Water management in a semiarid region an analogue algorithm approach for rainfall seasonal forecasting

Giampiero Maracchi, Massimiliano Pasqui and Francesco Piani

14.1

Introduction

Methods and results of this recent branch of atmospheric sciences must be the most simple and accessible as possible. For this reason, the Institute of Biometeo-rology, (part of the National Research Council, http://www.ibimet.cnr.it), has developed a physically - based statistical approach to obtain seasonal forecasts, regarding rainfall precipitation, over Sahel region.

tte method is based on the "similarity" conditions of the sea surface temperature (SST) in three areas of the world defined as: Niño-3 (5S-5N;150W-90W), Guinea Gulf(10S-5N;20W-10E), Indian Ocean (5S-15N;60E-90E) which, in literature, are indicated as the most important areas to drive the precipitation patterns: Indian Ocean and Southern Atlantic with regard to the trends of precipitation and Niño-3 with regard to the interannual time scale (Giannini et al. 2003).

tte importance of the sea surface temperature to force the long-term atmospheric anomalies, at least at a regional scale, is recognized with particular attention to the Pacific area affected by El-Niño Southern Oscillation (Dalu et al. 2006).

Many atmospheric Research Centers have developed their own methods to derive seasonal forecasts, based on the results of a large number of simulations of a Global (GCM) or a Regional (RCM) Circulation Model, namely "Ensemble Forecasts", or on statistical algorithms that relate the most important atmospheric variables, or both (see for example: http://www.ecmwf.int or http://iri.columbia. edu/).

ttis work describes a statistical method that relates the SST of three oceanic areas with the precipitation in the semi-arid region called Sahel. tte chance of forecasting a reliable rainfall field is, in many parts of Africa, dependent on prevailing patterns of sea surface temperature, atmospheric circulations, the El Niño Southern Oscillation and regional climate fluctuations in the Indian and Atlantic Oceans. A brief summary of scientific background of the method is the following: • West African Monsoon variability is strongly forced by the sea surface temperatures standardized anomaly (SSTAs) of the Gulf of Guinea. Warm Gulf of Guinea SSTAs generates a rainfall increase along the Guinean coast while the precipitation decreases over the Congo Basin, ttese features can be understood through the dynamical response of a Kelvin wave along the equator and a Ross-by wave to the west of the SSTA. tte first is associated with a weakening of the Walker circulation, while the latter tends to strengthen the West African Mon soon and the upward vertical velocity, tte effects of Cold SSTAs are opposite, but weaker (Vizy and Cook 2001).

• tte monsoon circulation influences the precipitation over the Sahel, in particular southern Sahel (10N-15N), in two main ways, tte moisture is transported by the low-level southerly flow, tte proximity of the monsoon circulation and circulation over Sahara generates a strong low-level convergence to force air parcels to rise vertically until the level of free convection (Vizy and Cook 2002).

• Positive SST anomalies in the Eastern Pacific and in the Indian Ocean, negative anomalies in the northern Atlantic and in the Gulf of Guinea are related with drought conditions over all the West Africa (Fontaine and Janicot 1996).

• Droughts limited to Sahel are due to a positive SST anomaly northward in the southern Atlantic and a negative pattern in the northern Atlantic. Floods along the West Africa are associated with positive anomalies in the northern Atlantic, while the floods limited to Sahel are related to different forcing: northward expansion of negative SST anomalies in the southern Atlantic, positive SST departures in the northern Atlantic, and development of negative SST anomalies in the eastern Pacific (Fontaine and Janicot 1996).

• tte Principal Components Analysis (PCA) performed on the summer precipitation in the Sahel region, demonstrates that the two leading principal components (PCs) explain almost half of the variability of the precipitation. Moreover two main patterns are present: the first along the Gulf of Guinea coast, between the equator and 10°N, dominated the interannual variability, the second associated with the continental convergence in the Sahel (between 10°N and 20°N) affected by the interdecadal variability, tte decomposition of these two leading PCs into high and low-frequency components shows the role of the SST of the Southern Atlantic and Indian Ocean for driving the long-term variability, while the interannual variations are driven by the ENSO (Giannini et al. 2003).

14.2

Methods and Dataset

In the method each "month" is defined by six variables: three are SSTAs while the other three take into account their respective tendencies (namely "Change Rates" or CRs). tte CRs are defined as the difference between the current SSTAs and those of the previous month, tte standardization is obtained with the subtraction of the 1979-2003 climatological mean and the division by the 1979-2003 cli-matological standard deviation, tte "similarity" to the current SST conditions is evaluated by means of the minimization of the Euclidean distance to find the most similar year (namely analogue) and assign the values observed in that year to the forecast rainfall field. Due to the specific dynamical behavior of the West African Monsoon this simple analogue characterization is able to catch main features of rainfall precipitation patterns during the JJA period and a validation of this approach, through analysis of forecast skills, shows encouraging results.

SSTs used in the method are from three oceanic areas: the Nino-3 area (5S-5N; 150W-90W), the Guinea Gulf (10S-5N; 20 W-10E), the Indian Ocean (5S-15N; 60E-90E) as in Fig. 14.1.

^e SST data have been obtained by the NCEP/NCAR Reanalysis dataset (2.5°x2.5° Lat-Lon, Kalnay et al. 1996; Kistler et al. 2001) while the precipitation data have been derived from the Global Precipitation Climatology Project (GPCP, Xie and Arkin 1996; Huffman et al. 1997; Xie et al. 2003) on a 2.5°x2.5° Lat-Long grid.

For each month and for each grid-point, the precipitation time series has been correlated to the SST time series, in order to have the relative weight of the three different oceanic areas with regards of the precipitation. Based on these weights and on the six variables defined above, the method searches for the most similar SST conditions in the past (the year obtained is called "Analogue Year"), assigning

Fig. 14.1 Oceanic areas from which SSTAs have been computed to be used in the method.

Fig. 14.1 Oceanic areas from which SSTAs have been computed to be used in the method.

e.g.: April 2005

e.g.: April 1989

i

e.g.: May 2005 = May 1989

MONTH+2

e.g.: June 2005 = June 1989

MONTH+3

e.g.: July 2005 = July 1989

1

1

CLIMATOLOGICAL AVERAGE

e.g.: May. June, July 1979-2003

ANOMALIES

Fig. 14.2 An example of the logical scheme applied for each precipitation grid point in order to derive the "Analogue" year.

the values observed in the closest year to the forecasts. Using 1979-2003 climatology, precipitation anomaly and percentage anomaly are then computed (Fig.14.2). tte first is derived by means of the subtraction of the climatological mean from the forecast precipitation, tte latter is obtained from the former, dividing for the climatological mean and it's represented as a percentage, tte dry regions of the world, identified by a monthly cumulative precipitation under 30mm, are blanked to avoid large values of anomalies and percentage anomalies (Table 14.1).

Table 14.1. An example of forecast maps of percentage anomaly for the year 2003 with the correspondent observed values from CMAP (http://www.cdc.noaa.gov/cdc/data.cmap.html) Janowiak and Xie 1999: for each month, June, July and August, forecasted maps were shown with different time lags.

June 2003 July 2003 August 2003

2 months ahead

3 months ahead

OBS CMAP

1 month ahead

14.3

Skill evaluation

A true validation strategy should be based only on data collected prior to the target month to be forecasted. Such a calculation should then be repeated for all available years. But the resulting skill depends on the amount of data used for each calculation. Another possible strategy is the adoption of a cross validation calculation. Each prediction was estimated using only data before and after that specific year, tte cross evaluation hindcast method is able to represent a good forecast skill measure if two conditions were satisfied: the climate statistics do not change among the period considered and there is a weak autocorrelation between neighboring years data. In order to perform the cross - evaluation analysis we select all summer time forecasts (June - July - August) in the 1979 - 2005 period. For each month the rainfall anomaly has been computed based on the 1979 - 2005 mean value for the region defined as Lat: 15°.0 - 17.5°, Long: -10° - 10°. For this area the entire monthly anomaly ensemble was divided into three categories below 33% percentile (Below hereafter), above 66% percentile (Above hereafter) and between them (Normal hereafter). Each monthly anomaly was aggregated in order to form three ensembles, tte same computation was performed for different forecast ranges: 3 months ahead, 2 months ahead and 1 month ahead respectively. For each observed anomaly group, by means Below - Normal - Above, all the forecasted anomalies were computed and showed as "chocolate wheels" graphs in Fig. 14.3 to Fig. 14.5 (Hayman 2000). As expected, using SSTA, for forecasting precipitation introduces a lag time in the peak performances of the method, by means: best performanc-

l Month Ahead 2 Months Ahead 3 Month Ahead

Fig. 14.3 Chocolate wheels for Sahel area evaluation. Numbers represent the percentage of historical cases in the lower (light grey), middle (white) and upper (dark grey) terciles for June. Table columns represent time lags and table rows represent different observed classes (Below - Above - Normal from top to bottom).

Fig. 14.3 Chocolate wheels for Sahel area evaluation. Numbers represent the percentage of historical cases in the lower (light grey), middle (white) and upper (dark grey) terciles for June. Table columns represent time lags and table rows represent different observed classes (Below - Above - Normal from top to bottom).

1 Month Ahead 2 Months Ahead 3 Month Ahead

Fig. 14.4 Chocolate wheels for Sahel area evaluation as in Fig. 14.3 for July.

es were obtained 2 o 3 months in advance. Same behaviors are present for all target months. Normal months were forecasted worse than Above or Below months, probably this is a link to choice of SSTA as predictors emphasizing strong rainfall anomalies.

One limitation of the evaluation strategy is the short period of time used: just 26 years. One single forecasted event can alter the statistics greatly. Further analysis will be focused on extending this period back in time increasing the statistical ensemble.

14.4

Conclusions tte weather predictions today are established on solid theoretical and practical bases, their reliability and accuracy are steadily increasing and their usefulness is widely recognized in a variety of fields and applications, often in the frame of automatic integrated prediction systems.

tte current state of the art of the weather forecasts allows the short-term prediction of rare and dangerous local events such as rainstorms, frost with high reliability and accuracy and very high spatial resolution, as well as the accurate medium range prediction, tte role of conventional weather forecasts is precisely defined as a strategic one, and as such, is considered the national as well as the regional weather services (Soderman et al. 2003). Having information on the future trend of precipitation three months or more in advance could be of extreme importance in many fields oflarge economic, social, environmental and strategic relevance: agriculture and forestry, land and landscape management (to forecast droughts or heat waves for example), international cooperation and catastrophy management (i.e. food shortages, droughts, production and distribution of energy).

I Month Ahead 2 Months Ahead 3 Month Ahead

Fig. 14.5 Chocolate wheels for Sahel area evaluation as in Fig. 14.3 for August.

Seasonal forecasts could answer these questions but not with the same efficiency, accuracy and reliability of the meteorological forecasts. Although many enhancements have interested this branch of atmospheric science in the last decade, large errors and uncertainties still affect this type of products. At the same time, seasonal forecasts have assumed a relevant role for planning and decision making.

Authors would like to underline the experimental character of the results of the method, ttey must be used just as an indication of the possible future trend of the precipitation.

References

Dalu GA, Gaetani M, Pielke RA Sr., Baldi M, and Maracchi G (2006) Regional variability of the ITCZ and ofthe Hadley cell. Geophy Res Abst 6:10-2-2004.

Fontaine B, Janicot S (1996) Sea surface temperature fields associated with West African rainfall anomaly types. J Clim 9:2935-2940.

Giannini A, Saravanan R, Chang P (2003) Oceanic forcing of Sahel rainfall on interannual to interdecadal time scales. Science 302:1027-1030.

Hayman PT (2000). Communicating probabilities to farmers: pie charts and chocolate wheels. In Petheram RJ. (ed.) Tools for participatory R&D in dryland cropping areas. pl33-136 RIRDC00/132. Rural Industries Research and Development Corporation, Canberra.

Huffman G I and co-authors (1997) ^e Global Precipitation Climatology Project (GPCP) combineddata set. Bull. Am. Met. Soc.. 78:5-20.

lanowiak IE, Xie P (1999) CAMS_OPI: a global satellite-raingauge merged product for real-time precipitation monitoring applications. I Clim 12:3335-3342.

Kalnay E and coauthors (1996) tte NCEP/NCAR 40-Year Reanalysis Project. Bull Am Met Soc 77:437-471.

Kistler R and coauthors (2001) Ue NCEP-NCAR 50-Year Reanalysis: Monthly Means CD-ROM and Documentation. Bull Am Met Soc82:247-268.

Soderman D, Meneguzzo F, Gozzini B, Grifoni D, Messeri G, Rossi M, Montagnani S, Pasqui M, Orlandi A, Ortolani A, Todini E, Menduni G, Levizzani V (2003) Very high resolution precipitation forecasting on low cost high performance computer systems in support of hydrological modeling. Prepr. 17th Conf. on Hydrology, AMS, Long Beach.

Xie P, Arkin PA (1996) Global precipitation: a 17-year monthly analysis based on gauge observations, satellite estimates, and numerical model outputs. Bull Am Met Soc 78:25392558.

Xie P, Janowiak JE, Arkin PA, Adler R, Gruber A, Ferraro R, Huffman GJ, Curtis S (2003) GPCP pentad precipitation analyses: an experimental dataset based on gauge observations and satellite estimates. J Clim 16:2197-2214.

Vizy EK, Cook KH (2001) Mechanics by which Gulf of Guinea and Eastern North Atlantic Sea Surface Temperature Anomalies can influence African Rainfall. J Clim 14, 795-821.

Vizy EK, Cook KH (2002) Development and application of a mesoscale climate model for the tropics: Influence of sea surface temperature anomalies on the West African Monsoon. I Geophys Res 107(D3), 10.1029/2001ID000686.

CHAPTER 15

0 0

Post a comment