The International Bathymetric Chart of the Arctic Ocean Version 4.0

Bathymetry (seafloor depth), is a critical parameter providing the geospatial context for a multitude of marine scientific studies. Since 1997, the International Bathymetric Chart of the Arctic Ocean (IBCAO) has been the authoritative source of bathymetry for the Arctic Ocean. IBCAO has merged its efforts with the Nippon Foundation-GEBCO-Seabed 2030 Project, with the goal of mapping all of the oceans by 2030. Here we present the latest version (IBCAO Ver. 4.0), with more than twice the resolution (200 × 200 m versus 500 × 500 m) and with individual depth soundings constraining three times more area of the Arctic Ocean (∼19.8% versus 6.7%), than the previous IBCAO Ver. 3.0 released in 2012. Modern multibeam bathymetry comprises ∼14.3% in Ver. 4.0 compared to ∼5.4% in Ver. 3.0. Thus, the new IBCAO Ver. 4.0 has substantially more seafloor morphological information that offers new insights into a range of submarine features and processes; for example, the improved portrayal of Greenland fjords better serves predictive modelling of the fate of the Greenland Ice Sheet.


Methods
Grid compilation. The IBCAO DBM compilation workflow, illustrated schematically in Fig. 2, contains six main steps.
Step 1 consists of assembling the different kinds of contributed depth data listed in Table 1 along with necessary metadata. The metadata follow the standard adopted by EMODnet Bathymetry 22 , with the additions shown in Online-only Table 1. Contributions to IBCAO come in various forms. Ideally, contributions are cleaned bathymetric data in the form of XYZ points representing spot soundings, single-beam soundings, nodes of high-resolution multibeam grids, or nodes of digitized contours from bathymetric maps. Gridded compilations derived from multiple sources have also been contributed (see sub-section 'Source data' and Online-only Table 2; the latter only available online) as well as raw multibeam bathymetry requiring processing. All gathered XYZ datasets are reviewed using QPS Qimera software. If necessary, additional post-processing is applied in Step 2 using tools available in Qimera including, for example, removal of outliers or adjustments of vertical levels where systematic offsets are evident. If datasets of relatively poor quality are found to be in conflict with other observations, they may be completely or partially removed. In Step 3, additional metadata are included; most importantly the version number of each dataset is incremented if it has been modified, permitting roll-back through the processing history.
In Step 4, the processed XYZ data are gridded using a modified version of the algorithm applied to compile IBCAO Ver. 3.0 15 . First, a low-resolution grid with a cell-spacing of 2000 × 2000 m is produced. The depth data passed forward are selected based on their quality prioritization within each 2000 × 2000 m grid cell. Multibeam data are generally prioritized before single-beam and spot-sounding data which, in turn, are prioritized ahead of digitized depth contours from charts. A block median filter is then applied using the Generic Mapping Tools (GMT) 23 . The block median filtered data are subsequently gridded using the GMT routine surface, which applies a continuous curvature spline in tension function 17 . The tension parameter is set to 0.34. This value was decided on after analyses of the gridding results over the course of the IBCAO-project. A value of 0 implies no tension of the spline surface, whereas a tension of 1 removes the curvature altogether by not permitting maxima or minima between constraining data points. The resulting 2000 × 2000 m grid is smoothed using a cosine filter over 6000 m in GMT to provide a smooth base over which higher-resolution data are merged. The smoothed grid is then resampled to 100 × 100 m.
Higher resolution datasets (i.e. multibeam surveys and some gridded compilations) are individually down-sampled (if high enough in resolution) to 100 × 100 m. If multiple contrasting depths exist for one grid cell, the depths passed forward to the block median filter at 100 × 100 m are selected based on the same prioritization as used for the 2000 × 2000 m grid cells. The final step in the preparation of the high-resolution data consists of a density filter, which only passes forward data if more than 30% of an area of 1000 × 1000 m is covered by depth values.
The final action within Step 4 consists of merging the high-resolution data passed forward from the procedure described above with the 100 × 100 m resampled 2000 × 2000 m smoothed grid by applying a remove-and-restore approach 24 . This involves the calculation of the difference between the 2000 × 2000 m grid resampled to 100 × 100 m and the high-resolution 100 × 100 m datasets remaining after applying the density filter. The differences, or residuals, are then gridded using the surface spline in tension function before they are added back onto the low-resolution 2000 × 2000 m grid (resampled to 100 × 100 m). This procedure results in a smooth merging of the high-resolution data onto the low-resolution resampled grid. To prevent introducing spline-function artifacts, the residuals are forced to be zero at a distance of 1000 m from the data. Finally, the entire grid is resampled to 200 × 200 m. The gridding algorithm is written in Python, from which the applied GMT routines are called.
Step 5 consists of a quality check of the final grid using a Stockholm University developed web interface along with Qimera and the Open Source Geographic Information System QGIS, version 3.8.3-Zanzibar, which has also been used to produce the maps displayed in this data description 25 . The web interface has a mark-up function permitting all members in the IBCAO Regional Mapping Committee to take part in the quality control. If issues are found and marked, the associated source data are passed back to Step 2 for further analysis and processing.
Step 6 in Fig. 2 is described in the following sub-section.
www.nature.com/scientificdata www.nature.com/scientificdata/ Calculation of statistics. Echo sounders mounted on surface vessels increase their ensonified area with increasing depth, thus decreasing their achievable mapping resolution with depth. Based on this principle, Seabed 2030 defined a set of target mapping resolutions: 0-1500 m, 100 × 100 m; 1500-3000 m, 200 × 200 m; 3000-5750 m, 400 × 400 m; and 5750-11000 m, 800 × 800 m 19 . Since IBCAO contributes to the Seabed 2030 project, the data coverage calculated in Step 6 uses the Seabed 2030 resolutions. For example, a depth sounding between 3000-5750 m is considered to map an area of 400 × 400 m whereas a sounding with a value between 0-1500 m only maps an area of 100 × 100 m. Where the source data are available in the form of multibeam, single-beam and spot soundings, it is thus relatively easy to calculate how much of the IBCAO grid is mapped or not. However, when the contributed data are compilation grids, the estimated surveyed area is uncertain as we do not know the underlying data coverage. Even if only the nodes of the contributed grids at their native resolution (i.e. before resampling) are counted, they will likely overestimate the mapped area. For this reason, gridded compilations are kept as a separate category (Fig. 1).

Data records
Source data. The IBCAO Ver. 4 is available for download from the British Oceanographic Data Centre 26 . The bathymetric source data for IBCAO Ver. 4 are listed in Online-only Table 2 along with references where available. Individual surveys are, in most cases, aggregated to one contributing organization. Each dataset is assigned a Source Identification number (SID) and Type Identification number (TID). The former links each dataset to its full metadata whereas the latter groups the data into the categories listed in Table 1. SID and TID grids are compiled within the workflow in   Table 1. (c) Close-up showing an area with single-beam soundings and digitized depth contours used in gridding. Since these data types occupy relatively few grid cells, they are difficult to see in the overview map shown in (b). (d) Summary statistics of the proportion of the IBCAO area covered by the different data types in Ver. 4.0 and 3.0. The data types "steering points" and "interpolated depths" are not shown in (a) as they are not counted as part of the depth data (Methods; Table 1). *Refers to "Isolated soundings", "ENC soundings" and "Mixture of direct measurement methods", which are merged with data type "Single-beam" sounding on the map as well as in the summary statistics shown in (d). LR: Lomonosov Ridge.
www.nature.com/scientificdata www.nature.com/scientificdata/ from within the fjords where a kriging algorithm is used to interpolate depths between the under-ice topography and available bathymetric data, including recent surveys along the Greenland coastline carried out by the NASA Earth Venture Suborbital mission named Oceans Melting Greenland 8,29 . We have masked BedMachine Ver. 3 so it is used from the outer coast of Greenland, resulting in a vastly improved fjord representation compared with other bathymetric models. Bathymetric data from Greenland coastal waters gathered since BedMachine Ver. 3 have been merged using the remove-and-restore approach. These include, for example, multibeam surveys of Petermann and Sherard Osborn fjords in northwest Greenland 30 and additional bathymetry collected and compiled within NASA's Ocean Melting Greenland 31,32 .
The area covered by "crowd sourced" bathymetry has increased substantially in Ver. 4.0 compared to Ver. 3.0 through contributions from fishing vessels and other ships using Olex (www.olex.no) and MaxSea (http:// www.maxsea.com/) mapping systems, the latter in Greenland waters only. Since 2012, when IBCAO Ver. 3.0 was compiled, numerous icebreaker expeditions mapping the seafloor with multibeam sonar in the sea-ice covered Arctic Ocean have been completed. These include expeditions with Canadian CCGS Amundsen and CCGS Louis S. St-Laurent, German RV Polarstern, Swedish icebreaker Oden, and USCGC Healy (Online-only Table 2).

Technical Validation
Validation: Comparison between IBCAO Vers. 3.0 and 4.0. The improvements in IBCAO Ver. 4.0 compared to earlier versions result from the large amount of new bathymetric data including gridded compilations, an improved gridding algorithm, and a higher resolution. This is best illustrated by specific examples, together with an overview map showing the depth differences between IBCAO Vers. 3.0 and 4.0, generated by subtracting Ver. 3.0 from 4.0, that highlights the most significantly updated areas (Fig. 5). The new multibeam bathymetry is readily visible in the difference map as well as in the improved representation of fjords along sections of the Greenland coast (Fig. 5). In general, the least updated areas in terms of absolute depth changes are located on the Russian continental shelf, in the Barents Sea between southern Svalbard and northern Norway, and on the Norwegian and Iceland continental shelves (Fig. 5). The lack of updates in Russian waters stems from the fact that no new multibeam data has been contributed from these areas, despite their collection during Russian efforts to map the extent of their juridical continental shelf. If we look at the updates as a function of how much the depth has changed relative to water depth (i.e. the percent depth change), the East Siberian and Laptev seas show some clear differences in Ver. 4.0 compared to 3.0 (Fig. 6). The updates result from the fact that individual soundings on charts were used, rather than digitized contours from charts, providing more bathymetric detail (Fig. 6). These soundings were digitized by Danielson, et al. 33 for the purpose of compiling the Alaska Region Digital Elevation Model (ARDEM). Areas that do not show large depth differences were already relatively well www.nature.com/scientificdata www.nature.com/scientificdata/  www.nature.com/scientificdata www.nature.com/scientificdata/ mapped in IBCAO Ver. 3.0. If the Barents Sea is examined carefully, the new additions from the MAREANO compilation are clearly visible (Fig. 5).
The incorporation of BedMachine Ver. 3 and additional merging of all bathymetry available since its release not only enhances the representation of Greenland fjords, but also highlights the complex coastal bathymetry (Fig. 7). This is particularly noticeable off the western coast of Greenland between about 55°N and 75°N, where IBCAO Ver. 4.0 reveals a rough submarine landscape characterized by criss-crossing channels that commonly occur where the seafloor is composed of igneous bedrock (Fig. 7). The transition to a smoother seafloor morphology on the outer continental shelf occurs rather abruptly across a near straight southwest-to-northeast trending line that fits well with geological maps showing change across a thrust fault from igneous rocks to a seafloor composed of sedimentary rocks further offshore 34 (Fig. 7).
Lack of depth data from the western Greenland inner continental shelf in IBCAO Ver. 3.0 resulted in a poorly constrained spline function causing undulations that do not represent the "true" seafloor morphology in this area (Fig. 7b). The Uummannaq Fjord of western Greenland is a good example, showing that submarine glacial landforms with spatial dimensions on the order of hundreds of meters, such as glacially streamlined drumlins and large mega-scale glacial lineations images using multibeam, are distinguishable in the IBCAO Ver. 4.0 DBM (Fig. 7d). This can only be the case when the gridding is based on high-resolution bathymetry, here collected by RRS James Clark Ross 35 .
The Lomonosov Ridge extends >1600 km across the central Arctic Ocean between the continental shelves of Northern Greenland and Siberia (Fig. 1). Details of the ridge came to light in the first published version of IBCAO 16 where it was drastically remapped compared to the GEBCO Sheet 5.17 36 , which had served as the authoritative international bathymetric map of the Arctic Ocean for nearly two decades before the IBCAO project began. Numerous multibeam surveys with icebreakers have been carried out over the Lomonosov Ridge since the release of IBCAO Ver. 3.0, (Online-only Table 2), leading again to a substantially improved bathymetry (Fig. 8). Examples include surveys that have been individually published revealing critical sills that influence water exchange across the Lomonosov Ridge 6 , ice-shelf grounding on the ridge crest 37 , and where the foot of the slope is located along the ridge flanks, identified for the purpose of substantiating Denmark's submission under Article 76 of the United Nations Convention on the Law of the Sea (UNCLOS) 38 .
The Science Ice Exercise (SCICEX) was a program utilizing US Navy nuclear submarines for systematic mapping under the Arctic Ocean pack ice between 1993 and 2001 39 . Of the eight completed expeditions, two (1998 and 1999) involved acquisition of swath bathymetry using the specifically designed sonar system Seafloor Characterization and Mapping Pod (SCAMP) 39 . This swath bathymetry was used in IBCAO Ver. 3.0, although in many areas newer multibeam bathymetry has now replaced the SCICEX data; for example along the Northern Alaskan margin and on Chukchi Borderland, where several mapping expeditions with USCGC Healy have been carried out to collect seafloor bathymetry in support of the establishment of a U.S. extended continental shelf under Article 76 of UNCLOS 40 . A major caveat with SCICEX/SCAMP data has been the problem of precisely geo-registering the swath bathymetry, which is particularly evident where areas have been systematically surveyed and the locations of seafloor features are noticeably offset on different tracks (Fig. 8c,d). To resolve this issue  Table 1. "Isolated soundings", "ENC soundings" and "Combination of direct measurement methods" listed in Table 1 are merged with data type "Single-beam" in this map. Note that contours and single-beam soundings hardly show at this scale.
www.nature.com/scientificdata www.nature.com/scientificdata/ in areas that were based solely on SCICEX/SCAMP bathymetry and appeared to show large 'fault offsets' , we used multibeam surveys that cross over the SCICEX tracks to re-position the swath data (Fig. 8c,d). These multibeam surveys were positioned using modern GPS implying a User Range Error (URE) commonly not exceeding 10 m. The result is not perfect but is a significant improvement in IBCAO Ver. 4.0 compared to Ver. 3.0.  www.nature.com/scientificdata www.nature.com/scientificdata/ Errors. Despite the fact that the IBCAO Ver. 4.0 DBM is a substantial improvement over previous versions, it is certainly not free of errors. The DBM remains limited by its underlying source database. The uncertainties associated with the depths of grid cells depend on a variety of factors including the approach used to correct soundings for sound speed, vertical referencing, navigation, and echo-sounder uncertainties. In addition, the gridding process will affect the final depth assigned to each grid cell. The random error component is thus a difficult parameter to derive, primarily because of lack of metadata on the widely varying data sources and the fact that some contributions are in the form of gridded compilations. In several areas we still rely on digitized contours from published maps for which the underlying source data are unknown. While the random error component of DBMs have been estimated using statistical modeling approaches 41,42 , we do not provide this for IBCAO Ver. 4.0 because the metadata are not sufficient to provide a classification to a large enough portion of the database. Instead, the accompanying TIDs and SIDs provide information that is useful for users when addressing the reliability of IBCAO Ver. 4.0. In addition, we have assembled two grids aimed to further assist users in assessing the reliability of the DBM: minimum and maximum depth grids. These grids report the minimum and maximum depth value for each grid cell, implying a depth range where the block median filter had several input depth values in one grid cell.

Usage Notes
The most common uses of the IBCAO DBM are map-making and/or geospatial analyses using GIS software and other tools capable of displaying geographic information. The DBM is provided in netCDF and GeoTIFF formats, which are readily imported into most standard GIS software, for example QGIS and ArcMap. The 'x' and 'y' variables within the netCDF/GeoTIFF grid files represent the grid cell positions, along the x and y axis, in Polar Stereographic projection coordinates (meters), with a true scale set at 75°N. For the DBM, the 'z' value represents elevation in meters, depths below the sea surface are negative and heights above the sea surface are positive. The horizontal datum for the dataset is WGS 84 and the vertical datum can be assumed to be Mean Sea Level (however, note that there may be vertical reference issues for older observations, which may be due to chart datum). For the TID grid, the 'band 1' value represents the TID code, describing the type of data on which the corresponding cell in the DBM grid is based. A list of TID codes is given in Table 1. The projection parameters are provided in the European Petroleum Survey Group (EPSG) database (https://epsg.io/) as code 3996. This database is used by standard GIS software implying that searching for EPSG 3996, or IBCAO, will provide the correct projection and datum for the IBCAO DBM.
The Polar Stereographic coordinates can be converted to geographic using the GMT   www.nature.com/scientificdata www.nature.com/scientificdata/ coordinates is achieved by adding -I to the command above. See http://gmt.soest.hawaii.edu/doc/latest/mapproject.html for more information.
The GDAL command gdaltransform can also be used to convert between the Polar Stereographic and geographic coordinates by calling for the EPSG   grid, now referred to as the 'IBCAO Ver. 4.0 Grid' , is available from https://www.gebco.net/. It is provided on behalf of the IBCAO project under the terms of the disclaimer information as given below.
The IBCAO Ver. 4.0 Grid, should NOT be used for navigation or for any other purpose involving safety at sea. The IBCAO Ver. 4.0 Grid is made available 'as is' . While every effort has been made to ensure reliability within the limits of present knowledge, the accuracy and completeness of the IBCAO Ver. 4.0 Grid cannot be guaranteed. No responsibility can be accepted by those involved in its creation or publication for any consequential loss, injury or damage arising from its use or for determining the fitness of the IBCAO Ver. 4.0 Grid for any particular use. The IBCAO Ver. 4.0 Grid is based on bathymetric data from many different sources of varying quality and coverage. As the IBCAO Ver. 4.0 Grid is an information product created by interpolation of measured data, the resolution of the IBCAO Ver. 4.0 Grid may be significantly different to that of the resolution of the underlying measured data.  (Fig. 1). (a) Systematic multibeam surveys in 2014 by Swedish icebreaker Oden mapped a trough formed in the ridge crest, Oden Trough, and a critical sill depth influencing water exchange across the ridge 6 . In addition, lineations were mapped on the ridge crest, interpreted to be formed by a grounded ice shelf during the penultimate glaciation at about 140 000 years ago 37 . None of these features could be seen in IBCAO Ver. 3.0 (b) because it was compiled in this area through gridding of bathymetric contours retrieved from the Russian map "Bottom relief of the Arctic Ocean" 43 . The 1500 m isobaths derived from Ver. 3.0 (white) and 4.0 (black) shown in b clearly illustrate the large bathymetric differences between the two versions in the area of the sill. (c) The portrayal of the two spurs extending from the Lomonosov Ridge at about 84°N 155-160°E, one of them named Senchura Spur, are improved in Ver. 4.0 compared to Ver. 3.0 (d) due to additional multibeam bathymetry and adjustment of navigational issues in SCICEX 1999 (see main text).