TABLE 1.
Boosted regression tree model performance.
| Hydrophone | n trees | cv.Dev | cv.Cor |
|---|---|---|---|
| Song (full) | |||
| MARU1 | 3300 | 0.794 | 0.893 |
| MARU2 | 5850 | 0.852 | 0.923 |
| MARU3 | 5900 | 0.789 | 0.885 |
| MARU4 | 3050 | 0.831 | 0.914 |
| MARU5 | 4650 | 0.779 | 0.883 |
| D calls (full) | |||
| MARU1 | 6200 | 0.708 | 0.837 |
| MARU2 | 4450 | 0.701 | 0.827 |
| MARU3 | 6100 | 0.733 | 0.847 |
| MARU4 | 4300 | 0.666 | 0.807 |
| MARU5 | 5500 | 0.697 | 0.850 |
| D calls (spring) | |||
| MARU1 | 3750 | 0.726 | 0.850 |
| MARU2 | 2600 | 0.236 | 0.467 |
| MARU3 | 4050 | 0.708 | 0.836 |
| MARU4 | 2150 | 0.605 | 0.832 |
| MARU5 | 4100 | 0.673 | 0.809 |
| D calls (fall) | |||
| MARU1 | 1000 | 0.466 | 0.630 |
| MARU2 | 1800 | 0.623 | 0.806 |
| MARU3 | 2700 | 0.509 | 0.734 |
| MARU4 | 1250 | 0.363 | 0.572 |
| MARU5 | 2150 | 0.359 | 0.634 |
| Song (fall) | |||
| MARU1 | 2750 | 0.522 | 0.733 |
| MARU2 | 2550 | 0.679 | 0.803 |
| MARU3 | 2350 | 0.679 | 0.822 |
| MARU4 | 2150 | 0.481 | 0.672 |
| MARU5 | 2850 | 0.266 | 0.537 |
Note: Evaluation of the boosted regression tree models fitted for each call type and hydrophone location. All models were fit with a learning rate of 0.005, a bag fraction of 0.75, and a tree complexity of 2. For song, the response variable is the daily song intensity index, fit with a Gaussian distribution. For D calls, the response variable is number of D calls per day, fit with a Poisson distribution, which is suitable for count data. Full models were first fit for each call type across all seasons at each hydrophone, and subsequently models were fit for within‐season peaks for each call type, at each hydrophone. Performance is assessed with two metrics, the cross validated percent deviance explained (cv.dev) and the cross‐validated correlation (cv.cor).