Different phytoplankton groups dominate ocean biomes and they drive differently the marine food web and the biogeochemical cycles. However, their distribution over most parts of the global ocean remains uncertain due to limitations in the sampling resolution of currently available in situ and satellite data. Information below surface waters are especially limited because satellite sensors only provide information on the first optical depth. We present measurements obtained during Polarstern cruise PS113 (May–June 2018) across the Atlantic Ocean from South America to Europe along numerous transects. We measured the hyperspectral underwater radiation field continuously over several hours from a vertical undulating platform towed behind the ship. Equivalent measurements were also taken at specific stations. The concentrations of phytoplankton pigments were determined on discrete water samples. Via diagnostic pigment analysis we derived the phytoplankton group chlorophyll a concentration (Chla) from this pigment data set. We obtained high resolution phytoplankton group Chla data from depth resolved apparent optical properties derived from the underwater radiation data by applying an empirical orthogonal function (EOF) analysis to the spectral data set and subsequently developing regression models using the pigment based phytoplankton group Chla and selected EOF modes. To our knowledge, this is the first data set with high horizontal coverage (50–150 km) and resolution (∼1 km) that is also resolved vertically for the Chla of major taxonomic phytoplankton groups. Subsampling with 500 permutations for cross validation verified the high robustness of our estimates to enable predictions of seven different phytoplankton groups’ Chla and of total Chla (R2 and median percent differences of the cross validation are within 0.45–0.68 and 29–53%, respectively). Our depth resolved phytoplankton groups’ Chla data reflect well the different biogeochemical provinces within the Atlantic Ocean transect and follow the distributions encountered by previous point observations. This verifies the high quality of our retrievals and provides the prospect to put similar radiometers on profiling floats or gliders which would enable the large-scale collection of vertically resolved phytoplankton data at much improved horizontal coverage relative to discrete sampling.