A LOCAL PRECIPITATION DOWNSCALING APPROACH Nora E. Ruiz* and Walter M. Vargas Departamento de Ciencias de la Atmósfera y los Océanos Universidad de Buenos Aires, Buenos Aires, Argentina Abstract synoptic-scale atmospheric circulation conditions and local-scale surface environment In the context of climate change it is (Barry and Perry 1973). The methods used in useful to downscale low-resolution large-scale synoptic climatology usually employ one of two features of models towards the high resolution different approaches: the circulation-to- small-scale features of environment and environment approach, which classifies and ecosystems which are modulated and driven clusters circulation data in some way and by weather and climate. In this work, we afterwards looks for links with local-scale address the problem of deriving local weather environment, and environment-to-circulation in terms of precipitation and the sensitivity to approaches, which structure circulation data be well-described in different areas of based on a criterion defined by the local-surface Argentina, through a fairly simple transfer variable (Yarnal 1993). When applying the function. circulation-to-environment techniques, the map- The performance of the dynamical- patterns are obtained independently of the statistical approach derived, that is based on a variable to be connected. Then, predicted local synthesis of significant anomaly circulation variable conditions associated with each map features linked to precipitation occurrence, is pattern must be estimated subsequently. Model assessed by means of examining frequency verification statistics between observed and distributions of results with two purposes: expected values of the environmental conditions validate the diagnostic study, and give a can be utilized to calculate the performance of representation of model uncertainty in the circulation-to-environment approach and to geographic terms through examining the evaluate the goodness of the relationship. When spatial variability as well as the interseasonal using the environment-to-circulation approach, variability. From these results, western and this predictive way cannot be employed, since southern areas are better predictable than the the obtained map-patterns are conditioned by humid pampas and the Chaco region. the variable of interest (Cannon et al 2002). By Therefore, the latter may be considered as this reason, though circulation-to-environment more vulnerable from the point of view of a and environment-to-circulation approaches are downscaling approach or forecast. In seasonal most commonly used, methods that jointly terms, wintertime and summertime have a consider circulation and environmental variables very similar level of predictability. have been developed. The study carried out in this work follows this line of investigation. The downscaling transfer functions Pioneer work by Klein (1965) examined developed in this study may be also specification models to search for relationships considered able enough to carry out future between synoptic-scale atmospheric flow and studies of precipitation vulnerability in view of variables related to the surface environment. potential scenarios coming from hypothetic More recently, Hewitson and Crane (1996), von climatic changes. Storch et al. (1993), Ruiz (2002) and Ruiz and Vargas (1998) applied other forms of empirical 1. INTRODUCTION downscaling with the same purpose. With specification or empirical downscaling As it is known, synoptic climatology techniques regression equations are used to investigates relationships between large- relate gridded circulation data to values of an environmental variable (Klein 1985, von Storch et al 1993). Even though the selection of grid- ∗ Corresponding author address: Nora E. Ruiz, Dpto. points and parameters in the specification Ciencias de la Atmósfera y los Océanos, FCEyN, Univ. de equations are not independent of the surface Buenos Aires, Ciudad Universitaria, Pab. II, 1428 Buenos Aires, Argentina; e-mail: [email protected] variable, i. e. it could be considered as an environment-to-circulation approach, the 599 resulting downscaled equation can be used to Rivadavia (45º47'S, 67º30'W). The predict values of the environmental variable only precipitation is accumulated from 1200 UTC of from values of the circulation ones; so, day i to 1200 UTC of day i+1. The data cover predictive properties are present in this method. the period 1983-1987. Models show difficulties in describing The meteorological stations selected properly the variability at scales of high represent different regions of Argentina: Salta, frequency precipitation. Moreover, precipitation the north-western region; Resistencia, the exhibits large spatial variability as well as non- north-eastern subtropical region; Córdoba, the normality. Additionally, it exhibits large spatial central region; Paraná, Litoral region; Santa variability as well as non-normality. In addition, Rosa, the semiarid pampas; Azul and Buenos the coarser the resolution, as in a general Aires, the humid pampas; Bahía Blanca, the circulation model, the higher loss of accuracy. coastal humid pampas; Neuquén, the northern Statistical techniques developing a concurrent Patagonia; Bariloche, the mountain Patagonia relationship between the upper-air circulation and Com. Rivadavia, the coastal Patagonia. and the surface weather variables may help in In this study the information is separated diagnosis/forecast problems (Klein 1965). The into two samples: the austral cold semester, objective of this work is to downscale results May to October (753 days), and the austral from a synoptic climatological study of warm semester, November to April (603 days), precipitation in some regions of Argentina based to take into account the annual cycle. There on 500-hPa geopotential vorticity fields (Ruiz are five winters and four summers in the whole 2004, Ruiz and Vargas 1998) by applying data. The length of this sample satisfies the empirical downscaling and statistical prediction conditions proposed by Carter (1986), who techniques to vorticity pattern predictors shows that three seasons with six months of exclusively, to obtain daily probabilities of data each one are needed at least to be able occurrence of precipitation. It is of main interest to derive stable regression equations to in this work to focus on a sole variable that evaluate precipitation probabilities. concentrates and captures the essential The variable to probabilistically downscale characteristics and physics of the circulation is the occurrence of daily precipitation at a given and, at the same time, involves direct location and it is considered categorical or non- implications on precipitation. Likewise, it is numerical. The occurrence and non-occurrence important to count on a variable with adequate of precipitation is defined as a binary variable. reliability level when dealing with predictions in Its value is 1 if the measurable precipitation the southern hemisphere (Hoskins et al 1989). concurrent with the day of analysis is ≥0.1 mm Another purpose of the downscaled variable and 0 otherwise. The empirical understanding climatology is to provide a minimum threshold of of the physical and mathematical relationships precipitation predictability for different regions of between predictors and predictand may Argentina as well as background for the optimize the results. This knowledge has to be evaluation and skill of numerical model applied during the process of selection of downscaling. predictors. For this reason, a synoptic climatology study of different tropospheric 2. DATA AND METHODOLOGY fields and their temporal-spatial patterns in relation with precipitation has been performed The data used consist of daily 500-hPa in previous works (Ruiz 2004, Ruiz et al 1999, relative vorticity fields at 1200 UTC calculated Ruiz and Vargas 1998) where the analysis of through 500-hPa geopotential height fields some linkages between atmospheric (Ruiz and Vargas 1998) elaborated at the circulation as model outputs and local/regional National Meteorological Service of Argentina precipitation time series was presented. It is a for the study region described in Fig. 1. This purpose of the present study to estimate the dataset is compatible with the available capacity of the map-patterns obtained in Ruiz precipitation dataset. Daily precipitation data at (2004) to distinguish occurrence of the following meteorological stations of precipitation. These regional classifications Argentina are used: Buenos Aires (34º35'S, are of fundamental, exclusive use in the 58º29'W), Salta (24º51'S, 65º29'W), original area of application. In a way, they Resistencia (27º27'S, 59º03'W), Córdoba characterize the region. 500-hPa fields and (31º24'S, 64º11'W), Paraná (31º47'S, particularly the derived variable: relative 60º29'W), Santa Rosa (36º34'S, 64º16'W), geostrophic vorticity, as one of the parameters Azul (36º45'S, 59º50'W), Neuquén (38º57'S, better linked to vertical motions and with better 68º08'W), Bahía Blanca (38º44'S, 62º10'W), signal on precipitation as compared with other Bariloche (41º09'S, 71º10'W) and Comodoro dynamical variables (Ruiz et al 1999), are used. It is emphasized that the focus is on vorticity in 600 itself (though many other variables are based on two statistical techniques: a) Multiple available) as a baroclinic eddy measure and Discriminant Analysis (MDA) (Miller 1962), and fundamental generator of precipitation b) Regression Estimation of Event processes. As in Wallace