Review for " Probabilistic precipitation downscaling for ungauged mountain sites: a pilot study for the Hindu Kush Karakoram Himalaya" by Marc Girona-Mata et al. submitted to EGUsphere (MS No.: egusphere-2024-2805).

General comments:

The authors present an innovative Generalized Probabilistic Regression (GPR) approach to Model Output Statistics (MOS) post-processing, specifically tailored for ungauged mountainous regions in the Hindu Kush Himalaya (HKH). By utilizing generalized linear models and artificial neural networks, the study convincingly demonstrates that GPR significantly improves the accuracy of daily precipitation estimates compared to raw regional climate model outputs and deterministic bias correction methods. This innovative approach of estimating daily precipitation addresses critical challenges posed by sparse precipitation data and the complex topography of the HKH region. This work contributes substantially to the scientific understanding of precipitation patterns in data-scarce mountain regions. The GPR framework not only provides a robust tool for estimating daily precipitation but also quantifies uncertainties for regions with limited data availability. This method has practical implications, offering decision-makers a valuable resource for improved water management and disaster preparedness. The overall quality of the paper is high, adding both theoretical and applied value to the field of hydro-climatology in the data spares and poorly understood region of the HKH.  While the paper is well-written, incorporating the suggested improvements below would further enhance its impact.

Specific comments:

1. How does the GPR perform under extreme precipitation?

While the GPR method improves precipitation estimation, its performance declines with increasing daily precipitation, as seen in Fig. 4 (30 mm/day). In the eastern HKH, especially during the monsoon season, daily precipitation often exceeds 30 mm/day. How would GPR perform under extreme precipitation conditions? The authors could explore a case study involving an observed extreme precipitation event and evaluate the GPR method specifically for such conditions.

1. How this can be used for winder benefits and decision making ?

The discussion section would benefit from elaborating on how the improved method can assist decision-makers. While the authors suggest that this method aids water resource management, the specifics remain unclear. For instance, could the authors detail practical applications such as how this method can be used in improving  water resources related decision making ?

Technical Corrections:

Here are technical comments P refer to page number while L refer to line number.

• P1L17: Include the names of the major rivers originating from the HKH region. P2L21: The term "Hindu Kush Himalaya" (HKH) is widely recognized. Clarify why the authors include "Karakoram" in the name while retaining the abbreviation HKH.
• P2L2021 and Fig 1: Add a background map showing the HKH's location in the context of the globe or Asia. Include major rivers. This would enhance the paper’s appeal to a broader audience.
• P2L25: What is mean by ‘this precipitation knowledge gap..’ not clear

This paper addresses an important topic of correcting output of daily precipitation from a regional climate model for sparsely gauged mountainous regions. The paper is well written. While the method is based on the unrealistic assumption of conditional independence spatially and temporally, it is understandable why the authors have done so. It was designed to make the best use of the very limited gauge data. The experiments were well designed. I have three main comments:

• On evaluation, while it is OK to use CRPSS to compare various probabilistic results, it is not useful for comparing the probabilistic results with the deterministic results (WRF_SF). Deterministic results tend to produce inflated CRPS and therefore should be compared with median or mean of probabilistic results.
• Also on evaluation, I suggest that the authors should evaluate the tails of the results (substantive or extreme events) by using twCRPS (threshold-weighted CRPS). It is important to know how good the results are on high precipitation.
• The probabilistic results are spatially and temporally unconnected, making it difficult for the users to apply the results. The authors should carefully consider this aspect and recommend how the results can be used when spatial and temporal connections are needed as the case in most applications.