The authors present SCAFET as a new framework to extract weather features from large climate datasets. SCAFET follows the standard paradigm of segment, filter, and track. The traditional approach to segmentation utilizes absolute thresholding of appropriate climate variables, which is well known to be sensitive to the particular model, climate state, and even spatial location. This makes it difficult to have a standardized detection algorithm that works uniformly across models and warming scenarios etc. SCAFET is instead shape-based, giving a relative and directional thresholding, making it a much more robust approach for weather feature identification. The authors demonstrate its utility by identifying and tracking atmospheric rivers, cyclones, fronts, and jet streams. SCAFET is a significant advance over the traditional absolute thresholding methods currently used by climate practitioners. With some minor revisions, see below, I recommend the manuscript for publication.

1. My main comment or question is related to how sensitive feature identification is to SCAFET parameters. You have shown that it is possible to identify weather features with SCAFET, which is great, but there is no discussion on how sensitive the results are. For example, how sensitive is the detection of ARs in Figure 4 to the parameters used in Table 1? On the one hand, it is intuitive to identify ARs as long, narrow shapes with (relatively) high IVT and precipitation. But concrete numbers must be used to implement that intuition. If you slightly change the SI threshold for Ridges, or the minimum length, or angle coherence, etc. does this totally change the kind of objects identified so that they no longer resemble ARs (I wouldn't think so, but perhaps), or does it slightly change the details of ARs detected? If it is the latter case, how did you decide on the exact values used in Table 1 for the best identification of ARs? I see there is one sentence, "The quantitative values for the properties are obtained from a consensus of previous studies referenced within each section." but I think this requires more elaboration.
2. My second question is, what are we supposed to take away from Section 4.1 on Jet Streams? It shows some proof-of-concept that the method can be applied, in principle, to 3D data. To my eye, I don't see a clear jet stream identified by SCAFET in (b), (d), and (f) of Figure 7. So while the method can be applied to 3D data, it is not clear that it is successful in identifying features in 3D data. 
3. Third, while I think SCAFET is indeed a significant advance, I believe there are some statements made in the paper which are not justified, or I have misunderstood what you are trying to say.
   1. Around line 40 there is discussion of dataset pre-processing, such as computing IVT fields for AR detection, and how this becomes infeasible for high resolutions and large ensembles. At first I read this as implying pre-processing as a downside of traditional methods, but something that SCAFET would bypass. However, SCAFET itself uses these pre-processed fields in the identification of ARs and cyclones.
   2. Starting at the end of line 327, there is the sentence "Due to its design, SCAFET does not require a priori climate information to identify features." I am not sure what is meant by this sentence. In the work presented, the shape-based component is only one piece of the full pipeline to identify weather features. Most obvious, the shapes are extracted from pre-processed fields IVT and RV, which are created from "climate information". Even knowing what generic shapes are appropriate for particular weather features I see as climate information.
4. Finally, the writing and grammar etc. of the paper need to be cleaned up. Below are some instances I found during my reading:

• line 15: "… and value (5Vs) of climate data (REFs) of climate data."
• starting in line 162: "In the current study, a simple radius is defined and the closest object within the given radius to each object at time n is clustered and identified from time n+1 as the same object in motion." I get the general idea of what you are saying here, but I found this sentence hard to parse.
• line 186-187: "…derive this threshold from dataset directly, …"
• line 204-205: "… each object is used as to filter …"
• line 274: "…, SSTFs are not tracked as ocean fronts are stationary rather than…"
• in the middle of the Figure 6 caption, "In the next step, ridges, caps, and domes are extracted from (b) and weak and small…", do you mean for this to be (a) instead of (b)?
• line 293-294: "Since the scope of this section is limited to the validation of the detection method, we have only shown jet detection in three selected time steps." I'm not fully sure what you are trying to say here. Do you mean that filtering and tracking steps have not been performed here?
• Figure 7 caption: "The 3D jet streams extracted for the corresponding time period is show in …"
• line 362-363: "change of direction of a along the curve."
• In Figure A2, please adjust the legend so it can be read more clearly