Self organizing maps (SOMs) are used to locate archetypal points that describe the multi-dimensional distribution function of a gridded sea level pressure data set for the northeast United States. These points -nodes on the SOM -identify the primary features of the synoptic-scale circulation over the region. In effect, the nodes represent a non-linear distribution of overlapping, non-discreet, circulation types. The circulation patterns are readily visualized in a 2-dimensional array (the SOM) that places similar types adjacent to one another and very different types far apart in the SOM space. The SOM is used to describe synoptic circulation changes over time, and to relate the circulation to January station precipitation data (for State College, Pennsylvania) in the center of the domain. The paper focuses on the methodology; however, the analysis suggests that circulation systems that promote precipitation have decreased over the last 40 yr -although January precipitation at State College has actually increased. Further analysis with the SOM indicates that this is due to a change in precipitation characteristics of the synoptic-scale circulation features, rather than to their frequency of occurrence.KEY WORDS: Self-organizing maps · Synoptic climatology · Downscaling · Climate change · Synoptic classification Resale or republication not permitted without written consent of the publisherClim Res 22: [13][14][15][16][17][18][19][20][21][22][23][24][25][26] 2002 some form of correlation, cluster, and/or eigenfunction analysis. In all cases, the approach of generalizing the circulation into characteristic modes or synoptic types required a fine balance between producing a small enough number of types to easily visualize and conceptualize the circulation, while avoiding so much generalization that the strength of any relationship to a local climate variable was lost. The major problems with this approach are due to the degree of within group variability produced. It is also common that days in the same synoptic type can often be associated with a very different local response, or that the same response can be obtained from different synoptic types. The fundamental characteristics of synoptic classification techniques are effectively summarized in Yarnal (1993).Underlying this traditional approach to synoptic classification is the premise that the continuum of weather states may be effectively divided into a small number of categories with clear discernable boundaries. This premise clearly has limitations: while typing weather systems gives a good first-order insight into the basic characteristics of the climate system, much of the information is inherently subsumed by the degree of generalization imposed.In response to this, Hewitson & Crane (1992a,b) proposed that the system can be treated as a continuum with a continuous function, and that quantitative relationships between the atmosphere and local surface variables can be developed in the form of a downscaling transfer function. This procedure is d...

Abstract. A range of different statistical downscaling models was calibrated using both observed and general circulation model (GCM) generated daily precipitation time series and intercompared. The GCM used was the U.K. Meteorological Office, Hadley Centre's coupled ocean/atmosphere model (HadCM2) forced by combined CO2 and sulfate aerosol changes. Climate model results for 1980-1999 (present) and 2080-2099 (future) were used, for six regions across the United States. The downscaling methods compared were different weather generator techniques (the standard "WGEN" method, and a method based on spell-length durations), two different methods using grid point vorticity data as an atmospheric predictor variable (B-Circ and C-Circ), and two variations of an artificial neural network (ANN) transfer function technique using circulation data and circulation plus temperature data as predictor variables. Comparisons of results were facilitated by using standard sets of observed and GCM-derived predictor variables and by using a standard suite of diagnostic statistics. Significant differences in the level of skill were found among the downscaling methods. The weather generation techniques, which are able to fit a number of daily precipitation statistics exactly, yielded the smallest differences between observed and simulated daily precipitation. The ANN methods performed poorly because of a failure to simulate wet-day occurrence statistics adequately. Changes in precipitation between the present and future scenarios produced by the statistical downscaling methods were generally smaller than those produced directly by the GCM. Changes in daily precipitation produced by the GCM between 1980-1999 and 2080-2099 were therefore judged not to be due primarily to changes in atmospheric circulation. In the light of these results and detailed model comparisons, suggestions for future research and model refinements are presented. IntroductionThe present generation of global general circulation models (GCMs) and higher-resolution limited area models (LAMs) of the climate system are restricted in their usefulness for many subgrid scale applications (including those to hydrology) by their coarse spatial resolution and the uncertain reliability of their output on timescales of months or less, especially for variables pertaining directly to the hydrologic cycle [Carter et al., 1994]. As Hostetler [1994] has observed, the parameterizations used in GCMs and in hydrological models are least reliable on the scale(s) at which these models interface. Hydrological models are frequently concerned with small, subcatchment scale processes and must parameterize regionalscale ones, whereas atmospheric models deal most proficiently with fluid dynamics at the planetary scale and parameterize many regional and smaller-scale processes.Climate model resolution issues have important implications

Atmospheric moisture transport over southern Africa and surrounding oceans is considered during wet and dry conditions over the South African summer rainfall region. Wet and dry synoptic spells within wet and dry austral summers are examined. A link between synoptic and seasonal timescales is investigated using seasonal statistics of wet and dry spells. Dry synoptic spells exhibit divergent moisture flux over South Africa, with inflow from the mid-latitude ocean regions to the south. Cyclonic features off the east coast may exist and attract moisture away from South Africa. For wet synoptic spells, there tends to be increased moisture flux from the tropical or subtropical southwest Indian Ocean (SWIO), either associated with ridging along the east coast or a deep low over the interior. Seasonal modulations of the intensity of the heat low over Angola/Namibia appear important for influencing early (OND) and late (JFM) summer rainfall over South Africa. This low may act as the tropical source for tropical temperate troughs and their associated cloudbands that are major synoptic rain-producing systems. Wet (dry) summers are often associated with a southward (northward) shift and strengthening (weakening) of the ITCZ over tropical southeastern Africa. Seasonal rainfall is found to be related to the distribution of wet and dry spells within the season, such that wetter seasons tend to have longer or more intense wet spells rather than a greater number of wet spells.

The current state of regional climate and climate change modelling using GCMs is reviewed for southern Africa, and several approaches to regional climate change prediction which have been applied to southern Africa are assessed. Confidence in projected regional changes is based on the ability of a range of models to simulate present regional climate, and is greatest where intermodel consensus in terms of the nature of projected changes is highest. Both equilibrium and transient climate change projections for southern Africa are considered. Warming projected over southern Africa is within the range of globally averaged estimates. Uncertainties associated with the parameterization of convection ensure that projected changes in rainfall at GCM grid scales remain unreliable. However, empirical downscaling approaches produce rainfall changes consistent with synoptic-scale circulation. Both downscaling and grid-scale approaches indicate a 10-15% decrease in summer rainfall over the central interior which may have important implications for surface hydrology. Climate change may be manifested as a change in variability, and not in mean climate. Over southern Africa, increases in the variability and intensity of daily rainfall events are indicated.

Self-organizing maps (SOMs), a particular application of artificial neural networks, are used to proportionately combine precipitation records of individual stations into a regional data set by extracting the common regional variability from the locally forced variability at each station. The methodology is applied to a 100 yr record of precipitation data for 104 stations in the MidAtlantic/Northeast United States region. The SOM combines stations with common precipitation characteristics and identifies precipitation regions that are consistent across a range of spatial scales. A variation of the SOM application identifies the temporal modes of the regional precipitation record and uses them to fill missing data in the station observations to produce a regional precipitation record. A test of the methodology with a complete data set shows that the 'missing data' routine improves the regional signal when up to 80% of the data are missing from 80% of the stations. The improvement is almost as pronounced when there is a bias in the missing data for both highprecipitation and low-precipitation events. KEY WORDS: Upscaling · Regional precipitation · Regionalization Resale or republication not permitted without written consent of the publisherClim Res 25: [95][96][97][98][99][100][101][102][103][104][105][106][107] 2003 precipitation in the SW USA / northern Mexico to examine variability in monsoon rainfall. Similarly, Waylen et al. (1996) examined the spatial and temporal response of annual precipitation patterns in Costa Rica to the Southern Oscillation. They first used cluster analysis to group stations with similar precipitation distributions into 4 major regions, and then they derived individually the response of each region's mean annual precipitation to the Southern Oscillation Index.In particular, station precipitation records tend to exhibit a large degree of spatial inhomogeneity due to small-scale events and local forcing, hence the need to identify homogeneous regions and to derive a regional precipitation signal. If the temporal characteristics of the record are also of interest, e.g. when identifying temporal patterns or trends in the data, it is also important that stations do not have a disproportionate influence on the temporal pattern either because of anomalous conditions at individual stations or because of differences in their length of record. The challenge, therefore, is to proportionately combine individual station records into a regional data set, with a methodology that removes the effects of missing data and produces a measure of the underlying regional precipitation, i.e. one that extracts the common regional variability from the locally forced variability at each station. The present paper develops such a methodology based on a particular form of artificial neural network known as the Kohonen self-organizing map (SOM), using station precipitation data for the northeast United States. SELF-ORGANIZING MAPSSelf-organizing maps, or SOMs (Kohonen 1989(Kohonen , 1990(Kohonen , 1991(...