 |
||||
|
||||
|
|
|
 |
||
|
||||
      |
Hydrologic Effects of Shifting Probability Functions for High Precipitation Events Project
Collaborators
Tom Croley, Great Lakes
Environmental Research laboratory (GLERL home)
Executive Summary
We attempt to incorporate the effect of the anticipated effect of increasing incidence of heavy precipitation events due to greenhouse warming. We previously discovered that, in comparing the spectrum of precipitation events in time periods covering the recent past vs. 90 years in the future, as simulated by the Canadian Centre for Climate Modelling and Analysis's general circulation model CGCM2, we do not necessarily find this phenomenon in evidence, and at some gridpoints find the opposite. We have now added similar investigation using the Geophysical Fluid Dynamics Laboratory (GFDL) model CM2.0, which gives somewhat more support for the premise that precipitation will increase proportionately more in heavy events.
This GOES-8 image shows a powerful storm system of precipitation across the midwest 2002-01-31. (image courtesy of NOAA)
2005 Accomplishments
This project was undertaken to test the validity and consequences of greenhouse warming causing precipitation to occur in more intense, although perhaps less frequent, episodes. Other things being equal, this would be expected to cause more water to run off from the surface, and thus cause greater river flow. This would bring more water to the Great Lakes and raise their levels.
We take a step toward incorporating this effect into offline hydrologic models of the influence of greenhouse warming on the Great Lakes water levels by stratifying precipitation events from heaviest to lightest and treating them separately In the previously used method, originally from Croley (1990), the mean precipitation for each month, from a future scenario of a general circulation model (GCM) of the climate, was divided by the same quantity derived from a present scenario to get a precipitation ratio. To produce a precipitation scenario for future times, all data points in that month within the observed time series of precipitation were multiplied by the same ratio, and used as input to an offline hydrologic model for comparison to a case using unperturbed observed precipitation. As an alternative to this, we propose ranking daily precipitation amounts in each of the GCM scenarios from highest to lowest, and deriving ratios from the similarly-ranked daily precipitation amounts of the two scenarios. These ratios would then be applied only to the observed data with the same quantile rank. That is, the highest-ranked precipitation day of a given month in the observed record would be perturbed using the ratio between the highest-ranked precipitation day in the future scenario and that in the present scenario. Likewise for the lowest-ranked precipitation day and all of those between.
a
b
c
FY 05 Figure 1 Examples, at different model gridpoints in the immediate vicinity of the Great Lakes, of ratios of like-ranked daily precipitation amounts during July for runs of the CGCM1 centered at 2095 and 1989.
This project has made some progress during FY 2005, and produced slightly more evidence that precipitation increases proportionately more during heavy precipitation events than lighter events. During FY 2004, ratios were derived from the Canadian Centre for Climate Modelling and Analysis's general circulation model CGCM2. Figure 1 shows samples of ratios of July precipitation at a few of this model's grid points. Because we used ten years of data from July only, the precipitation amounts are ranked from 1 to 310. If the simple statement were true that heavy rain events will intensify in the future, these charts should show the highest ratios of future to present precipitation in the highest-ranked precipitation days, with a monotonic progression to lower ratios for lower-ranked precipitation events. Instead, the example in Fig. 1a has a decreasing trend in ratio from the highest-ranked precipitation day to the subsequent rankings that extends through only a small portion of the entire ensemble, then a rise again, with a local maximum in the precipitation ratio at about the 190th rank, followed by a decrease to the lowest rank. The next example does not have the decreasing trend from the highest-ranked, but does have a local maximum near the 240th rank. Finally, the third example has entirely the opposite character from what was expected--the highest-ranked precipitation days have the lowest precipitation ratios and vice-versa.
FY 05 Figure 2 As in Fig. 1, but using data extracted from the GFDL Climate Model CM2.0.
Some roadblocks were overcome in data format problems with extracting the data from the GFDL model, and similar analysis was done on those data. The two panels on the left side of Fig. 2, corresponding to locations near Duluth, Minnesota, and Wawa, Ontario, show a pattern consistent with intensification of heavy rainfall events. However, a closer look at the scale on those two panels shows a general decrease in precipitation during July at those locations. The ratios approach 1 in some of the highest-ranked rainfall events, but are lower for other events. The two locations in the panels on the right-Flint, Michigan, and Buffalo, New York-show very different patterns. They show the highest ratios of future to present precipitation for July days ranked in the 200th or greater place. In the case of Buffalo, the ratio for the July days with the highest precipitation is only slightly greater than 1, while the largest ratios, at ranks around 215 out of 300, approach 7.
On the whole, these data from general circulation models (GCMs) do not consistently support the statement that precipitation will be concentrated in heavier events in the future. While this does not prove this statement to be false, it means that using the output of GCMs, which is considered a means of getting a plausible input for modeling hydrological impacts, will not allow one to test the impacts of these shifts in probability distribution of precipitation events.
2004 Accomplishments
This project was undertaken to test the validity and consequences of greenhouse warming causing precipitation to occur in more intense, although perhaps less frequent, episodes. Other things being equal, this would be expected to cause more water to run off from the surface, and thus cause greater river flow. This would bring more water to the Great Lakes and raise their levels.
We take a step toward incorporating this effect into offline hydrologic models of the influence of greenhouse warming on the Great Lakes water levels by stratifying precipitation events from heaviest to lightest and treating them separately In the previously used method, originally from Croley (1990), the mean precipitation for each month, from a future scenario of a general circulation model (GCM) of the climate, was divided by the same quantity derived from a present scenario to get a precipitation ratio. To produce a precipitation scenario for future times, all data points in that month within the observed time series of precipitation were multiplied by the same ratio, and used as input to an offline hydrologic model for comparison to a case using unperturbed observed precipitation. Instead, we have proposed ranking daily precipitation amounts in each of the GCM scenarios from highest to lowest, and deriving ratios from the similarly-ranked daily precipitation amounts of the two scenarios. These ratios would then be applied only to the observed data with similar quantile rank. That is, the highest-ranked precipitation day of a given month in the observed record would be perturbed using the ratio between the highest-ranked precipitation day in the future scenario and that in the present scenario. Likewise for the lowest-ranked precipitation day and all of those between.
Figure 1. Examples of ratios of like-ranked daily precipitation amounts during July for runs of the CGCM1 centered at 2095 and 1989.
This project has not had as much progress as anticipated, and the results that have been derived do not support the premise of this project. Namely, ratios have been derived from the Canadian Centre for Climate Modelling and Analysis's general circulation model CGCM2. Figure 1 shows samples of ratios of July precipitation at a few of this model's grid points. Because we used ten years of data from July only, the precipitation amounts are ranked from 1 to 310. If the simple statement were true that heavy rain events will intensify in the future, these charts should show the highest ratios of future to present precipitation in the highest-ranked precipitation days, with a monotonic progression to lower ratios for lower-ranked precipitation events. Instead, the example in Fig. 1a has a decreasing trend in ratio from the highest-ranked precipitation day to the subsequent rankings that extends through only a small portion of the entire ensemble, then a rise again, with a local maximum in the precipitation ratio at about the 190th rank, followed by a decrease to the lowest rank. The next example does not have the decreasing trend from the highest-ranked, but does have a local maximum near the 240th rank. Finally, the third example has entirely the opposite character from what was expected--the highest-ranked precipitation days have the lowest precipitation ratios and vice-versa.
Daily precipitation data from other GCMs have been sought to give similar treatment. However, difficulties with data format have presented a roadblock to further progress in this direction.
Project Scientific Rationale
One of the anticipated outcomes of global warming is that precipitation will increase, and that the predominant mode of this increase will be intense storms (Kharin and Zwiers 2000; Yonetani and Gordon 2001). If proportionally more precipitation were to occur during intense episodes, this is likely to increase runoff more than if precipitation from all episodes were to increase by a constant ratio. In past studies of hydrologic impacts of greenhouse warming (Croley 1990, Lofgren et al. 2002, Croley 2003), precipitation amounts were adjusted by constant ratios, one for each month of the year, that were applied to each precipitation report during that month for the entire historical record. These ratios were derived from the ratios of precipitation in general circulation model (GCM) simulations between present and future climate scenarios (actually time slices rather than separate scenarios in the case of transient simulations). This approach was taken because GCMs contain biases that can make it unrealistic to simulate hydrology using their output directly. However, the problem with this approach is that it preserves the temporal structure of storms as found in the historical record and does not use the storm structure in the changed climate GCM scenarios.
Because runoff in the long term is equal to precipitation minus evapotranspiration over land, let us also justify the anticipated increase in runoff from an evapotranspiration point of view. The soil has a moisture field capacity that is depleted by evapotranspiration and replenished by precipitation. When heavy rain occurs, the soil moisture becomes saturated, and the excess goes away as surface runoff or percolation into deep soil. Evapotranspiration can then occur, subject to the limitation that it stops once the soil moisture is entirely exhausted. In a hypothetical case in which the overall precipitation amount remains the same but comes in less frequent but more intense bursts, there is increased time between these episodes for the soil moisture to become exhausted and prevent evapotranspiration from occurring, hence less overall evapotranspiration and more runoff.
An approach that we would like to take to improve the method of hydrologic impact analysis would be to stratify the adjustment ratio according to the quantile rank of the daily precipitation. That is, if we compare two 20-year time slices from GCMs for July, we will get the (31x20=620) daily values of precipitation from the GCM for each of these time slices and rank them. A ratio will be calculated between the top-ranked precipitation events in the two GCM scenarios. Given a 49-year historical record, this ratio will be applied to the top two or three days' precipitation in the historical record. Many of the lowest-ranked daily precipitation amounts in the two GCM scenarios and (especially) the observations are likely to be zero values. To avoid problems with ratios that either have a value of zero or involve division by zero, I propose that for the quantile ranks in which either of the GCM scenarios has a zero value but the observations have a non-zero value, we revert to the default, the overall monthly ratio. The next step will be to normalize the data so that the overall ratio of precipitation between the original observed data and the synthesized dataset based on it will be the same as between the base period of the GCM and the future period. This normalization is necessary because, in general,
Here, xi are the individual daily observed precipitation values, yi are those for a future climate simulation, and zi are for a base case simulation.
Although this methodology is applicable to all of the meteorological variables used to drive hydrologic models, the basic motivation for this is in terms of precipitation, because of the anticipated consequences (enhanced runoff due to more intense rainfall events). A potential problem with using this methodology on multiple variables simultaneously is the mixing of cross-correlations (i.e. days with precipitation at a certain rank are likely to have temperature at a very different rank).
This methodology will help to capture the effects of changes in statistical distribution of precipitation on a daily basis. It is also likely to capture at least some of the effect of interannual variability in precipitation. This is manifest in this methodology because a period with higher than normal rainfall is likely to have many days among the highest-ranked. Thus, the ratios applied to these will be those between the high-precipitation times of the two GCM datasets. Likewise for the low precipitation time periods, thus capturing the range of precipitation regimes within the GCM datasets and transferring their relative strength to the corresponding periods within the historical record.
As an additional test of the reasonableness of this approach, we will do some time series analysis of the data. One way of looking at this will be to look at storm shape, that is, does the time of maximum precipitation within a storm period shift (i.e. from earlier in the storm to later) between the GCM's base case and future case? Another question is whether the frequency of precipitation events changes in the GCMs between the base case and future cases. Also, the magnitude of variability at different timescales will be examined. It is expected that, assuming that changes occur in storm shape, frequency, and magnitude of variability, the representation of overall magnitude of variability and of frequency of precipitation greater than a given threshold will be improved. However, it is unlikely that changes in storm shape or frequency of any amount of precipitation will be improved, and the details of variability at particular timescales may not be well captured.
It is hoped that this methodology will be viewed as an improvement over that used in Croley (1990), Lofgren et al. (2002), and Croley (2003), with only a marginal amount of added complexity and difficulty. As the former methodology remains standard practice within the community that examines the hydrologic impacts of global warming, an improved version using the same basic framework would stand in a good position to become widely used.
Literature Cited
Kharin, V. V., and F. W. Zwiers, 2000: Changes in the extremes in an ensemble of transient climate simulations with a coupled atmosphere-ocean GCM. J. Climate, 13, 3760-3788.
Yonetani, T., and H. B. Gordon, 2001: Simulated changes in the frequency of extremes and regional features of seasonal/annual temperature and precipitation when atmospheric CO2 is doubled. J. Climate, 14, 1765-1779.
Croley, T. E. II, 1990: Laurentian Great Lakes double-CO2 climate change hydrological impacts. Climatic Change, 17, 27-47.
Lofgren, B. M., F. H. Quinn, A. H. Clites, R. A. Assel, A. J. Eberhardt, and C. L. Luukkonen, 2002: Evaluation of potential impacts on Great Lakes water resources based on climate scenarios of two GCMs. J. Great Lakes Res., 28, 537-554.
Croley, T. E. II, 2003: Great Lakes climate change hydrologic impact assessment I.J.C. Lake Ontario-St. Lawrence River Regulation Study. NOAA Technical Memorandum GLERL-126, 77 pp.
Presentations
Lofgren, B. M., 2004: Discrepancies in greenhouse lake level predictions:
Reasons for uncertainty. 47th Conf. On Great Lakes Res., Int. Assoc. Great
lakes Res., Waterloo, ON, May 24-28, 2004.
Lofgren, B. M., 2004: Slippery when warmed: Predicting lake levels under
greenhouse warming. Purdue University–Departments of Atmospheric
Sciences and Agronomy Seminar Series, West Lafayette, IN, August 27, 2004.
Last updated: 2006-08-21 mbl