This is a great and valuable dataset, and I appreciate the authors hard work in aggregating and making this data publicly available. I downloaded the dataset and it was easy to navigate, visualize, and extract layers. The methods and uncertainties are well described in the paper. I have a few comments listed below:

-The gap assessment and spatial assessment in this study are both very useful for error quantification. However, I think there is an additional error term or test that could be helpful especially in context of your MethaneAir analysis - the gaps that exist in a "dense" country. Or another way, "if MethaneAir observes plume in the Permian, how close is the nearest piece of GIS infrastructure, and is it right to attribute it to that plume?" You show proof of concept with MethaneAir, but EDF has gathered extensive and independent facility-scale observations with attribution through PermianMap where you could do this test rigorously.

-Related to previous comment: in addition to randomly sampling GIS infrastructure and comparing to visible imagery, why not assemble an independent list of GIS infrastructure solely through visible inspection (say 100 of so in various basins/countries) and then see if your GIS database has elements near your visual list? This would be a more blinded test of gaps in the inventory and spatial accuracy. 

-The analysis of Permian emissions based on the inventory raises confusion. After application of literature-based emission factors, you get an emission budget/basin loss rate on par with previous atmospheric inversion studies (3.1 Tg). However, the conclusion that this budget is dominated by upstream low-emitting sources is a very different conclusion than what's been found in previous studies, some of which have been published by EDF. For example, from the abstract of Lyon et al. (2021) [https://doi.org/10.5194/acp-21-6605-2021]: "the Permian Basin is in a state of overcapacity in which rapidly growing associated gas production exceeds midstream capacity and leads to high methane emissions." Also, in your description of activity/emission factors, you use EPA GHGI emission factors for gathering pipelines - however from Yu et al. (2021) [https://doi.org/10.1021/acs.estlett.2c00380]: " In this study, we use methane emission measurements collected from four recent aerial campaigns in the Permian Basin, the most prolific O&G basin in the United States, to estimate a methane emission factor for gathering lines. From each campaign, we calculate an emission factor between 2.7 (+1.9/–1.8, 95% confidence interval) and 10.0 (+6.4/–6.2) Mg of CH4 year–1 km–1, 14–52 times higher than the U.S. Environmental Protection Agency’s national estimate for gathering lines and 4–13 times higher than the highest estimate derived from a published ground-based survey of gathering lines." Application these alternative emission factors which have been observed in the Permian would certainly change your conclusion about upstream/midstream. Without this context explicitly stated in your Permian analysis, this section reads as an attempt to arrive at a certain prescribed conclusion, which I don't believe is your intent. The preferred approach would be to make an ensemble of estimates based on the many emission factor distributions that have been recently observed. Or at minimum, the manuscript should state that upstream/diffuse result from your inventory is not totally consistent with other studies from the Permian. However, this seems outside the scope of a data description paper and may be worth total removal/saving for another analysis. 

-On the Zenodo DOI webpage it states that datasets for Russian compressors and VIIRS are not included in the dataset due to permissions. I did not see that description also written in the manuscript, where it should also be.

The oil and gas infrastructure mapping is very important to monitoring and modeling the GHG emissions for limiting climate change. This work provides the most complete dataset so far that collects infrastructure information from online sources. The infrastructure type and geolocation from this work can be very useful to the GHG emission inventory developments and modeling from facility to regional and global scale with remote sensing images. Also, it can also be used as the ground-truth dataset for the oil and gas infrastructure identification with remote sensing images and machine learning approach. The methods are clearly described, and the paper also provides a detailed bottom-up emission inventory case study using this dataset.

I’m very excited to download and look at the OGIM_v1 dataset (OGIM_v1.gpkg), however, I found the currently provided dataset should be further modified before it can be used in other research:

1. It would be better to provide the mapping from column names as well as the shortcuts used in the entries (especially for the “FAC_TYPE”) to their detailed meanings, maybe provide a table in the supplementary or in the description of Zenodo?
2. Figure 6. I see oil and natural gas infrastructures in some countries, such as China and India, but I cannot find them in the OGIM_v1 dataset. There is no “China” or “Indian” in the “COUNTRY”.
3. It will be interesting to see how the emissions of the natural gas compressor as well as the natural gas flaring changed from remote sensing images after the Ukraine war if the infrastructure information is provided by the dataset. So, if the Russian data are not included in the current version, it should be stated in the paper unless it will be available soon.
4. I tried to extract the pipeline from the OGIM_v1 dataset, however, there is no such type either from “FAC_TYPE” or “geometry”. The same issue also exists for the fields or basins.
5. Line 360-365: “We quantitatively assess data quality in each country for which open oil and gas data for these facilities are available in the OGIM_v1 database”. This will be another advantage of this dataset if all the entries are labeled with quality scores, and users with different research purposes can easily select them without any data cleaning processes. But I did not see the score from the dataset.

I uploaded a simple test code, maybe I missed them?