GFDL - Geophysical Fluid Dynamics Laboratory

Skip to main content

Runoff in a Changing Climate: "Stationarity is Dead"

Hydrologists use observations of streamflow to characterize the statistical distribution (the probabilities) of streamflow. Water planners (indeed, water users of all types: households, farms, industries, nations) typically assume that the historical statistics of streamflow will apply also during the future for which they are planning; this is the assumption of stationarity. Stationarity is an approximation, compromised by such processes as construction of dams, irrigation of crops, or regrowth of forests after abandonment of agriculture. The degree to which failure of stationarity is problematic naturally depends on the specific situation. One long-term, global-scale challenge to stationarity is contemporary human-driven climate change.

Read our short essay on the death of stationarity.

How will ongoing climate change affect streamflow? Will there be more water to use? Less water? Will the frequency of droughts change? What about flood risks? How will the answers to these questions vary around the globe? It is easy to state that climate change will affect the statistics of streamflow. It is difficult to give relevant numbers that are both precise and reliable.

It is the philosophy of this research project that the place to begin to address these issues is at the larger time and space scales. Projections of long-term regional changes in runoff appear more tractable than projections of local changes in flood risk. The former depend more on large-scale patterns of climate, which are the main focus of global climate models, while the latter depend more on storm dynamics and river-basin response, which are not currently well represented in such models.

Are global climate models up to the task of estimating the sensitivity of annual mean runoff to changing atmospheric composition? In our evaluation of global climate models, we found that they exhibited significant skill in translating changes in climate forcing (atmospheric composition, solar variability) into the global pattern of streamflow trends during the 20th century. This success has contributed, along with other analyses, to a perceived increasing credibility of hydrologic projections of climate models.

What is the current "best estimate" of future changes in runoff in various regions of the world? There is no short answer to this question. According to our 2005 analysis of climate-model experiments (figure below, click to enlarge), models generally agree that high latitudes of North America and Eurasia and some tropical regions will experience increasing runoff, while runoff is expected to decrease in southwestern North America, Mediterranean region, and southern Africa.

[CLICK GRAPHIC TO ENLARGE IN NEW WINDOW] Projected change in annual runoff volume by the middle of the 21st century, relative to the historical period 1900-1970. Color denotes percentage change (median value from 12 models). Presence of color indicates that at least 8 of 12 models agreed on the direction (increase or decrease) of runoff change under the Intergovernmental Panel on Climate Change (IPCC) "SRES A1B" emissions scenario. (Figure from Milly et al., 2008.)


Are numerical results available from our analyses? Yes. We have prepared a data set based on computations similar to those presented and documented by Milly et al. (2005). The data set contains estimates of the change in expected value of runoff, based on the median response of 12 climate models, for each year from 2016 to 2084 inclusive, under the SRES A1B scenario. The data set, in netcdf format, can be downloaded here. And here is an animation of this data set.