Seismic signal analysis for the characterisation of elephant movements in a forest environment

Monitoring of pachyderm movements in the neighbourhood of a forest is an important area of research for mitigation of human-wildlife conflict issues. This paper reports the feasibility study for the detection of elephants using seismic sensors inside a forest environment and further characterisation of the seismic signals. Seismic signals generated during elephant locomotion are analysed for various distances. Frequency analysis of these signals shows an average dominant frequency of 15.80 Hz (+/- 1.03 s.d.) and 16.52 Hz (+/- 1.20 s.d.) in the range of 0 m to similar to 20 m and 20 m to similar to 40 m respectively. Based on frequency information of the seismic signal, different filter bands were incorporated and the highest accuracy of detection was achieved corresponding to the filter band of 10-20 Hz. The paper also illustrates the statistical analysis of three signal detection algorithms; short and long time averaging (STA/LTA), amplitude spectrum of Fourier transform (ASFT) and continuous wavelet transform (CWT) as a function of the distance from the sensor and elephant group size. Comparative analysis of the preliminary dataset was carried out where CWT based detection approach shows F1-score improvement of similar to 28% and similar to 14% in comparison with STA/LTA and ASFT respectively for a radial distance 0-20 m whereas, similar to 2% for a radial distance 20-40 m from the sensor. Furthermore, CWT outperforms with a detection accuracy of similar to 90% on the preliminary dataset.

Automated summary

This paper presents a feasibility study on detecting elephants using seismic sensors in a forest environment and characterizing seismic signals generated during elephant locomotion. Frequency analysis shows the highest detection accuracy in the 10-20 Hz filter band. The CWT detection approach outperforms STA/LTA and ASFT with significant improvements in F1-score, especially in the 0-20 m radial distance range.