Skip to main content
. 2020 Apr 14;10:6393. doi: 10.1038/s41598-020-63367-z

Table 2.

Probabilities of identification errors while classifying each set of camera-trap photographs (capture event (CE) folders) of snow leopards (estimates are the mean ± SD of the posterior distribution of the expected mean error probability from Bayesian binomial models described in Appendix S1).

Error Overall Non-expert Expert P(Non> Expert)
Split 0.091 ± 0.016 0.098 ± 0.025 0.078 ± 0.021 0.709
Combine 0.016 ± 0.007 0.007 ± 0.007 0.019 ± 0.011 0.138
Shift 0.023 ± 0.008 0.042 ± 0.016 0.002 ± 0.001 1
Total split 0.111 ± 0.011 0.139 ± 0.028 0.078 ± 0.022 0.957
Total combine 0.037 ± 0.011 0.049 ± 0.017 0.019 ± 0.011 0.918
Exclude 0.087 ± 0.016 0.119 ± 0.027 0.053 ± 0.018 0.981
Exclude* 0.049 ± 0.013 0.044 ± 0.019 0.053 ± 0.018 0.344
CE error 0.125 ± 0.019 0.146 ± 0.029 0.099 ± 0.024 0.895
CE error + 0.208 ± 0.023 0.264 ± 0.036 0.151 ± 0.028 0.994
CE error* + 0.174 ± 0.023 0.191 ± 0.034 0.151 ± 0.028 0.782

Estimates are presented for the 8 observers (overall) and also divided according to their previous experience in snow leopard photo classification (non-expert vs. expert) with the probability that non-experts have greater errors than experts [P(non > expert) derived from the posterior distribution of the difference between observers]. Total split and total combine add the shift estimate to the split and combine estimates, respectively, since shifts involve a split and combination error. Capture event error (CE) relates to the total probability of a capture event being misclassified: this is presented for classification errors only (CE error) and when exclusions are considered as a classification error (CE error+). Because observer 2 excluded 30% of all capture events, some estimates are also presented where observer 2 has been removed from the analysis (*).