This manuscript addresses an important observational challenge over the Tibetan Plateau by developing an autonomous cloud detection algorithm that operates solely on infrared brightness temperature (IRBT) data. The novelty lies in eliminating auxiliary meteorological inputs and introducing a dust normalization procedure. The work is clearly motivated, well-structured, and the methodology is explained in detail. The results show promising agreement with radiosonde-based cloud detection, especially in the wet season. However, there are several points where methodological justification, quantitative validation, and broader applicability discussion could be strengthened.

Specific comments

1. My main concern is the selection of the several threshold values. These thresholds (e.g., 150% of clear-sky maximum for spectral test, SD > 0.3 for temporal test) are critical to the algorithm’s performance, yet their derivation appears empirical. I suggest the authors conduct a sensitivity analysis to show how performance changes with threshold variation and whether optimal thresholds are season-dependent.
2. I wonder what is the temporal resolution of the detection? Is it a minute basis or daily basis? The authors mention using a sliding window but also compare the BT with the diurnal cycle, so it is not clear whether cloud is detected for every measurement or over the window, or over the course of one day.
3. The dust normalization method is innovative, but its assumptions may not always hold—especially during days with persistent cloud cover, when the “daily minimum IRBT” may not represent a dust-free baseline. I suggest the authors separately examine the effect of normalization under strong dust contaminated days. The cloud recognition results should also be evaluated separately for dusty and dust-free days.
4. While radiosonde data are used for validation, the temporal (twice daily) and spatial (4 km offset) mismatch between radiosonde and radiometer observations is significant. This likely underrepresents actual performance. Is it possible to use collocated higher temporal resolution instruments (e.g., ceilometers, cloud radar, or even sky cameras) to validate the results? Or at least some quantitative discussion of potential errors caused by the mismatch should be provided.
5. I also suggest the authors discuss the performance of the algorithm for different cloud types and heights. This, combined with comment 2, might explain part of the poorer performance for the winter season.
6. The flow chart of Figure 7 is very important. However, this figure is too simplified lacking detailed information. For example, the normalization strategy and specific thresholds should be added.
7. While prior IRBT-based cloud detection work is cited, the manuscript could more explicitly compare its results with those of similar autonomous algorithms in other regions to highlight relative strengths and weaknesses.

This paper presents a novel cloud detection algorithm specifically developed for ground-based infrared radiometers operating on the Tibetan Plateau. The algorithm enhances detection accuracy by integrating spectral and temporal tests, which effectively mitigates the impact of dust contamination. This approach is valuable for observations that rely only on infrared radiometer data. The algorithm's performance is evaluated using various case studies. Consequently, this research offers a significant contribution, and the algorithm has a potential for broader application across the Tibetan Plateau in the future. Further improvements of the current manuscript could be considered by addressing the following comments.

Main comments:

1. Recent advancements of cloud detection in the Tibetan Plateau regions can be considered in the Introduction.
2. To enhance clarity, authors may consider presenting a comprehensive flowchart that displays the primary algorithm's logic and main steps. While Figure 7 illustrates the algorithm flow, it would be beneficial to include additional details described in the text, such as the steps of normalization and the calculation of the clear-sky IRBT diurnal cycle. This comprehensive visualization would assist readers in understanding the entire algorithm. In addition, Figure 7 can be repositioned to appear earlier in the methodology section.
3. In Section 2.1, it is recommended that authors present a figure or table summarizing the instrument information, its surrounding environment (e.g., location, elevation), and typical sky conditions.
4. Thresholds used in the section 3.2 and section 3.4 need to be clarified, such as 150% of the maximum and the lowest 5% of IRBT values. Authors may provide a brief discussion on how these thresholds were determined.

Some errors should be corrected:  

Line 58: ‘ too weak to reliably distinguish’ -> ‘too weak to be reliably distinguished’

Line 62: ‘that combines’ -> ‘that combined’

Line 248: ‘and are shown’ -> ’are shown‘