Thursday, November 19, 2026

Agricolture at a crossroads

On the occasion of the Festival della Meteorologia 2026, in the session Coltivare il Futuro, I gave a talk titled “L’agricoltura ad un bivio: prepararsi consapevoli al cambiamento” — Agriculture at a crossroads: preparing consciously for change. The slides are in Italian; this post collects what the talk says, the files, and all the material the QR codes on the slides point to. If you arrived here by scanning the QR code on slide 2, you are in the right place.

F. Depero, Carretto napoletano. Paese di tarantella, 1918, Mart 
The motto of the whole talk is one sentence: increase the intelligence per hectare. Not the quintals per hectare, not the litres of water per hectare: the amount of knowledge we put on every square metre of soil. The argument is built from three pieces of work done by students of our group, and then widens to what those pieces imply for training the people who will have to decide in ten years’ time.

Who is closing the stomata?

The first case is the bachelor thesis of Giulia Merler, supervised by me with Concetta D’Amato. A vineyard in Trentino, a WiseConn station measuring weather and soil moisture, and GEOSPACE — our one-dimensional simulator of the soil–plant–atmosphere continuum built from the GEOframe components WHETGEO, GEOET/Prospero and BrokerGEO. A one-metre soil column, one hundred control volumes, hourly time step, May to September 2023, 466 mm of rain. Two simulations were compared: a “potential” one, in which the vine feels only water stress, and a “total stress” one, in which the stresses due to temperature, radiation and vapour-pressure deficit are switched on as well. We were looking for water stress and found heat stress: seasonal evapotranspiration drops from 352 to 199 mm and transpiration from 306 to 148 mm, more than halved, while soil evaporation increases only slightly. More water therefore stays in the soil — a counter-intuitive outcome — but a vine that transpires less also photosynthesises less. Is that good or bad? It depends on what we want. The point I wanted to leave with the audience is another one: a model like this can anticipate such situations, if it is fed the right information.

The second case is the bachelor thesis of Marco Feltrin (poster on OSF), same model, another vineyard: Illasi, east of Verona, where WiseConn installed a weather station, a 60 cm soil-moisture probe and the module that drives irrigation, with data every fifteen minutes on the DropControl platform. Here the question is different: what happens to the vine if the year is dry? The model was run for one year, April 2023 to April 2024, under two rainfall scenarios — the wet year actually recorded, 1147 mm, and the previous dry year, 682 mm, little more than half. In the dry scenario water stress occupies the whole summer, transpiration between June and August is halved, and the strongest uptake is between 20 and 40 cm, where the roots are. The three applications Marco lists are the practical ones: reducing irrigation waste, assessing water availability for the crop and, as the next step, letting the model compute how much water to give, given the stress it estimates.

At the basin scale

The third piece of work changes scale, from the vineyard to the basin. It is the PhD thesis of Daniele Andreis, and what the slide shows is GEOframe Explorer, the interface through which one navigates the hydrological simulation of an entire Alpine basin, ten years, 2014–2024, sub-basin by sub-basin: temperature, precipitation, evapotranspiration, discharge, water budget, drought. Why does this matter for agriculture? Because the water that reaches the vineyard depends on what happens upstream — snow, reservoirs, aquifers — and on the other users: hydropower, drinking water, ecosystems. The same family of models links the soil column of Giulia and Marco to the basin of Daniele.

The crossroads

If we do not learn to make use of these and other tools, others will use them to their own advantage.

These tools exist, they are open, and we build them here in Trento. “We” are the farmers, the technicians, the irrigation consortia and the university. The alternative is that whoever sells the data, whoever sells the machines, whoever decides upstream will use them instead. From here the talk turns to people and training, in five steps.

Five challenges shape the training. The climate is changing: the grape harvest has moved from October to late August, and drought, frost and floods cost Italian farms about €5.6 billion in 2022. The demographic winter: Italian farms fell by about one third between 2010 and 2020, to 1.13 million, and farm heads under forty went from 11.5% to 9.3%. Soil degradation: Italy is the EU country with the largest share of soil affected by erosion — and the soil is the water reservoir that Giulia and Marco simulated. Value chains and farm income: long chains, brands, designations of origin, and farmers earn little. Markets: consumption of traditional products, wine first of all, is flat or falling, and competition is on quality, origin and trust. None of these five challenges is agronomic alone; that is why training must be broader than agronomy.

Resilience or transformation? This is the question we put to the C3A Advisory Board in April 2026. On one side, protecting traditional products, local knowledge and rural identity, and improving efficiency and water management within current systems. On the other, rethinking crops, varieties, technologies and business models under climate stress, accepting that some traditions may have to be redesigned to survive, and using biotechnology and process innovation where evidence supports them. The honest answer is: case by case. And deciding case by case requires professionals trained to do it — people who can innovate without destroying identity, and who can hold together scientific rigour, food safety and social consensus in a time of distrust towards new technologies.

Water at three scales. At the field scale the question is how much the crop transpires and how much water is in the soil, in order to irrigate when and as much as needed; the answer comes from soil and canopy sensors, satellite vigour and surface temperature (Copernicus Sentinel is free) and soil–vegetation–atmosphere models; the farm decides, with its technician. At the district and consortium scale: water transport in the network, losses, scheduling of irrigation turns; hydraulic models, flow meters, demand aggregated from the fields; the consortium decides. At the basin scale: flows between rivers, reservoirs, snowpack and aquifers and the conflicts between uses; GEOframe-type water balances, locally corrected weather forecasts and three-month seasonal outlooks, which are still an open research front; the Province and the basin authority decide. The digital twin of an irrigation district ties the three scales together: irrigation need is not rain minus evaporation, it needs the full balance, and the twin returns every day — with its uncertainty — the need per plot, the availability upstream and the consequences of a skipped turn. The bottleneck is not the data: it is their integration into models and decisions.

Four enabling conditions, or the technology stays unused. The farmer as an active subject, with co-design tables from the design phase and not only at validation. Training the last mile: who trains, with what resources, through which channels, with which contents — critical reading of data and model limits, not only tool use; innovation brokers and permanent living labs rather than one-off demonstrations. Verifiable data governance: farmers’ distrust is a rational response to the absence of guarantees, and it is answered with contracts, audits and documented cases. Integration with what exists: decision-support systems and models must plug into management software, cooperative platforms and tractor terminals. Other regions give the orders of magnitude: in the 2024 survey of the Tuscan AKIS, water (74%), precision agriculture (59%) and digitalisation (58%) were the top innovation themes, and respondents insisted that accompanying adoption matters more than financing machines; Emilia-Romagna, first in Europe for EIP operational groups, allocates over €51 million to AKIS in 2023–2027.

Two programmes in construction, with Fondazione Edmund Mach, are our answer: a first-level university master in digital agriculture and AI for agri-food, for bachelor graduates in agricultural, environmental, engineering or IT disciplines and for working professionals, open to participants from other regions and from abroad; and a digital-agriculture track in the MSc in Agricultural Sciences for Innovation and Sustainability, which forms the “bridge professional” in depth through applied theses with FEM, with the University’s modelling groups and with the companies of the territory. Three method choices run through both: teach to evaluate models, not only to use them — every tool comes with the question “when does it fail?”; work on real data and cases of the territory, linked to the digital twins public bodies are building; and treat governance not as a separate module — the graduate must also be able to advise the farmer on what not to sign.

Files and resources

Files are here:


The paintings that accompany the slides are, besides Depero’s cover: the months of August to November from the Ciclo dei mesi of Torre Aquila, Trento; Enrico Pedrotti and Fortunato Depero, illustration for the magazine Enrosadira, 1939; a Riccardo Schweizer painting (1925–2004); Umberto Moggioli, Primavera in Treporti, 1914; and a 2000 print by Ulrici on the closing slide. All the artists are from the Trento region. 

References

  • Bottazzi, M. (2020). Transpiration theory and the Prospero component of GEOframe. Doctoral dissertation, University of Trento.
  • D’Amato, C. (2024). Exploring the Soil-Plant-Atmosphere Continuum: Advancements, Integrated Modeling and Ecohydrological Insights. Doctoral dissertation, University of Trento.
  • Feltrin, M. (2024). Scenarios of evapotranspiration fluxes in grapevines: integration of the GEOSPACE ecohydrological model and WiseConn sensor technology. Bachelor thesis, University of Trento. https://osf.io/baqc4
  • Merler, G. (2025). Stima realistica della traspirazione in coltivazione calcolata con GEOSPACE. Bachelor thesis, University of Trento. https://osf.io/mk4qx
  • Tubini, N., & Rigon, R. (2022). Implementing the Water, HEat and Transport model in GEOframe (WHETGEO-1D v.1.0): algorithms, informatics, design patterns, open science features, and 1D deployment. Geoscientific Model Development, 15, 75–104. https://doi.org/10.5194/gmd-15-75-2022
  • Van Genuchten, M. Th. (1980). A closed-form equation for predicting the hydraulic conductivity of unsaturated soils. Soil Science Society of America Journal, 44(5), 892–898. https://doi.org/10.2136/sssaj1980.03615995004400050002x
  • ISTAT (2022). 7° Censimento generale dell’agricoltura. The European House–Ambrosetti (2024), on Eurostat and ISMEA data. C3A (2026). L’agricoltura digitale a un crocevia epocale, working document. https://osf.io/ht7xm/files/gkveb

Wednesday, November 4, 2026

The Kinetic Theory in detail

Work in progress 

In Sion I could show only a few slides of each step. This post collects the whole series Statistical physics of soil water (version 2.5.1): the decks, and for each one the Real Book that writes out every step, the storyboard and the notebook that computes its figures. The series is built on the preprint arXiv:2607.09416 and two companion papers, one on the Chapman–Enskog reduction and one on water age.

The idea in a few lines. Richards' equation is usually postulated: mass conservation, the Darcy–Buckingham law, and two curves fitted to data. The series derives it, as the Navier–Stokes equations are derived from the kinetic theory of gases. The state of soil water is the occupancy g(r, x, t), the filled fraction of each class of pores. Water moves among pore classes inside a volume and between neighbouring volumes, which gives a kinetic equation with two currents. When redistribution among the pores is fast with respect to the forcing (small Damköhler number), a Chapman–Enskog expansion returns Richards' equation: the retention curve at order zero, the conductivity at order one. Where the Damköhler number is not small, Richards' equation fails, and that is where dual permeability, preferential flow and hysteresis come from.

Deck 0. The whole story

A talk built on three ideas: the state is a function, Richards' equation is a limit, the age of water is written in the pores. An appendix gives two slides for each deck.
→ Slides · Real Book · Storyboard · Notebook

Part I. The pieces: geometry, potential, state

Deck 1. The nature of hydraulic conductivity. K(θ) from a network of pores: Hagen–Poiseuille in each pore, connectivity between classes, a cutoff fixed by suction. Burdine and Mualem are two bounds of one network; K vanishes at percolation.
→ Slides · Real Book · Storyboard · Notebook

Deck 2. Micro potential and macro potential. Capillary, adsorptive, osmotic and gravitational terms; the potential between two pores and between places. Suction is the level of the largest filled pore, defined only at equilibrium.
→ Slides · Real Book · Storyboard · Notebook

Deck 3. The occupancy as the state. Many occupancies share one water content; this degeneracy is where history dependence comes from. How the occupancy could be measured.
→ Slides · Real Book · Storyboard · Notebook

Deck 3b. Equilibrium occupancies. Gravity, films, solutes and temperature, one at a time, and what each does to the equilibrium step and to a measured retention curve.
→ Slides · Real Book · Storyboard · Notebook

Part II. From the pore to Richards' equation

Deck 4. The kinetic continuum equation. The budget of one pore class in one volume, the exchange between volumes, the limit to a point: one equation, two currents.
→ Slides · Real Book · Storyboard · Notebook

Deck 5. Mass conservation and the forbidden operation. Integrating over pore sizes gives mass conservation, exact and empty; integrating first and then looking for a flux law in θ is the shortcut of the classical theory.
→ Slides · Real Book · Storyboard · Notebook

Deck 6. The right procedure. Two clocks and their ratio, the Damköhler number; the shape of the state is slaved and eliminated.
→ Slides · Real Book · Storyboard · Notebook

Deck 7. Free energy and the Rayleighian. A Fermi–Dirac occupancy minimises the free energy; the kinetic equation descends it.
→ Slides · Real Book · Storyboard · Notebook

Deck 8. The Chapman–Enskog development. The expansion term by term: Richards' equation as a limit, and the band of conditions where it does not hold.
→ Slides · Real Book · Storyboard · Notebook

Deck 9. The spectrum. The conductivity as a sum over relaxation modes; the gap closes at percolation; Burdine and Mualem return as trial functions.
→ Slides · Real Book · Storyboard · Notebook

Part III. Structure and memory

Deck 10. Soil structure and dual permeability. A bimodal pore population splits the spectrum into a fast and a slow band: dual permeability derived, with the exchange coefficient computed.
→ Slides · Real Book · Storyboard · Notebook

Deck 11. Hysteresis. Wetting and drying do not commute: hysteresis as a holonomy, a loop that does not close.
→ Slides · Real Book · Storyboard · Notebook

A first look at water age

Deck 12A. Volume and flux. Tracers see the pores weighted by volume, the water balance by flux: why tracers find preferential flow everywhere while water balances close with Richards' equation.
→ Slides · Real Book · Storyboard

The rest of the work on water age and travel times (StorAge Selection functions derived from the occupancy, the age contrast between storage and discharge, and the lectures Travel times, kinetic edition) is not yet mature, and will have a post of its own.

Related. The MOOC on the hydrology of soil and the Lausanne lectures for the classical theory; the Sion lectures for the short path through it.

© 2026 R. Rigon, CC BY 4.0 (cover paintings excluded). Notebooks need numpy, scipy and matplotlib.

Monday, November 2, 2026

Slides, Storyboards, RealBooks

For a while now, my lessons have come with three objects instead of one: the slides, a storyboard and a Real Book. Each does a different job, and this post explains which. The first complete set is the one for The Hydrology of Soil, Lausanne version, which I am using for the soil lesson at the University of Lausanne (UNIL). The same scheme is used for the series Statistical physics of soil water.



The slides

The slides are what you see in the classroom, and they carry the story: one idea per slide, a figure or an equation, a few words. On their own they are deliberately incomplete. A slide shows the result of a derivation, not the derivation; it names an assumption, but does not discuss it. The colours of the boxes follow a code that is the same in all my recent decks:

  • blue for definitions and facts;
  • orange for consequences, assumptions and caveats;
  • green for what is new and for pointers forward;
  • red for open problems and corrections.

The PDFs are on OSF. Editable versions, PowerPoint files with real text boxes and LaTeX equations that open in Keynote or PowerPoint, are available on request.

The storyboard

The storyboard is the script of the deck. It is a table with one row per slide and five columns:

  • the slide number;
  • the spoken text, a suggested narration;
  • what is on the screen;
  • the speaker notes;
  • the mounting comments for whoever edits a recording: which build slides were merged, where a figure comes from, what was corrected.

Storyboards were born with the soil MOOC we produced for the EU project EDVANCE, where every episode had to be recorded and edited. I kept them because they answer the question a student asks when looking at a slide alone: what was he saying here? They are also the place where the changes between versions of a deck are listed. Each storyboard now ends with a bibliography in three parts: the sections of Freeze and Cherry's Groundwater (1979, openly available) to read with that deck, the works cited on the slides, and further reading.

The Real Book

The name comes from jazz. The Real Book is the collection of lead sheets that musicians actually play from: the melody and the chords, written out, no frills. My Real Books are the same for a lesson: every step, written out. Where a slide shows a result, the Real Book derives it. It also does the following:

  • keeps track of units, the most frequent source of error in this subject;
  • gives orders of magnitude;
  • states which hypotheses are being used, and where they fail;
  • lists the corrections to earlier versions of the slides (every deck has had its share of sign errors and missing factors).

The text uses the same colour code as the slides: Mathematics, Physics, Result and Correction boxes, plus Numbers boxes in the statistical-physics series. Each section starts by saying which slides it covers, and every Real Book opens with a table that maps each slide to the section where it is discussed. You can follow the lesson with the slides and open the Real Book only where a step is not obvious. Or you can read the Real Book from start to end, as a short textbook chapter.

How to use the three together

The three objects are meant to be used together:

  • In class, the slides.
  • To review a lesson, or to study it without having attended, the slides together with the storyboard.
  • To understand where an equation comes from, or to check it, the Real Book.
  • To go further, the bibliography at the end of the storyboard.

What they can serve

Beyond my own lessons, the three objects are meant to be reused.

  • Students can study a topic without having attended. They can check a derivation step by step and find the one or two readings that matter. The exercises at the end of each Real Book, with answers, help prepare for an exam.
  • Teachers can take a deck, or part of it, for their own course. They can cut it, reorder it or translate it, keeping the Real Book as the reference text behind the slides. The mapping from slides to sections makes it easy to see what is lost when slides are removed.
  • Anyone recording a lesson or a MOOC has, in the storyboard, the script and the editing instructions ready.
  • Researchers and reviewers have, in the Real Book, a place where every equation can be checked. Corrections are listed openly instead of being silently fixed.
  • Translators and adapters work from the LaTeX and editable sources rather than retyping from a PDF. This also includes adapting the material to another audience, for instance a shorter lesson for practitioners or a longer one for a doctoral school.

Since the Real Books and storyboards are written in LaTeX, they are easy to keep under version control and to update together with the slides.

Everything is released under a Creative Commons Attribution 4.0 licence (CC BY 4.0), except the paintings on the covers, which are copyrighted by the original authors and used according a FAIR use agreement. The materials are a work in progress, and will be improved after the lesson. If you find an error, a missing step or a better reference, please tell me: that is exactly what the Correction boxes are for.

Soil Kinetic Theory - Sion EPFL Lecture

Today NOvember 6st, I am in Sion, at the Laboratory of Catchment Hydrology and Geomorphology (CHANGE) of EPFL, to talk about soil water. The talk goes from the classical picture to the kinetic theory I have been working on in the last months: the state of soil water is a function, the filled fraction of each class of pores, and Richards' equation is the limit of a kinetic equation when water redistributes among the pores faster than it is forced.


There are seven decks. Each comes with its Real Book (every step written out), its storyboard (what I say on each slide) and the notebooks that make its figures. Here a more complete explanation.  A video of the lecture can be found here. 
  1. The energy of water in soil. The chemical potential, how it moves water and how it transforms it, pressures and heads, and the retention curve from the pore-size distribution.
    → Slides · Real Book · Storyboard ·  Notebooks: 1a, 1b, 1c
  2. The nature of K. The hydraulic conductivity as the conductance of the network of filled pores: bounds, Burdine and Mualem, percolation.
    → Slides · Real Book · Storyboard · Notebook
  3. Soil water budget. From mass conservation to Richards' equation, and the two hypotheses it rests on: instant redistribution and accessibility of all pores.
    → Slides · Real Book · Storyboard · Notebook
  4. The micro potential. The potential of water in a pore, and between two pores. Out of equilibrium there is no single ψ.
    → Slides · Real Book · Storyboard · Notebook
  5. Occupancy. Many states share the same water content: the state is the function g(r), and its degeneracy is the source of hysteresis.
    → Slides · Real Book · Storyboard · Notebook
  6. The kinetic equation. The budget of one pore class, the exchange between neighbouring volumes, one equation with two currents.
    → Slides · Real Book · Storyboard · Notebook
  7. The rest of the story. Two slides for each of the decks that follow: Richards' equation as a limit, the two clocks, free energy, the spectrum, dual permeability, hysteresis, and a first look at the age of water.
    → Slides · Real Book · Storyboard · Notebooks 7a, 7b, 7c, 7d, 7e, 7f, 7g, 7h

The first and the third deck come from my MOOC; the others from the series Statistical physics of soil water, built on the preprint arXiv:2607.09416 and two companion papers.

Here the video of the lecture given.

Related material. The MOOC on the hydrology of soil; the Lausanne lectures, on the classical theory; and the kinetic theory in detail, with all the decks of the series. Water age and travel times, only glimpsed here, will have a post of their own.

Slides, Real Books, storyboards and notebooks: © 2026 R. Rigon, CC BY 4.0 (cover paintings excluded). Thanks to the colleagues of CHANGE for the invitation.

Monday, October 26, 2026

Before the Kinetic Theory of water flow in porous media - What you need to know about soil - Lausanne (UNIL) seminar

I will give a long lesson on soil water at the University of Lausanne (UNIL) Hydrology lab. These are the slides I plan to use. They are the Lausanne version of The Hydrology of Soil, the course I prepared for the soil MOOC of the EU project EDVANCE, revised for this occasion. The whole course, included the parts which I have skipped here are available at the MOOC site. A last deck opens a window on the work I am doing now on the statistical physics of soil water and which is the topic of my talk at Sion EPFL Hydrology and Geomorphology lab. All the slides and the companion material are available as PDF files in the OSF repository. 

The story runs from what a soil is to the equation we use to describe water moving in it, Richards' equation, and then to the places where that equation stops working. Each deck below is followed by a few lines on what it contains.

Please consider all the material as a work in progress.  Though there should not be error in the hydrology, ll the material can be considered as a preprint to be subject to revision. Not sure, I will be able to tell everything, but parts can be skipped, and the whole amount of material is left to inspection by curious. 


Please find the slides at the links. You'll be maybe surprised to find two two other objects: a storyboard and a Real Book. This post explains what they are. 

0. What soil is

This is an introduction to what I assume to be known. It covers the five constituents that share the soil volume (solid particles, liquid water, ice, gas with its water vapour, and biota), soil profiles and horizons, and texture. That means sand, silt and clay, the particle-size distribution, and the texture triangle. It closes with the key definitions used in the rest of the lesson: masses, volumes and densities, volumetric water content, porosity, and relative and effective saturation.

Slides, Storyboard, Realbook, Notebooks I, II, III

1. The energy of water in soil

This is the longest deck, and the one on which everything else rests. It starts from the forms of energy that water in soil can have, and arrives at the chemical potential as the quantity that tells where water goes. From there it recovers the traditional pressures and heads (matric, osmotic and gravitational). It then treats capillarity: Young–Laplace, Jurin's law and the meaning of a negative pressure. Next comes the passage from single pores to the soil water retention curve, which is the pore-size distribution seen in suction coordinates. The deck ends with hydraulic hysteresis and the parametric forms of the retention curve (Brooks–Corey, van Genuchten, Kosugi, do not worry about, they are explained in the slides).

Slides, Storyboard, Realbook, Notebook

2. Darcy and Buckingham laws

The deck starts from Darcy's experiment, and asks whether Darcy's law can be derived from the Navier–Stokes equations (it can, under well-defined conditions). It then goes down to the pore with the Hagen–Poiseuille law, and back up to the Representative Elementary Volume. In unsaturated soil, Buckingham's law asks whether the flux can keep Darcy's form with a conductivity that depends on water content. Mualem's hypothesis of mutually accessible pores is what makes this work. The deck ends with the van Genuchten–Mualem conductivity and its siblings: Brooks–Corey–Mualem, Kosugi–Mualem and Peters–Kosugi–Mualem.

Slides, Storyboard, Realbook, Notebook

3. Soil water budget

This deck goes from a budget to a differential equation: conservation laws, the divergence operator, and the local form of the balance. Combining continuity, the Darcy–Buckingham law and the retention curve gives Richards' equation. Along the way it pays attention to which potential drives the flow. The excess of potential sits in the largest filled pores, and this is true only at equilibrium. The two foundational hypotheses are stated explicitly: internal redistribution is instantaneous, and all pores are mutually accessible. The deck closes with 1D simulations made with GEOframe (with Niccolò Tubini): sand, clay, ponding, and sand over clay.

Slides, Storyboard, Realbook, Notebook

4. Solving Richards' equation, and its limits

The deck starts with where the parameters come from: pedotransfer functions, from Rawls' regressions to today's tools (Rosetta, euptf and others). It then looks at what the soil column does not contain. Bedrock topography, pipes and macropores control much of the subsurface flow, and flow switches on above thresholds (Tromp-van Meerveld and coworkers; Koestel, Larsbo and Jarvis). It ends with the breakdowns of Richards' equation: hydrophobicity, a conductivity that grows with sample size, fingering and saturation overshoot. A few philosophical remarks on what a model leaves out close the deck.

Slides, Storyboard, Realbook, Notebook

5. The nature of hydraulic conductivity (statistical physics of soil water, Deck 1)

This last deck turns the question around: can K(θ) be derived from the pore space, instead of being postulated? It builds the conductivity from a network of pores: Hagen–Poiseuille in each pore, the connectivity between pore classes, and a cutoff fixed by suction. It then sums the network in parallel, in series and exactly (Kirchhoff's laws, effective medium). Burdine's and Mualem's formulas turn out to be two bounds of the same network. The conductivity vanishes at percolation, and water enters from the boundaries. The same question, one scale up, is the effective conductivity of stochastic hydrogeology. It is the first deck of a series built on the preprint arXiv:2607.09416, in which Richards' equation appears as a limit of a kinetic theory of pore water.

Slides, Storyboard, Realbook, Notebook


The slides of decks 0–4 have accompanying Real Book chapters where every step is written out, and the decks will be improved after the lesson. Comments and corrections are welcome. The paintings that open the decks are by Jay Stratton Noller and Mark Rothko. I thank Chiara Baldo and Lorenzo Duchi, who helped with the MOOC version, and the colleagues at UNIL for the invitation.

Daniele Andreis' PhD.

On 22 October 2026, in Mesiano (Trento), Daniele Andreis defended his Ph.D. thesis, Hyper-resolution monitoring and modelling of the water budget for improving water management. The thesis was written for the Doctoral School in Agrifood and Environmental Sciences (37th cycle) at the Centre Agriculture Food Environment (C3A) of the University of Trento. I supervised it together with Giuseppe Formetta.

\

The materials are open:

The question

Alpine water regimes are changing fast. Wet and dry years alternate, and we now see floods and droughts in the same year. In the Noce basin, the 2022 drought followed 2021 with a markedly different snowpack. Managing withdrawals among hydropower, agriculture, domestic use and tourism under this variability requires catchment-scale water budgets that we can trust. The usual answer is "hyper-resolution": resolve the variability with ever finer models. Daniele's thesis asks the question that usually stays implicit: how much spatial detail does an Alpine hydrological model really need, and for which purpose? It also asks a second one: how much does the way meteorological forcing is brought onto the model units matter, and does that depend on the model's resolution?

All the work uses GEOframe, in which a basin is partitioned into Geomorphic Units (GUs), sub-catchments defined by a contributing-area threshold. The threshold therefore sets the model resolution. The case study is the Noce basin in Trentino, an area of agricultural, hydropower and tourist interest, with a gradient from glacial terrain to forest to intensive orchards. The model was calibrated on discharge only. Its internal states were then checked against independent information: snow cover from MODIS, surface soil moisture from Copernicus, root-zone moisture from cosmic-ray neutron probes (CRNS) and from the MERIDA reanalysis. The water budget closes at every time step.

Three studies

1. Refine only where it matters. A new GEOframe component builds mixed-resolution partitions, with GUs refined (from 10 to 1 km²) only in selected areas. Discharge skill is comparable to the uniform 10 km² configuration. The internal states improve: the median Brier score for snow cover goes from 0.20 to 0.14, correlation with root-zone moisture increases in every elevation band, and in the 2022 drought summer at Vermiglio the KGE goes from −0.19 to 0.67. This study is submitted to Water Resources Research.

2. Forcing matters as much as delineation. Three ways of bringing precipitation and temperature onto the units (nearest station; kriging at the GU centroid; kriging on a grid and area-weighted averaging over the GU) were crossed with three delineations, giving nine configurations calibrated in the same way. The finest delineation with gridded, area-weighted forcing performs best for discharge (validation KGE 0.86 at Malé) and snow cover. Still, a parsimonious two-unit setup with centroid kriging already does well, and no single configuration wins on every variable. Summer actual evapotranspiration, and so the partition of the water budget, changes strongly with the forcing choice.

3. How fine is fine enough? Eight discretizations, from 90 to 3 km², with the same model and the same calibration protocol. The gain is not gradual. There is one large jump (60 → 30 km² at Malé, 30 → 15 km² at Vermiglio) and then small or inconsistent gains in validation. The same transition appears in snow cover, an independent variable. Coarse units overestimate volumes, while finer units mainly raise the computational cost. These thresholds are empirical and belong to this basin. There is no universal optimal GU size.

Daniele's own summary is the right one: spatial detail pays off when it brings information the model and the data can use; beyond that point it adds complexity, not skill. Choose the configuration by purpose.

Software and open science

I want to stress a part of the work that is easily overlooked: much of the thesis is software, and it is now part of GEOframe for everybody. Daniele wrote:

  • an automated kriging component that fits the semivariogram at every (hourly) time step and selects the best of five model families automatically, which makes long-series interpolation reproducible;
  • the mixed-resolution GU builder;
  • bug fixes and a new adaptive Runge–Kutta solver for the rainfall–runoff model;
  • a large part of the GEOframe refactoring (repositories, website, build system). The same code is used for the Po River District (AdbPo) datasets.

The modelling chain also runs as an operational prototype for the Noce basin. It updates daily from MeteoTrentino and FEM stations, stores its data in a self-describing GeoPackage database (developed with Andrea Antonello), and shows its results in a web application, GEOframe Explorer. It is a small, concrete step toward the participatory digital twins of hydrology (DARTHs) we have been advocating.

Code: github.com/geoframecomponents, Krigings, geoframe.zoom, ERM, geoframe-explorer.
Data: Study 1 doi.org/10.17605/OSF.IO/5SZ8A, Study 2 doi.org/10.17605/OSF.IO/J9ZVC, Study 3 doi.org/10.17605/OSF.IO/UAY8G, remote-sensing pipeline doi.org/10.17605/OSF.IO/47DW9.

Abstract

The hydrologic regime of Alpine regions is undergoing rapid change under global warming, characterized by an alternation of wet and dry years, and even by floods and droughts within the same year. In this context, accurate catchment-scale water balance estimation is essential. To manage water withdrawals and the competing demands of hydropower, agriculture, domestic use, tourism, and industry under such variability becomes difficult and requires new tools and multiscale answers. Therefore, this work investigates scale issues in Alpine catchments and implements new modeling components. The multiscale approach is based on the GEOframe system and its space partitioning into Geomorphic Units (GU), which is mixed with a grid approach. This has practically implied writing a tool to increase the resolution of sub-catchment extraction, tools for grid information averaging over the GU, and adaptation of many existing GEOframe tools. These improvements have also affected the kriging interpolator, whose new version accommodates the possibility of varying the semivariogram to capture the temporal variability of meteorological forcing at each time step. These developments also contributed to a broader refactoring of the GEOframe system, including bug fixes, the reorganization of its Git repositories and website, and improvements to the build system. The resulting code also supported the work carried out for the AdbPo project. The tools were extensively applied to the Noce River basin, an area of particular agricultural interest. The modeling and the analysis combined ground-based observations (weather stations and cosmic-ray neutron probes) with remote sensing products (snow cover and surface soil moisture). Summarizing the methodology, hourly semivariograms were used to estimate meteorological variables on a grid, subsequently aggregated as weighted area averages over the GUs and compared to the "centroid-based approach" formerly used, showing interesting improvements in simulating the river discharges. Besides, the Noce River basin was modeled at various resolutions, meaning with coarser and finer GUs. The multiscale approach also shows improved validation performance not only for discharge but also for soil moisture and snow cover when GUs are smaller. The modeling infrastructure was implemented according to FAIR principles and respecting the DARTHs blueprint and is currently operational for water budget analysis.

Congratulations, Daniele, and thanks for the many hours you gave to GEOframe and to all of us who use it.

Gaia Roati's Ph.D.

Gaia Roati has defended her Ph.D. thesis, Distributed hydrological datasets of the Po River Basin: geomorphology, catchments, and hydrological daily budgets for the 1991–2020 time period. The thesis was written for the Doctoral School in Agrifood and Environmental Sciences (37th cycle) at the University of Trento, and was co-funded by the Po River Basin District Authority (Autorità di Distretto del Fiume Po, ADBPO) and by the Center Agriculture Food Environment (C3A). I supervised it together with Giuseppe Formetta.



The materials are open:

The question

The Po River District covers about 87,000 km² across eight regions and one autonomous province, with the Swiss and French portions of the basin outside its jurisdiction. Twenty million people live there. It produces about 41% of the Italian GDP and about 55% of the national hydropower. After the 2022 drought nobody needs to be convinced that the Authority must know, for every subbasin and every season, how much water there is and where it goes. Gaia's thesis starts from that statutory question: what is the spatial and temporal distribution of water resources across the Po River District, and which pathways govern its hydrological response?

The answer cannot be measured directly. Gauges are discrete points, meteorological stations are sparse exactly where runoff is generated, and snow is essentially unmonitored above 2000 m. So the budget must be inferred with a model constrained by all the observations available. It must also be inferred continuously, not once, and in a way that people who did not build the system can audit and rerun. The thesis is therefore not about a model. It is about building an infrastructure, based on GEOframe, which the Authority now runs itself.

What Gaia did

This was a collective project involving the University of Trento, ADBPO, the regional environmental agencies (ARPA), Waterjade and the Politecnico di Torino. The thesis is explicit about what is Gaia's:

  • Data. She led the collection, harmonisation and quality control of precipitation, temperature, snow depth, discharge and abstraction records from all the regional agencies, plus Swiss and French sources for the transboundary areas. About 1,580 rain gauges and 1,550 thermometers survived quality control, and about 380 hydrometers were collected. It is the least visible and most consequential part of the work: every number in the thesis, and in the papers built on it, rests on it.
  • Calibration of the whole District. She organised and carried out the hierarchical calibration from the Alpine headwaters to the closure at Pontelagoscuro, including how the work was divided among institutions. This is her largest single technical contribution.
  • Budgets and operations. She co-designed the analysis of the simulated water budgets and moved the system from a research setting to a live operational setup at ADBPO. She also represented the modelling framework in front of the District's Permanent Observatory on Water Uses (Osservatorio Permanente sugli Utilizzi Idrici), the body that declares water-scarcity states and negotiates abstractions among users.

The products

1. A geomorphic decomposition of the District. With the Horton Machine, the District up to Pontelagoscuro was divided into 2,879 Hydrological Response Units, with target sizes of about 10 km² in Valle d'Aosta, 20 km² in Piemonte and 30 km² in the plain. The hydrometers were placed exactly on the extracted network, so that the HRUs close at the gauges.

2. Daily precipitation and temperature at 1 km, 1991–2020 (Salehi et al., 2026a). In leave-one-out cross-validation, mean KGE is 0.84 for precipitation and 0.88 for temperature. The error has a clear structure: it grows with elevation, by about 0.54 mm and 0.28 °C of mean absolute error per 1000 m. Compared with continental products, the local grid keeps the extremes. For Storm Alex (October 2020) the three-day peak in western Piemonte is 482 mm in the 1 km product and 277 mm in E-OBS, a 43% attenuation. A two-part correction (wet-day occurrence, then elevation-dependent intensity) reduces the wet-day bias above 2000 m from −17% to −8%.

3. A climate that is not stationary, and not uniformly so. The same fields, analysed by elevation band (Salehi et al., 2026b), show warming everywhere and three regimes: supply-constrained snow at 500–1000 m, persistence-limited snow at 1000–1500 m, and a thermal season that encroaches on the snow season above 1500 m. The key metric is the gap between the day snow disappears and the onset of the warm season. That gap is widening: the snow reservoir empties earlier relative to the time when water is most needed.

4. Thirty years of daily snow water equivalent at 500 m (Dall'Amico et al., 2025, Scientific Data; analysed in Wani et al., 2026). It was produced with GEOtop, with snow-depth and MODIS snow-cover assimilation. Basin-wide storage peaks at about 5.6 Gm³ in mid-March. In the last decade, compared with 1992–2011, snow volume fell by 28–38% below 1500 m. Above 2000 m it increased, but the thesis flags this as the least robust result, because at those elevations the forcing rests mainly on ERA5-Land. The result I find most interesting is about timing: at 1000–1500 m the season is about 21 days shorter, while the disappearance date moved by less than 3 days. The season is shortening at the front end. Snow arrives later in autumn, rather than leaving earlier in spring.

5. A District-wide, upstream-to-downstream calibration. Each zone is the set of HRUs between a gauge and the already-calibrated gauges upstream. Independent gauges are calibrated in parallel, and only parameter sets with KGE above 0.7 are propagated downstream. The calibration used LUCA (Shuffled Complex Evolution) on 17 parameters of the Embedded Reservoir Model, with KGE as objective function, over 2015–2018, a window that includes the 2017 drought. Mean KGE is 0.801 in Valle d'Aosta (12 stations), 0.834 in Piemonte (57 stations), 0.766 in Lombardia (21 finalised stations) and 0.823 for the 11 cross-border stations of Lake Maggiore, calibrated by the Politecnico di Torino team within LIFE CLIMAXPO. Emilia-Romagna (mean 0.672 over 35 stations) is declared preliminary: torrential Apennine regimes, gauges recording zero flow, and an hourly recalibration already planned. At Tavagnasco, the Valle d'Aosta outlet, the parameters inherited from upstream already give KGE 0.74 before any local tuning. A continuous run over 2002–2021 with the 2015–2018 parameters gives 0.739.

Much of the work in this chapter went into human influence on the observations. Where abstractions are measured, they were added back to reconstruct natural flows (Spineto Orco, on the Orco). Where regulation changed, the calibration window was moved (Rivolta d'Adda). Below the great regulated lakes, measured outflows are imposed as inputs. As the thesis puts it, below a regulated lake the discharge is a decision, not a hydrological response. GEOframe simulates the natural response. Diversions, licences, ecological flows and irrigation demands are handled downstream in RIBASIM, which ADBPO uses for its Water Budget Plan.

6. Closed daily water budgets by component and HRU, 1991–2020. Snow, canopy, root zone, groundwater and surface storages are resolved every day for every unit, and aggregated monthly, which is the timescale of the Observatory's decisions. One simple diagnostic deserves attention: aggregating by calendar year leaves spurious ΔSWE terms in alpine basins, while aggregating by hydrological year (October–September) brings net annual ΔSWE close to zero and the budget back to P ≈ ET + Q.

Honest about what is missing

What I appreciate most in the thesis is that it does not oversell itself. Chapter 2 scores the work against the thirteen tenets of the Digital eARth Twin Hydrology systems (DARTHs) and the table is, in the thesis's own words, "deliberately unflattering where the work is incomplete". The system is transparent, open, modular and operational, and it is bound to the basin by calibration. It is not yet a digital twin. There is no data assimilation in the coupled chain, uncertainty is not propagated, there is no formal provenance record, and it is not exposed through standard service interfaces. Equifinality is discussed openly (seventeen parameters against one discharge series). Evapotranspiration, the largest outgoing flux, is not yet independently validated. The residual-closure diagnostic and the split-sample validation are still being completed. Sensitivity analysis, assimilation of satellite soil moisture, eddy-covariance and satellite checks of ET, a two-step calibration of the snow module against the SWE product, and hourly runs are the agenda for the next years. Most of it will be carried out together with ADBPO.

Data and code

Abstract

In recent years, Italy has observed more frequent and intense drought events, with a particularly severe drought in 2022, leading to significant environmental, social, and economic damage. Also at a global scale, extreme events, floods and droughts, have been reported to be more likely due to climate change and environmental modifications. For this reason, already in 2021, the Po River Basin District Authority (ADBPO) started the implementation of the GEOframe modelling system on the entire territory of the District to update the existing numerical modelling for water resource management, and then to improve the planning activity of the Authority itself, producing a better quantification and forecast of spatial and temporal water availability. The GEOframe system is an open-source, semi-distributed conceptual modelling framework characterized by high modularity and structural flexibility, collaboratively developed under the leadership of the University of Trento. After a starting phase of data collection, validation, spatial interpolation of precipitation and temperatures (for the reference period 1991-2020), and geomorphological analysis, all the components of the hydrological balance (evapotranspiration, snow accumulation, water storage and discharge) have been simulated. Subsequently, a zonal calibration was carried out station by station over a four-hydrological-year window, using the Kling–Gupta Efficiency (KGE) as objective function: the characteristic model parameters were sought that best reproduce the discharge recorded at the hydrometric stations of each zone, proceeding from the headwaters downstream so that each calibrated zone provides the upstream boundary condition for the next. Calibration has been completed for Valle d'Aosta (12 stations, mean KGE 0.785) and Piemonte (49 stations, mean KGE 0.823), and is in progress for Lombardia and Emilia-Romagna; validation against independent periods is ongoing. The resulting product is a distributed dataset covering the reference period 1991-2020: the geomorphological decomposition of the District into subbasins and Hydrological Response Units, and, for each of them, daily time series of all the terms of the water budget (precipitation partitioning, snow water equivalent and melt, canopy interception and throughfall, evapotranspiration, root-zone and groundwater storage, surface and subsurface runoff, and discharge). Particular attention has been paid to the snow contribution, which has a major impact in alpine regions, dominating local and regional hydrology, strongly influencing vegetation growth and water availability and utilization (Sturm et al. 2017; Viviroli et al. 2007; Wu et al. 2015), as in the Po River Basin, which is bounded by the Alps along its entire course. The dataset is intended to support water-budget assessment, drought characterisation and water-resource planning at district scale.

Congratulations, Gaia. You did the hard, unglamorous work of making a whole District speak the same language, and now the Po has a water budget that anyone can open, check and rerun.