Issue 7.3, December 2003
REST in Bosnia: A Pilot Test of Detection Capability (Cont)

<- Back

Weather Variables

Significant effects were found for humidity (detection success was higher when humidity was lower, R2 = 0.145, F1,82 = 13.9, P = 0.0004; see Fig. 2) and for temperature (detection success was higher at higher temperatures, R2 = 0.061, F1,82 = 5.7, P = 0.02; see Fig. 3). Temperature and humidity are correlated variables (humidity declines as temperature increases, see Fig. 4), thus these two effects are likely to be linked. The effect of humidity was stronger than the effect of temperature, and the effect of temperature did not add significantly to the effect of humidity (multiple regression analysis, NS). The effect of temperature appeared to be attributable to low detection success at temperatures below 15°C, and it is likely that the effect indicates a threshold rather than a trend.

Figure 2: Relationship between the number of dogs finding a positive filter and humidity. Each point represents one filter, and humidity was recorded at time of sampling.

Figure 3: Relationship between the number of dogs finding a positive filter and temperature. Each point represents one filter, and temperature was recorded at time of sampling.

Figure 4: Relationship between humidity and temperature in Bosnia. Data are for the times at which the filter samples were made.

False Alerts

The samplers made 20 neutral filters at each sampling time, thus false alert rates (on neutral filters) in the testing laboratory were necessarily linked to the humidity and temperature at the time that groups of filters were made. Thus N in this analysis is for groups of 20 filters (i.e., N is the number of sampling events rather than the number of filters). Humidity was generally higher and temperature was generally lower at Sarajevo relative to Mostar. The false alert rates given by the dogs when testing filters were linked to these differences, with higher false alert rates being given in Sarajevo. Similarly to the results for positive filters, false alert rates declined with increasing humidity and increased with increasing temperature (see Figs. 5 and 6).

Figure 5: Relationship between false alert rates in the laboratory and humidity at the time of sampling at two field locations in Bosnia.

Figure 6: Relationship between false alert rates in the laboratory and temperature at the time of sampling at two field locations in Bosnia.

Discussion

Although this was a pilot study, we conclude that the REST concept can be used to find mines in Bosnia. More specifically, it can be used to find REST filters made over mines in Bosnia and therefore has potential for use as an area reduction tool. Issues of detection success and reliability require further research.

The sampling technology was robust with respect to the details studied here, as no effects on detection success were found for the following:

TMM1 mines are known to be difficult for dogs to find, but the Ns for this mine were small in this study and all were located in the area where detection success was lower overall (Sarajevo). Thus it is premature to draw any conclusions about this mine type.

Clearly, there needs to be further fine-tuning of the sampling process, particularly in relation to the effects of humidity. The effects of temperature may be less important, although operational field experiences6 combined with evidence from this study suggest that there is a minimum temperature below which sampling should not occur. For the moment, that temperature should be set at 15°C, although more data is clearly needed at lower temperatures before this is treated as a definitive value. NPA field dogs work at temperatures lower than 15°C, although it is known that detection becomes more difficult at these lower temperatures and dogs are worked very slowly. Field dogs cannot work at the highest temperatures at which sampling was undertaken here, so they can give no insight into the possibility that there is an upper limit above which detection success declines. Temperatures at ground level can be quite different from temperatures measured at 1.5 m; this is due to interacting effects of solar radiation, wind and vegetation. It might be possible to sample at lower ambient temperatures if, for example, the ground is exposed to direct sun.

Further tuning of the detection process is also clearly required because detection success here was considerably lower than is desirable for operational use of REST technology. In this study, four dogs were used, and the probability of detection by those dogs varied. For the moment, the cause of that variation is not understood. Certainly, an important requirement is to tune the dogs on prepared filters from the operational sites before detection begins, and there was only limited opportunity to undertake such tuning for this study. Any operational use of REST technology incorporates the assumption that all detectors (in this case, dogs) are working equally effectively. This assumption is possibly unrealistic, and there is a need to investigate whether variation in detection success is a random factor (implying that more detectors will give a higher probability of detection) or is correlated (implying that more detectors will not improve detection rate).

This is the first study to find a link between false alert rates and local conditions at time of sampling. In this study, the false alert rates were generally low but were still higher at Sarajevo than at Mostar. False alerts are not dangerous—the main cost is increased work for clearance teams. However, such costs decrease the overall value and efficiency of the technology and clearly it is essential that they be kept to a minimum. Recognising that false alert rates vary with local conditions is the first step towards minimizing their impact on productivity.

Experiences from the few organizations training REST detectors suggest that training takes between six and 12 months. The two species currently in training—dogs, rats (Cricetomys gambianus)8—both appear to require similar amounts of time to train. However, it is also clear that considerable work needs to be invested in tuning the detector in order to obtain high levels of detection success once the detector is operational. At the end of training, the dogs used in this study were giving detection success of about 95 percent suggesting that the success rate in this study could be considerably improved.9 Further attention to the problem of fine-tuning the detection process is clearly required.

*All graphics courtesy of the authors.

Acknowledgments

A. Sanjala (NPA-Angola) and E. Andersen assisted in the detection laboratory. Sampling was undertaken in collaboration with NPA, Bosnia. We thank T. Berntsen, G. Bjorsvik and S. Bryant for support. Elmir Tozo and Edin Avdic made the filters. The GICHD REST programme is funded by the governments of Norway, Sweden and the United Kingdom, as part of a broader programme of research on mine detection animals.

References

  1. R. Fjellanger, 2003. ”The REST (Remote Explosive Scent Tracing) Concept.” In GICHD (ed.), Mine Detection Dogs: Training, Operations and Odour Detection, GICHD, Geneva.
  2. H. Bach, I.G. McLean, 2003. ”Remote Explosive Scent Tracing: Genuine or Paper Tiger?” Journal of Mine Action, Issue 7.1. http://maic.jmu.edu/journal/7.1/features/gichd/gichd.htm.
  3. I.G. McLean, H. Bach, R. Fjellanger, C. Aakerblom, ”Bringing the Minefield to the Detector: Updating the REST Concept.” Proc. EUDEM2-SCOT conference on requirements and technologies for the detection, removal and neutralization of landmines and UXO, 15-18 Sept., Brussels.
  4. V. Joynt, 2003. ”Mechem Explosive and Drug Detection System (MEDDS).” In GICHD (ed.), Mine Detection Dogs: Training, Operations and Odour Detection, GICHD, Geneva.
  5. H. Bach, I.G. McLean, C. Aakerblom, R. Sargisson, 2003. ”Improving Mine Detection Dogs: An Overview of the GICHD Dog Program.” Proc. EUDEM2-SCOT conference on requirements and technologies for the detection, removal and neutralization of landmines and UXO, 15-18 Sept., Brussels.
  6. Personal communication with Terje Berntsen, NPA.
  7. R. Fjellanger, E.K. Andersen, I.G. McLean, ”A Training Program for Filter-Search Mine Detection Dogs,” International Journal of Comparative Psychology, in press.
  8. R. Verhagen, 2003b. Proc. EUDEM2-SCOT conference on requirements and technologies for the detection, removal and neutralization of landmines and UXO, 15-18 Sept., Brussels.
  9. R. Verhagen, C. Cox, M. Machangu, B. Weetjens, M. Billet, 2003a. ”Preliminary Results on the Use of Cricetomys Rats as Indicators of Buried Explosives in Field Conditions.” In GICHD (ed.), Mine Detection Dogs: Training, Operations and Odour Detection, GICHD, Geneva.

Contact Information

Rune Fjellanger
NOKSH AS
PO Box 57
N-5201
Os, Norway
E-mail: rf@noksh.com

Havard Bach
GICHD
P.O. Box 1300
CH-1211 Geneva 1
Switzerland
E-mail: h.bach@gichd.ch

Ian McLean
GICHD Researcher
7bis, Avenue de la Paix
Geneva 1, CH-1211
Switzerland
Tel: +41-22-9061676
Fax: +41-22-9061690
E-mail: i.mclean@gichd.ch
Website: http://www.gichd.ch