The influence of dynamic environmental interactions on detection efficiency of acoustic transmitters in a large, deep, freshwater lake

Table 3 Summary information for the best fitting five candidate models (including global model) summarizing the detection efficiency (DE) of V9, V13, and V16 tags deployed in northeastern Lake Ontario from 22 October, 2015 to 23 May, 2016 (215 days) as a function of environmental variables

Model	V9			V13			V16
Model	AIC	ΔAIC	Adj. R²	AIC	ΔAIC	Adj. R²	AIC	ΔAIC	Adj. R²
DE~ ti(D,th) + ti(D,v) + ti(D,i) + ti(D,t)+ ti(D,d) + ti(D,f) + s(D) + s(th) + s(v) + s(i)+ s(t) + s(d) + s(f) + tag	− 33,302.38	13.31	0.908	− 10,256.59*	15.31	0.916	− 38,410.15	13.44	0.907
DE~ ti(D,th) + ti(D,v) + ti(D,i) + ti(D,t)+ ti(D,d) + ti(D,f) + s(D) + s(th) + s(v) + s(i)+ s(t) + s(f) + tag	− 33,290.74	24.95	0.903	− 10,262.80*	9.10	0.916	− 38,417.44	6.15	0.907
DE~ ti(D,th) + ti(D,v) + ti(D,i) + ti(D,t)+ ti(D,d) + ti(D,f) + s(D) + s(th) + s(v) + s(i)+ s(t) + s(d) + s(f)	− 33,315.69	0	0.908	− 10,256.59	15.32	0.916	− 38,423.59	0	0.908
DE~ ti(D,th) + ti(D,v) + ti(D,i) + ti(D,t)+ ti(D,f) + s(D) + s(th) + s(v) + s(i) + s(t)+ s(f) + tag	− 33,183.61	132.08	0.876	− 10,271.90*	0	0.916	− 38,386.14	37.45	0.902
DE~ ti(D,th) + ti(D,v) + ti(D,i) + ti(D,f)+ s(D) + s(th) + s(v) + s(i) + s(t) + s(f) + tag	− 33,302.38	13.31	0.908	− 10,256.59*	15.32	0.916	− 38,410.15	13.44	0.907

DE is the daily probability of detecting an acoustic transmission. s() indicates a smoother and ti() indicates a tensor product interaction. Environmental variables included were distance between tag and receiver (D), thermocline strength (th), surface water velocity (v), ice thickness (i), temperature at 50 m (t), depth difference between receiver and tag (d), number of fish detections (f), and tag depth (tag). All models included an ARMA autocorrelation structure to account for temporal autocorrelation in data and tag–receiver combinations as a random effect. Akaike information criteria (AIC), delta AIC, and estimated adjusted coefficient of determination (Adj. R²) are summarized for each model. The lowest AIC scores are italicized for each tag type to identify the best fitting model. An asterisk (*) denotes models that did not include tag as a covariate due to the lack of tags present at more than one depth

ISSN: 2050-3385