Skip to main content

African elephants can detect water from natural and artificial sources via olfactory cues

Cite this dataset

Wood, Matthew; Chamaillé-Jammes, Simon; Hammerbacher, Almuth; Shrader, Adrian (2021). African elephants can detect water from natural and artificial sources via olfactory cues [Dataset]. Dryad.


Water is vital for mammals. Yet, as ephemeral sources can be difficult to find, it raises the question, how do mammals locate water? Elephants (Loxodonta africana) are water-dependent herbivores that possess exceptional olfactory capabilities, and it has been suggested that they may locate water via smell. However, there is no evidence to support this claim. To explore this, we performed two olfactory choice experiments with semi-tame elephants. In the first, we tested whether elephants could locate water using olfactory cues alone. For this, we used water from two natural dams and a drinking trough utilised by the elephants. Distilled water acted as a control. In the second, we explored whether elephants could detect three key volatile organic compounds (VOCs) commonly associated with water (geosmin, 2-methylisoborneol, and dimethyl sulphide). We found that the elephants could locate water olfactorily, but not the distilled water. Moreover, they were also able to detect the three VOCs associated with water. However, these VOCs were not in the odour profiles of the water sources in our experiments. This suggests that the elephants were either able to detect the unique odour profiles of the different water sources or used other VOCs that they associate with water. Ultimately, our findings indicate that elephants can locate water olfactorily at small spatial scales, but the extent to which they, and other mammals, can detect water over larger scales (e.g. km) remains unclear.


We conducted the study during the dry season from 20-Aug to 8-Sept-2019 at the Adventures with Elephants sanctuary near Bela Bela, South Africa (24°46'53.8"S+27°57'03.3"E). Adventures with Elephants operates as an education centre where people can interact with and learn about elephants. To ensure the safety and comfort of the elephants, the professional elephant handlers carried out the experiments and issued the verbal commands under our direction. To determine the extent to which elephants use olfactory cues to detect and locate water at small spatial scales, we conducted choice experiments on five semi-tame, adult African elephants (see details below). All experiments were conducted in a covered open area where the elephants are fed and where educational interactions take place. Each elephant has its own section within this area, which made it possible for us to interact with each elephant separately.


The elephants we used in the study consisted of two males (Chova 23 years of age, Chishuru 21 years), and three females (Shan 18 years, Mussina 17 years, Nuanedi 17 years). These animals are housed in a barn at night but free-range over the 500 ha property during the day. The handlers follow the elephants as they forage, but do not limit what and where they eat or how far they travel. Yet, during the day the elephants are brought to the covered interaction area at 08:30, 12:00 and 15:00. We limited our experiments to the 8:30 and 12:00 time slots.

Experimental design

To obtain the data, we followed a similar experimental design to (Schmitt et al. 2018). Specifically, we conducted scent-based choice experiments using four different sources of water. We collected the water from three locations on the property from which the elephants drink on a daily basis. We did this prior to the elephants arriving at the interaction area, thus they did not see where the water came from. The water sources comprised two dams (upper and lower) that refill via rain and runoff, and a plastic trough (artificial water source) that is refilled with underground water provided via a pump. The purpose of using these water sources was that each would likely have a different odour profile. Visually, there were noticeable differences between the water sources with the upper dam having a white cloudy appearance (due to the clay soil of the dam) compared to brownish water of the lower dam, and the clear underground water found in the trough, which also contained a small amount of green algae. Despite the visual differences, we could not detect any olfactory differences between the different water types. Finally, as a control, we used type II distilled water that was distilled at the University of Pretoria, as the distillation process would have likely greatly reduced the odour profile.

Experiment 1: Detecting water via olfactory cues

For the choice experiments, we used two identical 25-litre plastic buckets (32 cm x 32 cm x 40 cm) placed side-by-side (Fig. 1). The buckets each contained a 5-litre glass beaker (Super-duty, low form, CC Imelmann, Robertsham). We ensured that the beaker was held in the centre of the bucket by wrapping a blanket around the outside of the beaker (Fig. 1a). We wrapped the beaker to prevent it from moving, which could produce noise that the elephants may use to determine which bucket contained the water, and to prevent water from spilling out of the beaker when the bucket was moved between trials. Before inserting the beaker into the blanket, we covered the blanket with a 50-litre clear plastic trash bag to prevent the blankets from getting wet during the trials, and thus potentially providing an additional olfactory clue for the elephants.

In each experiment, we poured 3.5 litres of water into the beaker of one of the two buckets and left the beaker in the second bucket empty. To ensure that the elephants could only use olfactory and not visual cues to find the water, we drilled 28, 10 mm holes (separated by ~3 cm) into the lids of the buckets to allow any odours given off by the water to escape (Fig. 1a). Once the lids were attached, the two buckets (one with and one without water) were placed side by side ~5 m in front of the elephant being tested (Fig. 1b). To ensure that the elephant did not know which bucket contained the water prior to smelling, the handlers instructed the elephant to turn 180˚ and face away from where the buckets would be placed. Once the buckets were in place the elephant was instructed to turn and walk up to the buckets. As with (Schmitt et al. 2018), upon reaching the buckets the elephant was instructed to smell each bucket, remove its trunk, and then choose which bucket it wanted by placing its trunk on the lid of the bucket. To prevent the elephants from knocking the buckets over when they smelt, two trainers held the buckets in place on the ground (Fig. 1b). Once the elephant had made its choice, it was rewarded with the contents of the bucket it chose (i.e. water or no water), while the other bucket was taken away without them seeing its contents. This contingency trained the elephants to learn that whatever they chose, they received.

During the experiments, the trainers knew which bucket contained the water but were not aware of where the water came from (i.e. dams, trough, or distilled). As they knew the location of the water, it is possible that the trainers may have inadvertently signalled to the elephants where it was. However, if this was the case, then we would expect that the elephants would consistently locate the water and that there would be no differences in their ability to locate the different water types. As neither of these things happened (see results and supplementary material), we are confident that the trainers did not signal the location of the water to the elephants and that our results thus reflect the elephants’ ability to find water using olfactory cues. 

Over a three-day period, we tested the ability of all five elephants to locate the four water types using olfactory cues alone. On these days, we ran trials at 08:30 and at 12:00. To ensure that the elephants were thirsty, the handlers prevented them from drinking for a minimum of two hours prior to being tested. This was easy, as during the dry season the only surface water available on the property is found in the two dams and the trough. The water that we used in the different trials was collected ~30 minutes prior to each trial. For logistical reasons, we exposed all five elephants to the same water type during one of the time slots. Thus, we tested two water types each day. In each time slot, the ability of each elephant to locate the water was tested five times. This resulted in 25 separate choice tests in each time slot (i.e. 5 elephants tested 5 times). To control for any side bias or handler effect, the location of the bucket with water, as well as the handlers holding the buckets were randomized for each trial. In addition, we wiped down both lids of the buckets with the same cloth between each trial, and periodically swapped the lids between buckets. For each elephant, new lids were put onto the buckets, and the used ones were cleaned with water, rubbed down with sand, and left to dry in the sun. These were then used again in a later trial.

Training session

Prior to conducting the experiments, we pre-trained the elephants in the procedure. This pre-training involved first presenting them with the two buckets, one with a beaker with water from one of the dams on the property, one with an empty beaker. However, in these initial trials, we removed the lids so that the elephants could smell, touch, and see what was in each bucket. We followed the same experimental procedure as above (i.e. approach, smell, smell, choose). Once all five of the elephants were consistently choosing the bucket containing the water (~2 days), we then put the lids on and continued the training. It only took one day for the elephants to figure out the experimental design with the lids on and demonstrate that they could select the bucket containing the water (i.e. locate the water >75% of the time). As a result, we started data collection the following day.

Odour profiles of the different water sources

To determine the odour profiles of the four water sources (i.e. two dams, a plastic trough, distilled water), we collected 4 x 3 L water samples from each source on the day we exposed the elephants to the water from that source. We then stored these water samples in sealed glass bottles in a refrigerator for two weeks to prevent the VOCs from dissipating prior to analysis.

To determine the odour profiles, we decanted 300 µL from each sample into separate 1.5 mL glass vials with a septum and then incubated at 23 °C for approximately 24 h. The headspace of each vial was sampled for 40 min using a solid-phase-micro extraction fibre coated with 50/30 µm divinylbenzene/carboxen on polydimethylsiloxane solid support (Supelco, Merck). The fibre was injected into the inlet of an Agilent 7890 gas chromatograph coupled to an Agilent 7000D mass spectrometer (GC-MS) equipped with an Agilent J&W DB-Wax column. The inlet of the GC was set at 220 °C and the mobile phase flow rate at 1 mL/min. The initial temperature of 40 °C was held for 2 minutes after which the temperature was ramped up by 5 °C min-1 to a final temperature of 180 °C with an additional 5 min hold at 220 °C before re-equilibration to 40 °C. The volatiles eluting from the column were diverted to the ion source of the MS which was kept at 320 °C and ionized at 70 eV. We set the MS to scan mode with a low mass of 40 m z-1 and a high mass of 450 m z-1. Unfortunately, when we ran the first three samples (one each of the lower dam, trough, and distilled water), there was an air leak in the GC. As such we discarded the output of these samples, which left us with 13 samples in total. To view the data, we used MassHunter software (Agilent) and then used the NIST v. 98 library (National Institute of Standards and Technology, USA) for tentative identifications of the analytes. We identified potentially important metabolites by comparing the chromatograms and locating unique metabolites not found in the distilled water.

Experiment 2: VOCs associated with water

In addition to determining whether elephants could locate water via olfactory cues alone, we ran a second experiment to determine whether they could detect geosmin, 2-methylisoborneol (2-MIB), and dimethyl sulphide, olfactorily. As we were only interested in whether the elephants were able to detect these compounds, and not their degree of sensitivity to these different compounds, we standardised the compounds by creating a 1% solution of each by diluting them into dipropylene glycol, a relatively odourless solvent. For the choice experiments, 2 ml of the solutions were poured into separate Eppendorf tubes.

As with the first experiment, we ran an odour-based choice experiment. However, in this experiment we used two plastic boxes (29 cm x 21.5 cm x 16.5 cm) with 15, 1 cm diameter holes drilled into the lids. In each box, we taped an open 2 ml Eppendorf tube to the inside wall of the container, halfway up the side (~8 cm from the top and bottom). In one box, the Eppendorf contained 2 ml of the VOC solution, while the Eppendorf in the second box contained only 2 ml of dipropylene glycol. As in the first experiment, we had the elephants smell each box and then choose the one that they wanted. However, in contrast to the first experiment, the elephants were not given what was inside the box. Rather, if they chose the box containing the VOC, we gave them four animal food pellets (Alzu Game Feeds: Grazer, South Africa) as a reward. This acted as a positive reinforcement and thus, if the elephants made a significant number of correct choices, this would indicate that the elephants could detect the VOCs associated with water. Similar to the first experiment, the location of VOCs and handlers holding the boxes were randomised between each trial. Additionally, to ensure that the elephants were not associating the positive reward with another odour cue (e.g. the smell of the handler deposited on the box, or another random odour), we wiped down the boxes and lids with alcohol between each trial, as well as switching the lids randomly during each trial.

Prior to the experiments, we trained the elephants to learn that if they indicated to the box containing the VOC, we would give them the food pellets as a reward. This was done by first having the elephants smell the VOC and then immediately giving them the pellets. This helped them associate the VOC with a reward. We did this 10 times for each elephant for each VOC. We, however, only exposed the elephants to one VOC a day, thus it took 3 days to do the initial training. We then introduced the elephants to the two boxes and ran the experimental procedure 10 times for each VOC for each elephant during one day.


As both experiments comprised a series of binary choices (i.e. selection between two buckets or boxes), and we tested the same elephants multiple times, we treated the results from each individual elephant as repeated measures in generalized estimating equations (GEE’s; see Schmitt et al. (2018) for similar procedure). The GEEs incorporated an exchangeable correlation matrix and binomial error distribution with a logit link function. As such, the GEEs allow modelling the proportion of time elephants make a given choice and comparing it to a 50% distribution expected under random selection for that given choice. Data were back transformed from logit-scale for graphical representation. We used the means and 95% confidence intervals (CI) to determine whether the elephants’ choice differed from the expected 50% random selection for each of the water types.

To statistically analyse the odour profiles, we converted the GC-MS raw data to .mzXML file format using the msConvert GUI in the Protewizard tool (Kessner et al. 2008). We then uploaded the converted data to XCMS online (Tautenhahn et al. 2012) for peak picking and retention time correction. Once the retention time was corrected, we log-transformed the deconvoluted dataset and analysed it in R using the Metaboanalyst v. 3 (Chong and Xia 2018) platform.

We conducted a standard principal component analysis to determine the similarity of the four water sources. We then ran a random forest analysis to determine the specific analyte differences between the water sources (Biau 2012). Finally, we then ran separate 1-way ANOVAs to determine if the quantities of five key volatiles (see below) varied between the different water samples taken from the four water sources (N = 13 samples). For the principal component analysis, the random forest analysis, and the ANOVAs, we log transformed the data for normality.

Usage notes

All data are present and codes are found in the spreadsheets.