Effects of hydrological regime and land use on in-stream Escherichia coli concentration in the Mekong basin, Lao PDR

In the basin of Mekong, over 70 million people rely on unimproved surface water for their domestic requirements. Surface water is often contaminated with fecal matter and yet little information exists on the underlying mechanisms of fecal contamination in tropical conditions at large watershed scales. Our objectives were to (1) investigate the seasonality of fecal contamination using Escherichia coli as fecal indicator bacteria (FIB), and (2) establish links between the fecal contamination in stream water and its controlling factors (hydrology and land use). We present the results of (1) a sampling campaign at the outlet of 19 catchments across Lao PDR, in both the dry and the rainy seasons of 2016, and (2) a 10-day interval monitoring conducted in 2017 and 2018 at three point locations of three rivers (Nam Ou, Nam Suang, and Mekong) in northern Lao PDR. Our results show the presence of fecal contamination at most of the sampled sites, with a seasonality characterized by higher and extreme E. coli concentrations occurring during the rainy season. The highest E. coli concentrations, strongly correlated with total suspended sediment concentrations, were measured in catchments dominated by unstocked forest areas, especially in mountainous northern Lao PDR and in Vientiane province.

Mekong tributaries (Fig. 1). The sampling sites were chosen to ensure a broad geographical coverage of Lao PDR, and to represent a large range of geographical, topographical, and land use features. The sampling sites were also chosen for being logistically accessible from the road, in order to cover the majority of Mekong tributaries in a relatively short time ( Table 1, Table S1).
[E. coli] were found to be highly spatially variable across Lao PDR (Fig. 2), and 71% of the sampling sites displayed [E. coli] equal or greater to the lower detection limit of 38 MPN 100 mL −1 during both seasons. [E. coli] were below the detection limit in central and southern catchments (Nka, Ngn, Xbn) during the dry season and in central catchments (Nka, Xbi, Nhi) during the rainy season.
Moreover, [E. coli] were highly variable along the Mekong river mainstream (Fig. 2c). In the dry season, [E. coli] in the Mekong river ranged between 0 and 520 MPN 100 mL −1 , while it varied between 0 and 57,000 MPN 100 mL −1 in the rainy season. The highest [E. coli] were found at the sampling sites located near urbanized areas around Vientiane (MK_7 and MK_2), and to a lesser extent in the southern station near Pakse (MK_5). The lowest [E. coli] were found in the highlands of northern Lao PDR (MK_1) and in southern Lao PDR (MK_3, MK_4) (Fig. 2c).
In the following statistical analyses on Mekong tributaries, only the Nam Ngum outlet (Nng_1) among the Nam Ngum sampling sites, was taken into account, to avoid the overrepresentation of the Nam Ngum tributary among the data.
Overall, seasonal variations of in-stream [E. coli] in Mekong tributaries were observed at the majority of the sampled sites. [E. coli] were different between the dry and the rainy seasons (p < 0.05), and followed a lognormal Table 1. Description of sampling sites in Lao PDR: names of river, geographical coordinates of sampling sites (i.e., latitude and longitude in degrees, WGS 1984), sampling dates during field surveys in March and July 2016, and regular monitoring from July 2017 to December 2018, and catchment drainage area in km 2 . a Geographical coordinates in degrees (WGS 1984 www.nature.com/scientificreports/ to describe the geomorphological features of these catchments (Fig. 4a). To describe the land use within the watershed, we used dams' reservoir areas (Table S1), and areal percentages of unstocked forest, forest, paddy rice, grassland, urban, water and other agriculture areas in each catchment (Fig. 4b, Table S1), as well as human and livestock population densities per catchment (Fig. 4c,d,  (Fig. 5a, Table S2). The VIPs (Variable Importance in the Projection) for each explanatory variable of both components showed that areal percentages of forest and unstocked forest, [TN], and EC, contributed the most to the model (VIP > 1.5) (Fig. 5c). During the rainy season, [E. coli] was mainly explained by the first component (66%) and to a lesser extent by the second component (19%) (Fig. 5c, Table S2). Areal percentages of unstocked forest, turbidity, and [TSS], were highly influential on the model (VIP > 1.5) (Fig. 5d).
During both seasons, the first component associated [E. coli] with common factors like watershed characteristics including areal percentages of land use classes, population density, catchment area, and physico-chemical parameters ([TSS], [TN], [TPC], and turbidity) (Fig. 5a,b). However, their relative importance changed with the season (Fig. 5c, d).
During both seasons, [E. coli] was positively correlated to unstocked forest percentage area (r = 0.65 in dry season, r = 0.62 in rainy season, p < 0.05) and negatively correlated to catchment area (r = − 0.49 in dry season, r = − 0.70 in rainy season, p < 0.05).
During the dry season, [E. coli] was positively correlated with EC (r = 0.59, p < 0.05) and T (r = 0.47, p < 0.05), and negatively correlated with forest percentage area (r = -0.55, p < 0.05) and DO (r = − 0.45, p = 0.054). During  coli] was over two orders of magnitude and one order of magnitude higher during the rainy season compared to the dry season in both Nsu and Nou, and at MK_17, respectively. Likewise, the seasonal variability of [TSS] in all sampled watersheds is marked by higher values during the rainy season (p < 0.05).
[TSS] followed the same seasonal pattern as [E. coli], increasing over three orders of magnitude in Nsu, over two orders of magnitude in Nou, and over one order of magnitude at MK_17 during the rainy season. The opposite trend was noted for EC dynamics that showed higher values during dry season (p < 0.05). The lowest peaks of EC occurred during rainfall events and the highest values of EC were recorded during the dry season ( Fig. 6). In Nou, Nsu, and MK_17, water level was positively correlated to [TSS] and [E. coli], and negatively correlated to EC (p < 0.05) (Table S3). Likewise, [E. coli] in all three watersheds was positively correlated to [TSS] and negatively correlated to EC (Table S3).

Discussion
In this study, we investigated correlations between various environmental and water physico-chemical parameters and the occurrence of E. coli, as well as the seasonal variability of [E. coli] in the Mekong river and some of its major tributaries in Lao PDR.
Overall  46,47 . A decreasing EC is characteristic of an increasing overland flow contribution, whereas an increasing EC indicates a higher contribution of groundwater 39,45 . During the 2017 and 2018 rainy seasons, EC decreased, suggesting an increase in overland flow contribution to streamflow, further contributing to the dispersion of suspended sediment and of microbial contaminants such as E. coli along hillslopes and downstream. Suspended sediment and washed-off free-living and particle-attached E. coli can deposit in streambeds of rivers before being re-suspended during high discharge events 39,[48][49][50] . During the dry season, [E. coli] can be associated to other in-stream processes such as hyporheic exchange. In fact, during the dry season at baseflow, the stream-groundwater interactions through lateral flow and advective groundwater movement in the hyporheic zone may be responsible for remobilizing bacteria trapped in the porous space of streambed sediments 15,40 . Thus, the seasonal difference in terms of [TSS] and [E. coli] may be partly explained by seasonal streamflow regimes in response to meteorological patterns.
We noted similar trends in terms of positive correlations between [E. coli] and [TSS] during the rainy season of 2016 as compared to the rainy seasons of 2017 and 2018. Sources and dynamics of suspended particles were found to vary highly within a catchment (e.g., eroded sediments, re-suspended streambed sediments 18,39 ). Suspended particles are known to be carriers of adsorbed pollutants, nutrients, and microorganisms like E. coli, partly controlling their transport and fate 18,19,51 . Furthermore, suspended particles are not only vectors for bacterial transport, they could also provide optimal conditions for the survival of adsorbed coliform bacteria by protecting them from ultraviolet radiation and predators [42][43][44] . Bacterial decay rates can be influenced by the physico-chemistry of the stream water (e.g., pH, dissolved oxygen saturation, turbidity, EC and salinity 18,[52][53][54]. Water physico-chemical properties could be key drivers explaining the higher E. coli concentrations in the majority of the sampled tributaries during the rainy season.
In the present work, PLS analysis also helped to identify various correlations between land use classes and [E. coli] in stream water. During both dry and rainy seasons, [E. coli] was positively associated with the percentage of unstocked forest area and negatively associated with the percentage of forest and grassland areas, especially during the rainy seasons. Such correlations are consistent with reports from previous studies in similar tropical watersheds 14,26,33,55 . Higher FIB concentrations were measured in watersheds located in steep mountainous areas of northern Lao PDR as well as in the Vientiane plain, where the dominant land use is unstocked forest, and to a lesser extent, paddy rice and other agricultural land. The unstocked forest percentage areas are largely present in Southeast Asian upland catchments, due to the general pressure towards clearing forests for intensified annual crop production with reduced fallow period and suppressed understory cover 27 33,57,58 . These land use changes can affect soil erosion, water infiltration and overland flow 33,59-61 , due to reduced topsoil cover and reduced soil binding by roots 33,62 . In steep regions, deforestation may also lead to landslides due to the absence of stabilizing roots network and loss of soil stability 58,60 . Unstocked forest areas are thereby more vulnerable to soil erosion and landslides processes 29,63 , especially during the rainy season, increasing overland flow loaded with soil particles and attached FIB that end up in the downstream river.
In contrast, southern Lao PDR watersheds (Nka, Nhi, Xbi), characterized by high percentages of forest areas, had the lowest contamination during both seasons. Areas dominated by forested watersheds with understory cover have been shown to have high infiltration rates, low soil erosion, and high contaminant trapping efficiencies 64,65 . Land cover and management practices can be key factors controlling runoff production and surface soil erosion, and thus FIB contamination levels of rivers 52,66 .
In our PLS analyses, negative associations between [E. coli] and dams' reservoir area were noted during both seasons, yet they were not significant at the 0.05 level in our study case. The impact of dams on FIB could be different in each catchment, depending on dams' reservoir area, as well as on the distance between dams and sampling sites and on potential FIB sources in-between. Many studies focused on the potential hydrological effects of dams in the Mekong basin 67 . An important decrease in suspended sediment loads was noted in many tributaries following the construction of dams, for instance a 50% reduction at Pakse where the average suspended loads decreased from 120 to 60 Mt yr −168 . Along with flow alteration and sediment trapping, dams could also impact FIB fate and transport.
Escherichia coli sources vary greatly across Lao PDR, depending on human and animal density, as well as on the presence or absence of an operational wastewater collection system. It has been shown that mammalian presence closely influences the river's microbiological quality, particularly in rural areas of developing countries lacking sanitary infrastructure 15,24 . However, our study did not show any significant correlation between human densities and measured downstream [E. coli] in either the dry or the rainy seasons. Livestock densities were weakly correlated with [E. coli] during the rainy season (r = 0.45, p = 0.054). This can tentatively be ascribed to several reasons.
On the one hand, the population of Lao PDR is unevenly distributed across the country. About 70% of the population lives in rural areas. Additionally, there are significant urban-rural disparities in terms of access to improved sanitation facilities. Thus, some rural watersheds, although less densely populated, are more exposed to fecal contamination through point sources (direct release of untreated wastewater into river stream), as compared to other populated urban watersheds equipped with wastewater treatment systems. However, despite incremental improvements of the sanitation system in Lao PDR, open defecation remains a major issue, estimated to concern 32% and 45% of the overall and rural populations, respectively 69 . Along with the presence of wild animals or livestock, open defecation is a diffuse source of microbial pathogens, transferred to the stream with surface runoff, and contributing to the fecal pollution of rivers. Moreover, unstocked exploited forests in watersheds with low mammalian presence can be highly frequented by workers during specific periods and by villagers for their domestic needs, adding to the complexity of determining diffuse source inputs. On the other hand, the absence of strong correlations between human and livestock densities and FIB concentrations may also be due to the distance between primary sources (human and livestock) and water streams. This distance, in turn, affects survival rates of transferred FIB, which get increasingly exposed to environmental conditions as primary sources and streams are more distant from one another. During hot and dry periods, less favorable conditions for microbial development (e.g. more sunlight, less nutrients) might increase opportunities for their die-off 14,18,70 . In addition, longer transfer times from hillslopes to rivers due to disconnected flow paths might result in lower [E. coli] in downstream rivers due to the sedimentation of FIB-bound particles and FIB decay 55  (e) total rainfall recorded one week pre-sampling in March 2016 (mm week −1 ); (f) total rainfall recorded one week pre-sampling in July 2016 (mm week −1 ). Geographic coordinated system: WGS 1984, latitude and longitude in degrees. Altitudes of highest and lowest points in meters above mean sea level, from SRTM 90 m. Local populations and livestock per district were taken from the Lao PDR Population and Housing Census 2015. Rainfall data were obtained from the Multi-Source Weighted-Ensemble Precipitation (MSWEP V2); spatially distributed rainfall data was averaged per sampled watershed area. The land use map included land cover classes, namely: rock, water, grassland, and forest, and land use classes, namely: unstocked forest, paddy rice, other agriculture, and urban areas, making up a total of eight classes. According to FAO 2010, forests refer to areas of more than 0.5 ha with a canopy cover of more than 10% and trees higher than 5 m. Unstocked forests are forests with crown density lower than 20% resulting from exploitation for logging or shifting cultivation. Unfertile or degraded areas covered by grass are attributed to grassland category. Other agriculture refers to agricultural lands used for non-crop purposes like livestock grazing. Water class includes rivers and water reservoirs exceeding 10 m of width and 0.5 ha of surface area. Urban areas include permanent settlements like villages, towns, and roads having a width of more than 5 m. Maps generated with QGIS (version 2.6.1; https :// www.qgis.org) and edited with Inkscape (version 0.92.4; https ://inksc ape.org).   www.nature.com/scientificreports/ various factors and FIB, we were limited by the available data, especially the land use data. In future studies, when updated land use data will be available, it will be necessary to differentiate between planted and natural forests within the forest category at watersheds scale. The impacts of commercial tree plantations (e.g., teak trees, rubber trees) on hydrological response and associated increase in soil erosion, were pointed out by a few studies 28,33,76 . More accurate, complete, and higher land use data resolution would allow a closer understanding of the factors controlling bacterial contaminations and should be taken into account when addressing land management issues.

Conclusion
This study is the first to assess seasonal dynamics of fecal contamination in Lao PDR, based on a large physicochemical, microbiological, and geomorphological dataset, with the aim of identifying the relative importance of different controlling factors of in-stream E. coli concentrations during both the dry and the rainy seasons.
Our study consisting of (1) a spatial survey in 2016 during both the dry and the rainy seasons, and (2) a 10-day sampling monitoring from July 2017 to December 2018 at 3 stations, pointed out the following main findings.
• The seasonal variability of E. coli concentrations marked by higher and extreme values occurring during the rainy season, is noted in the majority of sampled Mekong tributaries. This is consistent with the increase in surface water turbidity during rainy season. • E. coli concentrations are positively correlated to total suspended sediment concentrations in both of the datasets, highlighting the potential role of suspended sediment dynamics in FIB transport, more particularly in catchments prone to soil erosion in a tropical setting. • E. coli concentrations are positively correlated with unstocked forest percentage areas, and negatively correlated to forest percentage areas, which points out the importance of land use/land cover as one of key factors impacting FIB dynamics at catchment-scale.
Our data provide new evidence that populations relying on untreated surface water resources of three northern watersheds in Lao PDR (Nam Ou, Nam Suang, and Mekong) are exposed to continuous fecal contamination all year round. Despite the World Health Organization (WHO) guidelines of 0 MPN 100 mL −1 of E. coli in drinking water, this remains an urgent public health issue in Lao PDR, putting lives at risk. The majority of sampled tributaries across Lao PDR in 2016 presented very high E. coli concentrations during the rainy season, exceeding 500 colonies per 100 mL, the threshold above which the WHO considers a 10% risk of gastrointestinal illness after one single exposure. This stresses the need for a better water quality assessment of the Mekong river and its tributaries, as well as a detailed evaluation of the risks posed by fecal waterborne diseases to rural populations directly depending on untreated water resources. In addition, given the rapid growth of hydropower plants in the Mekong basin 7, 8 , it will be necessary to investigate its impact on hydrology, sediment fluxes, and associated health issues like fecal contamination at watershed-scale in Lao PDR. It is also necessary to further investigate and quantify the factors controlling FIB survival and mortality in water and sediments under tropical conditions to understand the dynamics of these bacteria in this system.

Material and methods
Study site characteristics. Sampling sites were located in Lao PDR, a landlocked country in Southeast Asia, sharing borders with Myanmar, Cambodia, China, Thailand, and Vietnam. Lao PDR is mainly covered with mountains and forested hills, plateaus and plains along the Mekong river, where approximately 6.5 million people live on a 236,800 km 2 land 11 . About 70% of its population lives in rural areas and have a resourcebased economy, relying mainly on agriculture and forestry for their livelihood. The tropical wet and dry climate (Aw climate) is under the influence of monsoon regime, dividing the year into two seasons: a dry season from October to April, and a rainy season from May to October. The average annual rainfall in Lao PDR varies from 1300 to 2500 mm and exceeds 3500 mm in central and Southwestern Lao PDR (Fig. S1). The air temperature ranges from a minimum 15 °C in December-January to a maximum temperatures of 25 ± 30 °C from May to September 23 . The Mekong river runs about 4350 km through China, Lao PDR, Myanmar, Thailand, Cambodia and Vietnam, draining a 795,000 km 2 surface area 77 . The river flows from North to South of Lao PDR, forming a natural border with Thailand over 800 km.

Sampling design and watersheds characteristics.
Our study investigates seasonal dynamics of fecal contamination, based on field monitoring at multiple space and time scales. We used two different datasets ( Fig. 1; Table 1).

1.
A field campaign conducted in 2016 that sampled the Mekong river at six sampling sites, and 19 Mekong tributaries at 22 sampling sites including four sites along the Nam Ngum river (Fig. 1) located on mountains, hills and plains (Fig. 4a). The sampling was conducted once in March 2016 (dry season) and once in July 2016 (rainy season). The choice of sampling sites was based on a broad geographical coverage of Lao PDR between 15 and 20°N, and to encompass a broad range of catchment sizes (239-25,946 km 2 ), and a large range of geographical, topographical, and land use features. The sampling sites were also chosen to allow a large spatial sampling, covering the majority of the Mekong tributaries, in a relatively short time and logistically accessible from the road ( Table 1, Table S1). The surface area of the catchments of the Mekong tributaries ranges from 239 km 2 (A6) to 25,946 km 2 (Nou) ( Table 1). A range of eight different land use classes, grouped in eight main categories, is found across Lao PDR ( Fig. 4b; Table S1). The highest percentage of unstocked forest areas and the lowest percentage of forest areas were found in catchments of northern Lao PDR (Nou, Nsu, Npa, Nk20, and A6), and Vientiane Province (Nmi, Nsa, and Ntho), whereas forest areas dominated the catchments in southern Lao PDR (Nka, Nhi, Xbi, Xbg, Xbn, and SR). The highest percentages of paddy rice and other agriculture areas were found in the southern catchment near Pakse (SR) followed by catchments in Vientiane Province (Nmi, Nsa, and Ntho). The highest percentages of grassland and water areas were found in Vientiane Province (Nlik, Nng_3, Nng_4, Nng_2, and Nng_1). The livestock population is mostly present in catchments near Vientiane Capital followed by southern provinces of Savannakhet and Pakse ( Fig. 4c; Table S1). Human population density is variable across Lao PDR. The densest catchments are mostly found in the southern province of Pakse, followed by Savannakhet and Vientiane Capital (Fig. 4d, Table S1). Catchments are also highly variable in terms of rainfall (Table S1). During March 2016 (dry season), the rainfall was low (Fig. 4e). During July 2016 (rainy season), the highest rainfall was recorded in catchments located on steep terrain (Nlik, Nng_3, Nng_4, Nng_2, and Nng_1) and plains (Nmi, Nsa, and Ntho) of Vientiane Province ( Fig. 4f; Table S1).
Geographical analysis. The SRTM 90-m resolution digital elevation model (DEM) was used to draw the elevation map in the QGIS 2.6.1 software (https ://www.qgis.org/en/site/forus ers/). Based on this DEM and the geographical location of the sampling points, the catchment areas were then determined using the QGIS 2.6.1 software. We also computed surface area, perimeter, median slope, and median elevation of each of the catchment areas upstream of the sampling points from the DEM (Fig. 4a).
Data on land use and dams. We used information from the land use map provided by the Department of Agriculture Land Management (DALaM) of Lao PDR in 2013 (Fig. 4b). Technically, this land use map included land cover classes, namely: rock, water, grassland, and forest, as well as land use classes, namely: unstocked forest, paddy rice, other agriculture, and urban areas, making up a total of eight classes expressed in percentage of surface area. According to Lao Forestry Administration 78 and the Food and Agriculture Organization of the United Nations (FAO) 57 , forests refer to areas of more than 0.5 ha with a canopy cover of more than 10% and trees higher than 5 m. Unstocked forests are forests with a crown density lower than 20% resulting from exploitation for logging or shifting cultivation cycles. Unfertile or degraded areas covered by grass are attributed to grassland category. Other agriculture refers to agricultural lands used for non-crop purposes like livestock grazing. Water class includes rivers and water reservoirs exceeding 10 m of width and 0.5 ha of surface area. Urban areas include permanent settlements like villages, towns, and roads having a width of more than 5 m. Data on regional dams in Lao PDR represent maximum reservoir area (ha) of dams located upstream of our sampling sites. These data were taken from the dataset on the dams of the greater Mekong by Mekong Region Futures Institute 79 .

Data on livestock and local populations.
We obtained data on local populations and livestock per district from the Lao PDR Population and Housing Census 2015 conducted by Lao PDR Statistics Bureau 11 . Based on the percentage of area occupied by each district present in each catchment, we calculated the density of the human population and of livestock in each basin's catchment area (Fig. 4c,d). The data on livestock densities (cattle, buffaloes, pigs, goats, ducks, local and commercial chickens) were obtained from the Lao agricultural census 2010/2011 80 . In situ measurements and laboratory analysis. We measured a set of physico-chemical parameters in situ, including stream water temperature (T), pH, electrical conductivity normalized to 25 °C (EC), dissolved oxygen saturation (DO), using a Multi Probe System with a data logger (YSI 556 MPS). Concentrations of DO were transformed to oxygen saturations (%), using the Hua formula 81 . We collected 500-mL water samples 10 cm beneath the water surface in sterile plastic containers, and conserved them at a low temperature in the dark until laboratory analysis. A subsample of the collected water was analyzed within 24 h, for E. coli counts, using the standardized microplate method (ISO 9308-3 www.nature.com/scientificreports/ ral (3-hourly) and spatial (0.1°) resolution, which was evaluated and assessed on a global scale 84 and in Eastern Asia 85,86 . In our study, we used MSWEP cumulative rainfall data over a 1-week period (corresponding to the sampling campaign duration) before the sampling during the 2016 campaign, averaged per sampled watershed area. The daily rainfall measurements for 2017 and 2018 associated with the Mekong, Nam Ou, and Nam Suang sampling sites, were taken from three stations of the Lao PDR Government (Department of Natural Resources and Environment of Luang Prabang). The rainfall station in Nam Ou (20.081972, 102.264139) is located at the outlet of the catchment, at a 530 m distance from the stream sampling point, in Nam Suang (19.967111, 102.272444) at 3 km upstream to the sampling point, and in Mekong (19.898472, 102.16525) at 3 km distance from MK_17 sampling point. Water level data were measured at three gauging stations situated in Nam Ou (20.245556, 102.348806) at 21 km upstream the outlet, in Nam Suang (19.967111, 102.272444) at 3 km upstream the sampling point, and in Mekong (19.892361, 102.134167) near MK_17.
Statistical analysis. In this study, we first focused on identifying links between [E. coli] and discriminating variables like hydro-meteorological factors, land use, and geomorphological characteristics of 19 Mekong tributaries in Lao PDR. Therefore, we used an exploratory data analysis, namely the Partial Least Square (PLS) regression, which is a multivariate approach 87 . It is the most adapted to our dataset, as it is able to handle (1) a large database with more variables than observations, (2) nonlinear relationships between response and independent variables, and (3) possibly collinear variables 88 . Before applying the PLS algorithm, we centered and normalized the data, since the dataset included variables with different scales and measurement units. Moreover, turbidity, [TSS], and [E. coli], were log-transformed because the measured data were positively skewed. The variable importance in the projection number (VIP) was computed to determine the importance of variables. VIP greater than 1 are considered to be important in the analysis. The PLS analyses were done using the Microsoft Excel for Windows 2016 add-ins with XLSTAT version 2020.5.1 89 .
To test the significant difference between measured variables in Mekong and Mekong tributaries during dry and rainy seasons, we divided the 2017-2018 monitoring dataset into two datasets: a dry season from November to May, and a rainy season from June to October. We used the Wilcoxon-test, to test the significant difference of measured variables in 19 tributaries between dry and rainy seasons of the 2016 campaign, and of the 2017-2018 monitoring datasets, with statistical significance set at p-value < 0.05.
We calculated Spearman correlation coefficients to investigate relationships between measured variables and [E. coli] measured in 19 tributaries during the 2016 campaign, as well as during the water quality monitoring 2017/2018 at the outlet of three watersheds in northern Lao PDR (Nam Ou, Nam Suang, Mekong). In the statistical analyses on Mekong tributaries (PLS analysis, Spearman correlation and Wilcoxon test), only the Nam Ngum outlet (Nng_1) among the Nam Ngum sampling sites, was taken into account, to avoid the overrepresentation of the Nam Ngum tributary among the data. The Wilcoxon-test and the Spearman correlations were conducted using RStudio version 1.2.1335 90 . The original maps were created using QGIS 2.6.1 91 .

Data availability
The datasets generated during and analysed during the current study are available from the corresponding author on reasonable request.