Assessment of the morphological trends and sediment dynamics in the Indus River, Pakistan

LF, HF

Low flow, High flood

SSL, BL

Suspended sediment load, Bed load

AC, TC

Approach channel, Tail channel

RP, LP

Right pocket, Left pocket

MEP

Modified Einstein procedure

D₅₀

Median size of bed material

ACF

Autocorrelation function

a

Attribute before completion

C, D

Condition attribute, Decision attribute

F

Indiscernible classification

LIR

Lower Indus River

RST

Rough set theory

M_T

Million tons

TSL

Total sediment load

D/S

Downstream

U/S

Upstream

PAC

Partial autocorrelation

RP

Return period

Ǖ

Nonempty set

Ǻ

Nonempty finite set

River morphology is highly related to environmental conditions (Eaton et al. 2010; Rozo et al. 2014), and geometry of the alluvial channel changes due to variable rates of water and sediment inputs, climate variability, and human activities such as flow diversion, construction of dams, and deforestation (Dewan et al. 2017). A change of river geometry will result in an imbalance dynamic equilibrium (Petts & Gurnell 2005), thus perturbing the channel form and pattern. Consequently, these activities not only cause deterioration of the river conditions, such as hydro-geomorphology, bank erosion, and bank failure (Yang et al. 2015), but also adversely affect the ecosystem and biodiversity (Jain 2012). Therefore, assessment of river morphology and sediment balance is inevitable for the restoration and sustainability of the channel ecosystem (Ijaz et al. 2020). The trends of changes in the amount of scouring were examined, and the appropriate estimation/prediction of scour pit depth can be attained by varying the conditions containing change in wave height, current speed, and bed particles (Yamini et al. 2018). An innovative holistic method was developed by connecting the hydrology, morphology, and ecology characteristics based on the investigation of Sutla River and its biodiversity. It was observed that the ecosystem is directly linked with the river water level (Ćosić-Flajsig et al. 2020). The numerical approach was used for the modeling of flow hydraulics and results show that flow speed is reduced in the floodplain with increasing runoff in the main channel (Ahmad et al. 2020). Based on a literature review, understanding the channel form and pattern due to unstable river conditions is essential to develop modeling and management schemes for such rivers.

The temporal shift of riverine channels and river channel erosion have been core areas of study in geomorphology (Petts 1995). The measurement of riverbank and channel shifting have been made using different methods, such as sediment data, historic data, planimeter survey, periodic cross-profiling, erosive pins, and photogrammetry. According to the earlier research on riverbank change, four core techniques are divided into field measurement, remote sensing, historic data, and paleo (Yang et al. 2015). Each of these methods has its advantages and disadvantages as far as its size, precision, accessibility, and repeatability are concerned in change in river morphology (Grabowski et al. 2014). Remote sensing data give a direct, integrated, and synoptic view of a vast area in comparison with other methods. This approach is quite distinct from traditional methods, such as sedimentation, historical graphs, and cross-profiling, which are usually localized in extent (Yang et al. 2015). Remote sensing is an accurate and multi-temporal approach, which has been used to determine the change in riverine morphology in different rivers. Morphometric parameters, such as sinuosity index, channel surface area, channel erosion/deposition, channel centerline and channel width have been utilized in various studies to examine the river morphology with the help of remote sensing (Yang et al. 2015).

River channel erosion can be measured with two different methods. The first one is the quantitative technique, intended to measure the effective runoff, runoff that is transported mostly by sedimentation load (Pickup & Warner 1976). The concept of effective runoff is applied to suspended sediment load (SSL) (Wolman & Miller 1960), bedload (BL) (Barry et al. 2008), and total sediment load (TSL). The second method is the qualitative method, which mainly focuses on the hydrologic events to change the system morphology, called geomorphological effectiveness (Sloan et al. 2001). The quantification of the TSL combined with SSL and BL is still a challenging task in river engineering (Yang & Julien 2019). The BL and TSL is challenging to measure accurately. Wohl et al. (2015) have introduced the calculation of bedload which is bedload function. This technique for the calculation of sediment transport and total bedload is a remarkable development. The current study focuses on pre-flood and post-flood TSL computed from Einstein's procedure in the Indus River. The modified Einstein procedure (MEP) is applied, which is generally used to sand-bed rivers (Shah-Fairbank 2009). The runoff and sediment are the crucial factors linked to episodic flooding, infrastructure destruction, population movement, dam sedimentation, and extensive issues on water resources protection more usually.

The Indus River system is of deep concern because its reaches are prone to significant migration and erosion, threatening engineering structures which leads to social and environmental impacts that may be exacerbated by climatic change, land-use changes and agricultural intensification, and its banks are exposed to rapid population growth. Intense and frequent bank erosion causes significant economic and human losses, and river vulnerability studies reveal that the victims in Pakistan have been migrating and resettling in worse areas with a lack of necessities (Carling et al. 2018). Therefore, study of the Indus River is very important for flood protection and water managers need to keep a sharp eye to re-evaluate the procedures for river management. In this study, the Lower Indus reaches were selected where the outdated pond system operation drives fluvial morphology. The gates of left and right pockets remain closed almost all the year round. Thus, Sukkur barrage pockets become silted up, especially during the flood season and sediment also passes into the channels off-taking upstream of the barrage (Mahessar et al. 2020). It is hypothesized that the coarser particles play a role in channel morphology, especially in flood-based events.

This is the first study of its nature, designed to jointly examine the timeline of co-evaluation of morphology trends, sedimentary dynamics, and first-time application of the rough set theory (RST) to determine the dominating factors responsible for river shape changes. Specific objectives of this research are: (a) to examine the nature and extent of backline movement of the Guddu and Sukkur reach; (b) to explore the morphometric parameters, such as sinuosity index, migration rate of river centerline, braiding index (BI), river width and length; (c) to estimate sediment dynamics, such as erosion and deposition based on the daily observed sediment fluxes; and (d) to apply RST to compute the climatic based riverine dominating factors which affect the river shape changes. To analyze the sediment budget, erosion, and deposition, we have collected and assembled field, published, and unpublished data. We used the grain size particles, computed unit channel velocity, and carrying capacity of sediment load to examine the pre- and post-flood channel erosion in the Lower Indus River (LIR). Remote sensing data coupled with GIS are used to assess the river morphology and deep learning is used to estimate the riverine dominating factors responsible for river morphology. The present research deals with the conceptual lens to perceive the co-evolution of sedimentary-morphology dynamics, as well dominating factors responsible for change in river morphology. For the policymakers and environmentalists, these evidence-based solutions have a high potential for autogenic restoration, riverine regulations, and navigation safety against the looming threat of climate change.

This manuscript is divided into five major sections. The introduction discusses the literature review, problem statement, study gap, innovation and aim of the study. The description of the study basin gives a brief overview of the study area with graphical representation. The methodology and data collection is used to determine the river morphology-sedimentary dynamics and the RST is presented. In the results and discussion, the results of the Indus River using remote sensing coupled with field observation data and deep learning to determine the dominating river factors responsible for river morphology are presented. This section also discusses the practical implications of the present research. The conclusion presents the major crux of the study, limitations, and future recommendations.

The selected study reach is 670-km talweg-length of the Indus River of Pakistan, with a slope that varies between the Guddu to Kotri barrage from 0.00013 to 0.000082 (Figure 1). The annually observed values of runoff vary in size, although monsoonal succession timings are consistent (Yu et al. 2013). The observed peak monthly runoff values occur in July, August, and September (Figure S1), with exceptional five peak values occurring in the last 30 years due to climate change. The summer monsoonal peak flows are augmented by snowmelt, which transports sediments from Pajnad River to Indus River (Inam et al. 2007). The average annual monsoonal values (1962–2019) occurred in Guddu with a discharge value of 11,262 to 603,280 ft³s⁻¹, with an average upstream (U/S) water level 244.82 to 257.9 ft. Therefore, the average head difference (Guddu) between U/S and downstream (D/S) is 5.428 to 2.227 ft. Furthermore, the two-year return period (RP) value at Sukkur is 622,631 ft³s⁻¹. The exceedance probability P_i (X>=xt) and non-exceedance probability P_n (X=<xt) at two-year RP is observed to be 0.485 and 0.515. The peak flow values at 50-year RP is 1,249,200 ft³s⁻¹ with P_i (0.019) and P_n (0.981). Similarly, peak flow at 100-year RP is 1,337,200 ft³s⁻¹ with P_i (0.010) and P_n (0.990).

The discharge data, TSL, and SSL were collected from Provincial Irrigation Department Sindh, Pakistan. The hydraulic field survey was carried out near Sukkur (27° 40′ 50″ N; 68° 50′ 43″ E), upstream of the barrage, in 2019 (Table S1). We computed the average hydraulic parameters such as hydraulic resistance to flow ∼1.06, Froude number to flow ∼1.01, bed shear stress ∼10 Nm⁻², roughness coefficient ∼0.024, hydraulic radius ∼8.7 ft, Froude number ∼0.2, and sediment load by using the MEP (Shah-Fairbank 2009). The alluvial Indus River has little variation in the vegetation cover and sandy bed material from satellite images (Carling et al. 2018). We have observed that width range of the LIR is from a few kilometers to hundreds of meters with a few channels up to 8 km wide, and wetted course widths >2 km remain frequent in the dry season. The kinetic stream power ≤ 7,059 Wm⁻¹ with a unit stream power ≈ 3.2 Wm⁻² was observed. The sediment load is higher in post-flood compared with pre-flood which causes the banks to readily erode.

The multi-temporal Landsat data of peak flood was used for the analysis of channel dynamic (1986–2020). The 03 Landsat tiles for each year cover the entire study area and thus 15 total tiles covered the five flood events in the last 34 years (1986, 1992, 2010, 2015, 2020). Landsat 1–5 MSS (1986), Landsat 4–5 TM (1992), Landsat 5 TM (2010), Landsat 8 0LI/TIRS (2015, 2020), and digital elevation model (DEM) were acquired from the USGS website (http://earthexplorer.usgs.gov/). The cloud-free tiles for the HF periods were processed within the UTM 42-N projection that referenced to WGS84 datum and an atmospheric correction was not necessary (Song et al. 2001).

To gather information from both banks, the Indus River macro-channel centerline was calculated to divide the sub-reaches into the left or right bank features. Landsat images were transformed into raster data sets, reclassifying the scene into three categories: water, bare soil, and vegetated soil by using the unsupervised classification of Landsat in the Supplementary information (Figure S2). Bare soil mainly consisted of channel sandbars, floodplain sand, and just a few scour holes on the floodplain. Therefore, losses in any vegetated land (comparison of low flow sequences) are considered land lost areas due to the monsoonal flows. The Indus River positional changes, planform changes, river centerline, sinuosity index, braiding index, water surface area, channel width and channel length are evaluated by using the ArcGIS 10.2 software (Yang et al. 2015).

We divided the 670 km Indus River into two reaches: Guddu to Sukkur barrage and Sukkur to Kotri barrage, segmented by constructing transects at 5 km intervals. The right and left bank movement was computed for segments suggested by Kummu et al. (2008). The annual bankline migration for each period was computed from the net migration and number of years in the flood event/epoch. The relationship between erosion, accretion and classical meandering theory was investigated from riverine bank curvature (straight, concave, convex), which was recorded at the end of each transect for two flood events. Excel and MiniTab Inc. software was used for data series autocorrelation function (ACF) and partial autocorrelation (PAC) functions. For brevity, ACF and PAC function is not elaborated here but is presented in Gupta et al. (2013). The flowchart of the proposed methodology is represented in Figure 2.

The measurement sites were selected by conducting a reconnaissance survey of the proposed sites on canals and at the entrance of left pocket (LP) and right pocket (RP) of Sukkur barrage. Data of all benchmarks and their reduced levels as per Survey of Pakistan were collected from Provincial Irrigation Department Sindh. The Q, SSL, bed material sampling and boil sampling at D/S head regulator of canals were observed. Equilibrium experiments in head reach measurements were carried out to estimate the sediment load distribution in approach channel (AC), tail channel (TC), LP, RP, and off-taking canals. The SSL samples were collected with ten equi-spaced verticals in the cross-section of canals. The bed material samples were also collected from the canals by using BM-54 sampler. The collected samples were analyzed in the laboratory for sediment concentration as well as for particle size gradation.

Bed material samples for grain size distribution and D₅₀ were taken at 2,000 ft, 7,000 ft, and 9,000 ft upstream of Sukkur barrage. The suspended sediment and bed material samples were analyzed in the laboratory. The sediment coarser than 62 microns was analyzed by the visual accumulation tube or sieve method. The analysis of sediment finer than 62 microns was carried out by the pipette method. The concentration of suspended sediment samples was determined using the gravimetric method. The TSL was computed by using the MEP (Guo & Julien 2004), laboratory analysis of suspended sediment, and bed material samples (Vauchel et al. 2017; Zhang et al. 2017).

We have introduced an innovative deep learning RS technique for better understanding the dominating factors responsible for river shape changes. RST is a mathematical approach used to obtain knowledge from uncertain data and reduce the information by mining hidden data (Yan et al. 2020). RST deals with the data imputations based on indiscernibility (Yamany et al. 2016). Knowledge of the decision table in an expression system D=(Ǖ, Ǻ, ϓ, F) where Ǖ is a nonempty set, Ǻ is a nonempty finite set of the attribute, ϓ is attributed range value (a ɛ Ǻ), F: Ǖ×Ǻ→ϓ, for every a ɛ Ǻ, x ɛ Ǖ, then f (x, a) ɛ ϓ. If Ǻ (attribute set) is composed of C (condition attribute) and D (decision attribute) and satisfy C∪D = Ǻ, Ϲ ∩ D = ∅︀, and S (decision table) = (Ǖ, C ∩ D). Mostly, there is only one decision attribute, namely, S = (Ǖ, C ∩ {D}), where d ɛ C is the decision table. The prediction of channel morphology consists of the following steps (Figure S3): predicting process, attribute selection, data collection and processing, data collection of surrounding factors which makes it possible to affect channel shape, attribute discretization, and rule matching (Zhang et al. 2016). The attribute reduction was obtained after discretization [C1, C2, C3, C4, C5, C6, C7].

The superimposed channel maps for five periods, between 1986 and 2020 for both Guddu and Sukkur reaches (Figures S4 and S5). It is crystal clear from the Guddu reach that the river has braiding phenomena, formation of new channels, and tends towards the right side from downstream of Guddu reach. The water flows in a braided pattern in Guddu reach, dividing, and reuniting the flows. In addition, Sukkur reach examination represents meandering oxbow lakes, and expanding and contraction phenomena. It is evident that the erosion occurs in the outer parts having peak velocity, and deposition occurs along the inner bends where the velocity is low. The centerline of the Indus River regime at Guddu reach represents the steady state with a few typical reaches such as at Ghotki, Khanpur, Pano Aqal, and Rohari tending to a meandering form (Figure 3(a)). The highest bend occurs upstream in Ghotki with mean width (7 km) and bifurcation formation upstream of the Sukkur channel. The centerline of Sukkur reach represents an unsteady state, such as at Gambat, Kandiaro, Naushahro Firoz, Daulatpur, and Sekhat which were markedly changed in meandering and oxbow lakes from 1986 to 2020 (Figure 3(b)). The highest bend occurs at Sekhat with mean width of 4.2 km and Kotri upstream with mean width of 3.8 km.

The river pattern expression at different scales was obtained by using remote sensing and GIS. The bankline movement analysis was completed for two epochs by considering the year 1986 as the benchmark, and compared with other epochs such as 1992, 2010, 2015, and 2020, both for Guddu and Sukkur channels. The consideration of the two epochs was to examine whether the flow regulation had a significant impact on riverine migration or not. The study of river patterns such as the formation of straight, meandering, braiding, oxbow lakes, formation of new channel bar is assessed in detail (see Figure 4(a)). The mean channel width analysis for Guddu varies between 3.66 km in 1986 and 6.22 km in 2015. The actual channel width varies for all segments from 1.5 km to 10 km during the study period. The careful examination of Ghotki channel represented the meandering formation (8 km in 1992), and the river changed its course (2015). In addition, Pano Aqal and Rohri channels represent oxbow lake and meandering formation (1992), but these channels change their cross-section and have a braiding trend (2020).

The river pattern of Sukkur reach represents meandering, oxbow lake, formation of new channel bar, and changing both area and shape in the existing channel bar. The formation of maximum meandering occurred in Daulatpur (6 km in 2010), which cut off in 2015 to form an oxbow lake. The formation of oxbow lakes in Sukkur reach is due to the banks' unequal erodibility due to slope, soil structure, and geometry of adjacent bends that ultimately cut off the bends; details of the river pattern formations are given in Figure 4(b). The mean channel width for Sukkur varies between 1.9 km in 1986 and 10.22 km in 2015. The actual channel width varies for all segments from 2.1 km to 12 km during the study period.

The BI value for Guddu decreased by 8.54% in 1992 then increased to 18.1% in 2010, suggesting an increase in bars and islands in the channels. Evaluation of the percentage of the channel area dominated by bars and BI of Guddu shows that they are correlated positively with the linear coefficient of correlation of 0.75 (p = 0.049). The regression analysis showed a positive correlation between runoff and total erosion for Guddu reach (r² = 0.687, p = 0.01). Furthermore, Sukkur barrage evaluation of the percentage of the channel area dominated by bars and BI shows that they are correlated positively with the Pearson coefficient of correlation (r² = 0.867, p = 0.049). The regression analysis between runoff and total erosion is positively correlated (r² = 0.6938, p = 0.01) and is given in Figures S6(a) and S6(b).

The sinuosity index from Guddu to Kotri has an increasing trend, and it is evident that the Sukkur reach has more sinuosity than the Guddu reach between 1986 and 2020. The maximum value of the sinuosity index for Guddu and Sukkur reaches was observed to be 1.84 and 2.32. The river length has an increasing trend from Guddu to Kotri barrage, which varies from 660 to 688 km. The mean width of the channel increased from 11.05 to 12.9 km at high flow. The minimum water surface area at low flow and peak flow for the channel reach is 450 and 5,956 km² (see Figure 5).

The expansion of wetted areas from low to peak flows often involves forming new channels and dry channels are retained annually. It is evident from the low and peak flows' epochs that low flow channels are always occupied while peak flow channels are only occupied during the summer season. The lateral variations and declines occurred because of erosive phenomena of macro channels in a few years, with subsequent sediment fluxes on the opposing channels. It is clear that channel stability is retained at low flows, while channel changes occurred at flood events which carried huge sediment. The wetted channels at low and peak flows were observed from Guddu to Kotri reach from 1986 to 2015. The low flow channels are superimposed for five years of analysis obtained from 1986 to 2015. The channels are determined from 120 sub-reaches of 5 km length based on a previous study (Carling et al. 2018). It is evident that the study reach has a minimum of three and a maximum of nine channels in total for the given cross-section. The average channels for the low flow section are 4.75 with a standard deviation of 1.7, while the number of channels observed for peak flows is higher and is 5.7 with a standard deviation of 2.2 (not illustrated). The peak flow season has one more channel compared with the low flow season. Overall, the number of channels observed in the study reach is five, and four channels are wet throughout the year while five channels flow only in summer.

The riparian land loss on both right and left banks was calculated to consider the channel change more specifically. The land lost annually compared to the peak runoff for right and left banks was calculated for each 670-km spaced cross-section shown in a single cross-section in Figure S6(c). Despite the necessary runoff data up to 14,000 m³s⁻¹ and contrary to peak runoff which occurred in 2010 compared with other flood events, large peak signals appear. In all sections, 65% represents the positive linear relationships (Equation (1)), with 52% statistically significant (p < 0.04, r² > 0.35, N > 9). It was observed that 29% of sections have weak and non-significant relationships (p > 0.049), while 61% have positive relationships and tended towards intercept of zero land loss around runoff value of 5,899 cumecs. Additionally, all positive regression analyses were bound to intercept the x-axis where Q = 5,899 cumecs. The confidence limits (94%) almost zero-intercept regression curves showed that the true intercept lies between 0 and 12,000 cumecs (not illustrated). As monsoonal flood occasionally rises below 5,899 cumecs, intercept values of <6,000 cumecs can be reduced. Furthermore, as nine years of the study demonstrated land lost for high runoff <12,000 m³s⁻¹, it is not unacceptable to concede a threshold value of 5,899 cumecs for potential applications.

Hindcasts and forecasts are conceivable for each section, but there is a major endpoint control which is enacted on the slope of all positive regression lines at 27,000 m³s⁻¹, and 94% confidence limits were fitted in all cases to measure the ambiguity of land lost. The uncertainty throughout land loss was examined for any given runoff. To assess the acceptability of such a technique, additional land loss data (from satellite imagery) were also plotted for nine additional years (2004–2010 and 2014). Comments are critical on the negative or indeterminate regressions showing 30% of the sample. It is evident that 70% of regression analyses are positive. The macro channel begins to broaden at peak flows and widen considerably in peak flooding events such as 2010, without defined lateral accretion compensating the same event, and the system capacity increases as runoff increases. The AC and PAC results depict that between sections there is probability greater than 61% that results could be applied (10–20 km) to next-neighbor reaches. Thus, it is clear based on the above analyses that the channel changes through riverine banks' recession.

We have compared the results of the present research with a similar study which was conducted previously on the upstream of the Indus River (Carling et al. 2018). We noted that dry and wet channels are almost the same compared with the previous conducted study on Chashma-Taunsa reach. The river experienced contraction/expansion and vegetated land loss is directly linked with the current speed in both studies. Moreover, Taunsa-Chasma reach has braiding formation, but Sukkur reach has oxbow lakes' formation via neck cutoff, meandering, and channel bar formation. The previous study on the upstream reach of the Indus River concentrated on analyzing a single flood event (2010), while the present study extends to include the five major flood events. The present study deals with morphometric parameters which were not conducted in the Chashma-Taunsa reach. We have estimated the erosion and deposition based on sediment load during LF and HF periods, while most previous studies have estimated the erosion and deposition with remote sensing and GIS. As the runoff and sediment load are altered anthropogenically, or by climate-based flood events, the space needed by the riverine could be examined. Thus, the present study examined the co-evaluation of morpho-sedimentary dynamics and RST to determine the dominating factors responsible for river shape changes.

Hypothesis (1) is tested within this section.

The flow conditions at Sukkur barrage and seven off-taking canals were monitored pre- and post-flood 2019. It was observed from the data that river flows about 7–10% from the RP and 9–13% from LP during the HF period, and about 24–31% from RP and 32–51% from LP during the LF period were diverted to the off-taking canals from the Indus River, respectively. The total sediment load during the LF period at Sukkur barrage was 1,288 ppm, wherein computed bed load (BL) of the sand fraction (+62 micron) was 299 ppm, and SSL of the fine fraction (−62 micron) was 989 ppm. The average sediment load in the Indus River during the HF period was about 1.296 million tons/day (M_T day⁻¹), which comprised 77% fine sediment (clay and silt) and 23% sand. Thus, the river bed upstream barrage mostly comprised very fine and fine sand during the HF period. However, after the HF period, the average D₅₀ of bed material near the barrage (2,000 ft upstream) was 0.202 mm, and it was coarser than the average D₅₀ of bed material in the upper reaches (7,000–5,000 feet upstream), where it was 0.122 mm and 0.089 mm, respectively. In alluvial rivers, incision and narrowing occur when the ratio of supply to transportation decreases, and aggradation and broadening happen when the ratio increases.

During the HF period, the average sediment load (SL) was 1,379 ppm, 1,551 ppm, 1,402 ppm at AC, TC, and RP at Sukkur barrage. During the LF period, the average SL was 785 ppm, 601 ppm, 799 ppm at AC, TC, and RP at Sukkur barrage. However, SSL consisted of 3–13% of the sand fraction (+62 micron) and 87–97% of fine sediment (−62 micron). The bed material in the reach between AC and TC consisted of 52% sand and 48% clay and silt (Figure 6). It is evident from the analyses below that several tens of kilometer are eroded and deposition has taken place.

Longitudinal bed profiles were observed along each pocket and also extended up to 5,000 ft. These were observed at three equi-spaced locations and repeated on a monthly basis to find the trend of erosion or deposition in the pockets. The result indicates that the four longitudinal bed profiles taken at different times generally show a scouring trend in the upstream river reach, and also downstream of the submerged weir and deposition in the right pocket. This indicates that some of the sediment, which was eroded from the upper river reach has passed into off-taking canals and the remaining material was deposited in the RP. However, the profiles taken in the left pocket and its extension in the upstream river reach also indicate alternate deposition/erosion phenomena. The last two profiles taken in the months of September 2019 and October 2019 also show local scour in the river reach upstream of the LP. There is also a deposition trend in the LP and it is significant near the divide wall.

The Guddu reach represents the braiding formation and wandering belt configurations in various periods. The Sukkur reach represents oxbow lakes via neck cut-off, meandering, and channel bar formation in the Indus River (Figure 4). The sediment analysis represents that the average SL was observed to be more during the HF period compared with the SL during the LF period (Figure 6). Consequently, hypothesis (1), that the coarser particles played a responsible role for channel morphology, especially, in the flood-based events, is supported by the above analyses.

During the HF period, the average sediment concentration was 1,337 ppm, 1,324 ppm, 1,294 ppm, 1,286 ppm at the RP, and three of its off-taking canals (North West, Rice, and Dadu). During the LF period, the average sediment concentration was 772 ppm, 796 ppm, 830 ppm, and 800 ppm at RP and three of its off-taking canals. This indicates that, on average, canals were drawing slightly less sediment than the sediment load in RP during the HF period and causing RP deposition (Figure 7). During the LF period, the phenomenon was reversed and indicated erosion in the RP. Furthermore, SSL consisted of 3% sand, 58–70% silt, and 27–39% clay in the off-taking canals. The total sediment inflow during the HF period at the start of RP was 5.895 million tons (M_T), and the total sediment withdrawn from RP by the three off-taking canals was 5.506 M_T. This indicates the deposition of 0.389 M_T of sediment in RP. However, there was sediment erosion of 0.074 M_T during the LF period. Thus, there was a mixed trend of deposition and erosion in the RP with an estimated deposition of 0.315 M_T (4% of total sediment inflow).

During the HF period, the average sediment concentration was 1,525 ppm, 1,491 ppm, 1,741 ppm, 1,638 ppm, 1,586 ppm from the LP, and four off-taking canals (Nara, Kairpur Feeder East, Rohri, Khairpur Feeder West). During the LF period, the average sediment concentration was 642 ppm, 875 ppm, 778 ppm, 707 ppm, 737 ppm from the LP, and four off-taking canals. During the HF period, the sediment concentration in the LP and off-taking canals was almost the same. During the LF period, the sediment withdrawn by canals exceeded the LP sediment load, which indicates a mixed trend of deposition and erosion. The SSL consisted of 3–7% sand, 59–69% silt, and 24–38% clay in the off-taking canals. The total sediment inflow in the LP during the HF period is 10.595 M_T and the total sediment withdrawn by the four off-taking canals was 10.056 M_T. This indicates the deposition of 0.539 M_T of sediment in the LP during the HF period. However, there was sediment erosion of 0.411 M_T during the LF period. Thus, the net estimated deposition in the LP was 0.128 M_T (1% of sediment inflow) which showed that sedimentation in the LP was less than RP (see Figure 8(a) and 8(b)).

The total sediment inflow was 14.108 M_T at AC, while the sediment load at TC was 6.721 M_T and the RP was 8.085 M_T. This indicates that 0.695 M_T of sediment was eroded in the reach from the start of AC to its bifurcation into TC and RP channel. Although 6.721 M_T of sediment was diverted into TC, 0.489 M_T (7% of sediment inflow) was deposited in TC.

The channel morphology is controlled by dependent and independent factors such as hydraulics (velocity, roughness, shear stress, runoff, rainfall), sediment load (SSL, BL), channel characteristics (slope, size, shape, pattern), and bed and banks' materials (characteristic, composition) as described by many researchers (Morisawa 1985). We have collected the seven parameters based on the data available to ensure data effectiveness and consistency. Rosetta software was used to determine the dominating factors responsible for river shape changes. We established a prediction process and applied a method of attribute selection, data collection and processing, attribute discretization, reduction, and rule extraction. The test period was the year 2019, and the raw data with discretization and reduction is given in the Supplementary information (Tables S2 and S3). The condition attributes such as C1, C2, C3, C4, C5, C6, C7 represent the runoff, velocity, shear stress, roughness, slope, rainfall, and sediments. The decision attribute (D1) is erosion and deposition (Table 2). The attributes were reduced using the genetic algorithm with Rosetta software help, and minimum attributes [C1, C2, C3, C5, C7] were obtained. The RST showed that velocity, shear stress, slope, runoff, and sediment load factors are major contributors to the river shape changes (Table 3). For brevity, detailed discussion regarding how RST work is not elaborated upon here but is presented in Yan et al. (2020). The findings demonstrate that the RS reduction of unnecessary attributes could refine the sample structure and make it more accurate to classify dominating factors in the Indus River morphology. Thus, RST is a practical method and intelligently predicts the dominating factors based on the available data. In the future, more riverine morphology attributes could be selected for further study.

The key findings of this study, that is, that the alterations of both morphology and sediments have occurred during climate change-based episodic events, are the starting point for evaluating flood risk. The principle through which peak riverine runoff impacts the population can be deemed in two vital categories: first, the loss of riverine banks by erosive activity and second through the consistent flood events of residential and agricultural producing areas (Sharma et al. 2010). Bank erosion can lead to significant displacement, but supposing that high sand deposition is prevented, it does not certainly result in long depopulation. Thus, these research findings are considered from the standpoint of planning flood recovery. They describe the ability for detailed flooding, risk visualizations, and prioritizing flood prevention and land rehabilitation.

The bedload analysis highlights the relationship between flood duration and erosive activity. The erosive action is more on the HF events compared with the LF events. The runoff-SSC and runoff-sediment load dynamics provide a baseline for correction and an economical solution to the Sukkur barrage problems and guarantee the satisfactory barrage operation. RST was used to determine the climatic dominating factors which affect the river morphology. This is the first study of its nature that used RS to quantify the dominating factors, reduce unnecessary attributes, and improve the classification of more accurate influencing factors for riverine morphology.

The purpose of this study is to evaluate the influence of climate change-based flood events on the river channel incision, planform changes, and effective runoff for sediment load movement in the LIR. First, we examined the river morphology based on the multi-temporal images of Landsat from 1986 to 2020. Second, morphometric parameters, such as sinuosity index, migration rate of river centerline, braiding index, river width and length were explored. Third, sediment dynamics of the Sukkur barrage was estimated based on the observed daily sediment fluxes. Finally, RST was employed to understand the dominating factors that cause river shape change.

• The quantification of planform changes behavior of LIR showed that: (i) morphometric parameters, such as sinuosity index, surface area, channel length and width from 1986 to 2020 exhibited a generally increasing trend; (ii) BI value for Guddu reach varied from 3.47 to 7.18, and BI for Sukkur reach varied from 2.11 to 4.92; (iii) actual width of the study reaches varied from 2.1 to 12 km; (iv) no erosive activity for banks occurred for peak runoff value of <5,880 m³s⁻¹; (v) median size (D₅₀) of bed material during the HF period was 0.101–0.206 mm and during the LF period was <0.0625 mm.

• Observation-based sediment flux illustrates that intense bed incision occurred during the HF period because of sediment carrying capacity. Bed incision is greater for Sukkur than Guddu reach, because of rapid increment in its gradient with obvious meandering and oxbow lakes' formation. The average SL was 1,379 ppm, 1,551 ppm, and 1,402 ppm during the HF period, while the average SL was 785 ppm, 601 ppm, and 799 ppm during the LF period at AC, TC, and RP at Sukkur station. However, SSL consisted of 3–13% of the sand fraction (+62 micron) and 87–97% of fine sediment (−62 micron). The total sediment inflow was 14.108 M_T, 6.721 M_T, and 8.085 M_T at AC, TC, and RP. This indicates that 0.695 M_T of sediment was eroded from the reach from the start of AC to its bifurcation into TC and RP channels. Although 6.721 M_T of sediment was diverted into TC, still 0.489 M_T (7% of sediment inflow) was deposited.

• RST approach was applied to determine the dominating factors which affect the river shape changes. We established the prediction process, and rules were removed based on basic filtering with the condition that the coverage and stability be 0.05 and 0.5. Based on the attributes' selection, RST reduced the seven affecting morphological parameters to five parameters: velocity, shear stress, slope, runoff, and sediment.

• In particular, this research is based on flood-based events while further research requires to be done based on multi-temporal scales: (i) decadal-based and (ii) flood-based scale. The study deals with the sediment discharge data of a particular year, while more data for several years could be helpful to establish the relationship between SSL-water runoff. The SSL-water runoff relationship would be helpful to understand the co-evolution of river hydro-geomorphology, ecology, and sedimentary-morphology behavior (Dewan et al. 2017). The more riverine parameters such as width/depth ratio, bank material, and channel characteristics could be helpful to understand the role of major riverine factors for river shape changes.

• Moreover, the finding of this research provides the basis for consideration of flood risk management and the impact on the population exposed to the river banks. For the policymakers and environmentalists, these evidence-based solutions have a high potential for autogenic restoration, riverine regulations, and navigation safety. These findings address the potential for developing the flood vulnerability and risk graphs, and prioritizing flood protection and the establishment of temporary refugia in contrast to permanent displacement connected to the erosive land loss. This study will be helpful for water managers to understand the riverine banks' erosion processes and improve river banks' management. This morphological study has societal benefits to protect communities and croplands by artificial channeling, which will be useful for flood mitigation controls. The results are quite encouraging and illustrate that disruptions of their morphology-sedimentary dynamics are not inevitably irreversible.

This work was supported by the National Natural Science Foundation of China (Grant's # 41671455, 51879239); the major consulting project of Chinese Academy of Engineering (Grant# 2020-ZD-18-5); and the Think Tank Research Projects of Zhengzhou Collaborative Innovation with major funding (Zhengzhou University) (Grant# 2019ZZXT01).

All relevant data are included in the paper or its Supplementary Information.