Downscaling Large-Scale Atmospheric Information To Local Scales Using A Two Step Statistical Method.

BETTOLLI, MARIA LAURA; PENALBA, OLGA C; RIBALAYGUA, JAIME; TORRES MICHELENA, LUIS

Foz do Iguaçu, Brazil

Congreso; 2010 The Meeting of the Americas; 2010

American Geophysical Union

Regional and global atmospheric models properly describe climatic features at subcontinental scales but their use for local impact studies is limited. This is due to their low spatial resolution and incapacity to represent important characteristics at subscale resolution. An alternative to solve this problem is to apply empirical relationships between local climate and regional atmospheric systems. This procedure is known as downscaling. This work proposes a statistical downscaling methodology in order to derive local climatic estimates from large-scale atmospheric predictors in La Plata Basin (LPB) region.  This is a particularly suitable region for the application of downscaling methods, since climate variability and global changes have a strong impact on agricultural and hydropower production. The low resolution predictors used here were 1000, 850 and 500 hPa geopotential heights ERA40 Reanalyses at a 6-hourly resolution were used to derive daily mean geostrophic wind fields for each level and hourly 1000/500 hPa and 1000/850 hPa thickness. The domain chosen extends from 20.25°S to 55.125°S and 84.375°W to 42.625°W and it was defined to encompass circulation patterns that affect Southern South America. Series of daily rainfall and maximum and minimum temperatures located in the southern region of LPB were also used. Both reference (surface and atmospheric) datasets expand over the period 1960-2000. A two step statistical method has been developed and validated. In the first step, the analogue technique was used in order to select the "n" most similar days to the "X" day. The second step involves the estimation of daily rainfall and maximum and minimum temperatures. For temperature estimation multiple linear regression was applied. The potential predictors were hourly 1000/500 and 1000/850 hPa thickness above the meteorological station location (06 and 18 UTC for minimum and maximum temperatures respectively), a sinusoid function of the day of the year (seasonal factor) and a weighted average of the surface mean daily temperatures of the ten previous days of the n analogous days. For precipitation estimation two procedures were applied: average of precipitation values of a pre-selected number of analogous of X day, and rainfall distributions considering the n analogous and their weights (indirectly proportional to the distance to the X day), were built for each X day in order to improve the estimation of extreme precipitation values and the number of rainy days. The performance of the method is very good at estimating seasonal cycles and spatial and temporal variability. The spatial distribution of estimated fields is in agreement with the observed fields as well as the gradients. The method is good enough to allow its confident application over General Circulation Model outputs, in order to prospect possible climate evolution