Estimating pregnancy rate from blubber progesterone levels of a blindly biopsied beluga population poses methodological, analytical and statistical challenges

Beluga (Delphinapterus leucas) from the St. Lawrence Estuary, Canada, have been declining since the early 2000s, suggesting recruitment issues as a result of low fecundity, abnormal abortion rates or poor calf or juvenile survival. Pregnancy is difficult to observe in cetaceans, making the ground truthing of pregnancy estimates in wild individuals challenging. Blubber progesterone concentrations were contrasted among 62 SLE beluga with a known reproductive state (i.e. pregnant, resting, parturient and lactating females), that were found dead in 1997 to 2019. The suitability of a threshold obtained from decaying carcasses to assess reproductive state and pregnancy rate of freshly-dead or free-ranging and blindly-sampled beluga was examined using three statistical approaches and two data sets (135 freshly harvested carcasses in Nunavik, and 65 biopsy-sampled SLE beluga). Progesterone concentrations in decaying carcasses were considerably higher in known-pregnant (mean ± sd: 365 ± 244 ng g(−1) of tissue) than resting (3.1 ± 4.5 ng g(−1) of tissue) or lactating (38.4 ± 100 ng g(−1) of tissue) females. An approach based on statistical mixtures of distributions and a logistic regression were compared to the commonly-used, fixed threshold approach (here, 100 ng g(−1)) for discriminating pregnant from non-pregnant females. The error rate for classifying individuals of known reproductive status was the lowest for the fixed threshold and logistic regression approaches, but the mixture approach required limited a priori knowledge for clustering individuals of unknown pregnancy status. Mismatches in assignations occurred at lipid content < 10% of sample weight. Our results emphasize the importance of reporting lipid contents and progesterone concentrations in both units (ng g(−1) of tissue and ng g(−1) of lipid) when sample mass is low. By highlighting ways to circumvent potential biases in field sampling associated with capturability of different segments of a population, this study also enhances the usefulness of the technique for estimating pregnancy rate of free-ranging population.