This dataset comprises the global frequency, classification and distribution of marine heat waves (MHWs) from 1996-2020, in Chauhan et al. 2023 (https://doi.org/10.3389/fmars.2023.1177571). The classification was done based on their attributes and using different baselines. Daily SST values were extracted from the NOAA-OISST v2 high-resolution (0.25°) dataset from 1982-2020. MHWs were detected using the method presented by Hobday et al. 2016 (https://doi.org/10.1016/j.pocean.2015.12.014), and by using the 95th percentile of the accumulated temperature distribution to flag the extreme events. A shifting baseline of 8-year rolling period was selected between the years 1982-1996, since this period shows relatively stable maximum values of temperature across different ocean regions. The shifting baseline aims to account for the decadal changes of westerly winds, temperatures and ocean gyres circulations. The classification was done using the KMeans clustering algorithm to identify the relevant features of MHWs and classify them into separate groups based on feature similarities. This algorithm takes MHW features, namely duration, maximum intensity, rate onset and rate decline, as input vectors and applies clustering in the 4-dimensional feature space where each data point represents an MHW event. Note that all the MHWs features are standardized because unequal variances can put more weight on variables with smaller variances. This record comprehends the geospatial datasets of: Average number of MHW days per year (i.e., the sum of all MHW days divided by the total number of years, 1996-2020). Average cumulative intensity per year (i.e., the sum of cumulative intensity divided by the total number of years, 1996-2020). Total number of MHW events across the different periods averaged on the total number of years (1989-2020). The period 1982-1988 was only used as an initial baseline without calculating MHWs. Spatial distribution of three MHW categories: moderate MHWs, abrupt and Intense MHWs and extreme MHWs; displaying the total number of MHW days normalized by the number of years considered (i.e., 1989-2020). Distribution of Extreme MHWs across the different periods (A) 1989-1996, (B) 1997-2004, (C) 2005-2012, (D) 2013-2020. The relative frequency (γ) is a ratio of extreme MHWs in a specific period and all extreme MHWs in the same cluster for all periods.

Maps of potential biomass catches (tons/year) per surface unit (0.25º latitude x 0.25º longitude) based on 3-D probability of occurrence for the main commercial fish species of the Atlantic. To map potential catches, first, mean catches (tons/year) were calculated according to Watson (2020) Global fisheries landings (V4) database for period 2010-2015 and then the total mean catch value for each species was redistributed according to the occurrence probability value that was modelled in 3-D using Shape-Constrained Generalized Additive Models (SC-GAMs). Potential catch value of each cell integrates the catches along the water column (from surface until 1000 m depth). See Valle et al. (2024) in Ecological Modelling 490:110632 ( https://doi.org/10.1016/j.ecolmodel.2024.110632 ), for more details.

3-D habitat suitability maps (HSM) or probability of occurrence maps, built using Shape-Constrained Generalized Additive Models (SC-GAMs) for the 30 main commercial species of the Atlantic region. Predictor variables for each species were selected from: sea water temperature, salinity, nitrate, net primary productivity, distance to seafloor, distance to coast, and relative position to mixed layer depth. Each species HSM contains 47 maps, one per depth level from 0 to 1000 m. Probability values of each map range from 0 (unsuitable habitat) to 1 (optimal habitat). For depth levels below the 0.99 quantile of the depth values found on the species occurrence data, NA values were assigned. Maps have been masked to species native range regions. See Valle et al. (2024) in Ecological Modelling 490:110632 (https://doi.org/10.1016/j.ecolmodel.2024.110632 ), for more details.

These datasets contain 4D (x, y, z, t) weekly temperature and marine heatwaves (MHW) categories estimated from the surface up to 300-m depth, at a 0.25°x0.25° horizontal grid resolution and for 4 areas of interest that are: • Area 1 (around the Madeira Islands): 30°N-35°N, 15°W-20°W • Area 2 (Tropical Pacific Ocean): 30°S-30°N, 120°E-130°W • Area 3 (Mediterranean Sea): 40°N-45°N, 15°W-20°W • Area 4 (Global): 82.875°S-89.875°N, 0.125°E-359.875°E The weekly MHW are centered on the date of the file (±3days). For the temperature reconstruction, 2 approaches have been used: - for the regional areas, the temperature has been computed with a 2 steps method: a first estimate of the vertical temperature profiles by using a machine learning approach (Multi-Layer Perceptron (MLP)) and then, a combination of this field with in situ temperature profiles observations through an optimal interpolation algorithm. The Copernicus Marine Service ARMOR3D dataset was used as the targeted temperature field for the MLP. The input data used are: • First step: ◦ SST data are from daily OSTIA analyses [from Copernicus Marine Service: SST_GLO_SST_L4_REP_OBSERVATIONS_010_011 product] interpolated over the 0.25°x0.25° targeted grid resolution; ◦ SLA data are from the Copernicus Marine Service product SEALEVEL_GLO_PHY_L4_REP_OBSERVATIONS_008_047/dataset-duacs-rep-global-merged-allsat-phy-l4 • Second step: ◦ The in situ data are from the Copernicus Marine Service In Situ TAC and contains several observations type: CTD, Argo floats, drifting buoys, moorings, marine mammals). - For the global area, the temperature comes from the Copernicus Marine Service product ARMOR3D (MULTIOBS_GLO_PHY_TSUV_3D_MYNRT_015_012 (https://doi.org/10.48670/moi-00052). The MHW categories are derived from the Hobday’s method [Hobday et al.,2018] for the 4 areas. Each MHW event is classified among four categories (moderate to extreme), identified in terms of multiples of the local difference between the 90th percentile and climatological values, and defined as moderate (1-2×, Category I), strong (2-3×, Category II), severe (3-4×, Category III), and extreme (>4×, Category IV). When the category is zero, this means that there is no MHW. The period 1993-2021 is used as a baseline for defining the climatology to be as close as possible to the 30-year period suggested by Hobday. This choice is motivated by the need of altimetry data to constrain the vertical temperature reconstruction, which is required for most ocean reanalyses as well, therefore the baseline period slightly differs from the one used for the 2D atlas.