Rainfall intensity shifts to heavier rates at finer resolution, even with physics fixed

Why it matters

Differences in rainfall intensity between datasets or models of different resolution may come from resolution alone, not from better physics. Comparisons that ignore this can mislead. The authors suggest the framework could serve as a baseline for evaluating high-resolution models.

What they did

The authors took satellite-gauge products (IMERG at 0.1º, and four FROGS products at 1º) and coarsened them to grids up to 10º. They also coarsened 1-minute gauge records from two ARM sites to daily steps. Coarsening kept total rainfall fixed, so physics did not change. They built rainfall amount distributions at each resolution and measured the “shift mode”, the change in the rate above which half of the rain falls.

Key findings

• In IMERG, the rate above which half of rain falls drops from 19 mm/day at 0.1º to 16.38 mm/day at 1º and 10.85 mm/day at 5º.
• Globally averaged yearly maximum daily rainfall falls from 67 mm/day at 0.1º to 57 mm/day at 1º and 35.7 mm/day at 5º.
• The shift mode grows with resolution difference: 3.4% from 0.25º to 0.1º, and 144.8% from 10º to 0.1º. It also depends on the base resolution.
• In time, the shift is 3% from 2-minute to 1-minute and 1851% from 1-day to 1-minute at the tropical site. At the mid-latitude site the values are 5% and 2450%.
• The shift is larger over the tropics, land and high terrain, where rain systems are smaller. It works across the other satellite products, and the shift error for most IMERG pairs is 25% to 30%.

Limitations

• The shift mode does not recover heavy extremes well for large resolution gaps. At 0.1º, extreme rain is 80.90% larger than the adjusted 5º value.
• Spatial results depend on the effective resolution of merged products, and nothing finer than 0.1º could be tested. The gauge records also have known biases, and the two sites cover different time spans.
• Grid-point distributions are noisy and the records are short. Fit errors are large at many grid cells.

Glossary

• Precipitation amount distribution: Total rain falling in each rain-rate bin, normalized by number of days, on a log-spaced rate axis.
• Shift mode: A change in a distribution made by sliding it along the log rate axis without altering its shape.
• Conservative regridding: Coarsening a grid so that total precipitation amount stays the same.
• RX1day: The highest one-day rainfall total in a year.