The Perth Canyon is located on the south-western continental shelf, approximately 50 km west from Perth; it is 120 km long (Trotter et al., 2019). The passive acoustic recorders were deployed on the eastern plateau of the Perth Canyon, in depths between 400 and 490 m ( Figure 1 ). The canyon intrudes on the continental shelf and extends 4000 m in depth. It is home to a variety of deep-sea fauna including deep-sea cnidaria, echinoderms, fish, and cetaceans (Erbe et al., 2015; Trotter et al., 2019). The Leeuwin current runs over the Perth Canyon and contributes to the seasonal upwellings of nutrients. Long-distance migrants take advantage of the seasonally increased primary production (Feng et al., 2009; Rennie et al., 2009). Pygmy blue whales (Balaenoptera musculus brevicauda), Antarctic blue whales (Balaenoptera musculus intermedia), fin whales (Balaenoptera physalus), Antarctic minke whales (Balaenoptera bonaerensis), and humpback whales have been identified acoustically while migrating through the Perth Canyon (Erbe et al., 2015). Specifically, the Perth Canyon is a migratory corridor for the BSD population of humpback whales on their northward and southward routes (Jenner et al., 2001).

This study investigated humpback whale acoustic presence in the Perth Canyon using underwater acoustic recordings collected by the Centre for Marine Science and Technology (CMST), Curtin University. All recordings were collected with a CMST-built Underwater Sound Recorder (USR; McCauley et al., 2017), using HighTech Inc. HTI-90-U hydrophones (sensitivity –186 dB re 1 V/µPa, 2 Hz – 20 kHz). The recorders were calibrated with input white noise prior to deployment. Deployments were temporally staggered between 2002 and 2017. Recordings from 2009 onwards were collected as part of the Australian Integrated Marine Observing System (IMOS). Refer to Table 1 for deployment set numbers, start and end times, coordinates, sampling rates, and duty cycles.

Daily environmental data for the Perth Canyon (SST, Chl-a, and POC) and Antarctic Area IV (SST, Chl-a, POC, and SIC) were sourced from 2000 to 2018. A daily mean was calculated for each region utilizing the boundaries of the Perth Canyon Marine Park (Australian Government, 2013) and the Antarctic Area IV waters, south of 60° (IWC, 2007). SST was extracted from a single-sensor multi-satellite system using observations from Advanced Very-High-Resolution Radiometer (AVHRR) instruments on National Oceanic and Atmospheric Administration (NOAA) polar-orbiting satellites with a 0.02° × 0.02° grid cell size. Bias between satellite and buoy matchups is included for every cell. The SST dataset was summarized to only include cells with the level 5 highest quality. Data were obtained using the Australian Ocean Data Network (https://portal.aodn.org.au/ , accessed 6^th October 2021). Chl-a and POC data were provided by NOAA and obtained from the Environmental Research Division’s Data Access Program (ERDDAP; https://coastwatch.pfeg.noaa.gov/erddap/ , accessed 6^th October 2021). Data from 2003 onwards were collected in a 4 km grid on the National Aeronautics and Space Administration (NASA) Aqua Spacecraft using a Moderate-Resolution Imaging Spectroradiometer (MODIS). There was Chl-a availability prior to 2003 onboard separate satellites and using different equipment. A cross-correlation was used between two satellites for a comparable year, and data prior to 2003 were utilized if R² > 0.7. Therefore, Antarctic Chl-a data from 2000 to 2002 were collected in a 4 km grid on the NASA Orbview-2 satellite using a Sea-viewing Wide Field-of-view Sensor (SeaWiFS). Only Science Quality level 3 data were extracted. Daily Antarctic SIC was obtained from the National Snow Ice and Data Centre (NSIDC; https://nsidc.org/ , accessed 25^th March 2022; Fetterer et al., 2017). Data were collected in a 25 km grid using a Scanning Multichannel Microwave Radiometer (SMMR) onboard the NASA Nimbus-7 satellite. Concentration values less than 15% were removed from the SIC mean calculation due to data uncertainty from the passive microwave instruments.

Humpback whale population size for every year of acoustic data was computed based on the 13% annual growth estimate by Salgado Kent et al. (2012b), who undertook aerial surveys between 2000 and 2008.

Humpback whale vocalizations were identified based on repertoire descriptions of song and non-song catalogues in the literature (Payne and McVay, 1971; Recalde-Salas et al., 2020) and co-authors’ prior experience with this population. Humpback whale acoustic presence was visually and aurally determined every hour using spectrograms in Raven Pro (Version 1.6, The Cornell Lab of Ornithology, Ithaca, NY, USA), which were created with 2048-point discrete Fourier transforms, Hann windows, and 50% overlap. Spectrograms were scanned in hourly increments for humpback whale vocalizations between 10 Hz and 2 kHz. Presence per hour was determined by selections of non-song or song within each increment. Raven selection tables were exported to xls format.

Hourly presence/absence of humpback whale vocalizations was calculated in MATLAB (Version R2020b, The MathWorks Inc., Natick, MA, USA). Raven selection tables were read into MATLAB and the frequency of occurrence (FO) of humpback vocalizations was calculated as the number of hours that humpback whales were acoustically present per 24 h day. The FOs for the start and end days of deployment were computed as the number of hours with humpback whale vocalizations divided by the number of hours the recorder was in the water on those days.

The environmental variables in the Perth Canyon and Antarctica were averaged to get a daily SST, POC, Chl-a, and SIC value. The datasets were all resourced from satellite imagery, and therefore cloud cover resulted in missing data points.

The FO in the Perth Canyon is visualized in Figure 2 . Across the years, whales were acoustically present in the months of May to December. While this period captured two migrations (first northward, then southward), there was no gap in FO in between, so that the migration direction could not be separated at this recording site.

Between 2003 and 2009, migration A₂₅ were mid-June, whereas the most recent years (2010-2017) had humpback whale A₂₅ in late-May to early-June ( Figure 3 ). The earliest A₂₅ was the 23^rd May in 2013 and the latest A₂₅ was the 13^th June in 2005. The rate of change of A₂₅ was -1.4 days/year.

The D₂₅ were more variable across a longer temporal scale. BSD humpback whales departed the Perth Canyon in early-December to late-November in the early years (2002-2010), except for 2003, which had an unusually early D₂₅. The departures of the population in the later deployment years (2011-2017) were from early to mid-November. The earliest D₂₅ was the 16^th November in 2003 and 2011 and the latest D₂₅ was the 14^th December in 2002. The rate of change of D₂₅ was insignificant ( Figure 3 ).

The difference of FO between Year, Week of Year, and the interaction between these variables was significant by permutation (p-value ≤ 0.001; Table 2 ). A post-hoc pairwise analysis showed that the weeks of year in which humpback whales arrived differed significantly between years (PERMANOVA: p-value< 0.05). One of the earlier weeks of arrival (week 23) was in early-June and there were significant differences in humpback whale FO between the early years in the study (2003, 2004, 2005, 2008) and the later years in the study (2012, 2013, 2014, 2016, 2017). The last week of departure from the Perth Canyon was in week 50 and FO was significantly different between the earliest year of data (2002) and all other years (2003, 2009, 2010, 2011, 2012, 2013, 2016). There was variable significance in FO between years in the preceding weeks of departures.

The Draftsman’s plot found positive, approximately linear correlations between Population size and Year ( Figure S1 ), as well as between Mean and SD of POC and Chl-a ( Figures S2, S3 ). There was also a sinusoidal correlation between Week of Year and SST ( Figure S4 ). The variables POC Mean, POC SD, Population size, and Week of Year were removed from the DISTLM dataset to create the most parsimonious model.

The DISTLM results showed that the highest proportion of the variation in humpback whale acoustic FO was explained by SST Mean, SST SD, and Year ( Table 3 ). These groups were also statistically significant (p ≤ 0.001). The DISTLM model of best AICc fit utilized all selections, and these predictor variables explained 63.2% of the variation in the humpback whales’ vocalizations FO in the Perth Canyon (RSS = 120.65).

Out of the predictors, the GLM identified POC Mean as more significant than Chl-a Mean, and so Chl-a was dropped from the model. Year was a stronger predictor than Population size, and so the latter was dropped from the model. The effect of SST bias was insignificant, as was the effect of POC SD. The final model retained the variables SST Mean, SST SD, POC Mean, and Year, and the pair products listed in Table 4 . The normal probability plot of residuals is shown in Figure S5 . In order of decreasing deviance, the model identified SST Mean as the strongest predictor, followed by POC Mean, then the product of SST Mean and POC Mean, then Year ( Table 5 ). Given the predictor variables were normalized to standard Gaussians, the estimated coefficients were comparable; and so, the influence of SST Mean on humpback whale FO was more than four times stronger than that of the other environmental variables. A 1 °C increase in SST corresponded to a 4.4 h/d decrease in humpback whale acoustic presence—according to the model. At the peak of the humpback whale season in the Perth Canyon (when humpback whale sounds were detected 24 h/d), the mean SST was 19 °C.

Based on Eq. 1, with d _Ant =5100 km, d _WA =300 km, v _Ant =5.48 km/h, and v _WA =6.48 km/h, the estimated travel time from Antarctica to the Perth Canyon was ~40 days. Figure 4 shows the five time series and the values of the Antarctic environmental variables at the estimated departure dates from Antarctica. SST, Chl-a, and POC dropped past their peak in the early months of the year, months prior to the peak of humpback whale vocalizations being detected in the Perth Canyon. Sea ice concentration was lowest when the other environmental variables were at their peak. Humpback whales seemed to depart when SST was, on average, -0.33 °C (higher in recent than earlier years), Chl-a was 0.24 mg/m³, POC 75.0 mg/m³, and SIC 29.6%.

This study showed a gradual change in humpback whale acoustic arrival dates in the Perth Canyon over the years 2002-2017. Both arrival and departure dates in the later years of the study were weeks premature compared to those in the earlier years. However, departure dates exhibited great variability and there were fewer years with acoustic recordings at the time of departure, so that the shift in departure dates was statistically insignificant. These results, which are entirely based on passive acoustic monitoring, are consistent with visual observations in Geographe Bay, WA, finding an earlier shift of the humpback whale season by approximately four weeks over the last 15 years (C. Burton, personal communication, 21^st October 2021). The shift of humpback whale presence in the Perth Canyon alters the migration periods described in Jenner et al. (2001): Original estimations of northbound and southbound migrations were mid-late June and mid-October, respectively. The later years in our study estimate that the boundaries of the migration are now late-May to early-June for the arrivals and possibly early to mid-November for the last departures (pending further data collection to increase statistical power). These migration shifts are concurrent with the literature of humpback whale populations elsewhere, with varying rates of change or suggested influences (Ramp et al., 2015; Avila et al., 2020).

Extrinsic factors are posed as potential triggers for these shifts in migration. Different predictors for humpback whale presence have been found for different populations and geographic regions. At a summer feeding ground in the Gulf of St Lawrence (Canada), earlier arrivals related to sea ice extent and SST (Ramp et al., 2015). Avila et al. (2020) found no correlation to SST but instead suggested ice sheet mass as a predictor for an earlier migration arrival at breeding grounds in Gorgona National Park (Colombia). Possible predictors of humpback whale presence are not consistent in the literature and may be population or site specific.

Our model showed humpback whale acoustic FO in the Perth Canyon was largely driven by Year and SST. Call FO was temporally predicted by Year in the DISTLM, meaning the acoustic presence of humpback whales was not consistent between years. This could be influenced by population densities or behavior. The population growth was estimated as 13% per year by Salgado Kent et al. (2012b) based on population counts between 2000 and 2008, and this may be reflected in the species’ earlier (and ultimately longer) acoustic presence over the years. Others have speculated that a growing population shows earlier migration arrivals (Beazley, 2022). Alternatively, fluctuations in acoustic FO might reflect variability in the proportion of humpback whales migrating from Antarctica each year. Van Opzeeland et al. (2013) found pervasive acoustic detection of humpback whales in Antarctic waters year-round, suggesting that not all individuals complete a full migration each year, and the proportion of the population that undertakes long-distance travel may fluctuate between years based on environmental factors on their feeding grounds (van Opzeeland et al., 2013; Ramp et al., 2015; Avila et al., 2020). A high percentage of the migrating humpback whale population are males, implying that females might circumvent departure in particular years if they cannot afford the energy cost of long-distance travel (Brown et al., 1995; Alerstam et al., 2003). Humpback whales rely on high volumes of krill during the feeding season to build energy reserves for migration (Riekkola et al., 2020). Krill abundance is heavily influenced by SST, thus, environmental variability between years may affect their fitness for migration (Alerstam et al., 2003; Kawaguchi et al., 2013; Melbourne-Thomas et al., 2016; Riekkola et al., 2020).

SST was the strongest environmental predictor for humpback whale presence in the Perth Canyon and this finding aligns with observations elsewhere (Ramp et al., 2015; Chavez-Rosales et al., 2019). For example, along the eastern coast of Australia, humpback whales departed weeks earlier than normal during years of higher temperatures (Meynecke et al., 2017). Consequently, the parameters around migration timing and the presence of the BSD population in the Perth Canyon would be affected by future trends in SST and, on a larger scale, climate change. The southwest Australian waters have experienced an increase in SST over the last few decades (Foster et al., 2014). Coupled with our study results, the rising SST will likely impact the BSD migration in the future. The shifting arrival dates may already be an adaptation to climate change as hypothesised by Ramp et al. (2015). Looking at other taxa, pied flycatchers showed flexibility in migration timing, possibly to keep up with climate change (Both, 2010). Specifically, the birds shifted their migration to more closely align with a shifting food peak. Similarly, in our study, the shifting arrival dates of humpback whales in the Perth Canyon may be a behavioral response to triggering environmental variables on their Antarctic feeding grounds.

Assuming a constant (over the years) migration duration from Antarctica to WA, humpback whales would have left Antarctica when SST, Chl-a, and POC were well past their peak and SIC had started to increase again. We hypothesize that shifts in arrival in the Perth Canyon (during northward migration) may be linked to environmental conditions on the feeding grounds. Tsujii et al., (2021) found that the annual migration timing of bowhead whales was linked to both SIC and SST in the Arctic. Antarctic SST may be slowly increasing on the feeding grounds and acting as a trigger for the population to leave earlier (Mintenbeck, 2017).

Once again, the increasing population may create a natural temporal expansion of the migration parameters and pressure more individuals to initiate migration earlier. Conversely, a theorized earlier departure from Antarctica may be in relation to temporal shifts in the peak availability of food. Krill abundance may not thoroughly nourish the entire population sufficiently; an exhaustion of the food supply may encourage earlier departures from the polar waters. Antarctic krill populations are predicted to be negatively impacted by climate change in the future and this could lead to a negative fitness consequence for humpback whales undertaking their long-distance migration (Tulloch et al., 2019). Grey whales (Eschrichtius robustus) on their feeding grounds exhibited overall decreased body condition over a three-year period between the beginning and end of the study (Soledade Lemos et al., 2020). An earlier departure from Antarctica may therefore be indicative of diminishing food availability on the Antarctic feeding grounds.

This study was strictly focused on the acoustic presence and trends of the BSD population in a migratory corridor. While it seems plausible that changes in environmental conditions of the Antarctic feeding grounds send them on their way northward (and similarly, changes in environmental conditions of their breeding grounds send them on their way southward), no concurrent observations were available of humpback whale presence (arrival and departure) on feeding or breeding grounds. Acoustic recordings on the feeding grounds could substantiate theories of earlier migration shifts beginning in Antarctica. Otherwise, the humpback whale population could still be spending the same amount of time in each region but travelling at different speeds each year on their journey between the two (Zerbini et al., 2006; Andrews-Goff et al., 2018). For instance, humpback whales stagger their migration between travel and rest in the Australian coastal waters (Jenner et al., 2001; Bejder et al., 2019)—possibly differently in different years, leading to changes in overall migration duration. From this study, it is uncertain if the humpback whales’ feeding and breeding grounds are facing a similar temporal shift in arrivals and departures as the Perth Canyon migratory corridor. An acoustic comparison with Antarctica and northwest Australia would determine whether inter-annual shifts in arrival and departure also occur on feeding and breeding grounds and ultimately enable an investigation of the distribution of the BSD population throughout the entire migratory cycle.

The long-term nature of this study had shortcomings in the early acoustic datasets. Technology constraints at the beginning of the study condensed the available recordings to be much shorter in length and temporally sporadic. For the total 15 years of passive acoustics, there were only 10 years of arrival data and seven years of departure data. Additionally, only five years had both arrival and departure data. A more complete dataset with arrival and departure recordings every year would legitimize how migration patterns may develop (Laake et al., 2012). Hence, more long-term passive acoustic data for this region can better project future shifts in migration.

In addition, monitoring at a different location within the migratory corridor, where the northward and southward migrations are temporally separated, would be beneficial and provide both arrival and departure dates for both migration lags. In the Perth Canyon, the two lags were not separable acoustically; meaning that some vocalizing humpback whales were still migrating north when others were already on their return journey, migrating south. So we missed the departures of the northward migration from the Perth Canyon, and the arrivals of the southward migration. At a different location, such as Cape Leeuwin (the whales’ first contact with Australia on their northward migration and last contact with Australia on their southward migration), we expect the two migration lags to be temporally separated, but acoustic recordings with a sampling frequency appropriate for humpback whale sounds are unavailable.

During migration, all humpback whales can vocalize non-song sounds, but males also produce song (Recalde-Salas et al., 2020). Songs are much easier to detect because they run for longer periods of time and create recognizable patterns (Payne and McVay, 1971). We may have missed individuals in the population if they were not vocalizing when travelling within range of the recorder. Moreover, some cohorts (i.e., the mother-and-calf cohort) vocalize very quietly and far less frequently than the males (Félix and Haase, 2005; Salgado Kent et al., 2012a). These particular cohorts would likely be largely missed by the Perth Canyon recorders.

Migration timing is important to maximize presence in suitable habitat and needs to match up with environmental variables that inform desirability (Dingle, 2014). There is evidence of some migrant species adapting their behavior with climate change to maintain temporal and spatial synchronicity (Both, 2010; Sanders and Mennill, 2014; Mayor et al., 2017). Under increasing environmental pressure, short-distance migrants are more successful at balancing opportune habitats than long-distance migrants (Jenni and Kéry, 2003; Both et al., 2006). Long-distance migrants might struggle to keep arriving at their functionally different habitats at the optimum time (i.e., when habitat features are optimum). Furthermore, if the different habitats change out-of-sync, it might become impossible to be present in each at the optimum time. Some long-distance migrant birds are compensating for such environmental mis-match by leaving breeding grounds sooner to align with the phenology of insects on their feeding grounds (Jenni and Kéry, 2003). Similarly, a shift in humpback whale migration presence in Australian waters may be an indicator of shifting parameters elsewhere. Marine mammal migrants are less studied than their terrestrial counterparts and it remains undetermined whether a shift in humpback whale migration is detrimental to the population’s survivability. However, larger shifts increase the likelihood of habitat mis-match and may negatively affect the survivability, as witnessed with other long-range migrants (Both et al., 2006; Sanders and Mennill, 2014; Mayor et al., 2017).

Studies with similar shifts in migration arrivals and environmental predictors are also finding that the rates of change in migration are not the same between migrant species that occupy the same habitat (Ramp et al., 2015). For example, the increasingly earlier humpback whale northbound migration may temporally and spatially overlap migrant species that do not usually co-exist (Ramp et al., 2015). Another migrant whale species in the Perth Canyon is the pygmy blue whale (McCauley and Jenner, 2010). Although this species is listed as data deficient on the IUCN Red List of Threatened Species, other studies on a range of marine mammals have shown that shifting migration arrivals is not limited to humpback whales (Szesciorka et al., 2020; Huang et al., 2022). Therefore, the pygmy blue whales that utilise the Perth Canyon for feeding during summer could be affected by interspecific competition, should the humpback whale presence increasingly overlap their own. The humpback whale BSD population has recovered significantly over the last few decades; however, resource competition may add another threat to the growing number of concerns for the species.

This study was able to determine the arrival and departure parameters of the BSD population, identify a significant temporal shift of humpback whale migration, and provide supporting evidence of SST predicting their presence in the Perth Canyon. This research is novel for the BSD population in determining long-term trends of their presence. Future research should further investigate the Antarctic Area IV to determine the population’s geographic distribution and the contributing extrinsic factors in initiating their migration. Similar trends on the feeding grounds may pose additional threats to the BSD population in the face of climate change.