Overall this seems like a useful step forward in characterizing the distribution of hazardous materials sites in Alaska, and the meshing of OSM data with other data sets is clever. Some minor revisions, especially to make the figures more clear, would benefit the paper (see comments in the attached pdf.)

Review Kaiser et al., 2023

The combination of available infrastructure data sets to increase the semantic content is a good idea that has the potential to improve future analysis regarding contaminated sites and related hazards. Especially the increased number of usage categories is promising for many applications. The usage examples are well done and will be helpful for future users of the data set.

Below I outlined several points that would improve the manuscript in my opinion.

• Figure 2 may be valuable for users of the data but could go into the appendix. Instead, a figure like a flowchart or decision tree on how the input data sets were combined may be beneficial for the reader for a better understanding of the process.
• Regarding OpenStreetMap: As I understood from the text, the authors use OSM as an alternative source of linear infrastructure instead of the satellite derived Bartsch et al dataset. Previous publications including Bartsch et al 2021, which the authors heavily cite throughout the text, point out that OSM has some major drawbacks, including inconsistencies, and lacking some more recent infrastructure developments. In line 147-148 the authors note that they use OSM for the linear infrastructure, not SACHII, for buildings they describe a decision tree of combining the data sets (from line 207). Is there a reason a similar approach was not used for linear infrastructures, especially given the concerns previous publications have raised about OSM. Maybe just some additional discussion could be included on this.
• Figure 6: interbal – internal?
• Figure 10 could go into the appendix.
• Line 470: Is this issue not shown in Figure 8? Maybe include a reference here to this figure, just for improved understanding for the reader.
• Maybe I missed it in the text, but your input data sets have different extents? So would the accuracy of your output not differ between areas where both/all inputs were available and those that rely solely on OSM data?