Changjiang River (China) Suspended Sediment Concentration Estimation with Sentinel-2

1. Introduction

Sediment plays a crucial role in shaping fluvial systems, influencing the quality of habitats, determining the morphology of rivers, and impacting various uses of watercourses such as energy production, navigation, and drinking water supply. Therefore, it is essential to comprehend both the morphological changes in the river planform and the sediment transport processes. Quantitative data about the transported sediment and channel variations are necessary for understanding river dynamics at various spatiotemporal scales, as well as for engineering and ecological studies related to rivers. For instance, this information can be used to plan river restoration measures or hydraulic infrastructures and to investigate changes in water quality driven by sediment variations.

For many years, measuring the amount of suspended sediment in rivers has been an important way to determine erosion and the river's ability to shape the landscape [1,2,3]. The suspended sediment concentration (SSC) also has a significant impact on river ecosystems. It controls the amount of sunlight that reaches the water, which in turn affects the growth of vegetation and algae [4]. Additionally, SSC affects the suitability of habitats, fish migration, and the way estuaries form and change. Suspended sediment also carries carbon, nutrients, and contaminants [5,6,7,8]. Tracking SSC levels can provide insights into the dynamics of pollutants such as microplastics [9,10], which are the focus of innovative policies such as the European Zero Pollution Action Plan.

The impact of natural variations and anthropogenic activities on SSC trends is rarely discussed, even though the variability in SSC can significantly influence river habitats and floodplain development [11], and our knowledge about how human activities affect spatiotemporal SSC patterns in rivers is still rather limited [12,13,14]. It is evident that understanding sediment dynamics in rivers is crucial, but a major challenge is the lack of field information on SSC. Only a small percentage of the largest rivers on Earth have SSC monitoring sites, making it difficult to gather long-term SSC records and obtain spatial SSC coverage. While in-situ sensors can give insights into high-frequency dynamics, studying SSC in large rivers requires costly and time-consuming field campaigns with a lot of point measurements and observations. This results in an incomplete description of the spatial view of the process in the entire study region.

Satellites can provide a solution to the lack of data for developing long-term, spatially distributed SSC observations. Compared to local SSC gauging stations, satellite images can cover larger geographical areas with relatively high revisit time (i.e., temporal resolution) and detail (i.e., spatial resolution) for prolonged periods. This results in faster processing and reduced costs when compared with field campaigns. Recent studies have demonstrated the reliability of using satellite-derived information to investigate fluvial morphology and SSC in selected river reaches [15,16,17,18,19].

Previous research has primarily focused on large rivers, estuaries [20,21], lakes [22,23], reservoirs [24,25] and coastal [26,27] and marine areas, using remote sensing data with a spatial resolution of 500-1000 m [28,29,30]. However, more accurate information might be needed in the case of rivers characterized by large spatiotemporal variability in SSC, such as the Changjiang River, and therefore more high-resolute satellite images should be exploited. Therefore, the current work takes advantage of Sentinel-2 images. With respect to the medium-resolution images used so far (i.e., MODIS, MERIS and OLCI), data with the higher spatial resolution coming from the more recent Sentinel-2 Mission of the Copernicus European Earth Observation programme (10, 20 and 60 m spatial resolution) have opened new possibilities to investigate small targets [31].

In the Changjiang River Basin, most previous studies have focused on estuaries [32,33,34,35,36] and some large lakes [37,38,39], while there is still a lack of investigations on the main channel, mostly because of the lack of detailed field information and the challenges posed by retrieving SSC patterns in fast-changing conditions. The current study aims to contribute to this topic, providing a calibrated model that relates local SSC data with water reflectance, to eventually create SSC maps over the Zhijiang-Chenglingji reach in the middle Changjiang River mainstream, analyzing how SSC changed over the period 2017-2022 in correspondence to selected extreme events.

2. Materials and Methods

2.1. Study Area

The Changjiang River is one of the world’s largest rivers and has undergone a significant reduction in sediment load in recent decades, due to the effects of dam construction and soil conservation practices [40]. It ranks as the third longest river by length (~6,400 km), the fifth largest in terms of river discharge (~900 km³/y), and the fourth largest in terms of sediment load (~480 Mt/y) in the world [41]. It has a drainage area of approximately 1.9×10⁶ km², accounting for nearly 20% of the territory of China. The Changjiang River originates from the Qinghai-Tibet Plateau and stretches west-eastward from central China to the East China Sea. It is usually divided into three sub-basins with the demarcations at the Yichang and Hukou hydraulic stations, i.e., the upper, middle, and lower reaches, primarily based on its spatially varying landscape and climate patterns (Figure 1). The upper, middle, and lower reaches of the Changjiang River are about 4500, 955, and 938 km long, respectively. In this study, we considered the reach from Zhijiang to Chenglingji (~290 km), which belongs to the middle reach of the Changjiang River.

2.2. Field Data

Table 1 reports water levels, flow discharge and suspended sediment concentration at the two gauging stations of Shashi (SS) and Jianli (JL), measured on the same dates when cloudless Sentinel-2 images were available for the investigated reach. The data were obtained from the Changjiang Water Resources Commission (CWRC) (www.cjw.gov.cn/zwzc/bmgb/), and their consistency was verified.

It is worth noticing that water and sediments are continuously monitored in the two stations, as Figure 2 depicts.

2.3. Sentinel-2 dataset

Sentinel-2 data were used in the current study. The Copernicus Sentinel-2 mission consists of two polar-orbiting satellites positioned in a synchronized sun-synchronous orbit, with a 180° phase difference between them. The objective of this mission is to observe changes in land surface conditions. With its broad swath width of 290 kilometres and frequent revisits (10 days at the equator with one satellite, and 5 days with two satellites in cloud-free conditions, resulting in 2-3 days at mid-latitudes), it facilitates the monitoring of alterations on the Earth's surface [42].

Images, downloaded from the Copernicus database directly in the open software QGIS using the plugin Semi-automatic Classification Plugin SCP version 8.2.2 [43], belong to both Level L1C and Level L2A, being the latter atmospherically corrected and the first georeferenced. Images covering the period 2017-2023 with a cloud cover lower than 5% were selected to match the available SSC data. Images were then converted to TOA (Top of Atmosphere) reflectance. This accounts for both the reflectance of the Earth's surface and the influence of the atmosphere on the received light. This conversion normalizes the data for variations in solar irradiance (the amount of sunlight), making it easier to compare and analyze the information captured by the satellite across different conditions

The two satellites Sentinel-2A and Sentinel-2B are equipped with Multi-Spectral Instrument (MSI). This instrument captures data across various spectral bands (Table 2), including four bands at 10 meters resolution in the visible spectrum, covering the wavelengths of blue (490 nm), green (560 nm), red (665 nm), and near-infrared (842 nm). Furthermore, there are six bands at 20 meters resolution, four of which are narrow bands in the vegetation red edge spectral range (705, 740, 775, and 865 nm), along with two longer bands in the shortwave infrared (SWIR) region (1610 and 2190 nm). Additionally, there are three bands at 60-meter resolution dedicated to tasks like atmospheric correction (443 nm for aerosols and 940 nm for water vapour) and cirrus cloud detection (1380 nm) [44].

To provide a comprehensive evaluation, three Sentinel-2 granules were selected given the large extent of the study area. These granules are also called tiles, which are the minimum indivisible partitions in the MSI sensor reference frame of a given number of lines along the track and detector-separated [45]. Specifically, the tile IDs are 49REP, 49RFP, and 49RFN, which extensively covered the study area from west to east.

2.4. Reflectance Models

Multiple models able to relate field data and image reflectance exist and were tested in the current application (Table 3) by comparing SSC field and satellite-retrieved information. For a full description of the models readers can refer to the cited literature.

3. Results

3.1. Satellite Image Calibration

To retrieve a relationship between measured SSC and band reflectance, images should be first pre-processed to account for atmospheric correction, and then the reflectance values could be derived. To correlate measured SSC and reflectance, 9 pixels neighbouring each station were analysed in QGIS, and the reflectance value was then averaged to reduce local errors. Values were then tabulated and different relationships for each model (see Table 3) were tested, to select the best-fitting model for both stations (i.e., higher R²).

As visible in Table 4, all models present an R² lower than 0.5, with the ratio R665/R560 (band 4/band 3) being the best model.

The model R665/R560 performs better than the other, and it was therefore selected for further application using eq. (1):

$$SSC=0.2139\frac{R665}{R560}-0.1033$$

(1)

3.2. Application to Extreme Events

The algorithm was applied considering six different conditions, characterized by high water discharge and/or high SSC measured at the gauging stations (Table 5). As visible, events having a significant water discharge might mobilize a relatively low amount of sediments, as happened on August 30, 2021.

From Figure 3, it is evident that high flow conditions generally transport significant suspended sediment along the whole reach, covering both the main channel and the secondary ones. Similarly, low flow conditions are unable to transport a significant amount of material, therefore resulting in lower concentration transported only in the middle of the channel.

It is worth noticing here that, in July 2017, SSC increased in the lower part of the study area, starting from about 10 km upstream of JL. This might be due to significant riverbank erosion in this area. Such instability is mainly caused by a decrease in SSC during normal flow conditions due to the cascade reservoirs present upstream of the study reach [55], which represents a current worrying risk to flood protection in the area [56]. In fact, according to the statistics provided by the Changjiang Water Resource Committee in 2021, a total of 1046 riverbank collapses occurred, with a cumulative length of approximately 760 km in the middle-lower reaches of the Changjiang River from 2003 to 2021 [57].

Looking at the other events analysed in the current study, no significant spatial differences are visible, meaning that no localized sources of sediment impacted the study area during such extreme events.

4. Discussion

4.1. Uncertainties in SSC Retrieval from Remote Sensing

It has been found that only about 10% of the signals that were received by the satellite sensors are from water bodies. The rest of the signals are mainly originating from atmospheric scattering, topographic and soil brightness effects, as mentioned in studies conducted by Huete et al. [58]), Borgogno-Mondino et al. [59], Boothroyd et al. [60], and Duan et al. [18]. Thus, it is crucial to perform atmospheric correction to derive accurate colour-related water parameters [61]. As rivers generally exhibit high turbidity and complex optical characteristics, the assumption of zero water-leaving radiance in the near-infrared band [62] might be no longer valid in such environments. Previous research has shown that correcting for atmospheric effects using the short-wave infrared band can help deal with the issue of water-leaving signals in satellite imagery. However, in the case of very turbid waters, like those in the Changjiang River during floods, there is still a noticeable water-leaving signal in the short-wave infrared band [17].

In addition to the challenges posed by turbid waters, the effects of land adjacency may significantly impact the signal in the short-wave infrared band, even in cases where the water-leaving radiance is negligible [61], especially in narrow rivers crossing urban centres [17]. Additionally, the reflection of direct sunlight at the air-water interface, known as sunglint, may also have an impact on the retrieval of water-leaving radiance [63]. The combination of all these challenges often leads to uncertainties in estimating remote sensing reflectance after atmospheric correction, further affecting the results of SSC retrieval [18].

Besides uncertainties connected to atmospheric correction, also differences in sediment particle size may contribute to the uncertainties in SSC retrieval. As past investigations have shown [34,64], particle size and distribution could impact surface reflectance altering the backscattering coefficient, with differences increased with increasing wavelength. SSC particle size was highly dynamic in this area as a result of river scouring and the replenishment of fine sediment [65]. In addition, as the erosion has lasted for over 20 years, the replenished and restored fine riverbed particles as suspended sediment showed an overall decreasing trend. To further corroborate the results presented in the current study, future studies should consider the distribution of sediment particle sizes and their influence on remote sensing models.

It is also worth noticing that, with respect to more still water present in small/medium lakes and reservoirs, a large range of hydrological conditions should be considered in estimating SSC in rivers, as water optical properties might have significant variations in the dry and wet seasons [17]. Therefore, selected calibration and validation data should be sufficient and cover as many different seasons as possible. At the same time, the need for covering a broader spectrum of hydrological conditions SCC values might result in less accurate models, as suggested by the current research, especially in watercourses affected by significant variations of SSC due to climate variations and human intervention.

SSC models derived by using band ratio, as made in the current work (eq. (1)) could implicitly compensate for some uncertainty in remote sensing reflectance estimation [49]. Thus, since the spectral variation in the backscattering coefficient of suspended sediments is smooth, band ratios can mitigate the impact of differences in sediment particle size [66]. To further mitigate the uncertainty in reflectance retrieval, values were derived from multiple pixels located around the gauging stations, eventually deriving a spatial average.

4.2. Driving Forces of SSC Transport

SSC in rivers is usually positively correlated with water discharge, as also demonstrated by past investigations focused on the middle reach of the Changjiang River [67,68] and the Yellow River [18] in China, or on the Indian Chandra River [69] and the River Rhine and the Celone River in Europe [70].

However, anthropogenic alterations such as the presence of reservoirs could alter this trend, reducing the correlation between water discharge and SSC, therefore complicating the estimation of SSC as a function of water discharge via classical sediment rating curves [71]. For example, Zhang et al. [23] reported that, after the construction of the Xiaolangdi reservoir on the Yellow River, the correlation between water flow and SSC decreased at several hydrological stations, as the reservoir has a minor impact on annual water flow, but a significant impact on sediment transport in the lower reach, leading to a substantial decrease in SSC. For a river system undergoing significant sediment load decline, river channel self-adjustment is an important factor driving the recruitment of SSC from bed and bank erosion. It was reported that from the impoundment of TGR in 2003 to 2021, the whole middle reach of the Changjiang River was scouring, resulting in a gradual SSC recovery from the dam site downstream to Chenglingji [72].

5. Conclusions

In the current work, remote sensing was combined with field data to investigate SSC dynamics along the Zhijiang-Yueyang reach of the Changjiang River, and how the quantity of transported sediment changed in response to extreme high and low flow conditions, considering six selected events during the period 2017-2022.

Despite uncertainties in calibration, this application has shown the potential of using satellite images and band reflectance to map spatiotemporal changes of SSC, eventually overcoming limitations due to the very local nature of using field sampling. Multiple models combining band reflectance were tested to derive the most suitable one for describing the observed SSC variability, highlighting the strengths and weaknesses of the proposed approach. Indeed, monitoring from above highly variable conditions of SSC, which are typical of riverine environments, might be very challenging, and a large volume of information is needed to derive good-fitting relationships. To address this point, future investigations will consider a longer reach and extended temporal horizons, eventually combining satellite data at different spatiotemporal resolutions.