Mars today is a frozen desert — barren, thin-aired, and brutally cold. Yet carved into its surface are river valleys, delta fans, crater lakes, and what may be ancient shorelines: geological fingerprints of a world that once ran with liquid water. For decades, planetary scientists have asked the same stubborn questions: where exactly did that water flow, how did it connect across a planet scarred by hundreds of thousands of impact craters, and where did it ultimately pool? A new computational model may finally offer answers at a scale never before attempted.
A planet that remembers its water
Mars’ geological record is surprisingly talkative. Networked river valleys, deltaic fans, carved crater lakes, and potential palaeoshorelines all point to a warm, wet past during the Noachian and early Hesperian epochs — roughly 3.7 to 4.1 billion years ago. The central question has always been: where exactly did that water accumulate, how did it connect across the planet, and where did it ultimately go?
Answering that computationally has been brutal. Mars hosts more than 385,000 impact craters, each a natural water-trapping basin. Standard Earth-based hydrological tools either work at too coarse a resolution, treat ocean boundaries as fixed, or simply collapse when asked to compute millions of overflowing crater lakes simultaneously across a high-resolution planetary grid.
Building a hydrological engine for an entire planet
Lead author Alexandre Gauvain and colleagues at the Laboratoire de Météorologie Dynamique tackled this by building a global hydrological model from scratch. Their approach centers on a depression hierarchy graph and a fill-spill-merge algorithm — water flows downslope, fills a basin, then spills into the next one downstream, merging lakes as it goes.
Topographic data came from the Mars Orbiter Laser Altimeter, at roughly 1,519 feet per pixel near the equator. From that, the team pre-computed a database of approximately 12 million depressions — nearly 6 million individual craters and 6 million merged meta-depressions.
The real breakthrough was computational. Rather than recalculating terrain geometry at every time step, the team built lookup tables linking lake volume, surface area, and elevation for every basin. That single design choice cut processing time for over 2 million active lakes from nearly two hours down to 4.75 seconds — a speedup of more than 1,400 times.
What 48 simulations revealed about ancient Mars
The team ran 48 global simulations to equilibrium, varying the total water inventory from 3 to 3,000 feet of global equivalent layer and testing a range of evaporation rates and starting water distributions. One result stood out immediately: regardless of where water began — spread evenly or dumped into a single basin — every simulation converged to the same steady-state distribution for a given water volume.
At very low inventories, water stays trapped in the cratered southern highlands. Between 3 and 30 feet of global water equivalent, something shifts: mid-latitude lakes begin overflowing northward, and a contiguous Northern Ocean forms. At 300 feet, that ocean holds half of all planetary surface water. Push to 3,000 feet and it holds three-quarters, while major southern basins like Hellas and Argyre account for another significant share.
Rivers larger than the Congo — and craters that held lakes
The model identified four primary trunk river networks funneling highland runoff into the Northern Ocean. The largest drained the Tharsis and Olympus Mons regions toward Marte Vallis, producing a simulated steady-state discharge of roughly 2,330,000 cubic feet per second — the modern Congo River, for comparison, averages around 41,000. A second major network through Simud and Lobo Valles drained Valles Marineris at nearly 1,059,000 cubic feet per second.
Individual landing sites came into focus too. At Jezero Crater — where NASA’s Perseverance rover operates — two converging river channels each delivered roughly 1765 cubic feet per second. At Gale Crater, explored by Curiosity, a combined river system fed over 8,475 cubic feet per second into the crater lake.
When the team ran simulations on reconstructed pre-True Polar Wander topography, the Northern Ocean shrank significantly and water shifted toward equatorial regions. Ancient crustal elevation, it turns out, matters just as much as ancient climate when reconstructing where water went.
Beyond Mars: a tool for any wet world
The model’s most consequential feature may be its flexibility. It can couple directly with 3D Global Climate Models, enabling feedback simulations where evaporating lakes drive localized rainfall over orbital timescales spanning hundreds of thousands to millions of years — something previous approaches couldn’t do dynamically.
Simulated river discharges can be checked against sediment transport rates in preserved deltas. Predicted lake levels can be compared against mineral deposits — phyllosilicates, hydrated sulfates — mapped by orbiting spacecraft. That gives scientists a concrete way to test what the model predicts.
The algorithms aren’t Mars-specific. The same framework could model liquid methane lakes on Titan, ancient surface water on Venus, or hydrological conditions on rocky exoplanets — making it a versatile tool for any wet world, past or newly discovered.
For more information, visit the complete report from the European Geosciences Union.
Read the whole thing?
Get the week's signal, not the noise
Our sharpest reporting on energy, climate and nature — free, once a week.
