Rainfall Looks Heavier at Higher Resolution — Even When It’s Not

When a storm drops rain over a city, the reported intensity depends not just on the clouds, but on how finely you’re looking. A new study reveals that rainfall appears more intense at higher spatial and temporal resolutions — even when the actual physical conditions haven’t changed. The finding isn’t about weather; it’s about measurement. And it has profound implications for how we compare climate models, interpret satellite data, and prepare for extreme events.

At 0.1 degrees resolution — roughly 11 kilometers near the equator — the most common heavy rain rate in global satellite data is 19 mm/day. But coarsen that same data to 5 degrees (about 550 km), and the peak shifts down to just 10.85 mm/day. That’s not because less rain fell. It’s because the way we aggregate data systematically alters the shape of the distribution. According to the study by Rajeev, Masleyev, and Pendergrass (2026), this distortion follows a simple rule: the entire precipitation intensity distribution shifts in logarithmic space without changing its shape. They call this the shift mode — and it explains up to 80% of the difference between resolutions.

This means that when climate models simulate more intense rainfall at finer scales, we can no longer assume it’s due to better physics alone. Some — perhaps much — of that increase may simply be an artifact of resolution. For cities planning flood defenses, or scientists tracking changes in extreme precipitation, mistaking a scaling artifact for a physical signal could lead to costly errors.

The Science

The researchers set out to isolate how spatial and temporal resolution affect precipitation intensity — independent of real-world atmospheric changes. To do this, they used observational datasets and artificially coarsened them, ensuring that no new physical processes were introduced. This methodological rigor allowed them to study only the effect of scale.

For spatial resolution, they used the IMERG satellite-gauge merged product, which provides daily precipitation estimates at a native resolution of 0.1° — among the finest available globally. They regridded this data conservatively to coarser grids: 0.25°, 0.5°, 1°, 2°, 5°, and 10°. Conservative regridding preserves total precipitation volume across scales, meaning the same rain is just distributed over larger boxes. They repeated this process with five other datasets from the FROGS database, including GPCP, CMORPH, and PERSIANN-CDR, all regridded from 1° to coarser resolutions.

For temporal resolution, they turned to ground-based measurements from the U.S. Department of Energy’s Atmospheric Radiation Measurement (ARM) program. These include 1-minute rainfall records from two sites: a tropical location at Manus Island (TWP) and a mid-latitude site in Oklahoma (SGP). From these high-frequency observations, they aggregated the data into longer intervals — 2-minute, 5-minute, up to 24-hour averages — again preserving the underlying signal while changing only the time step.

At each resolution, they constructed precipitation amount distributions: histograms showing how much total rain falls at different intensity bins, logarithmically spaced to capture the full range from drizzle to deluge. This approach, adapted from Pendergrass and Hartmann (2014), allows comparison across scales by focusing on the distribution of rain mass rather than event frequency.

They also introduced joint precipitation amount distributions — a novel tool that maps how high- and low-resolution rain rates co-occur. This helped them trace exactly how fine-scale bursts of rain get smoothed out when averaged into larger grid cells or longer time windows.

What They Found

The results were strikingly consistent. As spatial resolution decreased, the peak of the precipitation amount distribution shifted to lower intensities. At 0.1°, the median rain rate (the rate above which half of all rain falls) was 19 mm/day. At 1°, it dropped to 16.38 mm/day. By 5°, it was just 10.85 mm/day — a 43% reduction (Rajeev et al., 2026).

Median Rain Rate vs. Spatial Resolution (IMERG)

Label	Value
0.1°	19
0.25°	18.37
0.5°	17.7
1°	16.38
2°	14.5
5°	10.85
10°	8.2

Source: Rajeev et al. (2026), derived from IMERG precipitation amount distributions.

The same pattern held for extreme rainfall. The globally averaged RX1day — the wettest single day each year — was 67 mm/day at 0.1°, but only 35.7 mm/day at 5°. That’s not a small discrepancy: it’s the difference between a disruptive downpour and a manageable storm.

Even more revealing was the behavior of the distribution’s shape. While the peak shifted, the overall form of the distribution remained nearly unchanged — especially for small resolution differences. This suggested a simple transformation could bridge scales: a pure shift in log-space.

They tested this idea using the directly-calculated shift mode, defined as:

$$b_o=\frac{r_{1/2}^H-r_{1/2}^L}{r_{1/2}^L}$$

where $r_{1/2}^{H}$ and $r_{1/2}^{L}$ are the median rain rates at high and low resolution. This value quantifies the fractional change in the distribution’s center due to resolution alone.

When they applied this shift to the coarse-resolution distribution, it aligned remarkably well with the fine-resolution data. For example, shifting the 1° distribution by $b_o = 0.16$ brought it within 5% of the 0.1° distribution across most rain rates.

Global Average RX1Day by Spatial Resolution

Label	Value
0.1°	67
0.25°	64
1°	57
5°	35.7

Source: Rajeev et al. (2026), IMERG data. Extreme precipitation intensity drops significantly with coarser resolution.

The joint distributions confirmed this.

shows how high-resolution rain rates (y-axis) relate to their coarse-gridded counterparts (x-axis). The mass of precipitation clusters along the diagonal, but systematically offsets toward higher intensities at fine scales. This isn’t noise — it’s a predictable redistribution.

The same logic applied to time. At Manus Island, 1-minute data showed a median rain rate of 0.25 mm/hr. When aggregated to 1-hour averages, it dropped to 0.18 mm/hr — a 28% decrease. The shift mode captured this change with similar accuracy.

Shift Mode Magnitude by Temporal Resolution (Manus Island)

Label	Value
1→2 min	0.12
1→5 min	0.18
1→10 min	0.22
1→30 min	0.26
1→1 hr	0.28
1→3 hr	0.32
1→6 hr	0.35
1→12 hr	0.38

Source: Rajeev et al. (2026), ARM gauge data at Manus Island. Larger temporal aggregation leads to greater shift in median rain rate.

Across all datasets and locations, the shift mode explained the majority of resolution-dependent differences. In some cases, like IMERG’s transition from 0.1° to 0.25°, the error after shifting was less than 2%. Even for large jumps — say, from 1-minute to 1-hour — the shift accounted for over 70% of the change.

Why This Changes Things

For decades, climate scientists have interpreted higher-resolution models as capturing more realistic physics — better convection, sharper fronts, more intense storms. And often, they are. But this study shows that part of what we’ve been calling “improvement” may just be geometry.

Consider a thunderstorm 5 km wide. At 0.1° resolution (~11 km), it fills about half a grid cell. Its rain is averaged with surrounding dry areas, diluting the peak intensity. But at 0.05° (~5.5 km), the storm fits neatly in one cell, preserving its full strength. No new physics — just better alignment.

This has real-world consequences. Many climate impact studies rely on gridded precipitation data. Urban planners use it to size stormwater drains. Hydrologists use it to model river flows. If a model reports 50% more heavy rain at high resolution, is that due to climate change — or just grid size?

The answer matters. Overestimating future rainfall intensity could lead to overspending on infrastructure. Underestimating it could leave cities vulnerable.

The shift mode offers a correction. By applying a simple logarithmic offset, researchers can now translate between resolutions — not perfectly, but with high fidelity. This is especially valuable for model intercomparisons, like those in the Coupled Model Intercomparison Project (CMIP), where models run at wildly different resolutions.

It also reframes long-standing debates. For years, scientists have argued whether satellite products like IMERG or GPCP are “more accurate.” But this study shows that even within the same dataset, resolution alone changes the reported intensity by tens of percent. Comparing IMERG at 0.1° to GPCP at 1° without correcting for scale is like comparing a zoomed-in photo to a wide-angle shot and concluding one camera sees more light.

Even trends in extreme precipitation — a key indicator of climate change — may need re-evaluation. If coarser-resolution datasets systematically underestimate intensity, apparent increases over time could be partly due to improving observational resolution rather than warming climates.

What’s Next

The shift mode is powerful, but not universal. The study found that for very large resolution jumps — say, from 1° to 10° — the distribution’s shape does change slightly. The tails broaden, and the fit degrades. This suggests that while the shift captures the dominant effect, secondary processes like nonlinear averaging or intermittency play a role at extreme coarsening.

Future work should test this framework in climate models. Do simulated precipitating systems follow the same scaling? If so, we could develop resolution-invariant metrics for model evaluation. If not, the deviation itself would signal where model physics go awry.

Another frontier is regional variation.

shows that the shift mode isn’t uniform globally. In the tropics, where convection is frequent and intense, the effect is stronger. In dry regions, it’s muted. This hints at interactions between resolution effects and local climate regimes — a complexity the current framework simplifies.

There’s also the question of why the shift occurs. The paper doesn’t derive it from first principles, but the pattern aligns with scale-invariant theories of turbulence and cascades (Lovejoy, 2023). Rain, like river networks or cloud fields, may follow fractal-like statistics across scales. If so, the shift mode could be a manifestation of deeper physical symmetries.

For now, the takeaway is clear: resolution matters — not just for detail, but for magnitude. When we say “rain is getting heavier,” we must ask: Compared to what, and at what scale?

As satellite constellations improve and models push toward kilometer-scale grids, this insight will only grow more important. The next generation of climate services — from early warning systems to agricultural advisories — must account for scale not as a technical footnote, but as a fundamental driver of what we observe.

The rain hasn’t changed. Our lens has. And now, for the first time, we have a way to focus it.