Environmental determinants of E. coli, link with the diarrheal diseases, and indication of vulnerability criteria in tropical area (Kapore, Burkina Faso)

In 2017, diarrheal diseases were responsible for 606 024 deaths in Sub-Saharan Africa. This situation is due to domestic and recreational use of polluted surface waters, deficits in hygiene, access to healthcare and drinking water, and to weak environmental and health monitoring infrastructures. Escherichia coli (E. coli) is an indicator for the enteric pathogens that cause many diarrheal diseases. The links between E. coli, diarrheal diseases and environmental parameters have not received much attention in West Africa, and few studies have assessed health risks by taking into account hazards and socio-health vulnerabilities. This case study, carried out in Burkina Faso (Bagre Reservoir), aims at filling this knowledge gap by analyzing the environmental variables that play a role in the dynamics of E. coli, cases of diarrhea, and by identifying initial criteria of vulnerabilities. A particular focus is given to satellite-derived parameters to assess whether remote sensing can provide a useful tool to assess health hazard. Samples of surface water were routinely collected to measure E. coli, enterococci and suspended particulate matter (SPM) at a monitoring point (Kapore) during one year. In addition, satellite data were used to estimate precipitation, water level, Normalized Difference Vegetation Index (NDVI) and SPM. Monthly epidemiological data for cases of diarrhea from three health centers were also collected and compared with microbiological and environmental data. Finally, semi-structured interviews were carried out to document the use of water resources, contacts with elements of the hydrographic network, health behaviors and conditions, and water and health policy and prevention in order to identify the initial vulnerability criteria. A positive correlation between E. coli and enterococci in surface waters was found indicating that E. coli is an acceptable indicator of fecal contamination in this region. E. coli and diarrheal diseases were strongly correlated with monsoonal precipitation, in situ SPM, and Near Infra-Red (NIR) band between March and November. Partial least squares regression showed that E. coli concentration was strongly associated with precipitation, Sentinel-2 reflectance in the NIR and SPM, and that the cases of diarrhea were strongly associated with precipitation, NIR, E. coli, SPM, and to a lesser extent with NDVI. Moreover, the use of satellite data alone allowed to reproduce the dynamics of E. coli, particularly from February to mid-December and those of cases of diarrhea throughout the year. This implies that satellite data could provide an important contribution to water quality monitoring. Finally, the vulnerability of the population is found to increase during the rainy season due to reduced accessibility to healthcare and drinking water sources and increased use of water of poor quality. At this period, surface water is used because it is close to habitations, free and easy to use irrespective of monetary or political constraints. This vulnerability particularly impacts the Fulani, whose concessions are often close to surface water (river, lake) and far from health centers, a situation aggravated by marginality.

Introduction diarrhea). Concerning, the bacterial risk, studies have shown that E. coli can be used to 137 indicate a probable presence of fecal borne pathogens of bacterial origin (i.e. enteric 138 pathogens) [8][9][10]. In temperate areas, improvements in E. coli detection methodology made 139 the analysis of thermotolerant (fecal) coliforms unnecessary to assess water quality 140 monitoring [11]. E. coli is therefore considered to be the best indicator of fecal contamination 141 or FIB -Fecal Indicator Bacteria [11][12][13]  [61]. Its analysis requires a holistic approach that integrates biophysical parameters and the 201 characterisation of the pathogens together with the social dimensions. The latter includes 202 exposure (socio-spatial organization and temporal dynamics), sensitivity or susceptibility 203 which refers to social and ecological fragility (practices, water uses, access to drinking water, 204 7 to a healthcare network, existence of prevention policy, etc.), and forms of resilience [62]. 205 Ultimately, to protect against a risk, a person must perceive the risk as such, know how to 206 protect herself or himself against that risk, and have the means and resources to do so. In 207 addition, inequalities depend on several determinants, in particular, distance [63][64], the size 208 of the household [64] which increases the likelihood of encountering difficulties in paying for 209 care, and the cost of treatment. In Burkina Faso,Haddad et al. [63] showed that socio-210 economic inequality in access to care is most evident in rural areas: 20% of the richest go to 211 the health center on the first day of illness, whereas 20% of the poorest wait until the 5th day 212 of this work were 1) to compare the dynamic of E. coli and enterococci in a rural sub-Saharan 220 system, 2) to characterize the hydro-meteorological parameters playing a role in the dynamic 221 of E. coli, 3) to reveal the diarrheal epidemiology/incidence in the Kapore region and the 222 associated environmental factors, and 4) to identify the initial criteria of socio-health 223 vulnerabilities. 224

Study Area 226
The Bagre Reservoir is located in the Center-East region of Burkina Faso (Fig 1A). The 227 average area of the reservoir is 20,000 ha and the maximum area is 25,500 ha, with a volume 228 of 1.7 billion m 3 , corresponding to 14% of freshwater resources in Burkina Faso. The 229 reservoir is located in a basin covering 33.500 km 2 and belongs to the Nakambe River  It is characterized by two well marked seasons: a rainy season from June to September, with 241 July, August and September contributing to 67% of the annual rainfall on average, and a dry 242 season from October to May (Fig 2)  This site is suitable for satellite monitoring by Sentinel-2 (spatial resolution of 10 -20 meters) 269 throughout the year. The sample site is located near three farming villages (Dango, Dierma 270 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) concessions near to Kapore (Kapore Peul) which are located between 5.5 and 10 km from the 276 Dierma HSPC and between 2.5 and 7 km from boreholes and wells of Zella (the nearest 277 neighborhood of the village of Dierma, Fig 1B). There is only one borehole at Kapore Peul 278 and some concessions are located more than 3 km away (Fig 1B) the Lengha wells ( Fig 1C). The Yakala Peul concessions are located between 3 km and 5.5 288 km from the first wells of Lengha, 7 km from HSPC and 5 km from the Yakala boreholes (Fig  289   1C). However, there is a well at Yakala Peul used by nearby concessions but not by those 290 further away ( Fig 1C). Finally, Yakala has 3 boreholes and is located 9 km from Lengha (Fig  291   1C). This situation in the Lengha region forces certain populations, mainly Fulani, to use 292 surface water, including the lake. In addition to the services provided by the HSPC the 293 population also consults traditional healers. 294 Populations living near the banks of the Bagre Reservoir depend on nearshore water for 295 drinking, cooking, dish and clothes washing, showering or bathing and recreation (Fig 3 A). 296 Land use around the sampling site is dominated by croplands (rain-fed subsistence) with some 297 rangelands, where herds are led to graze and drink. During "normal" rainfall years, the 298 animals gather at the lake in April, when the other water points have dried up, and stay there 299 until June-July. The banks of the lake and the shallow waters are also used for market 300 gardening from March onwards and rice is cultivated during the rainy season. During wet 301 years, the lake water level is sometimes too high to permit market gardening. While 2017 was 302 a particularly dry year (Fig 2), in 2018, the shallow waters and lake banks were used from 303 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. parameters at different sites around the lake. Samples were also collected from "puisards", 317 open holes dug in the sand of dry riverbeds and banks of the lake where water is drawn for 318 domestic use (Fig 3C). 319 SPM was measured on duplicate sub-samples (30-100 ml) of water that were filtered on to 320 glass fiber filters (Whatman GF/F 0.7 µm nominal pore size). Filters were pre-weighed after 321 oven-drying at 105°C for 1 h 30 min. After sample filtration, the filters were dried and re-322 weighed. SPM was calculated as the difference between the dry weights after and before 323 filtering divided by the volume of filtered water. 324 E. coli and enterococci counts were determined using the standardized microplate method (NF 325 EN ISO 9308-3 and NF EN ISO 7899-1 for E. coli and enterococci counts, respectively). 326 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. High frequency satellite-derived rainfall data were averaged over the Kapore area, including 355 the two main tributaries located upstream of the sampling point ( Fig 1A). We used the GPM 356 IMERGHHV6 data product (30mins), which is suited for monitoring rainfall in tropical 357 regions [72][73][74]. Rainfall data were cumulated per day and also over 8 day periods to be 358 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 25, 2021. ; https://doi.org/10.1101/2021.04.21.21255867 doi: medRxiv preprint consistent with the timing of the lake water sampling. Data were collected through the 359 Giovanni website (https://giovanni.gsfc.nasa.gov/giovanni/). 360 Altimetry data from Jason-3 were used to monitor the water level of the Bagre Reservoir from 361 01/03/2018 to 01/04/2019. Jason-3 provides 10-day data for the track that crosses the Bagre 362 Reservoir (Fig 1A). They were collected from the Dahiti website [75], 363 https://dahiti.dgfi.tum.de/en/943/water-level-altimetry/). 364 Sentinel-2 surface reflectance products were used to observe the dynamics of surface water 365 reflectance and area at Kapore. Data  Sentinel-2 Surface reflectance products were also used to monitor the seasonal dynamics of 376 the vegetation close to Kapore by averaging the land Normalized Difference Vegetation Index 377 (NDVI) over two areas: a "small" area which potentially impact the dynamics of E. coli at the 378 Kapore site and a "large" area to study the link with the dynamics of diarrhea cases over the 379 whole health area of the three HSPC ( Fig 1A). NIR reflectance was used to monitor the 380 dynamics of SPM following 76]. 381 Finally, the Land Surface Temperature (LST) and Emissivity product MOD11A2 (MODIS 382 sensor on-board TERRA) which provides land surface temperature at 1km resolution, was 383 used to extract the daily surface water temperatures (24-hour average from day and night 384 data). MODIS LST 8-day composite data for daily and nightly overpasses (around 10:30 am 385 and 10:30 pm local time) were extracted using Google Earth Engine 386 (https://earthengine.google.com/). Because of its spatial resolution, LST was extracted over an 387 area slightly downstream from the sample site, where the lake is larger (Fig 1A). 388 389 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The Ethics committee on non-interventional research (CERNI) of the University of Nantes 413 approved the study (approval number: n°19112020). Before starting each interview, we asked 414 for oral consent from each respondent. 415 416 Statistical analysis and software 417 FIB (E. coli and enterococci) data were log 10 transformed to achieve a normal distribution. 418 We calculated correlations and regressions to identify explanatory variables predicting E. coli 419 and cases of diarrhea. First, standard Pearson analysis was performed for correlations between 420 log 10 transformed E. coli and enterococci, and between in-situ SPM and the Near Infrared 421 (NIR) band. Pearson's correlation (r) is preferred to study these relationships, because a linear 422 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 25, 2021.    Table 1). Two specific "outlier" groups can also be noted.

468
. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 25, 2021. ; https://doi.org/10.1101/2021.04.21.21255867 doi: medRxiv preprint The water samples from the "puisards" are consistent with the relationship between E. coli 469 and enterococci in the Bagre Reservoir (Fig 5). Values of E. coli and enterococci in the 470 "puisards" range from 100 MPN 100 mL -1 during the dry season to 3000-4000 MPN 100 mL -471 1 during the rainy season. The values always exceed guideline values (0 MPN 100 mL -1 ) for 472 water for domestic use. The FIB counts observed during the dry season in "puisards" are 473 similar to those observed at the lake routine sampling point. During the rainy season, the 474 values measured in the "puisards" are lower than those measured at the routine point (1200 475 MPN 100 mL -1 versus 5100 MPN 100 mL -1 ). 476

Dynamics of E. coli in relation to environmental variables 477
E. coli is strongly correlated with precipitation (r s = 0.8 for weekly rainfall, r s = 0.64 for daily 478 rainfall), with SPM (r s = 0.66), and with Sentinel-2 satellite NIR band data (r s = 0.77) (Fig 6,

493
. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Weekly values of SPM and E. coli display close seasonal variations (Fig 6H, I). Starting in 494 mid-May, both SPM and E. coli values increase in parallel with the arrival of the first heavy 495 rains ( Fig 6A) and peak during peak rainy season. Infra-seasonal fluctuations of SPM and E. 496 coli during the rainy season seem to correspond to variations in precipitation (Fig 6A, B). In 497 addition, the leading role of precipitation is illustrated by the strong relationship observed 498 between E. coli and cumulative rainfall and between SPM and cumulative rainfall from 499 March 13 to July 31, 2018: r s = 0.9, and r s = 0.88, respectively ( Fig 6C, Table 1). Interestingly, 500 although an increase in bacteria number and SPM concentration is observed when land NDVI 501 is low, both variables decrease when vegetation cover is maximum (September-October) 502 despite significant rainfall (Fig 6F). explained above 70%, S1 Table, Fig 7). This is in line with results of the one-to-one 520 regressions, although the variable ranking is not in the same. All components have been 521 retained for the E. coli predictive model (Fig. 8), resulting in R² equal to 0.66 and RMSE of 522 1.66 log MPN 100 mL -1 (Fig 8A). Using the three main explanatory variables only (S1 523 Table), R² still reaches 0.54, with a RMSE of 1.81 log MPN 100 mL -1 (Fig 8B). 524 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. One of our objectives was to question the use of satellite data to monitor the dynamics of E. 536 coli to investigate the potential of tele-epidemiological approaches. We therefore chose to 537 perform a PLS regression retaining only the key satellite data (Weekly rainfall and the NIR 538 band, S1  (Fig  542   8C). 543 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 25, 2021. ; https://doi.org/10.1101/2021.04.21.21255867 doi: medRxiv preprint The model quality limits are linked to individuals 31 to 34 which correspond to the limit of 544 detection of E. coli (less than 50 MPN 100mL -1 for which it was arbitrarily decided to assign 545 the value "1") (Fig 9). These very low values make the relationship between E. coli 546 concentration and the explanatory variables nonlinear. If these data are removed, the quality 547 of the model is improved (S2 Fig), highlighting that  In the Center-East region, these diseases correspond to 6% of the consultations in the HSPC 566 and 1.7% in hospitals. 567 in Lengha. When the data from the 3 HPSCs for 2018 was pooled, a first peak was observed 584 in February, this was followed by a decreasing trend in March and April that thereafter 585 increased to reach a maximum in August after which the values decreased to a minimum in 586

Incidence of diarrheal diseases in the
December-January (Fig 10). When observed individually by site, the data from the HPSC of 587 Dango differed slightly from this bulk trend: the first peak occurs in April instead of 588 February, and its minimum is in October. Finally, a time lag between the start of the increase 589 and the main peak differed between HPSC: May and June-August for Dierma, June and 590 August-September for Lengha, June and August for Dango. 591 592 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Relationship between diarrhea incidence and environmental variables and E. coli 620
The numbers E. coli and cases of diarrhea were significantly correlated over the period March 621 2018 -March 2019 (r s = 0.70, Table 2). There was also a significant correlation between 622 diarrhea cases and SPM values (r s = 0.72, Table 2) with a one-month lag between the peak of 623 SPM in July and the peak of diarrhea cases in August. We also observed a significant 624 correlation between diarrheal cases and precipitation (r s = 0.71, Table 2) and diarrheal cases 625 and surrounding land NDVI (r = 0.68, Table 2). 626 The seasonal dynamics were similar between April and December. A one-month lag between 627 the peaks of NIR band and cases of diarrhea (r s = 0.63), occurring in July and August 628 respectively, was observed similar to that observed for in-situ SPM (Fig 11). 629 630   Table,  642   Fig 12), and to a lesser extent the SPM and NDVI (explained variance equal to 68% and 67% 643 respectively, S2 Table, Fig 12). Compared to the results of the one-to-one regressions, the 644 PLS shows that the NIR band variable has a more important role than SPM, and that the 645 NDVI has a less important role. We decided to retain these 5 variables as the main 646 explanatory variable to set up the simplified model. As done for the E. coli analysis, all 647 components have been retained for the predictive models. When all the variables are retained, 648 the R² of the model is equal to 0.81 with an RMSE of 32.08 cases of diarrhea (Fig 13A and  649   Fig 14). When only the main explanatory variables are used (Monthly rainfall, NIR Band, E. 650 coli, SPM, and NDVI), the model presents an R² = 0.82 with an RMSE of 31.00 cases of 651 diarrhea (Fig 13B and Fig 14). The use of satellite data that best explain PLS component 1 652 variance (S2 Table, Fig 12) provides an R² = 0.76 and an RMSE = 35.14 ( Fig 13C and Fig 14,  653 Eq. (2)) and the following model: 654 Cases of diarrhea = 1.63 + 0.163 · Monthly rainfall + 0.013 · NIR band + 400.232 · NDVI (2) 655 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Socio-health vulnerabilities 671
The interviews revealed the different types of water supply used for domestic purposes: 672 borehole, "puisard", lake and / or tributary, and combinations of these different types of 673 water. People, especially women, often need to draw water 2-3 times a day to fulfil the 674 requirements of the household. 675 676

Environmental and spatial vulnerabilities 677
The first criterion of vulnerability is environmental. The evolution of the vegetation cover 678 during the year being similar for the 3 health areas, the environmental vulnerability is mainly 679 linked to the direct proximity to surface water, including Bagre Lake. Thus the higher number 680 of cases of diarrhea recorded at the Dierma health center could be linked to the fact that this 681 health area has the most inhabited areas (Dierma village, Zella, Kapore Peul) in close 682 proximity to surface water (Fig 1). The population would therefore be in contact with surface 683 water (microbiological hazard) all year round, and particularly so in July and August. This 684 situation concerns also Fulani settlements near Lengha (Ouazi Peul, Yakala and Yakala Peul). 685 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 25, 2021 This vulnerability linked to the proximity to surface water, increases with distance from 686 drinking water points and with the use of "puisard" for part of the year. This mainly concerns 687 the Fulani settlements, in particular Ouazi Peul, as well as certain sectors of Kapore Peul and 688 Yakala Peul, located far from boreholes or wells. "I use the water from the lake when I don't 689 have time to go to the borehole", explained a woman working on a vegetable field (Kapore 690 Peul), and "the majority goes to the lake or dig 'puisards'" at Ouazi Peul. This situation is 691 aggravated during the rainy season when access to a borehole becomes difficult: the tracks are 692 less passable, and rivers are full of water, forcing people to bypass them. Moreover, 693 temporary streams and ponds located in proximity to the habitations lead to direct use of 694 surface water, or to digging of "puisards", thus increasing the use of this type of non-potable 695 water at this time of year. During a so-called "normal rainfall year", the populations living far 696 from boreholes mainly use the lake between December (the backwaters and tributaries are dry 697 from December-January) and June. Both "puisards" and the lake or the tributaries close to the 698 habitations are used in July and November, with mainly the "puisards" being used between 699 August and October. "Puisards" are often dug close to where people live. The water quality is 700 considered to be better than that of the lake: "puisards make people less sick" (B.A, Yakala 701 Peul), "water from the puisards is better than that of the lake" (D.M, Ouazi Peul). Indeed, the 702 E. coli values observed in the "puisards" are lower: at Kapore "puisard" the values are 480 703 MPN 100 mL -1 without rain and 3 300 MPN 100 mL -1 after rain; in comparison, the values on 704 the Kapore routine site in the Bagre lake at the same dates were equal to 5300 and 15000 705 MPN 100 mL -1 . Nevertheless, E. coli numbers are still high for a water source used for 706 drinking water. 707 Distance also plays a role in vulnerability in terms of access to healthcare. The Lengha HSPC 708 action plan in 2018 reveals that 41.5% of the population resides between 5 and 10 km, 709 including some populations of Ouazi and Zamsé Peul, and 8% at more than 10 km, mainly 710 Yakala Peul, which limits access to healthcare. Moreover, these difficulties in accessing 711 healthcare are accentuated during the rainy season. The data collected from the action plans of 712 the HSPC in Dierma reveal that from July to December 37.11% of the population had 713 difficulty in accessing the HSPC due to the river that crosses the village (Fig 1B). CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Social vulnerability 721
The lack of financial resources is also a vulnerability criterion for access to water and 722 healthcare. Access to drinking water may be subject to charging: in Dierma it is 500 FCFA / 723 woman / concession, and in Zamsé Peul it is 5000 FCFA / year. The lack of financial 724 resources can therefore lead some people to use the water from the lake: "this is why we are 725 going to draw water from the lake" (B.D). The issue of the cost of care, between 1500 and 726 5000 FCFA, can also be a limiting factor. Financial difficulties can lead to self-medication or 727 the late arrival of some people at the HSPC with conditions "often more serious and 728 In summary, the epidemiological study and the interviews revealed five criteria of 750 vulnerability: 1) proximity to surface waters and the distance to a potable water point and 751 health center; 2) financial difficulties; 3) age: children who are more fragile and who drink 752 surface water while playing; 4) type of activities: shepherds who leave their habitation with 753 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 25, 2021. ; https://doi.org/10.1101/2021.04.21.21255867 doi: medRxiv preprint drinkable water and then drink water from the lake, and some people who cultivate on fields 754 far from concessions; 5) population in a situation of subjugation. 755

756
The microbiological health risk and its particularity in 2018 757 The microbiological health risk associated with diarrheal diseases is higher during the rainy 758 season. The values of E. coli are the highest, the use of surface water is more important 759 (closer and easier to access relative to the pumps for people living far from the villages), 760 access to health centers is more difficult (impassable roads, less financial availability). This 761 risk is reinforced for the Fulani by the territorial organization and the associated power 762 relations. 763 The number of cases of diarrhea was higher in 2018 suggesting that the risk was even higher 764

Dynamics of E. coli in relation to environmental parameters 778
We have shown that water at the Kapore site is polluted by bacteria of fecal origin throughout 779 the year, with higher values observed during the rainy season. Our E. coli numbers are within 780 the range of values found by Akrong et al. [86] in the Bosomtwe Lake in Ghana (E. coli 781 ranging from <1 to 5400 cfu / 100ml and for enterococci from <1 to 8400) and by Tyner et al. 782 [87] in Lake Malawi (E. coli range from 0 to 21200 cfu / 100 ml). 783 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Ultimately, the use of satellite data represents a major challenge with a very cost-effective 864 There are many studies that highlight the link between diarrhea incidence and meteorological 904 parameters such as rainfall [52,[43][44]121]. More specifically in Sub-Saharan Africa, both 905 Alexander et al [114] in Bostwana, and Alemayehu et al [122] in Ethiopia highlighted this 906 relationship. However, to our knowledge, the study presented here is the first to illustrate the 907 link between diarrhea incidence and satellite-derived variables. Some explaining variables 908 probably have causal links with diarrhea cases (E. coli and rainfall for instance), while others 909 may only show spurious correlation (NDVI probably). At this stage, the monthly time-step 910 and the well-marked seasonal cycle of different variables prevents more conclusions to be 911 drawn. A possible time-lag between SPM, NIR and cases of diarrhea cases may impact the 912 correlation value given. The number of data is too small to give solid evidences so far, 913 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 25, 2021. ; https://doi.org/10.1101/2021.04.21.21255867 doi: medRxiv preprint although such a lag is expected because of delays in contaminants peak, contamination, and 914 consultations in health center. 915 The validity of these models over several years and their capability to predict interannual 916 variability will have to be assessed in further studies. 917 918 Epidemiology and socio-health vulnerabilities 919 In Dierma and Dango, consultations are more frequent in the dry season than in the rainy 920 season. This is probably linked to the agricultural calendar and to the fact that the inhabitants 921 of Dierma, and to a lesser extent of Dango, cultivate crops on the other side of the lake. 922 Ouedraogo and Janin [66] also observed seasonal differences in the use of healthcare for all 923 pathologies in the region studied. 924

925
Our interviews highlight a more important use of surface water during the rainy season. This 926 is due to the easier accessibility and availability of surface water as opposed to borehole 927 water, similar to the conclusions of Schweitzer et al. [64] for Burkina Faso. In future research, 928 it would be interesting to determine the proportion of people using the different types of water 929 (lake, other surface water, "puisard") and to investigate the seasonality of FIB and SPM 930 across a wider range of water surface points including a more comprehensive investigation of 931 the "puisards". 932 933 Distance and financial availability can be limits for seeking healthcare. This is in line with 934 results by Haddad et al. [63], who have shown that people living more than 10km away from 935 a health center went less to health facilities: 25% went there on the 7th day of the illness 936 against 40% for those living at less than 10km. Haddad et al. [63] also revealed that the 937 maximum cost envisaged to pay for treatment for a family member is 2537 FCFA in rural 938 areas. However, in our study region, the cost varies between 1500 and 5000 FCFA, which can 939 further limit the use of care. Finally, our interviews revealed that financial availability for 940 treatment varies according to the harvest sales periods in agreement with Schweitzer et al. 941 [64]. 942 943 Social, physical and financial inequalities, problems of marginality and power imbalances 944 also play an important role in determining access to water resources and health care [123-945 125]. They are therefore important determinants of socio-health vulnerabilities. Regarding 946 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 25, 2021. ; https://doi.org/10.1101/2021.04.21.21255867 doi: medRxiv preprint access to water, this refers to "political ecology" which "recognizes that access and control 947 over resources, including water, are rooted in local stories and social relations" [126][127]. 948 Indeed, our interviews reveal that the notions of territory, marginality, and social exclusion 949 shape the landscape of water resources. This theoretical framework can also be extended to 950 include access to care. Thus, social and political interactions between resource managers and 951 health actors on the one hand and potential users on the other hand are often very limited. Ultimately, the socio-health vulnerabilities and therefore the associated health risks observed 957 in our study region combine social, political and physical factors. The study of the health risk 958 linked to diarrheal diseases requires an interdisciplinary analysis using both qualitative and 959 quantitative methods. However, these approaches are still rare as already pointed out by 960 Batterman et al. [130]. 961 962 Finally, we would like to underline the limits of this case study. Our study concerns a small 963 region (120 km²) with a single in-situ monitoring point over one year and some occasional 964 measurements in "puisard". We have only monthly data on diarrhea cases from 3 health 965 centers, and interviews with inhabitants of the 3 villages, Fulani settlements in the area and 966 health personnel. It would be interesting to have a longer time series of data and to conduct a 967 quantitative vulnerability survey. Despite these limitations, our work allowed to 1) test 968 hypotheses on the link between E. coli, environmental variables and cases of diarrhea in a 969 tropical context, 2) develop a simple prediction model for E. coli from variables that can be 970 estimated using satellite data; 3) propose a first step in the interdisciplinary analysis of the 971 health risk by combining environmental, microbiological, epidemiological data and 972 interviews. 973

974
This study highlighted the positive correlation between E. coli and enterococci in tropical 975 rural West Africa with higher numbers observed during the rainy season. E. coli and diarrheal 976 diseases were strongly correlated with precipitation, in-situ SPM, and reflectance in the NIR 977 band between February and mid-December implying that satellite data can provide an 978 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. In the study area, diarrheal diseases represent the 3rd cause of consultation, with children 984 under 5 years old the most impacted. The vulnerability of the population depends on income, 985 age, activity, and worsens during the rainy season due to reduced accessibility to drinking 986 water and healthcare in conjunction with increased proximity of water sources of poor quality. 987 The importance of the power relations, which have a direct consequence on the territorial 988 organization for access to water and healthcare, were also highlighted. Overall, the 989 microbiological health risk associated with diarrheal diseases is more important during the 990 rainy season from June to September. Therefore intervention though public 991 education/awareness should be increased as the rainy season approaches. 992 Cross-referencing of data from interviews and environmental parameters helped to clarify the 993 causes of the large number of cases of diarrhea in 2018, illustrating the importance of 994 interdisciplinary approaches. In order to have a more detailed understanding of the processes 995 and interactions at play, it will be necessary in the future to investigate the co-variability of 996 environmental, microbiological and epidemiological data over finer timescales and longer 997 periods. This should be coupled with a more in depth quantitative survey (large number of 998 interviews) to obtain a complete vision of health issues related to fecal contamination over the 999 whole Bagre area through an EcoHealth approach. 1000 Although this is an initial case study, this work provides is a first step in the interdisciplinary 1001 analysis of health risk. The approach proposed here provides a base for developing future 1002 programs that explicitly integrate global, social and human mechanisms that modify habitats, 1003 modes of transmission, pathogen survival, and ecosystem functioning, as well as access to 1004 drinking water and healthcare. It is only by explicitly recognizing the complexity of the 1005 problem and by adopting a more equitable approach to problem solving that we can hope to 1006 provide solutions and to improve the livelihoods of the affected populations in Burkina Faso 1007 and elsewhere. 1008 1009 Kpeke Nestor Kambiré and Yasmina Karambiri for collecting part of the health data and part 1013 of the location of village concessions, boreholes and wells, and the regional health office of 1014 Tenkodogo for the use of its premises. 1015 1016