Time-series records of surface water salinity were collected by the Bangladesh Water Development Board (BWDB) from a monitoring network of 119 tidal stations used to monitor surface water salinity. The BWDB reports surface water salinity in terms of electrical conductivity (EC in units of μS/cm) and chloride concentrations (Cl in units of mg/L) for both high and low tides. Data were collated over three decades between January 1990 and January 2019. The frequency of the collated data is inconsistent and varies from weekly, bi-weekly, to monthly; data gaps also occur in the dataset. Following a careful review of surface water salinity times series records, 86 surface water salinity monitoring stations (Figure 2A) were selected for further analysis. A district headquarters map is also provided in Figure 2B to relate patterns and processes revealed by the surface water salinity data to an administrative level and facilitate risk-informed policy making. The interrogated dataset contains 39,464 observations of EC. Between 1990 and 2000, the number of stations (~10) and continuity of records are much more limited relative to the subsequent two decades and especially between 2001 and 2017 (Figure 3). Further, data gaps are consistently observed during the peak of the monsoon season (July to October) and are attributed to the challenges of conducting field measurements during monsoonal flooding in Bangladesh. We report salinity observed at BWDB’s surface water monitoring stations across coastal Bangladesh which largely comprise rivers but exclude shrimp-farming ponds and inland (closed) water bodies.

For the purpose of integrated coastal zone management, the coastal region of Bangladesh is delineated into two distinct sections: “exposed” and “interior” coasts (Figures 2A,B) based on their degree of interaction with the sea in a north–south direction such as tidal fluctuations, soil salinity, surface water or groundwater, and their vulnerability to cyclones and storm surges (Uddin and Kaudstaal, 2003). Areas situated to the north of the coastal boundary are classified as “inland.” Water Resources Planning Organization (2001) divided the coastal area of Bangladesh into four hydrological units (Figures 2A,B) that include South West (SW), South Central (SC), South Eastern (SE), and Eastern Hills (EH); SE and EH parts are collectively referred to as the SE coast in this study.

Exploratory (scoping) analyses of surface water salinity data were employed to assess its spatiotemporal structure across coastal Bangladesh. Applied statistics include long-term mean, standard deviation, and decadal means. Long-term mean and standard deviations were calculated between the 1991–2019 periods at the mean tide (i.e., arithmetic means of high tide and low tide salinity) to explore the spatial structure of the surface water salinity dataset. Similarly, decadal means were calculated over the last three decades (i.e., 1990–1999, 2000–2009 and 2010–2019) to assess changes in spatiotemporal patterns. Surface water salinity for individual stations was also plotted to identify erroneous values and assess trends.

The analysis of salinity trends focused on a subset of the surface water salinity dataset from 2001 to 2017 during which there are fewer data gaps. Linear regression is a parametric model commonly applied to detect trends in time series data but assumes that analysed datasets are normally distributed (McBean and Motiee, 2008). Many hydrological variables do not follow a normal distribution (Viessman et al., 1989) but show a pronounced “right skewedness” as is evident from the surface water salinity dataset in coastal Bangladesh (Supplementary Figure S2). Consequently, a non-parametric measure of trend, Sen Slope Estimator (Sen, 1968), was also employed. For the calculation of Sen Slope Estimator, data gaps were infilled using “MissForest”, a random forest imputation algorithm for missing data and converted to a uniform monthly time series (Stekhoven and Bühlmann, 2012). Additionally, to identify trends and seasonal patterns in surface water salinity, we applied another non-parametric technique, seasonal-trend decomposition procedure based on Loess (STL) (Cleveland et al., 1990; Shamsudduha et al., 2009). The STL method detects nonlinear patterns in long-term trends that cannot be assessed through linear trend analyses. It comprises a series of smoothing operations with variable moving window widths chosen to extract different frequencies within a time series and can be regarded as an extension of classical methods for decomposing a series into its individual components.

We further applied a partitional (non-hierarchical) clustering method (K-means) with a pre-assigned number of clusters as it is computationally more efficient and faster than hierarchical clustering with high-dimensional (i.e., number of features) time-series data (Aghabozorgi et al., 2015; Javed et al., 2020; Shamsudduha et al., 2022). K-means clustering is fundamentally different from hierarchical clustering in that it is a form of partitional clustering where data are divided into K number of partitions based on a priori knowledge about the optimum number of clusters in the dataset. To determine the optimum number of clusters, we applied several techniques to the groundwater-level dataset ranging from the sum of squared error (SSE) or “elbow” method to the Caliñski and Harabasz (1974) criteria. These methods indicate that the optimum number of clusters is 6 by the “elbow” method, and between 2 and 6 clusters by the Caliñski–Harabasz index. Here, we considered 6 clusters for grouping surface water salinity monitoring records from 86 stations. To bring further homogeneity to the automatically performed clustering, we reviewed all the time-series data in 6 clusters and moved some stations between clusters. The supervised re-classification of the original clusters was performed by objectively looking at the individual time series plots and reconciling patterns revealed by the composite cluster plots. It is based on the proximity and magnitude surface water salinity monitoring stations and works as a proxy indicator to infer about key processes that governs surface water salinity regime at a local level such as complex hydro-morphology and anthropogenic factors. These localised process and patterns enable us to comment about the major surface water salinity pathways at the regional level.

Analytical outcomes from the analyses were mapped using a Geographic Information System (GIS). The confluence of the Rivers Ganges, Brahmaputra and Meghna in the north, Bay of Bengal in the south, and the national boundary between India and Bangladesh in the West and coastal zone boundary in the west demarcate the area of interpolation used in this study. Here, we applied the Empirical Bayesian Kriging (EBK) interpolation algorithm available in the ESRI ArcGIS desktop environment, which is a geostatistical interpolation method that automates the most difficult aspects of building a valid kriging model (Krivoruchko and Gribov, 2019). Other kriging methods in Geostatistical Analyst in ArcGIS require manual adjustments of parameters to generate accurate results whereas EBK automatically calculates these parameters through a process of sub-setting and simulations. Empirical Bayesian Kriging differs from other kriging methods by accounting for the error introduced by estimating the underlying semi-variogram. This approach has an advantage over other kriging methods in that it calculates the semi-variogram from known data locations and uses this single semi-variogram to make predictions at unknown locations. Semi-variogram analysis of surface water EC data in this study shows a strong spatial dependence up to a distance of 50 km suggesting that the EBK algorithm is suitable for interpolation.

The spatial distribution of surface water salinity defined by long-term mean EC is presented in Figure 4. For contouring purposes, long-term mean surface water salinity is divided into five categories based on EC: (i) very low—freshwater (≤1,500 μS/cm), (ii) low (1,501 to 5,000 μS/cm), (iii) moderate (5,001 to 10,000 μS/cm), (iv) high (10,001 to 15,000 μS/cm), and (v) very high (>15,001 μS/cm). Surface water salinity is consistently high in the southwest (SW) coastal region (e.g., tidal rivers in Khulna, Satkhira and Bagerhat districts). Here, moderate to high surface water salinity is observed in the exposed coastal part of this region whereas the interior and inland coastal area to the north shows very low to low salinity. The surface water salinity contour of 5,000 μS/cm reaches up to 100 km inland, just north of Khulna (Figure 4). The highest surface water salinity (14,500 μS/cm) observed in the SW coastal region is in the River Shibsa in Koyra, ~40 km southwest of Khulna. In contrast to the SW coastal region of Bangladesh, very low to low surface water salinities are observed in the south-central (SC) coastal region and include tidal rivers in Barisal, Barguna, Patuakhali, Priojpur, and Bhola districts where surface water salinity is predominantly below 1,500 μS/cm (i.e., freshwater), except for one station located near the SW coastal boundary.

Surface water salinity in the southeast (SE) coastal region is higher and more variable, influenced by some extreme values (Figure 4). Very high surface water salinity is observed around Chittagong and Cox’s Bazar where monitoring stations are close (<25 km) and well connected to the sea. Low to moderate surface water salinity is observed in the interior part of Chittagong and Hill Tracts and exposed part of Noakhali and Laxmipur. Surface water salinity monitoring stations located near the Meghna basin outlet at Naokhali and Laxmipur (Figure 4) show low salinity (1,500 to 5,000 μS/cm).

At most surface water salinity monitoring stations in the coastal region of Bangladesh, mean EC values are below 10,000 μS/cm, except for three stations (Supplementary Figure S3A): River Shibsa in Koyra, Khulna (14,500 μS/cm); River Sangu in Banshkhali, Chittagong (20,400 μS/cm) and Maheshkhali Channel in Maheshkhali, Cox’s Bazar (32,900 μS/cm). The latter two stations are located very near the open sea and their EC values are close to that of seawater (~50,000 μS/cm). An increasing trend in the lower 5th percentile is also observed in Maheshkhali Channel and River Shibsha (Supplementary Figure S3B).

Statistical clustering identifies six distinct patterns (i.e., supervised clusters) in the time series of observed salinity changes at 86 stations in coastal Bangladesh (Figures 5A,B; Supplementary Figure S4). These statistically characterised time series patterns demonstrate a clear spatial structure (Figure 6 and Table 1A).

Clusters 1 (N = 5) and 2 (N = 6) describe observed changes in surface water salinity in the SW interior coast. Here, the mean salinity is generally low (mean= ~450 μS/cm; SD = ~200 μS/cm) and varies minimally (Figures 5A,B). A small but distinct positive deflection (i.e., rise) in salinity is observed in 2007 in both clusters. Time-series records in cluster 1 show evidence of recovery from this rise in salinity whereas time series records in cluster 2 do not (Figures 5A,B).

Cluster 3 comprise 28 time-series records from monitoring stations in the SE exposed and interior coastal regions (Figures 2, 5, 6). Mean salinity in these records is almost an order of magnitude higher (mean = ~3,100 μS/cm; SD = ~1,700 μS/cm) with a high seasonal range (~80,000 μS/cm) and a step-change rise in salinity in 2007 (Figures 5A,B) affecting seasonal lows and highs such that the amplitude of seasonal oscillations remain consistent (Figures 5A,B).

Cluster 4 (N = 8) comprises time series records with a mean salinity of ~1,000 μS/cm (SD = 1,200 μS/cm) that is found in a transition zone in the SW (Figure 6) between the exposed and interior coastal regions (Figure 2). Rises in surface water salinity are observed in 2007 and 2014 but there is also evidence of a partial recovery to lower salinities as well as a shift to more frequent higher seasonal salinities (Figures 5A,B).

Cluster 5 (N = 12) includes stations found in the SW coastal region, just north of the Sundarbans (Figure 6), where the mean surface water salinity is brackish (~6,500 μS/cm; SD = 4,500 μS/cm). Here, there is a noticeable increase in peak surface water salinities in 2007 and 2014; this peak salinity continues to rise, increasing the seasonal amplitude in observed salinity as seasonal low salinities remain effectively constant (Figures 5A,B).

Cluster 6 (N = 27) is found in the south-central coastal area (Figure 6) in both the exposed and interior coastal regions (Figure 2A). As noted above (Figure 4), mean surface water salinities are low, consistent with freshwater (mean = ~750 μS/cm; SD = 600 μS/cm). Rises in surface water salinity are visible in 2007, 2011 and 2017 (Figures 5A,B; Supplementary Figure S5); seasonal oscillations in surface water salinities recover quickly but there is evidence of a trend toward rising salinities over the time series as reported above.

STL decomposition of time-series records of surface water salinity reveal the respective contributions of seasonality, trend and residuals to the overall variance observed in the coastal region of Bangladesh (Figure 7). Variance in surface water salinity in the SW exposed coastal region (cluster 5) is dominated (>80%) by seasonality. Seasonal components continue to explain some of the variance in salinity observed in the SW interior coastal region (cluster 4) just north of Khulna where monitoring stations are connected (proximate) to major distributaries. In SW inland coastal region (clusters 1 and 2) remote from major distributaries, trend (and residual) components explain almost all of the observed variance in surface water salinity. In the SC coastal region (cluster 6), much (50 to 60%) of the variance in surface water salinity is explained by the trend component. In the SE coastal region (cluster 3) no consistency is observed in the dominance of seasonal, trends or residual components.

Surface water salinity time-series exhibit seasonal regimes of varying amplitudes across the 86 stations. For clusters 1, 2, 3 and 6 in the SW inland, SC, and SE coastal regions, surface water salinity is highest during the dry season between February and April and lowest during the monsoon from June to August (Supplementary Figure S6), consistent with simulations by Bricheno et al. (2021) and Bricheno and Wolf (2018). The amplitude of these seasonal variations in surface water salinities is small (<30%) relative to clusters 4 and 5 in the SW exposed and interior coastal regions (>100%) where peak salinities occur later in the dry season in April and May and lowest salinities are observed in October and November.

During the post-monsoon season from September to November (Figure 8A), surface water salinity across the entire coastal region is below 10,000 μS/cm and remains fresh (<1,500 μS/cm) in the SC region. During the dry season, surface water salinities rise considerably in the SE and SW exposed coastal regions, the latter just north of the Sundarbans (Figure 8B). Here, the contour between fresh and brackish water migrates between 20 and 40 km northward during the dry season (Figure 8C). In terms of magnitude, surface water salinity increases 3 to 5 times that observed in the wet season. In the SE, surface water salinity doubles at some monitoring stations near the sea whereas in the SC coastal region surface water remains fresh (<1,500 μS/cm), though the spatial extent of this freshwater zone shrinks slightly during the dry season.

Spatiotemporal changes in surface water salinity depicted in Figure 9 show how surface water salinity has evolved over the last two decades (2000–2019); the very limited number and distribution of observations from 1990 to 1999 prevent meaningful representations of salinity prior to 2000 (Supplementary Figure S5). Two major changes that can be observed by comparing Figures 9A,B are: (i) a generalised increase in salinity in the SW coastal region; and (ii) shrinking of the freshwater (<1,500 μS/cm) “pocket” in the SC coastal region.

Long-term trends in surface water salinity over the last two decades 2001–2017 are mapped in Figure 10 and summarised in Table 1A. Overall, there is a rising trend of surface water salinity in the coastal region. Moderate to high positive trends (+100 to +450 μS/cm per year) in surface water salinity are observed in the exposed SW coastal region up to Khulna and parts of the SE coastal region near Chittagong. Lower positive trends in surface water salinity (+1 to +100 μS/cm per year) are observed in the rest of the coastal region including the SC region; marginally negative trends (i.e., reducing salinity) are computed in the interior and inland SW coast north of Khulna and western part of SC coast south of Comilla; these are thought primarily to derive from uncertainty and data gaps in the times series observations. Substantial increases in salinity are observed primarily in the SW coastal region, north of The Sundarbans. Highest trends are observed near Khulna (e.g., stations SW258—River Shibsa in Paikgacha, SW28—River Bhadra in Dumuri) and further south (e.g., SW107.2—River Gorai-Madhumati in Sarankhola Bagerhat, SW 108—River Gorai-Madhumati in Barguna, Patharghata). Most stations in the interior coast show marginally increasing (<70 μS/cm per year) salinity; marginally increasing salinity is also observed in the south-central exposed coastal region.

From 2006 to 2009, five tropical cyclones hit the coastal region of Bangladesh including severe cyclones Sidr in 2007 and Aila in 2009 (Supplementary Figure S4; Figure 11). Notwithstanding sharp increases in surface water salinity observed in 2007 within each cluster across the coastal region, these rises preceded the landfall of cyclone Sidr in mid-November of 2007. They are pronounced in stations within clusters 1, 2, 4 and 6 where mean surface water salinities are comparatively lower than stations within clusters 3 and 5 (Table 1B). Figure 11 compares mean surface water salinity between the period of 2001 to 2005 and that of 2006 to 2010. An increase in the magnitude of surface water salinity and area with a mean salinity exceeding that of freshwater (EC = 1,500 μS/cm) is observed between these periods in the SW coastal region where the freshwater contour migrated between 11 and 28 km northward and between 4 and 22 km eastward.

Variations in river discharge (Q) across the coastal region of the Bengal Basin provide a compelling explanation for the spatial structure in observed surface water salinity in Figure 4. Surface water salinities are higher and more variable in the SW exposed (mean: 6,500 μS/cm, SD: 4,500 μS/cm) and interior (mean: 1,000 μS/cm, SD: 1,200 μS/cm) coastal regions, north of the Sundarbans, relative to the freshwater conditions that persist in the SC (mean: 750 μS/cm, SD: 600 μS/cm) coastal region (Table 1A). Freshwater fluxes to the SW coastal region derive from distributaries within the moribund delta draining the River Padma (see locations of gauging stations on Rivers Gorai (SW99) and Mathavanga (SW205) in Figure 6) with a combined mean Q (1967–2016) of ~1,200 m³/s (Supplementary Figures S7–S11). In contrast, freshwater discharges to the SC region within the active delta are supplied by the combined contributions from the Rivers Padma (Ganges), Jamuna (Brahmaputra), and Meghna with a combined mean Q of ~37,000 m³/s (Sarker et al., 2003; Palash et al., 2018). Surface water salinities are also high in the SE exposed coastal region (mean: 9,730 μS/cm, SD: 5,760 μS/cm, Table 1B) where the magnitude of freshwater fluxes is constrained by smaller, localised catchment areas (e.g., River Karnaphuli) that drain the Chittagong Hill Tracts to the Bay of Bengal.

Very considerable seasonal variations in the combined discharge of Rivers Padma, Jamuna and Meghna (flood Q: ~140,000 m³/s, dry-season Q: ~5,000 m³/s; see Figure 3 and Table 1 in Palash et al., 2018) occur and explain, in part, observed seasonal variations in the salinity of surface waters in the coastal region, which can vary from 1,000 to >20,000 μS/cm at individual stations (e.g., SW250 and SW1 in Figure 6B; Supplementary Figure S3). Across the coastal region during the monsoon season, observed salinity at all stations but one lie below 10,000 μS/cm (Figure 5A) whereas during the dry season mean EC values at seven stations exceed 10,000 μS/cm (Figure 5B). Dry-season (5th percentile) river discharge (~55 m³/s) to the SW coastal region (SW99, Supplementary Figures S7, S8) reduces to just 1.5% of the mean discharge (1976 to 2015); this seasonal reduction in flow coincides with a shrinking of the SC freshwater zone and seasonal (northward) migration of the freshwater contour (1,500 μS/cm) by ~20 to ~40 km (Figure 5C). Across coastal Bangladesh, observed seasonality in EC is greatest in the SW exposed and interior coastal regions, represented by clusters 4 and 5 (Supplementary Figure S6), and lags behind other coastal zones (Figure 8). It is here within the southern part of the moribund delta where distributaries are least well connected to seasonal freshwater fluxes of the Rivers Padma, Jamuna and Meghna, and their distributaries.

Observed reductions in dry-season (5th percentile) discharge of the River Padma and its distributaries including the River Gorai (Supplementary Figures S7, S9, S10; Ali and Hasan, 2022), the principal dry-season source of freshwater to the SW coastal region (Rahman and Rahaman, 2018), are expected to have enhanced the seasonal (northward) migration of salinity. Hydrological records at Hardinge Bridge (SW90 in Figure 6) on the River Padma reveal a more than 50% decrease from 2,080 m³/s (1963–1975) to 910 m³/s (1976–2015) in the dry-season discharge (Supplementary Figures S7C,D) following the commissioning of the Farraka Barrage in 1975 (Mirza, 1998; Rahman and Rahaman, 2018; Sherin et al., 2020; Feist et al., 2023). The barrage, 18 km upstream of the River Padma (Ganges) from the border between India and Bangladesh, diverts flow from the River Padma (Ganges in India) to the River Hooghly in India. As argued by Goes et al. (2021) and is detectable from time series records of river discharge (Supplementary Figures S7C,D), dry-season flow of the River Padma declined somewhat steadily to ~500 m³/s from 1975 to the mid-1990s; implementation of the 1996 Ganges (Padma) Treaty between India and Bangladesh then led to a partial recovery in the mean dry-season flow to ~1,080 m³/s (1996 to 2015). The reduction in the dry-season flow of the River Padma disproportionately affected downstream distributary discharges including the River Gorai (SW99, Supplementary Figure S8), which occasionally ceased to flow during the dry season between 1988 and 1998 (Monirul Qader Mirza and Sarker, 2004); mean estimated dry-season flow 1988 to 1992 was ~1 m³/s. This non-linear reduction in river flow has been attributed to both the development of a sandbar inhibiting flow at the mouth of the River Gorai and dry-season extraction of surface water to supply the Ganges Kobadak Irrigation Project (Goes et al., 2021).

Trends in surface water salinity exist within the dynamic seasonal regimes observed in the coastal region of Bangladesh. The freshwater zone in the SC exposed and interior coastal regions has shrunk primarily from the moribund delta in the SW exposed zone (Figure 9C); here the lateral extent of the freshwater zone reduced by ~13 to ~16 km between the periods of 2000–2009 to 2010–2018. Within the active delta, proximate to contemporary freshwater fluxes from the Rivers Padma, Jamuna and Meghna, the lateral reduction in SC freshwater zone is lower, between 4 and 9 km over the same time periods. Notwithstanding small rising trends (Sen slopes) in EC detected most notably at monitoring stations in the exposed SW and SC coastal zones (Table 1B and Figure 10), the freshwater contour in SW interior coastal zone remained largely unchanged between the periods of 2000–2009 to 2010–2018 (Figure 9C). Similar to reductions in dry season discharge, monsoonal flood (95th percentile) discharges of the Rivers Padma and Gorai decreased (Supplementary Figures S7A,B, S9) with a disproportionate (non-linear) 32% reduction in the mean flood discharge of the River Gorai between the periods 1976 to 1995 and 1996 to 2015, relative to a 12% reduction in the flood discharge of the River Padma over the same two periods. Over this latter period (1996–2015), the annual flood discharge of the River Gorai (mean: ~3,030 m³/s, Supplementary Figure S5) declined by ~50 m³/s per year (p-value: 0.04).

The influence of sea-level rise (SLR) on surface water salinity in the coastal region of Bangladesh is itself affected by changes in land elevation due to variable sedimentation rates and subsidence (Steckler et al., 2016; Becker et al., 2020; Paszkowski et al., 2021; Hanebuth et al., 2022; Steckler et al., 2022). Mean rates of SLR derived from satellite altimetry in coastal Bangladesh are consistent with and, in places (e.g., SE Bangladesh), slightly exceed the global mean of between ~3.3 and ~3.6 mm/year (Akter et al., 2016; Ubelmann et al., 2021). Observed SLR varies from ~3.0 to ~3.5 mm/year in the SW coastal region to ~3.5 to ~4.0 mm/year in the SC, and ~4.0 to ~5.0 mm/year in the SE (Figure 6). Subsidence rates, recorded from in situ observations (Steckler et al., 2022), also vary over the coastal region from 5.5 to 8.0 mm/year in the SW exposed and interior coastal region (moribund delta) to between 3.7 and 4.6 mm/year in the exposed SC coastal region.

The combined influence of SLR and subsidence, the twin drivers of the northward intrusion of saline waters from the Bay of Bengal to the delta, is highest in the SW coastal region where rising trends in surface water salinity in the exposed and interior coasts are greatest (Table 1B and Figure 10); it is also in the SW exposed coastal region where land elevation is lowest (e.g., Pinchoff et al., 2019). Trends in surface water salinity in the exposed and interior coasts of the SC region are lower where combined rates of SLR and subsidence are lower. SLR is greatest in the SE coastal region but its impact on surface water salinity is constrained by a higher land elevation.

Sea level and subsidence vary seasonally (Ubelmann et al., 2021; Steckler et al., 2022) with lower values occurring in the dry season (e.g., March) when the northward migration of surface water salinity peaks (Figure 8). As the Bengal Basin is a tide-dominated delta (mean tidal range exceeds 4 m, Bricheno et al., 2016), it is also during this period when tidal hydrodynamics can serve to drive seawater furthest inland (Bricheno et al., 2021). Although spatial variability in the magnitude of observed trends in surface water salinity in the coastal region (i.e., higher salinity trends in the SW than the SC coastal region) is consistent with observed trends in the combined influences of SLR and subsidence, substantial changes in surface water salinity in the SW coastal region between periods 2000–2009 and 2010–2018 are not observed (Figure 9C) as noted above.

Salinity rises observed in the dry season preceding the 2007 monsoon, follow in 2006 the lowest monsoonal river discharge ever recorded to that date at River Gorai (SW99, Supplementary Figure S7A) and the lowest monsoonal river discharge recorded at River Padma (SW90, Supplementary Figure S7B) since 1989. Annual rainfall in the Bengal Basin in 2006 was also the lowest recorded to that date since 1994 (Supplementary Figure S12). In the SW coastal region in 2014, sharp rises in surface water salinity are again observed in clusters 4 and 5 (Figure 5; Supplementary Figure S4) independent of the landfall of any tropical cyclones. These increases occur during the dry season that follows monsoonal discharge at River Gorai (Supplementary Figure S7A) that is even lower than in 2006 and the lowest in the entire time series from 1972 to 2016.

Although these two intervals represent pre-cyclone and cyclone periods (Figure 11C), the disconnection between the timing of cyclones and salinity rises means that the salinity incursion between these two time periods is better explained by a 20% reduction in the median monthly discharge of the River Gorai. Reductions in dry-season (February to April) river discharge of ~26% not only influence the salinity of intruding water from the Bay of Bengal (Chaitanya et al., 2015) but may also be driven, in part, by freshwater capture through groundwater-fed irrigation (Sherin et al., 2020; Shamsudduha et al., 2022). In the SE coastal region, increases in the salinity of surface water are largely limited to near-sea areas and, with few exceptions, mostly unchanged between these two periods. Full to partial recovery from sharp rises in salinity is observed in clusters 1, 2, 4 and 6 (Figure 6), which are better connected to more voluminous seasonal river discharge than clusters 3 and 5 (e.g., Supplementary Figure S10: cluster 5—SW105, 107.2, 258, 28, and 243) located in the SW and SE coastal regions. The precise processes by which these episodic rises in surface water salinity persist, require further investigation (Ayers et al., 2017). The impacts of tropical cyclones and storm surge inundation on river-water salinities can be short-lived as our observations show. However, the impact of tropical cyclones may be more substantial for inland water bodies especially within polders, in case of their breach, and coastal water-logged wetlands due to inhibited drainage and evaporation effects.

Human interventions such as polders and upstream diversions alter sedimentation patterns in the coastal region of Bangladesh. These are largely thought to interrupt sedimentation inside polders, reducing subsidence but also a relative elevation through impeded deposition (Auerbach et al., 2015). Subsequent breaches in the form of storm surges and cyclones inundate these areas with brackish water from the surrounding tidal channels and deposit salt-rich sediments, leading to the salinisation of surface water bodies in polders. Shrimp farming in southwestern Bangladesh during dry season also leads to increases in the salinity of water and soil around shrimp ponds (Ayers et al., 2017; Mahtab and Zahid, 2018). The impact of these human activities on surface water salinity monitored outside of polders and remote from shrimp ponds remains unclear. As cited above, dry-season irrigation of Boro rice through groundwater abstraction reduces baseflow contributions to surface water through freshwater capture (Shamsudduha et al., 2022), lowering river discharge, as reviewed above in section 5.1, that is most visible during the dry season. Use of cross-dams in tidal distributaries to trap fresh surface water for dry-season irrigation also serves to increase measured dry-season surface water salinity (Rashid et al., 2013).

The outcomes presented here are rooted in the analysis of the largest number of time series observations of surface water salinity in the Bengal Basin to date. Uncertainty in these observations and their processing exist and include most notably:

Seasonally occurring data gaps: During the monsoon season in Bangladesh, which usually peaks between July and October, gaps can occur in observations due primarily to the logistical challenges of collecting field data during severe flooding. Missing data were imputed using a Random Forest algorithm, MissForest, to enable the computation of Sen slopes (trends) as these cannot be calculated if there are gaps in the dataset.

Geospatial interpolation of point data: This process involves the estimation of values at unsampled locations based on the values measured at a finite set of points. The reliability of the interpolated values depends on the quality and representativeness of the original data as well as the interpolation method used. We used the Empirical Bayesian Kriging (EBK) for interpolation as it calculates the semi-variogram from known data locations and uses this single semi-variogram to make predictions at unknown locations. Semi-variogram analysis of surface water EC data in this study shows a strong spatial dependence up to a distance of 50 km, suggesting that the EBK algorithm is suitable for interpolation.