Seasonal Summer rainfall prediction in Bermejo River Basin in Argentina

GONZÁLEZ, MARCELA HEBE; MURGIDA ANA MARIA

Climate Variability - Some Aspects, Challenges and Prospects

INTECH

Lugar: reading; Año: 2012; p. 141 - 160

Bermejo River Basin is located in the Chaco Plains in northern Argentina. The river has an extension of 1450 km and the basin area covers 16048 km2, comprising the north of Salta and the entire Formosa and Chaco provinces. Its principal tributary is San Francisco River which brings mountain waters. Two different sections can be detected in Bermejo River: the upper and the middle-low Bermejo. Vegetation is wooded with more plains to the east and with the presence of isolated yungas. The worst areas are historically inhabited by indigenous communities with extensive farming practice. Historical data show that the region has been the scene of frequent hydro-meteorological disasters (floods and droughts) and the impacts of these events have had a strong impact on the welfare of the population, productive activities and infrastructure. There is ample evidence that climate change impacts are already being observed today and that policies that seek the best ways to meet them are essential for the development and welfare of the community. The Chaco region is one of the regions that, as a result of the change in land use, presents “hotspot "or critical areas in recent times (from 1980). They are the result of the implementation of deforestation as a technology for the advancement of agriculture and intensive farming. In this region the climate is subtropical with a mean annual rainfall cycle showing a minimum in winter, which is more pronounced in the west, with dry conditions prevailing from May to September. The Andes chain lies along the west of Argentina and prevents the access of humidity from the Pacific Ocean. Therefore, the flow is governed by the South Atlantic High and as a consequence, winds prevail from the north and the east. The great interannual rainfall variability generates the requirement to understand the large circulation patterns associated with different hydric situations. Some remote sources affect the mentioned interannual variability. Subtropical South America is known to be one of the regions of the world with an important El Niño-Southern Oscillation (ENSO) signal in the precipitation field. This signal varies along each of the ENSO phases, and it differs between sub-regions. Although ENSO is the most important remote forcing without a doubt, the variability originated by other regional or remote sources cannot be disregarded. The scientific basis of the seasonal climate predictability lies in the fact that slow variations in the earth’s boundary conditions (i.e. sea surface temperature or soil wetness) can influence global atmospheric circulation and thus precipitation. As the skill of seasonal numerical prediction models is still limited, it is essential the statistical study of the probable relationships between some local or remote forcing and rainfall. In this paper an example of seasonal rainfall prediction is presented for Bermejo River basin. As the maximum rainfall season was summer, from January until March (JFM), this period was used to study interannual rainfall variability and predictors in the previous December could be defined, for each one of two sub-basin (Lower-Middle Bermejo and Upper Bermejo). Athough it is a small area, some differences were detected all over the basin. Therefore, two mean rainfall series were constructed as the average of monthly precipitation of nineteen stations in the upper Bermejo river basin (UB) during the period 1982-2007 and fourteen stations in the lower and middle Bermejo River Basin (LB) during the period 1968-2007, in order to be representative as from the precipitation over each one of the basin regions. Different period were considered in each sub-basin because of the availability of data. Simultaneous and one month lagged correlations were calculated to find the existing relation between summer (JFM) rainfall in LB and UB and sea surface temperature, 1000 Hpa, 500 Hpa and 200 Hpa geopotential height SST and 850 Hpa zonal and meridional wind. The results allowed to define some predictors in previous December, which were used to develop a statistical forecast model using the forward stepwise regression method, which retained only the variables, correlated with a 95% significance level. Forward stepwise regression is a model-building technique that finds subsets of predictor variables that most adequately predict responses on a dependent variable by linear regression, given the specified criteria for adequacy of model fit. The basic procedures involve identifying an initial model, then predictors are added one-by-one with the remaining candidate predictor that reduces the size of the errors, and this process continues until the errors cannot be significantly reduced. Linear regression models were developed for both, UB and LB. The correlation between observed and forecast rainfall time series derived from cross-validation was 0,6 and the linear regression model explained the 49% of the variance of EFM LB rainfall. However, the summer rainfall in UB depended mainly of Pacific and Indian Ocean sea surface temperatures. The final model explained the 46% of the variance of JFM rainfall in UB and the correlation between observed and forecast series was 0,49. Results indicate that the two mayor factors that influence summer precipitation in Lower Bermejo were the South Antarctic Oscillation and the weaken Atlantic High. In Upper Bermejo it depends mainly on Pacific and Indian Ocean sea surface temperatures. The efficiency of the method was proved by calculating some statistics like the hit rate, the probability of detection and the false alarm ratio. Results in LB are better than in UB. The probability of above normal rainfall events is in general, better than the probability to detect below normal rainfall ones. The probability to give a false alarm in a below normal rainfall event is greater than in the above normal cases. Additionally, the probability functions resulting from estimated and observed JFM rainfall resulted similar at the 95% confidence level and reveal that the method underestimates the most extreme cases. These results are promising and encourage further work in order to examine new techniques to better estimate rainfall, especially the extremes, and to investigate other predictors which could affect precipitation in summer.