A Statistical-Empirical Forecast of October-December 2002 Precipitation in Uruguay-Rio Grande do Sul (Brazil) Based on the ENSO State

Rainfall in the part of South Eastern South America (SESA) which comprises Uruguay-Rio Grande do Sul, UY-RS, (approx. 25º-35ºS, 49/54 º-58ºW, see Fig. 1) occurs during the 12 months of the year. Midlatitude fronts reach this subtropical region especially during winter and also fall and spring. Convective rainfalls affect the region during summer but also in spring and fall, especially in the North and Northwest (continentally influenced) subregions. The complex interaction of several rainfall-generating mechanisms produces a wet climate with a “flat” annual cycle of precipitation. This is the case particularly in the southern part of Uruguay (UY) and SE Rio Grande do Sul (RS), influenced by the Rio de la Plata and the Atlantic Ocean. In the northern and northwestern subregions, fall and spring rainfall exceeds that of winter and summer (Prohaska, 1976; Ratisbona, 1976; Terra and Pisciottano, 1994; Pisciottano et al. 1994; Diaz et al., 1998).

In spite of this complex climatic situation, several studies have confirmed a significant influence of ENSO on UYRS ‘s interannual rainfall variability, with a tendency for above normal rainfall during warm Pacific events, especially in November and December when the event is near its mature phase (Ropelewski and Halpert 1987; Aceituno et al. 1988; Pisciottano et al. 1994; Diaz et al. 1998; Montecinos et al. 2000; Grimm et al. 2000; Barros et al 2002). There are spatial gradual variations both, in the annual cycle and, also, in the relationships of the subregional rainfall with the ENSO phenomenon, which include the “timing”, “intensity” and “persistence” (measured e.g. through statistical significance), of the ENSO “response”. Mechanisms explaining this association have to do with changes in the subtropical southern westerlies by remote forcing of the large scale upper SH circulation and local interaction with the regional-scale mechanisms. The location (e.g. N-S position, exit regions), structure, intensity and other features of the anomalous jet stream determine both the baroclinic developments and the probability of convective events in SESA. Cazes et al. (1996, 2000) showed that changes in circulation and precipitation patterns affecting Uruguay, during November of warm Pacific events, can be simulated through an AGCM and well compared with adequate composites of past events.

Diagnostic studies have also revealed that:

-there is a significant tendency for less than normal precipitation in Uruguay and partially in RS during La Niña episodes, especially from October to December, which is reanalyzed here, together with the former one, via statistically significance test of correlation fields; a relationship between Pacific SST (indices or fields) and precipitation in UYRS that shows some real skill on seasonal (bimonthly to quarterly) time scales (Diaz and Studzinski 1994; Cazes et al. 1994); methods based on canonical correlation analysis (CCA) and stratification techniques show the greatest skill for the seasons around November, for the UYRS region or subtropical South America precipitation field (Diaz and Studzinski 1994; Montecinos et al. 2000); statistically, the best relationships are the simultaneous ones; by using forecasted SSTs we could use these to predict Oct-Dec/04 regional precipitation; zero or one month lead/lag between the SST “predictor” and the seasonal precipitation yield similar “cross-validated” relationship strengths (Diaz and Studzinski 1994; Cazes et al. 1994); thus for short lead times (up to 2-months) we do not need to use forecasted SST values, probably due to the slow timescales of the ocean rather than the atmospheric response.

Based on those results we present here a seasonal forecast for subregional precipitation for a region comprising Uruguay and Rio Grande do Sul (UYRS) by using similar statistical techniques of our former predictions (see Pisciottano and Mendina, 2002, Pisciottano et al., 1999 and previous ones) which has been evaluated (see Pisciottano et al. 2000, 2001; Mendina et al. 2001).

For diagnostic studies and for results presented here, we use the 1950-1998 time series of monthly values of the Nino 3.4 Index of SST anomalies (ºC) in the central tropical Pacific (N3.4), available from NOAA. Our precipitation data are monthly values (mm), from 1950 to 1998, from 17 rainfall stations in Rio Grande do Sul - Brazil (RS) and 21 in Uruguay (Fig. 1) The data, for RS, come from the meteorological network of FEPAGRO (RS-Br) and the 8^th DISMET/ Instituto Nacional de Meteorologia (IMMET/Br), and, for Uruguay, from Direccion Nacional de Meteorologia del Uruguay (DNM/Uy). We could have used longer period records but we prefer to use only the most recent data after 1950 (49 years) in order to avoid interdecadal climatic variations from the first decades of the XX century to the latest ones. Then we focus on the interannual variations of the precipitation based on the 1950-1998 climatology (see Fig. 1).

For diagnostic and prediction purposes we have grouped the stations in four regions in UYRS (see Fig. 1) based on the annual cycle homogeneity (see Terra and Pisciottano, 1994; Diaz et al., 1998), the spring interannual variability (Berger et al. 1996) and ENSO-relationship of the rainfall anomalies, as described in many of the above mentioned studies and the field of correlation coefficients between July-August-N3.4 and October-December local precipitation shown in Fig. 2. Note that for 49 data (years) correlation coefficients values above 0.294 are 95% significant, and values above 0.402 are 99% (or “very”) significant. Fig. 2 shows lagged correlation coefficient field (up to approx. 0.52) because we will use July-August average N3.4 as “predictor” for Oct-Nov-Dec regional precipitation to be predicted; however simultaneous correlation reach higher values, up to 0.6 or higher (not shown). For each region, we calculated a monthly regional precipitation value for each month in 1950-98. Time series for October-December are formed from the monthly values. In summary, the northernmost region (number 4 in Fig. 1 and Tables), in north RS, is the rainiest in UYRS (more than 400 mm in the quarter), but it is, also, the one for which the ENSO-rainfall relationship are the weaker one. The southern most region (number 1), in south and southeast Uruguay has the climatological lowest value (260 – 270 mm) and also weak ENSO-rainfall relationship. The “central” regions (number 2 and 3) have a climatological intermediate value but the strongest ENSO response on rainfall. Table 1 shows correlation coefficients between Oct-Dec regional rainfall and July-August N3.4.

We defined fixed equiprobable categories (quartiles) of the predicted variable (Oct-Dec regional precipitation). Changes in the probabilities given a specific climatic event (e.g. an ENSO event) are calculated for each category and used to form a probabilistic prediction. First, the statistical distribution parameters (first quartile, median, and third quartile; called qcl, mcl and Qcl respectively) for the whole record are determined for October-December. These statistics characterize the rainfall climatology (cl) of the respective region. In the absence of any climate-determining information, a value “around mcl” is expected and the interval (qcl, Qcl) has a large (50%) chance of containing the real observed value. In this case the “forecast” would be mcl and an “error bar” would be (qcl, Qcl).

To incorporate the knowledge provided by the ENSO-rainfall relationships, we compare the statistical parameters of the “whole population” (all the values of the October-December precipitation) with those of the October-December values for years (“subpopulation”) whose July-August N3.4 mean was similar to that value for the current year, in the sense that the value of this index (“predictor”, July-August N3.4) hit a window of values around the value for the current year (defined below). In agreement with previous studies, we find that in the so defined “subpopulation” there is a (weak) tendency for higher rainfall. We test this statistical tendency with a hypergeometric distribution using the frequencies of above and below median rainfall in the population versus the frequencies for the “subpopulation”. We note a direct relationship between the “width” of the window, the size of the “subpopulation” and the statistical significance of the rainfall differences (in the sense of Ropelewski and Halpert 1987; Pisciottano et al. 1994; Cazes et al. 1994). A small window produces a small number of “events” similar to the current one and there are problems to test any possible shift. Previous evaluation of an ensemble of regional rainfall predictions done for regions in Uruguay by using the same methodology, show that significance values around 97% are crucial in order to assure skill (see Pisciottano et al. 2001). Current values of N3.4 (Jul-Aug/2004) is approximately 0.7 ºC (July, 0.6ºC; August 0.8 ºC). After some analysis, we choose a window around this value defined as: 0.34 ºC < Jul-Aug-N3.4 < 1.24 ºC , which define a “subpopulation of 12 cases, 6 at each side of the current year value. Small changes in the selection of this window yield some changes of the results of the test, but in any case, regions 2 and 3 show a weak tendency for precipitation above normal, which reach enough significance in the statistical test but regions 1 and 4 show a not significant weak tendency. It could be that, by using longer dataset, would document the tendency for above normal rainfall even in the number 4 region, but long period trends or decadal variations can be involved. Using that window we estimate the statistical parameters q*, m*, and Q* for the “subpopulation” ( 0.34 ºC < N3.4 (Jul-Aug) < 1.24 ºC ) for October-December for the 4 regions. The procedure is similar to that used by Ropelewski and Halpert (1996) and previous predictions for Uruguay and UYRS. The statistical parameters (qcl, mcl Qcl) and (q*, m*, Q*) for the four regions are shown in Fig. 3. Table 2 illustrates the rainfall distribution shift associated to years having N3.4 (Jul-Aug) higher than 0.34 ºC and smaller than 1.24 ºC, through frequencies of occurrence of each of the four (climatological) quartiles for each of the four regions.

These results enable us to “predict” the expected regional precipitation as the median of the “subpopulation” (m*) rather than the climatological median (mcl). The interval (q*, Q*) from the subpopulation is the “error bar” of the prediction. This technique circumvents problems associated with the nonlinearities of SST – rainfall relationships and skewness of the rainfall distributions, and has been used during the past several years to predict rainfall and streamflow in Uruguay. Those forecast were used by rice growers and hydroelectric producers for water resource management.

Based on the currents values of the Pacific SST anomalies and the distributional shifts discussed above, we issue the following rainfall predictions for the period October-December 2004 for the 4 regions covering Uruguay-Rio Grande do Sul (UYRS, see Fig. 1):

Region 1- Southern and Southeastern Uruguay. A value around 310 mm of rainfall is expected with 50% chance of between 262 and 411 mm. The probabilities of rainfall in the 3^rd and 4^th quartile of the climatological distribution are 25% and 50%, respectively; and of rainfall above the median (267 mm) is 75%.

Region 2- Northeastern Uruguay and Southeastern Rio Grande do Sul. A value around 403 mm of rainfall is expected with 50% chance of between 344 and 461 mm. The probabilities of rainfall in the 3^rd and 4^th quartile of the climatological distribution are 25% and 58%, respectively; and of rainfall above the median (311 mm) is 83%.

Region 3- Central and Northwestern Uruguay and Southwestern (inland) Rio Grande do Sul. A value around 428 mm of rainfall is expected with 50% chance of between 368 and 504 mm. The probabilities of rainfall in the 3^rd and 4^th quartile of the climatological distribution are 50% and 42%, respectively; and of rainfall above the median (354 mm) is 92%.

Region 4- Northern Rio Grande do Sul. A value around 426 mm of rainfall is expected with 50% chance of between 384 and 570 mm (climatology), with a weak (not significant) tendency to rainfall above the median.

In summary, a wetter than normal October-December/2004 period is expected in Uruguay-Rio Grande do Sul, particularly around the regions close to the Uruguay-Brazil boundary. This tendency to a “wet” period is weak in Southern Uruguay and it is not distinguishable from climatology in Northern Rio Grande do Sul.

Acknowledgments: To Prof. Moacir A. Berlato fron UFRGS-Br, for the dataset from RS stations and DNM-Uy for the UY dataset.

References:

Aceituno, P, 1988: On the functioning of the Southern Oscillation in the South American sector. Part I: Surface Climate. Mon. Wea. Rev, 116, 505-524.

Barros, V. y G. E. Silvestri, 2002: The Relation between Sea Surface Temperature at the Subtropical South-Central Pacific and Precipitation in Southeastern South America. J. Climate, 15, 251-267.

Berger, L., G. Pisciottano y R. Terra, 1996: Variabilidad espacio-Temporal de la Precipitación en la Región Sudeste Sur América (SSA). Climet VII, Congremet VII, Buenos Aires-Argentina, pp 337-338, Setiembre 1996. Versión completa disponible en IMFIA-FI-UR, Montevideo-Uruguay.

Cazes, G., J. L. Genta, G. J. Pisciottano, 1994: Generación de información hidrológicamente relevante a partir de información y diagnóstico climático. Aplicación en Uruguay Mem. XVI Congr. Latino-Americano de Hidraulica. Santiago, Chile; 7-11 Nov. 1994. Pp 121-132.

Diaz, A.F. and C. D. Studzinki, 1994: Rainfall forecast in Uruguay and Southern Brazil using cononical correlation analysis. VIII Congresso Brasileiro de Meteorología / II Congreso Latino-Americano e Ibérico de Meteorlogía.

Diaz, A.F., C. D. Studzinki, et C. R. Mechoso, 1998: Relationships between precipitation anomalies in Uruguay and Southern Brasil and sea surface temperature in the Pacific and Atlantic Oceans. J. Climate, 11, 159-171.

Grimm, A. M., V. Barros y M. E. Doyle, 2000: Climate Variability in Southern South America Associated with El Niño and La Niña Events. J. Climate, 5, 1532-1539.

Mendina M., G. Pisciottano, A. Diaz, J. L. Genta y G. Cazes, 2001: Evaluacion detallada de pronosticos climaticos (IMFIA) regionales de precipitacion en Uruguay. Relación significancia/grado de “acierto” y variaciones espaciales. Resumenes de las IX Jornadas de Jovenes Investigadores de la A.U.G.M, Rosario-Santa Fe-Argentina, Setiembre 2001. Editorial de la Universidad Nacional de Rosario, 374 pp. Version completa disponible en CD.

Montecinos, A., A. Diaz and P. Aceituno, 2000: Seasonal Diagnostic and predictability of Rainfall in Subtropical south America Based in Tropical Pacific SST. J. Climate., 13, 746-758.

Pisciottano,G., A. Diaz, G. Cazes and C. R. Mechoso, 1994: El Niño-Southern Oscillation impact ranifall in Uruguay. J. Climate., 7, 1286-1302.

Pisciottano,G., G. Cazes, A. Diaz and J.L. Genta, 2000: A revision and evaluation of the IMFIA-UR Seasonal Rainfall Forecast Method based on the ENSO State. 9 A.2, 6th Int. CHSM&O, Santiago, Chile, Abril 2000; AMS. Pp 211-212.

Pisciottano, G., A. Diaz, M. Mendina, J. L. Genta y G. Cazes, 2001: On the relationship between skill (expost) and statistical significance (exante) for an ensemble of regional-seasonal rainfall forecasts issued by IMFIA-UR (Uruguay). Apliccatons of Climate Forecasting for Better Decision-making Processes in Agriculture. XII Foro Regional de Perspectivas Climáticas para el Sudeste de América del Sur, Passo Fundo-RS, Brasil, Abril 2001. Ed., G: R. Cunha, J. C. Haaz and M. A. Berlato. Pp 327.

Pisciottano, G. and M. Mendina, 2002: A Statistical-Empirical Forecast of October-December 2002 Precipitation in Uruguay-Rio Grande do Sul (Brazil) Based on the ENSO State. ELLFB-COLA. V 11, N 3. Sep. 2002.

Ropelewski, C. F. and M. S. Halpert, 1987: Global and regional scale precipitation patterns associated with the El Niño-Southern Oscillation. Mon. Wea. Rev., 115,1606-1626.

Ropelewski, C. F. and M. S. Halpert,1996: Quantifying Southern Oscillation-Precipitation Relationships. J. Climate., 9, 1043-1059.

Terra, R. and G. Pisciottano, 1994: Regionalizacion del Uruguay segun el ciclo anual de precipitaciones mediante “cluster analisis”. Mem. XVI Congr. Latino-Americano de Hidráulica. Santiago, Chile; 7-11 Nov. 1994. Pp 227-236.

Figures captions:

Table 1: Correlation coefficients for each of the four regions in UYRS (as shown in Fig. 1) between the July-August N3.4 and October-December regional precipitation time series.

Table 2: Frequencies of occurrence of (Oct-Dec) rainfalls in each one of the quartiles of the climatological distribution, given a “weak-warm” year (0.34 ºC < (Jul-Aug N3.4) < 1. 24 º C), for each of the four regions in UYRS.

Figure 1.: The 17 rainfall stations in RS and the 21 in UY (indicated by *) used in this study, and the four regions in UYRS. Also isolines of the climatological (1950-98) October-December total precipitation

(in mm) are shown from 250 to 500 mm in the quarter.

Figure 2.: Correlation coefficients field in UYRS between the July-August N3.4 and October-December local precipitation time series, based in rainfall of each station shown in Fig. 1.

Figure 3.: a) Medians and quartiles of the distribution of precipitation for the period October-December, in the region 1-S and SE UY, for all the cases (Climatology, left hand bar) and for the subpopulation of cases when the Nino 3.4 index averaged in the previous July-August period was above 0.34 ºC and below 1.24 ºC (which is the “season 2004 forecast”, right hand bar). The data record spans from 1950 to 1998. b) Same as a), for the region 2-NE UY and SE RS; c) Same as a), for the region 3-Central and NW UY and SW inland RS; d) Same as a), for the region 4-N RS (note that for this region number 4, there are not statistically significant differences between the two bars and the forecast is climatology).

*Region*	1	2	3	4
*Corrcoef*	0.45	0.62	0.59	0.44

Table 1: Correlation coefficients for each of the four regions in UYRS (as shown in Fig. 1) between the July-August N3.4 and October-December regional precipitation time series.

		*REGION*
*Quartile*	1	2	3	4
N1^th (%)	2 (17 %)	0 (0 %)	0 (0 %)	2 (17 %)
N2^th (%)	1 (8 %)	2 (17 %)	1 (8 %)	1 (8 %)
N3^th (%)	3 (25 %)	3 (25 %)	6 (50 %)	4 (33 %)
N4^th (%)	6 (50 %)	7 (58 %)	5 (42 %)	5 (42 %)
N	12	12	12	12

contributed by Gabriel Pisciottano, Mariana Mendina and Alvaro Diaz