One of the main objectives of the study was to developed a statistical model to predict rainfall anomalies over Cordoba region. For this, several analysis were performed along the study in order to better understand the climatology and the weather patterns in the region.

Data Analyse

First of all, Precipitation data from 90 weather stations in Cordoba region were obtained from IDEAM (Institute of Hydorlogy, Meteorology and Environmental Studies – Colombia) and were analysed. From this stations, it were selected the ones which had 10% or less of missing data(Lau and Sheu 1988).  From the 90 stations, only 45 presented less than 10% of missing data and these ones were used in the first part of this study. After this first step, an analyse of the climatology of the 45 remain stations were made (with a histogram) to verified what kind of distribution each station presented (Unimodal or Bimodal) and also to verified the average of all of them in order to classified each station in above, below or in the average (normal). Considering the average approximately 1600mm, the remain stations were divided in 6 groups: Unimodal/Above, Unimodal/Below, Unimodal/Normal, Bimodal/Above, Bimodal/Below and Bimodal/Normal.

After that, a correlation analyses within each of these groups were made in order to verified what stations were better correlated (to then be chosen)to proceed with the study, but no good correlations were found. For this reason, the stations were then, divided only in two groups: Above and below the average.  After this, another correlation within both groups was calculated and better values were obtained, reducing the number of stations to 12 from the below the average group and 15 above the average.  Thus, in total, this study will analyse 3 different groups of stations: The group Below the average, the group Above the average and the group with All the stations.

IDEAM weather stations in Cordoba. Below the average: Green Points; Above the average: Yellow points


Monthly precipitation data from 1980 to 2013 of those stations underwent a quality control procedure and the missing data were disregarded.

Canonical Correlation

The interanual and seasonal climatic variability of a certain region is partly associated with the interanual variability of the Sea Surface Temperature (SST) on the different oceans all over the world (Diaz, Studzinski et al. 1998). The variability on those oceans is extremely important and influences the climate patterns in the 5 continents. There are many different sources of this variability within those oceans but the most famous one is the climatic variability over the Pacific Ocean known as ENSO – El Nino Southern Oscillation which affects the weather and the climate in different areas of the world by causing sensitive anomalies on the atmospheric flow through an anomalous heating in the Equatorial Pacific waters associated with some variations on the pressure field measured by the South Oscillation Index (SOI) (Ropelewsky and Halpert 1989).

In order to investigate what regions of the world most influence the climate patterns in Colombia, especially in Cordoba, a canonical correlation analysis between the global SST and the weather data from Cordoba was made. For this, the software “Climate Prediction Tool” (CPT) developed and provided by the International Research Institute for Climate and Society (IRI) was used (IRI 2015). This software has a package for windows for the construction of seasonal climatic models offering the statistical options of Principal Component Regression (PCR) and the Canonical Correlation Analysis (CCA) being the latter, the chosen to study the relationship between the global SST and the Cordoba rainfall data.

The data used as predictors were the global SST average since 1980 to 2013 obtained along the IRI website with a spatial resolution of 2º x 2º (latitude and longitude). According to the rainfall distribution’s graphs generated with the data from Cordoba weather stations, the SST data were then, divided in wet (May- October) and dry season (November to April). On the other hand, the predictand data used in the study were the monthly rainfall anomaly from Cordoba weather stations obtained according to the formula present in Porfirio da Rocha 2013.






The analysis that CPT (Climate Prediction Tool) makes optimizes the number of EOF modes, once it uses EOF1 to make a cross-validated forecast to then calculate a goodness index summarizing how good all the forecasts are (Mason 2013). As close to 1.0 the goodness index is the better. The CPT uses EOF1 and EOF2 to remake cross-validated forecasts and calculates a new goodness index for these, and so on until all five EOFs have been used (Mason 2013). At each step, CPT makes a comparison between the goodness indices retaining under the column “Optimum”, the highest goodness index and the corresponding number of EOFs (Mason 2013). After this, CPT uses this number of EOFs to build the model.

From the maps obtained along with CPT, it was possible to verify 5 major areas of influence of SST in Cordoba region, which means that the areas with stronger correlations (both positive and negative), are related with climate indices. Those areas are the ENSO region in the equatorial Pacific, the IOD region in the Indian Ocean, the PDO region in the Northern Pacific, the CAR (Caribbean) in the Caribbean Sea and the NAO region over the Northern Atlantic.



The results obtained within this study led us to analyse the relationship between these 5 indices and the precipitation in Northern Colombia (Cordoba region) at monthly time scale and time lags from 0 to 12 months.




The graphics above shows respectively the monthly correlations between the indices chosen (CAR, ONI, NAO, PDO and DMI) and the monthly precipitation from Cordoba regions for the All stations group.From those graphics, it is possible to identify the best indices and months suitable to predict rainfall in each month. For example,from the graphics, it is possible to observe that the CAR index of October/November/December shows a strong positive correlation with the rainfall of April while ONI index of July/August/September/October/November/December has a strong correlation with the rainfall of January. On the other hand, the NAO index for November has a good correlation with the rainfall of February, the PDO index of March presents a strong correlation with November and the DMI index of February shows a good correlation with the rainfall of June. However, these correlations may not be so significant to construct the statistical model. For this purpose, the correlations must be analysed according to each month. Taking April as an example, the CAR index from October (CAR6), the ONI index from December (ONI4), the NAO index from August (NAO8), the PDO index from July (PDO9) and the DMI index from July (DMI9), does not present a very strong correlation with the month, although the best model developed for this group considers those indices with those respective lags (this topic will be further discuss in text).


Temporal Analysis

In order to verify the temporal variability of the data from Cordoba, a temporal analysis was made. The aim of this temporal analyse was to verify how often extreme, or more intense precipitation events occurred in the data series of Cordoba’s weather stations to investigate which meteorological phenomenon would be interfering in the region to cause those extremes. For this, the Jaziku software was used. The Jaziku software was developed by the “Weather and Climate” branch of IDEAM (Institute of Hydrology, Meteorology and Environmental Studies – Colombia) and was designed to find relationships through composite analysis methodology between the time series of the dependent variables (observed meteorological variable) with the index that represents the phenomenon of the climate variability (Corredor Llano). Jaziku is free and open software so then, can be improved and adapted for different purposes. The software is used for exploratory data analysis to detect changes in the homogeneity of the time series and analysis outliers. Besides, based on predicted probabilities of IRI models for the ENSO and using the forecast module, for each month a seasonal forecast is made for Colombia and the results are added to other methodologies such as dynamic models and canonical correlation analysis for example (Corredor Llano).



The graphic above was made with the data from one weather station from Cordoba (Ayapel). From this graphic, it is possible to verify that  approximately each 12 months, there is a peak happening along the entire time serie. This means that every 12 months, there is something, some weather pattern that might be interfering directly in the precipitation over the region.

This phenomenon with a 12 month periodicity, presents a very clear signal in the serie, showing its direct relation with the climatic patterns of the region. Analysing the graphic, it is possible to verify that the signal in 12 months is very evident with a strong correlation happening between 1995-1997 period in which a La Nina event happened, in a way that this correlation is, probably, associated with this phenomenon. Besides this correlation, it is also possible to verity another strong correlation between 2009-2012, an El Nino event period (which is probably also associated with this high correlation value).

Regression Model

In order to construct a statistical model, firstly an equation relating the forecast variables need to be postulated and then an estimative of the parameters in the model need to be made to minimize the error of the predictions assuming that the past statistical relations will be maintained in the future (DelSole and Shukla 2002). The regression analysis is a statistical technique to study and model the relation between variables, being used for several different applications (Data description, parameter estimation, prediction and estimation and control) (Montgomery, Peck et al. 2006). A simple linear regression model presents a model with a single regressor (x) which has a relationship with an answer and where the relationship is a straight line.

tabletable 4_october



  1. Discussion and Conclusions


Many studies related with climatic patterns and their relation with rainfall amounts over a certain region were previously developed and their main focus was on ENSO impacts over South America (Ashok, Guan et al. 2001; Poveda, Rojas et al. 2001; Cobb, Charles et al. 2003; NOAA 2005; Porfirio da Rocha 2013). However, fewer studies have attempted to better understand teleconnections patterns and their potential impacts over the continent (Wang, Kucharski et al. 2009). This MScCCAFS Researhc project study has investigated links between Sea Surface Temperature (SST) variations in global oceans and rainfall index over Colombia. The study had taken into account seasonal and monthly correlations between SST and NCR (Northern Colombia Rainfall) from zero to two seasons of lag and from zero to twelve months of lag.


In this MScCCAFS Researhc project study it was possible to observe that several climatic and meteorological patterns interfere on the rainfall regime over Cordoba. The climatic patterns are more related to the indices and to the teleconnection patterns in a global scale while the meteorological patterns are related to more local mesoscale patterns.


The climatic indices that are calculated from SST anomalies around the globe are related with rainfall anomaly data from the IDEAM weather stations in Cordoba presenting a good correlation (both negative and positive ones) with them. From the correlation graphics for example (for All the station group), it was possible to observe that both ONI and DMI index for September and October display a strong correlation with the rainfall of January. The fact that these two indices present a strong negative correlation for the same period is a good thing once several other studies have shown in the last years, that both indices are correlated with each other. A study from the rainfall patterns in East Africa for example, had shown that there is a good evidence that the observed teleconnection between the rainfall in East Africa and ENSO phenomenon is actually a manifestation of a link between ENSO and IOD once each one of the four ENSO warm phases associated with high rainfall in East Africa coincided with a dipole event (Black 2005). According to (Taschetto and Ambrizzi 2012) this can also have a physical meaning. The positive phase of the Indian Ocean Dipole is also known as Indian Ocean basin-wide warming (IOBW) and presents a periodicity of approximately 4 years that can be associated with El-Nino events. The dynamical/thermodynamical link between IOBW and the El-Nino Southern Oscillation phenomenon (ENSO) can be verified because, during the development of an El-Nino event for example (from June to November), the eastern Indian ocean is relatively colder and in an essentially positive IOD state (Taschetto and Ambrizzi 2012). By the time of the arrival of the Australian monsoon, the trade winds get weaker and become westerly. This process also weakens the oceanic upwelling, reducing surface cooling via changes in latent heat flux. From November to December, when the Nino is in its mature phase, it generates subsidence and an anticyclone wind stress anomaly in the eastern Indian Ocean through an anomalous Walker circulation reducing convection and consequently the formation of clouds. This also enhances the heat flux into the ocean, warming the eastern Indian Ocean basin (Taschetto and Ambrizzi 2012). The anomaly in the wind stress decreases the wind speed leading to oceanic downwelling of Rossby waves which propagates western and is responsible to create a uniform warming of the Indian Ocean 3-4 months after the peak of El-Nino (Taschetto and Ambrizzi 2012).  The IOBW co-occurs with El Nino events that are larger in amplitude and also extension being justifiable the fact that scientists focus primarily on the Pacific warm episodes. The Indian Ocean basin-wide (IOBW) is not an independent mode of variability in a physical sense but it is and indirect effect of El Nino events (Taschetto and Ambrizzi 2012). This warming of the Indian Ocean may act as a source of heating, being exchange between the ocean and the atmosphere. The Indian Ocean remains anomalously warm during several months after the peak of El Nino which means that while El Nino normally peaks at the end of austral spring and summer, the ocean continue anomalously warm until the following austral winter, when the events in Pacific starts to dissipate. So, in addition to the Walker circulation anomaly, the ENSO phenomenon can significantly modulate rainfall across the South America subtropics as well via the Pacific-South American pattern (Taschetto and Ambrizzi 2012). Some studies developed in the last years have shown that over the austral spring, the Indian Ocean Dipole (IOD) can remotely affect the atmospheric circulation in South America. This generates a significant impact in precipitation distribution and intensity once the warm SST anomalies in subtropical Indian Ocean rises the South America teleconnection pattern potentially inducing changes onto the circulation over the continent during austral summer (Taschetto and Ambrizzi 2012).


Although the best correlations are found in different months, it is possible to verify that noticeably, they are concentrated in the month’s correspondent to the dry season (November-April). From all the indices analysed, CAR was the one which presented the strongest positive correlations in the end of the dry season (April). The positive values of anomaly correspond to an increase in precipitation while the negative ones are related with a decrease in rainfall amounts. This means that in El Nino years for example, when the temperatures over the Pacific gets warmer and the anomalies are positive (which means positive values of ONI) the precipitation over Cordoba region decrease, that’s why the correlation between the two  variables are negative. The same happen with the DMI, NAO and PDO. The graphics from Above average and Below average group does not present a very different behaviour from the graphic of the All stations group. Both of them show almost the same pattern with CAR, ONI, NAO and DMI and a small difference in relation to the PDO. For the Above average group, PDO index of March showed a better correlation with the rainfall in March (Lag 0) and in the Below average graphic for example, it was possible to observe that the PDO index of November showed a good correlation with the rainfall in November (also Lag 0) rather than with the rainfall in December.  Besides this difference within PDO index, it is possible to verify that both groups also present the best correlations in the months that cover the dry season (November-April).


Besides these global patterns, there are other local ones that can also influence in the climatology of the region. It was possible to determine, via the Periodograms, that the Intertropical Convergence Zone has a more direct influence on the precipitation in the region. The conclusion related to the wavelets are the same for both groups. All the stations present a very evident periodicity of 12 months, with different years of occurrence of the highest correlations (although all of them are also associated with El Nino or La Nina events). Although they present very good correlations in distinct years, it is clear the 12 month periodicity signal showed for the stations. This signal is related to the occurrence of a meteorological/climatological phenomenon that also presents a 12 month periodicity. So, it is possible to conclude that the Intertropical Convergence Zone (ITCZ)  is the phenomenon that most interfere (and presents a more direct influence) over the rainfall distribution in Cordoba. However, there are some other factors that need to be taken into account. As it was possible to observe in the correlation graphics, the climate indices (and consequently some different regions of the world) also present a correlation with Cordoba’s rainfall distribution. Also as speculated before, this correlation may not happen in a direct way. This means that those indices may present a better correlation with the ITCZ, both related with its intensity and position, and the ITCZ in its turn, has a more direct relation with the rainfall distribution over Cordoba.


The relation between the ITCZ and ENSO variability for example, has been deeply investigated in last years. Some studies verified that the position of the ITCZ is related with an increased prevalence of temperature anomalies caused by El-Nino/La Nina events in the eastern tropical Pacific though the Holocene (Cobb, Charles et al. 2003). Besides, some other paleoclimatic studies also verified that the movement of the ITCZ southward lead to an increase/decrease in the rainfall distribution over the Amazon for example, and to an shift in the position of the South America Summer Monsoon (SASM) during the periods known as the Little Ice Age (LIA) and the Medieval Climate Anomaly (MCA) (Cobb, Charles et al. 2003; Vuille, Burns et al. 2012).  More precisely, some paleoclimatic reconstructions made for South America Summer Monsoon (SASM) for the past millennium showed a weaker monsoon during the Medieval Climate Anomaly (MCA) and a stronger monsoon over the Little Ice Age, presenting an anti-correlation with some reconstructions of the Southeastern Asia monsoon and the Northern America and Northern Asia monsoon for the same period, being this theory supported by modelling studies (Vuille, Burns et al. 2012).


Some models simulations also suggest that the ocean transport towards north during the Medieval Climate Anomaly (MCA) presents a relation with a cross-equatorial temperature gradient over the Atlantic and also over the movement of ITCZ. Supplementary, a modelling study developed by Lee Chiang in 2011 (Lee, Chiang et al. 2011) verified that in North Atlantic, when the temperatures are cooler than normal (like in the Little Ice Age), the Atlantic ITCZ moves southward, strengthening  the northern Hadley cell over the austral summer, moving its rising branch towards south. Another point worth highlighting is that some of the paleoclimatic reconstructions made also suggest that the anomalously temperatures observed during the Medieval Climate Anomaly (MCA) and the Little Ice Age (mostly in the North Atlantic), are able to modify the position of ITCZ in South America, affecting consequently, the South America Summer Monsoon (SASM).


As mentioned before, the Periodograms  demonstrated that in an approximately 12 month range a precipitation peak appeared in the graphics (ITCZ). However, by analysing in more details the Periodograms, it is possible to observe smaller peaks within the 12 month range. During this MScCCAFS Researhc project study project, due to lack of time, it was not possible to more deeply investigate the phenomena related to those smaller peaks although it can be speculated that they may be related with west waves, Madden Julian Oscillation or even cold fronts. Through the Periodograms it was also possible to observe that not necessarily the peaks related to the Intertropical Convergence Zone occurred in El Nino or La Nina years, but when they did, they peaks presented a higher signal. This leads us to conclude that the climatic indices and the teleconnection patterns, have an influence in the intensity and position of the ITCZ and that the ITCZ, in turn, is related in a more direct way with the rainfall in Cordoba.


Our MScCCAFS Researhc project study also found significant lagged seasonal correlations with different areas all over the globe. Those areas are related with ENSO phenomenon (ONI index), the Indian Ocean Dipole (IOD/DMI index), the Northern Pacific Ocean (PDO index), the Equatorial Atlantic Oscillation (NAO index) and the Caribbean Sea (CAR index). It is known that all of those areas are related with atmospheric circulation patterns (Hurrell and Van Loon 1997; Penland and Matrosova 1998; Ashok, Guan et al. 2001; Hurrell, Kushnir et al. 2001; Mantua and Hare 2002; Cobb, Charles et al. 2003; Wang, Kucharski et al. 2009; Mochizuki, Ishii et al. 2010; JAMSTEC 2012; Taschetto and Ambrizzi 2012; Jianping Li; Cheng Sun 1 2013).


An analysis of the relationship between SST and rainfall in Cordoba, allowed the elaboration of a linear regression model to make monthly predictions of rainfall over the Cordoba region. It was possible to observe robust and highly significant lagged correlations between SST indices and rainfall index. From the table above for example (Linear regression for April and October), it was possible to determine that both models (from April and October) are really good to predict rainfall anomalies as they present a very small p-value (especially the model from April that presents a p-value<1%). For April, CAR, ONI and DMI are positive on the equation which means that they have a positive relation with the rain. From the developed model, those variables have a positive weight, which means that they correspond to positive rainfall anomalies for the forecast in this month. A curious point about this is the fact that both ONI and DMI present this relation. As previously mentioned, both indices are related with each other and mostly, are responsible to generate less rain over Colombia when are in its positive phase. Positive values of ONI represent an anomalous warming in the Pacific waters and are responsible for the appearance of the El Nino phenomenon. This phenomenon starts to develop between June and November (during the wet season in Cordoda) having its peak occurring on December (over the dry season). Thus, the fact that these indices have a positive value in the model for April (last month of dry season) means that, even being responsible for rainfall deficits, the indices present a quite strong signal on the wet months of the region.  A positive ONI and consequently, an El Nino phenomenon, is not synonymous of a completely lack of rain in Cordoba during the wet and dry season, in a way that what occurs is that the rainfall amount decrease, presenting below average values. For October, the indices NAO and ONI are the most significant ones and then, have a higher weight on the model. Unlike April, that presented a positive signal for ONI, October presents a negative one.  In opposition to the positive signal presented in April, the negative signal indicates that the ONI has a negative relation with the rain for this period and then, is related to a decrease on the rainfall regime. Since October is the last month of wet season, this means that the signal captured on the model is related with the beginning of the decrease in rainfall on the region. Another aspect to be considered is that not necessarily a positive or negative signal on the ONI index is related to Nino or Nina phenomena. What happens is that even in neutral years, without the occurrence of one of those phenomenons, it is possible to observe small variations on the SST anomalies (both positive and negative) which generate the opposite signals on the models for the dry and wet month.


In this study, it was also found that despite being a remote response to ENSO events, the Indian Ocean has the potential to feedback onto the atmosphere in order to induce tropical and extratropical teleconnections over South America. This was previously suggested by (Taschetto and Ambrizzi 2012) whose results present evidence that the rainfall over South America can be modulated by Indian Ocean SST variability through remote mechanisms. In addition, (Wang, Kucharski et al. 2009) have also showed teleconnections patterns between the Indian, the Atlantic and the Pacific Oceans and rainfall variability in South America. One of the most import results of (Wang, Kucharski et al. 2009) is the finding that the rainfall in Peru for example, is influenced by other areas besides ENSO areas, with several months in advance, which are related with known indices like the Indian Ocean Dipole or the Equatorial Atlantic Oscillation. These results are extremely useful and open new possibilities to understand and also to monitor the rainfall over South America.


It is also necessary to conduct further research on the phenomenon known as El Nino Modoki and its impacts on South America rainfall variability. Some studies present different oceanic and atmospheric patterns when compared to canonical ENSO. The impacts on the rainfall levels in South America indicate that it is really important to study the two types of phenomena in a separated way.


Finally, it is proposed that further studies should be conducted in order to verify the mechanism involved on the correlation obtained in this study using the atmospheric general circulation model (AGCM) and the regional climate model (RCM). The first one can assess the influence of SST in the circulation under Cordoba providing boundary conditions to the RCM and the second can provide physical link between SST and rainfall. This would be useful since the AGCMs have difficulties in analyse the distribution of precipitation, being a dynamical downscaling needed. This would deeply improve the forecast system in the region facilitating for example, the decision-making by farmers (since the region is the biggest corn producer in Colombia) and may be considered as an adaptation measure facing climate change.