Climate Change: Simple, Serious, Solvable › Models and Data
Where the Rain Goes: The Water Balance in 16 Climate Models
Rain and snow fall on the land. Some goes back to the air, some runs off to the rivers, and some waits in the soil. Sixteen of the world's climate models keep those books month by month. Pick a region and see what they say about now and about late in this century under high emissions.
Loading the data…
- 1995–2014
- 2081–2100, SSP3-7.0
| 1995–2014 | 2081–2100 | Change | Models that agree |
|---|
| Month | P | E | R | P − E − R | Soil water |
|---|
- 1995–2014
- 2081–2100, SSP3-7.0
| Model | 1995–2014 | 2081–2100 | Coast filled | ||||
|---|---|---|---|---|---|---|---|
| P − E | R | Gap | P − E | R | Gap | ||
Over a year, the books close. In the wet months, more water falls than leaves, and the rest goes into the snow, the soil and the ground. In the dry months it comes back out. Averaged over 20 years, what goes in comes out, so P − E should equal R. The gap is what a model's water doesn't add up to here.
These are the climate models' own water books. Each model works out evaporation and plants' water use with its own weather, soil and plants, and the plants respond to the extra CO₂ as well as to the heat. That's why this page and the watershed model can disagree. The About tab explains.
Each scenario is a what-if. SSP3-7.0 is high emissions. It isn't a forecast.
About this page
Chapter 8 of the book follows the water through a watershed: P − E = R + ΔS. Precipitation, P, minus evapotranspiration, E, equals runoff, R, plus the change in stored water, ΔS. Climate models keep the same books for every patch of land, every half hour. This page shows what 16 of them say, month by month, for the IPCC's land regions, in 1995–2014 and in 2081–2100 under the high emissions of SSP3-7.0. They're the same 16 models, and the same runs, as the book's Figure 8.4.
On the page as it opens, it's the Mediterranean. The average of the models gives it 15% less rain and snow in a year, 10% less evapotranspiration, and 33% less runoff, from 105 to 70 mm a year. All 16 models have less of each.
How to use this page
Choose a region. "The year" shows its twelve months in four graphs: precipitation, evapotranspiration, runoff, and the water in the top 10 cm of soil. Gray is 1995–2014 and red is 2081–2100. The bold lines are the average of the models, and the thin lines, if you tick the box, are the models one by one. "Pick out one model" draws one in black. The first table sums up the year, with the change and how many models agree on its direction. The second gives the average month by month, for either period.
"Do the books close?" checks the water balance. Each dot is one model's year: across, P − E; up, R. If the books close, the dot sits on the dashed line. The graph below it shows P − E − R month by month: water going into storage and coming back out.
"Back to where the page opened" puts everything back. The address in your browser changes as you work, so you can bookmark a view or send it to someone. Nothing is sent anywhere as you work.
This page is part of the website for Climate Change: Simple, Serious, Solvable, an open textbook by Scott Denning. It goes with Chapter 8, "Future Climate Change", sections 8.3 and 8.4.
The arithmetic
The page answers one question: where do the models say the rain and snow go, and how will that change? Water is in millimeters: a millimeter of water spread over the whole region.
1. A month's water
The models give each month's precipitation, evapotranspiration and runoff as a rate, p, in millimeters a day. A month's total, here its precipitation P, is the rate times the days in the month, d:
A year's total is the sum of its twelve months. The water in the soil is an amount, not a rate, so the page shows it as it is, and its year is the average of the twelve months.
2. The books
What falls and doesn't leave is stored:
Month by month, ΔS rises and falls as snow piles up and melts and the soil fills and dries. Averaged over 20 years, the stores end about where they started, so over a year
3. The average of the models
The bold lines and the tables are the plain average of the n models that count in the region (usually all 16), month by month. Every model counts the same.
4. The change, and how many agree
The change, c, is the models' average in 2081–2100, F, against their average in 1995–2014, B, as a percentage of B:
This is how Figure 8.4 measures change. It isn't the average of each model's own percentage. "Models that agree" counts the models with more in 2081–2100, or with less, whichever is greater.
A worked example
Choose the Mediterranean and pick out MPI-ESM1-2-LR, from the Max Planck Institute for Meteorology.
- One month (step 1). In 1995–2014, its January precipitation averaged 1.5175 mm a day. January has 31 days, so P = 1.5175 × 31 = 47.0 mm.
- The books (step 2). Over the year, this model has P = 350.4 and E = 297.7 mm, so P − E = 52.7 mm. Its runoff is 52.8 mm. The books close to a tenth of a millimeter.
- The change (step 4). In 2081–2100 its runoff is 34.1 mm. The change is 100 × (34.1 − 52.8) ÷ 52.8 = −35%.
"Do the books close?" lists those numbers, and the summary under the first graph says −35% for the picked model.
Where the numbers stop
- No snow. The model output this page uses has no snowpack, so the snow shows up only in ΔS: water stored in winter and given back in spring. For the snow itself, use the watershed model.
- One scenario. SSP3-7.0, high emissions, late in the century. Lower emissions would mean smaller changes.
- Runoff from each patch of land, not river flow. The page adds up the water that leaves the soil. It doesn't follow it down the rivers, through reservoirs or out to farms and cities.
- One run of each model. Some of a model's difference from the others is just its weather.
- Big regions, coarse models. The models' grid squares are about 80 to 300 km across, so a mountain range is a few squares, and a region's average hides its mountains and coasts. A grid square that is partly sea takes the numbers of the nearest square that is nearly all land, so the sea's evaporation isn't counted as the land's.
- The books don't close perfectly in every model. In a few models and regions, the runoff is more than P − E by up to a few hundred millimeters a year. The page doesn't know why. It may be how a model keeps its lakes, wetlands or ice, or how the page picks out each model's land. The page shows the gap, so you can see it.
Why the watershed model says something else
The book's watershed model does the same bookkeeping with a few lines of arithmetic. Its evaporative demand rises about 6% for each degree of warming, from temperature alone, and its plants take what the air demands if the soil has it. On the French Broad River in North Carolina, 5 °C of warming and 10% more rain and snow cut its runoff from 828 to 733 mm a year. Over the same kind of land, eastern North America, the climate models give 11% more rain and snow, 12% more evapotranspiration, and 10% more runoff, and 13 of the 16 models have more.
The biggest difference is the plants. With more CO₂ in the air, leaves open their pores less and lose less water. In 16 climate models of the generation before these, under high emissions, that almost cancels the extra pull of warmer, drier air on plants: where water isn't short, evapotranspiration rose only 3.6% from 1861–1960 to 2071–2100, because the drier air added 8.5% and the plants' response took away 8.1% (Yang and others, 2019). A drought index worked out from temperature, sunshine and humidity, leaving the plants' response out as the watershed model does, shows drought stress rising over more than 70% of the world's land; the climate models' own water balance shows it over 37% (Swann and others, 2016).
So the watershed model is good for seeing how the books work and how snow changes the timing, and this page for how much wetter or drier the models say a region will get. In the Mediterranean the two agree on the direction, because there it's the rain and snow that fall.
Things to try
- Dry gets drier. In the Mediterranean, evapotranspiration falls even as the air warms: there's less water to take. Every model agrees.
- Wet gets wetter. In South Asia, the rain and snow rise 18% and the runoff 32%, and all 16 models have more of both. Look at the monsoon months in the runoff graph.
- Earlier water. In western North America, the month with the most runoff moves from March to February, and the spring months lose runoff while the winter gains it.
- Storage. On "Do the books close?", choose northern Europe. In 1995–2014, water goes into storage from September to March and comes out from April to August, most of all in May, as the snow melts. By 2081–2100 it starts coming out in March, and May's loss is smaller.
- The soil dries almost everywhere. Choose all land. The rain and snow rise 6% and the runoff 9%, but the water in the top 10 cm of soil falls in all 16 models.
Where the data come from
The models' monthly output is from CMIP6, the experiments that climate modeling centers ran for the IPCC's Sixth Assessment Report: precipitation, evapotranspiration, total runoff and the water in the top 10 cm of soil, from one run of each model, for 1995–2014 and for 2081–2100 under SSP3-7.0. They're the inputs the book's Figure 8.4 was made from, read from the public CMIP6 archive on Google Cloud, with every file's fingerprint kept. The models' data are CC BY 4.0. The modeling centers are named in the "Pick out one model" list.
The regions are the IPCC's reference regions:
Iturbide, M., et al. (2020). An update of IPCC climate reference regions for subcontinental analysis of climate model data: definition and aggregated datasets. Earth System Science Data, 12, 2959–2970. doi:10.5194/essd-12-2959-2020
How plants' response to CO₂ changes the picture:
Swann, A. L. S., F. M. Hoffman, C. D. Koven, and J. T. Randerson (2016). Plant responses to increasing CO₂ reduce estimates of climate impacts on drought severity. Proceedings of the National Academy of Sciences, 113, 10019–10024. doi:10.1073/pnas.1604581113
Yang, Y., M. L. Roderick, S. Zhang, T. R. McVicar, and R. J. Donohue (2019). Hydrologic implications of vegetation response to elevated CO₂ in climate projections. Nature Climate Change, 9, 44–48. doi:10.1038/s41558-018-0361-0
How it was checked
- The data, against the source. Every file read from the archive was checked against its fingerprint from when it was downloaded for Figure 8.4, and the page's files are made from the reduced numbers by a script whose output is checked the same way.
- The books. For all land, every model's P − E and runoff agree to within 5% of its precipitation, in both periods.
- The arithmetic, against a second copy. The tests have the page's arithmetic written out a second time, in another programming language, and compare them for many regions.
- This tab, against the data. Every equation above is tried with the page's own numbers, the worked example is done again from the rounded numbers a calculator would use, and each number in the text is worked out again from the data.
The code
Two files do all the work, and they use no outside software.
numbers.js is the arithmetic: months, years, the books, the averages and the changes.
app.js is the page: the controls, the graphs and the tables.
To keep a copy, save numbers.js or app.js.
Credits
By Scott Denning, for Climate Change: Simple, Serious, Solvable. Claude, from Anthropic, built the page from the models' data and checked it against the book. The modeling centers that ran the models are named in the "Pick out one model" list.