Comment on nhess-2021-212

The authors try to test three filtering methods to reduce scatter in landslide monitoring data. They check in several testing frameworks using synthetic time series whether filtering would lead to losing an early warning pulse. They also demonstrate their method with a real-world case. The article is well written and fluent. The study's motivation and implementation are of interest to the landslide research community, and they are adequate for NHESS. Before being published, I recommend some minor revisions to increase the potential impact of the research.


Introduction
The introduction is well written. Authors could consider extending their literature review by including some other examples. A few citations seem to appear oddly frequent. I also find series of citations after a particular statement relatively inefficient; authors should consider elaborating why they cite a specific paper (this comment applies to other sections as well).
Another aspect that could increase the fluency of the text is to use prefixes, for example, using "infeasible" instead of "not feasible". Would you please check other possible places where this point could be improved?
Lines 52-60 sound like a method, and I recommend rewriting this section a little more towards a "problem statement".

Methodology
There are a series of abbreviations used in the article, some of which make sense not to repeat each sentence. However, some of them decrease the fluency of the text, e.g., NASD or BR.
Line 134: Do the authors mean residuals when stating "ratio of scattering amplitude".
Lines 156-162: mentioning the GWMA acronym once might help the reader link the abbreviation to the subsection.
Line 169: Here, does the word "above" relate to GWMA"? Line 170: Authors mentioned "evenly spaced" data. Does the method also work with missing data or unevenly spaced data? Later on, in line 262, a similar issue is mentioned: when a data point is classified as an outlier, it is replaced with a new interpolated data point, which is not a proven practice. It could be an option to leave it as a NaN, or some bootstrapping and randomization process needs to be implemented.
Lines 181-186: I found this part of the text somewhat confusing; please consider reformulating the text.
Line 241: even if it is commonly used, please state the open form of GNSS.

Results and Discussion
Would it be possible to introduce sub-sections to distinguish the effects of direct and indirect filtering on the synthetic analyses? Line 373: "even", is it "event"?
Line 436: Could authors perform RMSE or another approach to quantify the trends in the data instead of visual inspection.
Line 464: Could an example of the outlier removing process be supported with a figure to visually demonstrate which type of data point authors considered an outlier. Tables   Table 1: Could table 1 be integrated into Figure 1 to save space and better relate to one another? Table 2: "60-s reading" could be mentioned as "1-m reading" to have synchrony with "1-h reading".  Currently, Geocubes are not explained in the figure. Resolution is also low to see the individual features correctly. Also, in subplot (a), the location of the study site is not highlighted in the map of British Columbia. Figure 6: Would it be possible to combine subplots to compare them better. Y-ticks are tough to relate to at the moment. Figure 9: Instead of a linear y-axis, an option could be y^2 (e.g., 2 4 8 16) or y^3 (e.g., 2 8 32 128) style access to distinguish the lines from one another for better visibility. Similarly, please consider this approach also for figure 17.