Amazonia and Nordeste, Brazil are the regions of interest for this study. These regions are subject to differing large scale circulation patterns seasonally as well as complex local forcings of topography, land-use, and land-sea contrasts. The combination of large scale and local forcing results in a spatial distribution of seasonal and annual rainfall which shows sharp horizontal gradients over a few tens to a few hundred kilometers. In addition, variations in rainfall from year to year, particularly in Nordeste, can be as large as contrasts between normal wet and dry seasons. Anomalous seasonal rainfall in Nordeste is related to the inter-hemispheric Atlantic SST gradient, while in Amazonia rainfall anomalies have been related also to inter-annual SST variability in the eastern Pacific (El Nino/La Niņa).

Successful simulation of seasonal precipitation in TSA requires adequate representation of both large scale and local surface forcings. Large scale circulation anomalies evolve in response to SST forcing while local forcings include rapidly varying terrain, particularly that of the Andes, which is not well resolved by current general circulation models (GCMs) and land surface characteristics. To study both large scale and local influences on warm season rainfall in these regions a nested climate modeling system is proposed. A global model (CSM/CCM3) will be used to simulate SST forced large scale circulation for anomalous warm seasons. An increase in regional resolution, which better captures the effects of complex terrain, regional circulations, and landuse, will be obtained by a high resolution limited area model (RegCM2). Both models incorporate state of the art land surface parameterizations.

PROJECT GOALS:
Our objectives are to: 1. Develop a nested modeling system suitable for seasonal integrations in the tropical Americas. 2. Study and quantify the mechanisms which cause inter-annual variability in rainfall in Amazonia and Nordeste. 3. Evaluate the ability of this modeling system to predict seasonal rainfall in Amazonia and Nordeste.

METHODOLOGY:
In order to fulfill these objectives we have outlined six Tasks. 1. Regional model development and adaptation to the Brazilian region, 2. Global model performance in simulating inter-annual variability with prescribed observed SST forcing, 3. Regional model performance in simulating inter-annual variability with prescribed observed atmospheric and SST boundary forcing, 4. Sensitivity of regional simulations to soil moisture initialization and landuse changes, 5. Nested model performance in simulating inter-annual variability with predicted atmospheric (CCM3) and prescribed SST boundary forcing, 6. Preliminary study using the fully coupled CSM with predicted SSTs to perform an experimental forecast with the nested modeling system. Eight sets of experiments are defined for the implementation of these components.

RESULTS AND ACCOMPLISHMENTS:
In the first phase of the development of a nested modeling system based on the NCAR global (CSM) and regional (RegCM) climate models, each model is evaluated independently for its seasonal predictive skill in the PACS region. This involves Tasks 1 and 2 above and experiments EXPAR and EXDOM which are described below.

From Task 1, the RegCM configuration and domain were defined.

In Task 2, an ensemble of 5-45 year integrations of CCM3, forced by NCEP global SSTs (1950-1994) has been employed to evaluate the model annual cycle in the PACS region and seasonal anomalies for the 1982/83 warm event and the 1986/87 cold event. These `global hindcast' simulations have been evaluated using NCEP reanalyses and Xie-Arkin rainfall observations in the region of tropical Americas. The upper level anti-cyclone (Bolivian high) is well simulated in January (in the model mean annual cycle), and in July, westerly flow south of the equator with easterlies at 10 N compare well with reanalyses. Not surprisingly, the largest circulation errors occur in the low level (950 mb) winds over the eastern Pacific where the cross equatorial flow results from radiative convective-oceanic mechanisms which are not well understood. The model captures the seasonal shifts in the ITCZ rainfall in both the Atlantic and Pacific, including the split ITCZ in April in the Pacific. CCM3 circulation anomalies for the warm event of 82/83 compare quite well with those seen in the reanalyses in both upper level (200 mb) and low level (950 mb) winds and precipitation. These results were anticipated as most GCMs capture the ``local'' responses to SST anomalies seen in the tropics. The anomalies during the cold event in 86/87 are also simulated, though less coherently due to the weaker SST forcing during this period. We are encouraged that our nested model experiments will provide positive results.

From Task 2 two cases are selected, a warm event (1982/83) and a cold event (1986/87), for seasonal integrations with RegCM. If time is permitted we plan to perform tests for the 1997/98 warm event as well.

In Task 3, the first RegCM seasonal integrations (EXREG) or `regional hindcasts' are being performed using NCEP global analyses to drive the domain lateral boundaries. The aim of this study is to evaluate the capability of the RegCM in simulating interannual variability over the region given ``perfect'' forcing conditions. The analysis focuses initially on seasonal and monthly precipitation (Figures 1 and 2 show January-April precipitation differences 1988-1983 from Xie-Arkin merged precipitation and the RegCM simulations, respectively) time series analysis of daily rainfall events and related atmospheric and surface fields. These simulations will be later used to compare against the nested CCM3/RegCM2 integrations.

Task 4 considers boundary conditions at the earth's surface which have been shown to be important in determining interannual variability in monthly and seasonal climate, particularly in the tropics. Analysis of simulations for the United States indicates that warm season precipitation is sensitive to landuse specification and soil moisture initial conditions. Because soil moisture is not a standard measurement nor is it readily derivable from satellite data, few compilations are available on the continental scale. Soil moisture initialization in the U.S. has been based on educated guesses and modeling experience. Due to the uncertainty in the specification of these surface characteristics it is important to assess the model sensitivity to both landuse specification and soil moisture initialization.For this purpose, a set of simulations (EXLAN, EXMOI) are designed to test the sensitivity of seasonal rainfall and climate to changes in land use and soil moisture conditions. These experiments also employ NCEP reanalyses to drive the RegCM boundaries for the two anomalous periods defined above. We will use comparisons of the EXLAN/EXMOI experiments with the EXREG experiments to assess the importance of local versus remote controls over the simulated precipitation patterns.

FUTURE WORK:
The previous experiments were designed to evaluate the regional and global models separately in their ability to simulate interannual variability. The experiments defined in the third phase of this work (Tasks 5 and 6) will employ the nested modeling system to explore the predictability of seasonal and interannual rainfall. An ensemble of 4-6 CCM3 realisations using observed SSTs (EXGCM) will be employed as lateral boundary forcings for the selected RegCM2 domain. These ensembles of regional simulations will be performed for the two anomalous periods (EXNES) selected in 3.3. The `nested model hindcasts' will be evaluated against observations to assess the skill of the model prediction system. The `predicted' rainfall will be verified against observed seasonal total rainfall and spatial distributions. The nested results will also be evaluated against the global model results (EXGCM) in order to determine the value-added of the higher resolution nested modeling system over the region. In addition to the seasonal total precipitation, the frequency of occurrence of daily rainfall events will be analysed and compared between the nested and global models. Finally, the EXNES experiments will be compared with the EXREG experiments to determine the importance of the large scale forcing fields.

Based on the success of these hindcasts, the next step will be to nest the RegCM2 in the NCAR CSM wherein the SSTs are predicted. Three possible methods will be considered for this experiment (EXSST). The most simple method would be to employ observed SSTs forcing at the start of the prediction experiment and continue to use those SSTs assuming ``persistence'' similar to that of Graham et al. (1994). A second method is currently being used for interannual studies with the coupled ocean-atmosphere version of the NCAR CSM. A technique is in development where the atmosphere is forced for a period preceeding the prediction with monthly observed SSTs. The ocean model is also integrated for a preceeding period with observed monthly winds and fluxes. At the time the model components are coupled initial conditions for the prediction are thus tied to observations. A coupled integration is then performed to ``predict'' the succeeding seasonal climate. It is expected that this method will be much refined by the third year of this proposed effort (Joseph Tribbia, personal communication). If this method is employed,ensemble realisations from this technique will be employed to force the RegCM2. The third possibility will be to incorporate this same method, but nesting the RegCM2 in a `tropical strip' version of the coupled CSM which has active tropical Pacific and Atlantic basins. This model is in development and would be less computationally intensive than the global fully coupled CSM for the prediction experiment.

Anji Seth
seth@iri.ldeo.columbia.edu
phone: (914) 860-4419
fax: (914) 860-4864

International Research Institute for climate prediction
Lamont-Doherty Earth Observatory
Columbia University
Palisades, NY 10964

http://iri.ldeo.columbia.edu

Return to CLIVAR PACS Home Page