While the completion of estimates for summer SIE on a pan-Antarctic scale is desirable, it was decided to concentrate on a restricted, but more sampled, area; thus, the Weddell Sea sector (60∘ W–20∘ E, 76–60∘ S) was selected as the region of study, but note that Bellingshausen's route wobbles slightly in and out of the northern limit of this notional study area (Fig. 2). Over the past century, the Weddell Sea has exhibited highly variable sea ice trends, making it of significant interest (Yang et al., 2021; Fogt et al., 2022). In addition, following the early expeditions of explorers such as Cook and Bransfield, routes through which vessels could navigate the treacherous Southern Ocean became well established, particularly within the Weddell Sea. As a result, there exists a wealth of historic records containing data and observations for the Weddell Sea, thus making it an ideal region of study.

Two factors determined the selection of the time period for this project. One was the pre-existing literature, and the other was the availability of historical records that were deemed suitable for this study, according to the project criteria (outlined in Sect. 2.2). Very few studies have examined records from the 18th and early 19th century (Parkinson, 1990; Wu et al., 1999; Martin et al., 2022; Martin, 2023). Given the recent effort to digitise historical marine records, there are now many more available from this time period than there were prior to 2000. Research conducted by Edinburgh and Day (2016) consisted of an analysis of the records from 1897–1917, so here we attempted to extend their work backwards to the first half of the 19th century, namely over 1820–1843 CE. A future study will consider a comparison of this record with those from the first circum-Antarctic navigation of Cook and Furneaux in the 1770s. Modern-day satellite remote sensing data were used for comparison with the historical reconstructions.

The bootstrap algorithm and satellite-derived sea ice concentration dataset, obtained through the National Snow and Ice Data Center (NSIDC, 2017), was selected for this study. This global dataset comprises daily ice concentration data recorded at a 25 km resolution. Data used in this study were available from 1978–2017. With a dataset of this nature, it is common practice to define the sea ice edge using a threshold of 15 % concentration (Edinburgh and Day, 2016). The selection of this dataset was partly based on a study conducted by Worby and Comiso (2004). Their study highlighted the potential for an offset in sea ice edge position, as recorded by qualified human observers and satellite imagery. As detected via satellite imagery, the sea ice edge position was found to be approximately 0.75∘±0.61∘ to the south, relative to in situ observation. The bootstrap algorithm was chosen for this study, as it delivers the greatest consistency between satellite-derived ice edge position and in situ observation (Worby and Comiso, 2004; Edinburgh and Day, 2016). However, it is important to acknowledge any errors that may exist within this dataset; therefore, this offset will be considered following the completion of the statistical analysis.

A total of nine records from eight Antarctic expeditions have been utilised in this study; the details of each record have been presented in Table 1. Specifically, these records comprise ship logbooks, meteorological registers, charts and journals. These records were selected based on their fitting of five criteria. These criteria are as follows:

Records must come from expeditions that took place, at least in part, through the Weddell Sea sector.

Additionally, prior to extracting information regarding sea ice observations, the various terminology used in each record was analysed and established as being representative of sea ice or not. Within each record, a variety of terms are used in place of sea ice, which was only identified once in one journal account. Interpretation of these various terms can be challenging due to the often ambiguous way in which they are used. However, through context, it was possible to deduce the meaning of different terms and thus distinguish between observations relating to sea ice and other ice forms, such as icebergs or shelf ice. In cases where the meaning can be clearly deduced, the term has been allocated its relevant meaning (“sea ice” or “not sea ice”). A list of these terms and their classification is presented in Table 2. It should be noted that where the use of a term was ambiguous, it was classified as “not sea ice” to keep errors to a minimum.

A sea ice index was created using the data extracted from each historical record. The aim of the index was to collate relevant sea ice information from each record and ultimately produce a set of summer sea ice edge position data points to be used for statistical analysis, as will be outlined in Sect. 2.5. From each record in which a sea ice observation was logged, five key points were noted.

These points are as follows:

the time of day (usually noon) and ship's position at the time of the observation;

Next, the method of recording ship position was noted upon analysing each record. It is important to know exactly how ship positions were recorded so as to estimate the accuracy with which this was done. For example, it is well established that where the method of dead reckoning has been used, there are likely to be significant errors in the recorded longitude (Jackson et al., 2000). However, it was established that each ship position included in the sea ice index was recorded using a chronometer, which is generally a very accurate instrument and capable of leading to longitude estimates with an error of 6–19 km (Sobel, 1996). To validate the accuracy of these positions, each route was divided into “legs”, with each leg comprising a section of the route between landmarks (for which an estimated position was always noted). Then, using a British Antarctic Survey digital map of the Antarctic region, each landmark's true latitudinal and longitudinal positions were ascertained and compared to those indicated in the historical record. Where there were inaccuracies, positions along the next leg of the route were adjusted accordingly. Perhaps surprisingly, the majority of positions were very accurate, meaning that the routes needed minimal adjustment. The only route that required significant adjustment was that of the ship Tula.

Following these adjustments, the data regarding sea ice observations were compiled to create the sea ice index and a set of 173 summer sea ice position data points. Of these data points, 76 were subsequently used for statistical analysis. Only data points relating to explicit sea ice observations were used for analysis. The full index is stored at the UK Polar Data Centre at https://doi.org/10.5285/53000D3B-3069-494B-8FA9-F9C7504CAE25 (Love and Bigg, 2023). It is worth noting here that the sea ice positions recorded from ship are invariably northernmost ice edges and do not contain information about potential open water or polynyas further south in the Weddell Sea at the time of any specific observation. The Weddell Sea is well known now to have such open-water regions within the sea ice extremes (e.g. see Martin et al., 2022, for a recent example and the inset in Fig. 2), but the early explorers were unaware of this and assumed that reaching sea ice pack stopped any further poleward travel.

As outlined above, each record utilised in this study contains specific information regarding the time of day, latitude and longitude, which are noted at regular intervals. It is therefore possible, using this information, to reconstruct each route taken through the Weddell Sea. To complete this task, each ship position falling within the Weddell Sea sector was logged in each historical record, and adjustments were made where necessary, as outlined in Sect. 2.3. Using Google Maps, each route was then delineated onto a basemap using the mapping tool. Within many of the historical records were original sketches of the expedition route with corresponding values for latitude and longitude. Therefore, these maps were used, where possible, to validate the route reconstructions. The reconstruction of the routes of the eight voyages is shown in Fig. 2.

The initial requirement of this work was to produce estimates for summer sea ice edge position across the Weddell Sea sector, according to the month and decade of the 19th century (1820–1843) time series. These positions were then compared to those from the recent satellite (1987–2017) time series and the early 20th century time series comprising data obtained from Edinburgh and Day (2016). The methodology associated with estimating summer sea ice edge position, as adapted from that developed by Edinburgh and Day (2016), comprises three key stages. The three stages are as follows:

obtaining relevant satellite-derived data;

An estimate of the Weddell Sea SIE was also made from the historical ship observations. For each historical ship observation, the mean sea ice edge latitude difference – calculated from the paired analysis – was assumed to occur across the mean sea ice edge boundary of the Weddell basin for that day of the year, leading to a first-order estimate for the change in SIE between the modern era and the date of each ship-observed sea ice position. By combining the values for each month, or each decade, first-order estimates of mean SIE change between the early 19th century and today can be calculated.

The highest numbers of ice edge positions were recorded during January (43 values), followed by February (16 values), and then December (11 values), and the lowest number of positions was recorded during March (6 values). Figure 3 presents the number of observations made per decade, including the number of those observations recorded during each given month. The highest number of ship-observed ice edge positions was recorded during the 1840s (47 values), including observations from January, February and March. During the 1820s, 21 positions were recorded, including observations from all months. Finally, the lowest number of ship-observed positions was recorded during the 1830s (eight values), including observations from January and February only.

Table 3 and Fig. 4 present the results of the paired analysis for a comparison of the summer sea ice edge position between the 19th century and modern time series for each decade. Table 3 shows that, on average, the summer ice edge position was further north in each decade of the 19th century time series relative to today. All results were significant at the 95 % confidence level. The greatest difference in the summer ice edge latitude between the two time series occurred during the 1830s, although there were far fewer observations in that decade. Figure 4 illustrates the position of each ship-observed ice edge point in relation to the average satellite-derived ice edge position. Mostly, the ship-observed ice edge data points fall north of the satellite-derived ice edge as expected, given the paired analysis results (Table 3). However, this is not consistently so for either decade or longitude, so it is instructive to examine the monthly comparisons to explore the historical contrast further.

Table 4 and Fig. 5 present the results of the paired analysis for a comparison of summer sea ice edge position between the satellite and historical time series for each month of the historical time series (DJFM). Results in Table 4 show that during December and March the 19th century ice edge latitude was essentially similar to that of the modern time series across the Weddell Sea. However, during January and February, the ice edge in the 19th century time series was further north by a statistically significant 0.16 and 2.36∘, respectively. However, Fig. 5 shows that the ship-observed ice edge data points that fall north of the satellite-derived ice edge in January and February are essentially from the central part of the Weddell Sea rather than its western or eastern peripheries, meaning that in the core Weddell Sea the sea ice melted much slower in the 19th century decades than today.

There is a limited amount of ship-observed sea ice latitude data from the Weddell Sea in the Heroic Age, and it is from only 3 years, some of which may have been exceptionally ice-heavy due to a simultaneous El Niño event (Edinburgh and Day, 2016). However, Edinburgh and Day (2016) found the Weddell Sea sector sea ice to be significantly further north than during the satellite era; this is also the case when comparing with the observations considered here. Only the months of February and March have sufficient data to be comparable, but the Heroic Age sea ice latitude is significantly further north than either the 1820s–1840s or the modern era (Table 5).

The sea ice extent (SIE) estimates presented here, whose methodology was explained in Sect. 2.6, are approximate and rely on local, 1 d differences being representative of that day for the whole Weddell Sea sector. However, they are generally consistent with the sea ice edge latitude analysis. For the months of December–March, only the relatively well-observed month of February had a 19th century SIE greater than today's mean, while the other 19th century summer months had a similar SIE to today (Table 6). Nevertheless, all 3 decades of the 19th century time series had marginally to significantly greater SIEs, on average, than today (Table 7).