Rumbling elephants: Eavesdropping on bi-modal signals for elephant tracking

Project Details

Key Questions

Can we remotely monitor elephants by eavesdropping on their bi-modal rumble vocalisations?


Technologies to monitor wildlife have the potential to contribute to effective conservation strategies by monitoring and responding to behaviours of welfare concern, for example a real-time alarm system to limit poaching threat. The gold standards for monitoring the behaviour of terrestrial wildlife should be technologies suitable for remote locations, with minimal interference with wildlife or their habitat. It should provide information-rich real-time data in manageable quantities with maximum accuracy across a range of biological and physical contexts. Seismic monitoring has excellent potential across these gold standards for monitoring the behaviour of large terrestrial mammals that generate high seismic forces, including elephants. Elephants generate seismic cues incidentally as they move around, but elephants also purposefully generate seismic signals when they vocalise: their infrasonic ‘rumbles’ co-generate both an acoustic and seismic component in the frequency range under 20 Hz, which are thought to be used for long distance communication between herds.

Aims of the Project

The first aim of this project is to detect, locate the source, characterise and quantitatively compare the co-generated seismic and acoustic components of elephant rumbles using field data collected in Kenya. The second aim is to correlate these measures with physical factors, including propagation distance, geology, wind speed and direction, acoustic/seismic noise level and biological context, including behaviour and social context. Combined, this develops an approach to monitor and track elephants in the field using seismic and acoustic recordings, and gives quantitative measures for what physical and biological factors influence the accuracy and practical implementation of this method. Detection and localisation algorithms have been developed from previous work by our team, trained with data from the same location. Specific objectives given in full project description.

Project Description

The project goal is to develop and evaluate the use of a dense geophone array, deployed in conjunction with acoustic sensors and camera traps, as a tool to monitor elephant behaviour, specifically elephant rumble vocalisations. Seismic monitoring is routinely implemented in the geophysical sciences, utilising sensitive instrumentation and clever data processing. Here we make use of ultra-dense arrays of seismic nodes. A dense array provides high signal to noise and spatial resolution, which is required for accurate detection, characterisation and localisation of the rumbles. A large data set with high spatial resolution is also required to robustly correlate the analysis outputs with the relevant physical and biological factors. It also enables seismic imaging techniques to be used which allow the local geology to be estimated with high spatial resolution. If dense arrays can be used to give better signal-to-noise ratio and tracking information, this opens up opportunities for real-time monitoring of elephant behaviour for fundamental biological research into elephant behavioural ecology and conservation applications. Specific objectives are given below, but there is also scope to modify the specific objectives to suit the interests of the student. Obj. 1: Collect field data. a. Contribute to data collection at Mpala Research Centre, Kenya, using an ultradense array of seismic nodes, along with acoustic sensors and camera traps. b. Use behavioural observation in the field to collect data on social and behavioural context of rumble generation. c. Collect data on dynamic environmental variables: temperature, wind speed and direction for the timings of elephant rumbles. Collect data on static environmental parameters: geology, topology, vegetation. Obj. 2: Detecting the two components of elephant rumbles a. Run elephant rumble detection algorithms on the seismic and acoustic data recordings. b. Assess detection levels across a dense array of geophones and the acoustic sensors. Obj. 3: Comparing localisation ability of elephant rumbles from seismic and acoustic recordings a. Run localisation algorithms based on a time-difference-of-arrival-method. Verify location with camera trap images or observation data, if available in that location. b. Characterise and compare the rumbles from seismic and acoustic recordings that are successfully localised: including frequency content, harmonics, modulation and duration properties. Obj 4: Correlating with naturally-varying physical factors a. For the events where location is successfully calculated, calculate the propagation distance and correlate events with relevant static environmental parameters (geology, topology, vegetation). Estimate maximum propagation distances of the rumbles across each modality. b. Determine signal-to-noise-ratio measure for each sensor within 1 km of each successfully localised event. c. Statistically determine how propagation distance, static and dynamic environmental variables, biological context and signal to noise ratio correlates with: detection rate, localisation rate and accuracy, rumble characteristics; across both seismic and acoustic data. The outputs from the proposed project will provide a quantitative comparison of the co-generated acoustic and seismic components of elephant rumble signals, how far each propagate within the bounds of the array, how either could be used for real-time localisation and monitoring, and how properties of these signals (frequency content, signal to noise ratio) changes with physical factors and biological context.

Methods to be used

Sensors – geophones (seismic nodes), custom made acoustic sensors with infrasonic capability, camera traps. Other field data – geological and ecological survey, behavioural observation of elephants. Data management and setting up analysis pipeline for TBs of multimodal data. Machine learning algorithms – rumble detector from time series data. Rumble source localisation algorithms – based on a time-of-arrival approach. Rumble characterisation algorithm development – codes to quantify the frequencies, time patterns and amplitudes of detected and localised rumbles. This could be used for further analysis on elephant ‘language’. Behavioural analysis – coding defined behaviours and social contexts from observational data. Statistical analysis – correlating different data streams together, testing specific hypotheses using the different data streams.

Specialised skills required

Interest and motivation to do an interdisciplinary project; experience using statistical methods; experience with machine learning methodologies is desirable; knowledge of coding in Python is desirable; some experience with seismic instrumentation is desirable; otherwise full training will be provided

Please contact Beth Mortimer on if you are interested in this project