see attached file

Review ‘Observed global ocean phytoplankton phenology indices’

This work documents the process of creation of a dataset of phenological indexes of phytoplankton blooms on the global ocean, using the 26-years timeseries of the satellite-derived chlorophyll gathered by the OCCCI. The dataset could be potentially useful to feed other analysis. However, in its current state I see two main weaknesses, it is not validated (1), and it does not provide error estimates (2).

General comments

1. The manuscript shows, technically speaking, a great data analysis. The authors have done a rigorous job collecting data, filling gaps and applying globally-appropriate bloom detection methods. Their results look very neat, and it would be very interesting to see a deeper analysis. However, I do not see so clear the potential of these data being useful in the future to other scientists and therefore being published as an ESSD dataset.

The OCCCI Chl-a is itself a satellite-derived product, based on the disaggregation of the world ocean in a certain number of optically-homogeneous water classes. However, global algorithms, even blending different waterclasses, do not compare necessarily well to observations in certain regions, where regional algorithms are proposed (e.g. Johnson et al. 2013 in the Southern Ocean [https://doi.org/10.1002/jgrc.20270]; Volpe et al. 2019 in the Mediterranean Sea [https://doi.org/10.5194/os-15-127-2019]). To the best of my knowledge OCCCI do not consider regional-specific algorithms.

And on top of that it is the uncertainty of the phenological analysis performed (which is also not very well documented, see my next comment). With such level of derivation I do not see how these metrics provided could be considered observed data. This issue could be overcome if the authors present some comparison to in situ observations of Chl-a timeseries, observed phenology or other common standards, but that is not done in the current version.

Have the different bloom detection methods been validated with observational data on their own? Since obtaining global-scale validation data could be challenging, maybe one option is to perform a more formal analysis on the agreement/disagreement among methods considering in which temporal/spatial domain they have been validated independently.

2. The quality of the presentation is high. The dataset is accessible and straightforward to interpret. However, another big concern is that the dataset does not include any estimate of error associated to the metrics given. It is of utmost importance to provide such an error, considering that the trends on such metrics seem to be on the range of 5-10 days per decade. Dispersion metrics around the mean for each pixel (in the 9km and 25km versions) are also missing. I think these can be provided since phenology indexes are computed in the 4km version and later regridded (L115). There is no discussion about the potential sources of errors and limitations of the bloom detection methods, only references to other works (L135). Maybe the authors could mention the potential caveats of the methods when they elaborate on the agreement/disagreement between methods (L267).