 |
 |
 |
 |
 |
 |
|
|
|
|
|
|
To convert the SST predictions to climate impacts and to understand the physical processes involved, we have used the GISS GCM Model II' in a hindcast mode with globally observed SST (Hansen et al., 1997). We assessed: (1) the influence of the ENSO phenomenon on seasonal anomalies in the study regions; (2) the number of GCM simulations needed for producing a seasonal climate prediction; (3) the effect of including climate forcings other than SSTs; (4) the improvements offered by increasing the vertical or horizontal resolution of the GCM; and (5) the use of remote sensing techniques to assist in forecast evaluation on global and regional scales.
The first set of experiments (Druyan et al., 1999) evaluated the GCM's responses to observed global sea surface temperatures specified monthly over the time period 1969-98. SSTs for 1979-88 were taken from the AMIP data sets, and those for subsequent years from the blended analysis approach of Reynolds (1988). Each of 5-10 simulations using this version of the model was initiated using a unique perturbation of the atmospheric state in order to gauge the model's range of responses to initial conditions, but each was forced by identical observed SSTs. Results were compared both with remotely sensed temperatures from the Microwave Sounding Unit (MSU), by way of a microwave postprocessor to the GCM results calculating the brightness temperature that would be observed by the MSU radiometer if it were orbiting the simulated GCM atmosphere (Shah and Rind, 1998). Additional comparisons were made with ground-based temperature and precipitation observations.
 Figure 1: Top - Five-month running means of precipitation rate anomalies (mm) for Nordeste. 'Model' refers to ensemble values that are departures from 23-year monthly means, averaged over the area 4-12S, 37.5-47.5 W. Observations are departures from the base period 1951-1980, based on precipitation measurements averaged over the same area and archived at NCDC. Bottom - Eleven-month running means of the standard deviation (mm/day) between the modeled precipitation rates of the five simulations over the same area (above). |
The second set of GCM experiments included simulation of changes in trace gases, volcanic aerosols, stratospheric and tropospheric ozone anomalies, stratospheric and tropospheric sulfate aerosol anomalies, and tropospheric carbon aerosol anomalies over the same time period (Shah et al., 1999). The forcings were added individually and in various combinations. Results were compared with those obtained from the ensemble runs, and were evaluated relative to the mean and standard deviations from those runs. Again comparison was made from both remote sensing and regional ground-based observations.
The final set of GCM experiments investigated the utility of increasing horizontal and vertical resolution. First the 4°x5° 9-layer model was increased to 4°x5° by 18 levels. Then the horizontal resolution was increased to 2°x2.5°, with 18 levels. Finally, the last run was made with 2°x2.5° by 32 layers. Each model was identical to that used in the previous set of simulations except for the horizontal/vertical resolution change. All simulations were made using observed SSTs but not other forcings.
For the time period investigated, results showed that mid-tropospheric Tb correlations reached as high as 0.66 globally and 0.84 in the tropics. As the geographic region moved further away from the tropical Pacific, the correlation in general diminished. Surface air temperature correlations showed similar tendencies, with somewhat lower values, due to the heterogeneity of the surface-level temperature patterns.
|
|
|
Figure 2 (left): Model ensemble means of the time-longitude distribution of zonal wind component anomalies at 850 mb between 1971-1991 averaged over 5S-5N between 100E and 280E (80W), in m/s. Figure 3 (right): Model ensemble means of the time-longitude distribution of Outgoing Longwave Radiation anomalies averaged over 5S-5N between 100E and 280E (80W), in W/m2. |
|
Of the several agricultural regions studied, northeastern Brazil was simulated best, for both temperature and precipitation. This is consistent with observations indicating that sixteen out of seventeen warm ENSO phase episodes from 1911 to 1983 coincided with below-normal rainfall over northeastern South America (Ropelewski and Halpert, 1987), presumably associated with subsidence as part of a meridional circulation induced by the SST changes (Moura and Shukla, 1981; Marengo and Hastenrath, 1993). Use of the additional forcings had their primary impact over land at higher latitudes, although they also allowed the model to better simulate the observed precipitation extremes in the Nordeste region, as did finer resolution (Fig. 1). Also shown (Figs. 2 and 3), are two aspects of the model's ability to simulate ENSO events over the nearby Tropical Pacific.
The number of simulations in the ensemble improves the correlation with observations. The improvements are especially large going from one simulation to three, and then become progressively smaller. It appears that five simulations are sufficient for a relatively homogeneous variable such as temperature in the middle troposphere. In the next phase of the project, we will test the optimal number of simulations for precipitation, and the use of additional influencing factors, such as the phase of the quasi-biennial observation (QBO).
We are conducting a coordinated analysis of the 1997-2000 El Niño/La Niña, examining the predictions, oceanic manifestations, climate teleconnections, marine and terrestrial biological responses, and fishery and agricultural ecosystem outcomes. To assess how well the GISS GCM can predict climate parameters of consequence for agricultural output, we've been running hindcast experiments for the latter half of the 1990s using observed SSTs. We compare the accuracy of forecasts for temperature and precipitation in the chosen agricultural target areas when running models with finer horizontal or vertical resolution in comparison to coarser grid models.
The apparent advantages that finer resolution models have include a more accurate numerical depiction of the fundamental equations (so that storms move faster, for example), and parameterizations of subgrid scale processes using variables determined at a scale closer to the relevant one (for processes such as convection). The disadvantage is that many more simulations can be done with the coarser grid model.
Five simulations have so far been performed with a 4°x5° (lat × long) model with 12 vertical layers, for the decade of the 1990s. An additional simulation has been made for a 2°x2.5° model with 32 vertical layers for the same time period. We concentrate on determining the accuracy of the depiction of temperature and precipitation variables for the agricultural regions during the intense El Niño of 1997/98, and the subsequent La Niña. One research question is: Will the mean for the five coarser grid runs be a better representation than the single finer grid run? In progress are further simulations with each resolution, but the coarser grid model will obviously be able to be used for many more simulations.
Current work involves adding a third model to the simulation experiments, a 4°x5°, 53-layer model whose top is at the mesopause. The ability of wave/mean flow interaction involving the stratosphere to affect interannual variability has recently been a focus of much research here at GISS (e.g., Shindell et al., 2001), and we will test whether having the full stratosphere improves the global teleconnections associated with ENSO events from the perspective of agricultural impacts.
The 15-layer regional climate model (RCM) at GISS/CCSR uses a Cartesian grid with 50-km spacing for dynamics and incorporates special treatment of soil moisture. The RCM has previously been applied to studies of African wave disturbances (Druyan et al., 2000, 2001) and monthly mean climates over southern Africa (Fulakeza et al., 2001).
Prescribed lateral boundary conditions (LBCs) are merged with the predicted RCM evolution by weighting them with progressively decreasing weights inward within a buffer zone that completely surrounds the domain of interest.
We are participating in a regional model intercomparison effort coordinated by the International Research Institute for Climate Prediction (IRI). In this context the RCM has been used to downscale NCEP reanalysis fields over South America (hereafter "DNR" runs), beginning with February 20, 1997. Several initial RCM simulations for March 1997 were used to "tune" precipitation rates by adjusting several parameters and by adopting a better moisture initialization procedure. A single simulation from February 20, 1997 and until December 31, 1997 was completed and results (monthly precipitation and surface temperature) are posted on an ftp site hosted by the Experimental Climate Prediction Center at Scripps, UCSD.
We explored the question of model variability by adding four additional simulations covering the March-May 1997 season. Each of these simulations was begun from one of four different NCEP reanalysis data sets, February 21-24, 1997. Together with the original run, this provided an ensemble of five runs, all forced with identical LBCs.
Additionally, we evaluated the potential for seasonal climate prediction by making a parallel set of five simulations from the same initial conditions, but using GISS GCM predictions for LBCs (hereafter "PRED" runs). This experiment incorporated persisted SST anomalies in place of the observed March-May 1997 SST used in the first set.
Creation of identical ensembles of five DNR and five PRED simulations is underway for the period March-May 1985, which was a rainier season over northern Brazil. By focusing on the 1997-1985 differences in climate fields, the performance of the RCM can be evaluated without systematic biases.
Simulated precipitation fields were validated against station rain gauge data as well as CMAP distributions. (CMAP data is recognized as an imperfect proxy for actual precipitation rates.) For both 1997 ensembles, RCM March-May precipitation fields validated better than individual monthly mean fields. Both the March-May 1997 DNR and PRED precipitation rates underestimated observations over Northeast Brazil, with PRED making slightly larger errors. Both DNR and PRED overestimated the observed rainfall over Ecuador and southeastern Brazil, but DNR errors were distinctly larger over the latter area. These common errors probably reflect RCM modeling biases. DNR Amazon Basin rainfall was quite realistic, except for exaggerations near and just west of the Amazon Delta. PRED was too rainy just south of the Amazon River and too dry where the ITCZ crosses the Brazil coast. These different outcomes may be a result of the differences between NCEP reanalysis versus GCM forcing or a consequence of differences in the SST used for each set of experiments. Testing may sort this out.
With completion of the 1985 simulations, the 1997-1985 differences in several climate fields can be formed. The potential for seasonal climate prediction will be better assessed since RCM systematic biases will be filtered. Additionally, the long simulation will be continued into 1998, as planned for the model intercomparison study. Analysis of a complete compliment of simulated climatological fields will be made to better understand RCM performance, weaknesses and strengths. Experiments will be designed to determine the relative impact of spatial resolution, domain size, method of lateral boundary forcing, quality of large scale forcing data and the importance of surface boundary information, such as SST, vegetation cover, soil moisture, etc.
CAFE Home *
Summary *
Team
Components:
ENSO,
Climate,
Marine,
Vegetation
Education *
Instruments *
Acquisitions *
Publications
|
|
|
