in the tails of the concentration distributions. To increase the accuracy, the concentration distributions were cut at the point when concentration dropped below 0.5 % of the maximum concentration in the given cross-section, following the experience and recommendation of other scholars e.g..

The influence of the vegetation cut on the mean velocity were characterised as UNV/UVEG, where the subscript NV refers to the non-vegetated and VEG to the vegetated conditions, respectively.

The discharge was approximately double and sub-reach mean velocities were 3-4 times higher in the fully cut conditions when compared to the vegetated scenario (see Tables  and ). Before the maintenance, the vegetation coverage was mostly very high (>90 %), except for sub-reach B (68 %). The vegetation coverage computed for the entire reach (i.e. between the P1 and P5 cross-sections) according to Eq. (1) was equal to 93 %. The water depths were comparable between the two scenarios, ensuring that the vegetation coverage was the most significant factor causing differences in other hydraulic and mixing parameters. Thus, the fully cut conditions that reduced the coverage to 0 % notably improved the conveyance, as was expected based on e.g.  and . The increase in the velocity ratio UNV/UVEG was approximately linearly dependent on the vegetation coverage (Fig. ). If we assume that UNV/UVEG=1 when V=0, linear regression analysis indicates that under study conditions, the influence of the vegetation cut on the flow velocity can be approximated as UVEG=UNV/(0.03V+0.9). The formula remains the same (considering the coefficients' accuracy to two decimal places) if we include additional data points for vegetation coverage and sub-reach mean velocity, computed using Eqs. (1) and (3), respectively. Additional points (green triangles in Fig. ) include the values obtained for the entire reach (called the ABCD sub-reach) and selected from possible combinations of sub-reaches, i.e. ABC (P1–P4) and BC (P2–P4). The ABC and BC sub-reaches were selected as having the computed V differing the most from the already plotted points, equal to 92 % and 85 %, respectively. We assume that the linear dependency between velocity change and vegetation coverage can be extended as a first-order approximation to other trapezoidal channels with such high vegetation coverages > ∼68 %. However, the slope coefficient of the formula likely depends on channel geometry and flow forces, and the formula should be evaluated against a substantially larger dataset to derive more general conclusions. It should be emphasised that the dependency may deviate from the linear relationship at coverages lower than the ones presently investigated.

We are not aware of previous studies explicitly quantifying the relationship between the mowed vegetation coverage and enhanced conveyance. However, qualitatively similar results can be inferred from , who reported an approximately doubled mean velocity when vegetation coverage was reduced from ∼80 % to 0 %, and from , who found that vegetation removal from the coverage of 90 % to 0 % decreased flow resistance to one fourth, indicating a substantially enhanced mean velocity. Since the vegetation, in our case, was mostly emergent, the planform and cross-sectional blockage by vegetation are approximately similar, indicating that the results are in line with studies reporting a strong relationship between flow resistance and the cross-sectional vegetative blockage e.g.. However, as common for field conditions, it was not possible to control all the variables that may influence the flow discharge in a channel. Besides the major influence of vegetation removal on the results, some impacts may come from other origins. Water depth was somewhat lower, particularly in the two most downstream sub-reaches in Exp. 2 compared to Exp. 1, which partly explains why the flow velocity increased more than the discharge (Table  vs. Table ). The reported flow velocities based on the tracer data may slightly differ from the mean velocity classically determined as discharge divided by flow area (e.g. due to the low number of measured cross-sections and not well-mixed conditions). The presented image analysis method may not recognise very small patches or submerged vegetation and is not directly applicable to such conditions.

Table  shows longitudinal dispersion coefficients (DL) for each sub-reach and for the entire reach. Similarly to the flow velocities, the longitudinal dispersion coefficient values were significantly higher in the second experiment (fully cut conditions) compared to the vegetated conditions (see Fig. ). The highest values of U and DL under vegetated conditions were found for the least vegetated area, i.e. sub-reach B.

Considering different vegetation coverages in particular sub-reaches in the first experiment, it is worth analysing how change in vegetation coverage affects longitudinal dispersion coefficients. The relationship between obtained longitudinal dispersion coefficient (DL) and vegetation coverage (V) are presented in Fig. . The dispersion coefficients decrease with the increase of the vegetation coverage. The line fitted to the obtained values for each sub-reach (circles) indicates a linear relation in the analysed range of vegetation coverage.

Similarly to the velocity ratio, the additional values may be computed for the entire reach ABCD and chosen sub-reaches: ABC and BC. The obtained values of dispersion coefficients are 0.38, 0.42, and 0.61 m2 s-1 for the entire 467 m long reach and for the ABC and BC sub-reaches, respectively. These additional values of DL and V are added to Fig.  (green triangles) and they lie close to the line fitted to the previously obtained points (circles).

In non-vegetated open-channel flows, mixing parameters are often scaled against bed shear stress and water depth e.g., allowing for comparison of non-dimensional dispersion coefficients for different flow rates. However, the applicability of the traditional scaling of the longitudinal dispersion coefficient by the shear velocity for the vegetated flows is debatable. In artificially vegetated conditions, this is no longer appropriate, as the bed is not the dominant source of turbulence . Therefore, despite different attempts and investigations under laboratory conditions e.g., DL scaling in naturally vegetated channels remains an open question. The problem is incredibly complex in small natural streams with very diverse, extensive vegetation. Large datasets from further observations for different flow conditions, including detailed hydrodynamic measurements, are needed to address this question. However, to compare the data obtained from both experiments, we scaled the DL coefficient values against the mean sub-reach velocity (U) values for each experiment (Fig. ). In the case of Exp. 1, the value of the DL scale against the U decreases with the increase of the vegetation coverage V, similar to that presented in Fig. . In the case of Exp. 2, except for the value obtained for the sub-reach A (affected by non-well mixed conditions over the channel width in the P1 cross-section), the obtained values of DL/U are comparable for the B, C, and D sub-reaches (≈13±0.6 m). The obtained results suggest that although we may expect a linear relationship between the vegetation coverage and dispersion coefficient for highly vegetated conditions, the relation may be different for channels with low vegetation coverage.

The present values of dispersion coefficients and their relation with the vegetation coverage agree with previous findings obtained with uniform vegetation e.g. confirming that the presence of high vegetation coverage can diminish longitudinal dispersion. Our study shows that the decreasing effect of plants on dispersion extends from fully vegetated conditions down to the vegetation coverage of 2/3. As past investigations found that the dispersion can increase at lower coverages, particularly if the vegetation clumps, further experiments are needed to confirm the present conclusions and extend the obtained relationship to vegetation coverage below 68 %, as well as considering different vegetation arrangements and various flow conditions.

The vegetation cutting that reduced the coverage from 68 %–98 % to 0 % substantially influenced the flow and transport processes. The mean flow sub-reach velocity increased by about 3–4 times and the passage of the concentration peak was 4–5 times faster (see Fig. ), while the mean water levels remained comparable. In addition, the cutting moderately increased the peak concentrations (Fig. ). Thus, extensive cutting of vegetation can lead to harmfully high concentrations in small agricultural channels receiving large inputs of nutrients and agricultural chemicals from the fields. The fast flushing of the contaminants to receiving downstream water bodies is exacerbated by sub-surface drainage, typically used in northern and central Europe, which creates very flashy hydrographs e.g.. The limited residence times under non-vegetated conditions (Fig. ) decrease the likelihood for instream retention and may manifest as increased nitrate and suspended sediment loads e.g. to downstream water bodies after extensive cutting. In addition to decreasing instream retention, vegetation removal may increase erosion and mobilisation of e.g. heavy metals and phosphorus from the channel bed .

The relative changes were lower for the smaller reduction in vegetation coverage, suggesting that less extensive vegetation removals create less severe impacts on the transport of harmful substances while substantially enhancing the flow conveyance (Fig. ). Leaving some vegetation in the channel, e.g. close to the banks , likely guarantees acceptable water levels while allowing solutes and particulate matter to have a longer time to be permanently trapped or processed into less harmful forms. There is a need to evaluate the impacts of less intensive cutting scenarios, such as different spatial patterns of cutting and heights of vegetation, and of different channel designs and geometries e.g. on transport and mixing. In addition, the most suitable timing of cutting based on different criteria should be accurately determined, as observed that the conveyance enhancement by summertime cutting of aquatic vegetation could be short term.