This paper greatly attempts to map the annual global land cover change using all Landsat observations and the CCD approach. The paper is well written with clear logic, while some issues need to be clarified for publication as a scientific data paper.

1. The CCDC approach is very sensitive to the parameters adopted, which may identify pseudo changes. This may directly relate to the derived temporal stable regions, and further impact the results of global changed areas (as well as the statistics). As a scientific data paper, this part should be enhanced with quantitative analysis, especially the uncertainty of change detection across different cover types.

1. As a global land cover change dataset, the accuracy assessment regarding those changed pixels should be significantly improved. The confusion matrix just shows the accuracy in 2020, while those changes in different years with field samples as well as global product comparison (e.g., ESACCI and MODIS) should be given and discussed.

The authors present a very detailed manuscript on the generation of a global 30-m land cover product. It is original in the spatio-temporal density of Landsat satellite imagery used to generate annual maps over nearly 30 years. I recommend this paper for publication. There is one major issue that needs to be addressed and a number of minor or editorial issues to address:

Major change:

Both the Continuous Change Detection and Classification (CCDC) Algorithm and the Random Forest Algorithm for subsequent land cover classification use several hyperparameters. Both models will be sensitive to the hyperparameters selected. As a minimum, the hyperparameters selected for both models need to be clearly defined and justified (this is already partially done for the CCDC algorithm). However, to fully justify the use of hyperparameters, sensitivity analysis should be provided of the values used, and validation that the optimum or a favourable set of hyperparameter values have been selected.

Minor or editorial changes:

Introduction/ methods – various mentions of model ‘accuracy’ is used. This includes, but is not limited to lines 31-32, 69 and 155. Please be specific on the accuracy metric(s) used.

Line 24- The use of the phrase ‘In specific’ is awkward and I suggest changing e.g. ‘Specifically’ or ‘In particular…’

Line 201- ‘The first time series validation set was assessed the performance…’ remove ‘was’.

Line 205- Change ‘It developed by combining…’ to ‘It was developed by combining…’

Figure 2: This Figure is very useful for help the reader understand the main processes carried out in this project. Please add the shortened names of each dataset to the flow chart to make it even easier for the reader to follow the text.

Figure 2: You refer to masking ‘poor quality’ pixels. Please be more specific on this. Does this just include applying a cloud mask, or does it also consider issues with the Scan Line Corrector on Landsat 7, for example. What cloud mask was used. How did you account for pixels that may be under light cloud/ haze which may not be picked up by a cloud mask (e.g. does the CCDC intend to overcome this?)

Table 1: Please add the abbreviations for each land cover type to this table (at later points you refer to Table 1 as containing these).

Line 357- please provide more information on the indicator function, or at least a reference.

Figure 7- the very thick lines corresponding to pixels with a stable land cover overwhelm this image and make it difficult for the reader to decipher the most dominant types of land cover change. Please either rescale the image or consider removing the lines corresponding to no land cover change to make it easier for the reader to assess the dominant types of land cover change.

Table 2 and 3- please use a method to highlight the relative performance of your algorithms. For example use a colour ramp or make particular values bold.

Results and discussion are very thorough although there is no mention to coastal regions which will be areas of major change detectable at 30 m resolution.

Overall, I enjoyed reading this paper. The analysis was very thorough and easy to follow.