Allen Hunt's answer to Chaopeng Shen Commentary

[Riccardo Rigon]

I would like to respond to Professor Chaopeng Shen's post. From last year's paper, C O M M E N T A R Y

"The physics and the biology of the water balance: A personal

journey through the critical zone into the water balance," in Hydrological Processes, I quoted then Chairman of the NRC Frank Press' Foreword to the NAS book, "Opportunities in the Hydrologic Sciences," Press (1991). “Its pragmatic focus has left fundamental hydrologic sci-

ence lagging behind [compared] with other geosciences. The result is a

scientific and educational base in hydrology that is incompatible with

the scope and complexity of many current and emerging problems.” and added the comment,

"What else is machine learning but the ultimate pragmatism?"

The number of citations generated through such conferences is truly staggering, but I am reminded a second quote of Klemes who agreed with another author who  "criticized the common practice whereby ~speculative a s s e r t i o n s . . , become scientific f a c t . . . [by a] proper number of

citations in the literature." Trying to impress me by the number of citations when it is in the hundreds of thousands does not work; it sounds more like bot-amplified misinformation numbers. If that is the future of science, I am glad that I was educated in another generation, when the term educate still meant "to pull out of." In this line, I expect our article (with Jasper Vrugt and Gabriel Katul) on science and education in hydrology to be published in the opening issue of "Hydrology and Water Resources" with Yongqiang Zhang as Editor-in-Chief.

Nevertheless, I can comprehend the need for practical solutions in parallel with advances in understanding, and I believe that authors such as Vit Klemes could have, too, but I have looked on appalled at the prospect of a future with no science education in hydrology at all. Klemes in 1988 was appalled at the thought that 80% of the discipline could not be considered science, but was more nearly fudging data. Now, in spite of the existing rush to AI, hydrologists are being chided for showing too much allegiance to the science of hydrology? Well, I cannot say that I did not see it coming, but the members of the above-mentioned committee, including Peter Eagleson, Ignacio, V.K. Gupta, Tom Dunne, Syukuro Manabe, Gary Sposito, Donald Nielsen, Wilfried Brutsaert, and the rest probably would not have seen it in Professor Shen's way, and I sincerely believe that I would be speaking for some of those members. I have a close contact with one of the authors to back up the statement.

In any case, I will not be attending. But everyone must make their own decisions.

Allen Hunt

[Joe Janssen]

I love that Vit Klemes has been brought up in this discussion because as a lowly PhD student (who has also published/reviewed for these ML conferences), I deeply admire his work. With love and respect, I would like to caution Dr. Hunt (and many other hydrologists) against judging a book by its cover. Indeed, not all of machine learning is about predicting the next token. If you look through the top ML conferences (i.e., Neurips, ICML, ICLR, AISTATS, UAI, etc...) there are many papers about causal inference, system identification (i.e., Dr. Shen's paper), and optimization which CAN be used for advances in understanding. Dr. Klemes seemed to be very excited about the prospect of using causal inference instead of empirical modeling back in the 80s, though he seemed to think that causality was limited to physically-based models. Since the 80s, advances in causal inference have been widespread and have shown that we can indeed learn processes from observational data under certain circumstances.

One of my favorite quotes from Dr. Klemes is: "While this danger is serious and universal, it remains dormant until the models take the shape of easy-to-use software packages. At that stage the unholy trinity of good intentions, ignorance, and efficiency closes the circle and a hydrologic misconception becomes a virtually insurmountable obstacle to progress in hydrology”. To me, this quote speaks about the need for hydrologists to learn about the methods they are using (and all the assumptions) before using a model, rather than ignoring reductionist models all together. 

Also, it is important to note that the two papers that Dr. Shen shared probably represent the most important scientific advances in the past decade, so I think their 100k+ citations are deserving. In fact, one of those papers is about diffusion modeling, a subject that may even be of interest to you Dr. Hunt (and other fluid dynamics people)! While that particular paper is about image generation, ML researchers are also very interested in diffusion models and system identification related to the fokker-planck equation (advection-diffusion). See for example the two papers (from ML conferences) below which may actually be very applicable to hydrology:

Janoos, F., Denli, H., & Subrahmanya, N. (2014). Multi-scale graphical models for spatio-temporal processes. Advances in Neural Information Processing Systems, 27.

Terpin, A., Lanzetti, N., Gadea, M., & Dorfler, F. (2024). Learning diffusion at lightspeed. Advances in Neural Information Processing Systems, 37, 6797-6832.

Anyway... hope this discussion continues,

Joseph Janssen

https://hydroml.github.io/

---

From: abouthy...@googlegroups.com <abouthy...@googlegroups.com> on behalf of Riccardo Rigon <abouthy...@gmail.com>
Sent: June 13, 2025 4:47 AM
To: AboutHydrology <abouthy...@googlegroups.com>
Subject: [AboutHydrology] Allen Hunt's answer to Chaopeng Shen Commentary

 

--
For see all the Job and Positions give a look at: https://hydrojobs.blogspot.com/
To see the webinars announced: https://hydrowebinartribune.blogspot.com/
---
You received this message because you are subscribed to the Google Groups "AboutHydrology" group.
To unsubscribe from this group and stop receiving emails from it, send an email to abouthydrolog...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/abouthydrology/393afbce-959d-465b-826b-aa56c3a54560n%40googlegroups.com.

[Solomon Vimal]

I agree with Joseph Janssen, that the citations are deserved. This was in fact brought up in the conference HydroML where Prof. Praveen Kumar, a keynote speaker, argued that they should be cited even more! Prof. Shen is making our field proud and we should really read his papers closely! 

To view this discussion visit https://groups.google.com/d/msgid/abouthydrology/BYAPR01MB4760A23FB0E9EB4BEE0DCD04AA77A%40BYAPR01MB4760.prod.exchangelabs.com.