Dear reviewer,

Thank you for reviewing the paper and for your constructive comments.

Your suggestion is highly relevant, and machine learning is indeed a promising way to integrate geophysical data in hydrological studies. So far we are only discussing machine learning approaches for prediction purposes and spatial heterogeneity representation. There are some recent papers that investigated the use of machine learning for petrophysical relationship (e.g. Gottschalk and Knight, 2022) for airborne data, which opens the possibility to inform the spatial variability of aquifer at the catchment (or even beyond) scale.  We will seek to also address such innovative approaches and expand on the use of machine learning in the revised manuscript.

Best wishes,

For the authors

Thomas Hermans

Dear reviewer,

Thank you for reviewing the paper and for your constructive comments. Below, we provide a response to your comments.

• Concerning the detailed level of information. We realize that the different subsections are relatively long with a lot of details provided. We think this level of details is necessary to stress the challenges. However, we also recognize that there is some redundancy, including between the different sections, and we will try to reduce as much as possible the length of the text. We wrote a small summarizing paragraph at the end of each section to deliver a simplified message, and we will try to strengthen this in the revision.
• On delivering opinion about when 4D is needed. In the current version, we provide a general discussion on processes requiring a 4D characterization in the concluding remarks, rather than discussing every process individually, as it is impossible to provide an exhaustive list of these processes. We recognize that this is coming a little bit late and we will strive to discuss important elements earlier in the manuscript. In this section, we stress that 4D is key for process understanding, and a necessary step for upscaling processes at larger scales. We also propose to generalize the use of (global) sensitivity analysis to investigate the role of spatial heterogeneity and time variations, so that simplifications used are based on scientific evidence. However, we agree that this message could be stronger, and we will follow your suggestion to extend this part.
• On the suggestion to use more computing power. Our main message did apparently not come across: we indeed identify cloud computing as one avenue to allow the inclusion of more complexity when it is needed, but this is only one of the recent innovation that is listed among others that are dealing with the modelling approach itself (use of proxy, upscaling methods, Monte Carlo based approaches). We will revise the manuscript to make it clear that cloud computing is just one of the proposed innovation.
• About the scale. We agree that the scale is of uttermost importance. For the four processes we have selected, understanding the implications of variability at a very small scale (pore-scale to meter scale) is key to be able to upscale and simplify the processes at a larger scale. For example, mixing and reactions are often not adequately represented by macro-dispersion. The idea of a figure is appealing, we tried to include it in figure 1, but what your propose would be indeed clearer. We will strengthen the message about the scale issue in more detail in the revised manuscript.

Best wishes,

For the authors,

Thomas Hermans