Non-stationary associations between rainfall intensity and large-scale variability in an east-Andean city over the last 108-years

doi:10.21203/rs.3.rs-3150138/v1

Download PDF

Research Article

Non-stationary associations between rainfall intensity and large-scale variability in an east-Andean city over the last 108-years

https://doi.org/10.21203/rs.3.rs-3150138/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

This work explores for the 1912–2019 period the interannual associations between the maximum daily rainfall (Rx1) in San Miguel de Tucumán (Argentina, 26.8°S, 65.2°W) and the large-scale variability. The flood hazard in this east-Andean city is high due to the presence of lowlands and urban stream channels with poor maintenance. We analyzed the interannual relation between Rx1 and El Niño Southern Oscillation (ENSO) and the Southern Annular mode (SAM), and the non-stationarity of these associations as a response to Pacific Decadal Oscillation (PDO) and Atlantic Multidecadal Oscillation (AMO). We compute the moving correlation for all possible window lengths between Rx1-ENSO and Rx1-SAM. As a new approach we based our analysis on the periods with most significant correlations (i.e., with minimum p-value). The aim was to obtain a period when there exists an association with maximum statistical confidence. Results suggest that Rx1 variability shows the highest statistical significance in its association with ENSO for the 1945–1974 period and with SAM for the 1974–2007 period. The overall non-stationary nature in the relations can be attributed to the phase shifts of PDO. Negative PDO and positive ENSO phases favor higher Rx1 values, while this also occurs with both positive PDO and SAM. Inverse phases favor lower Rx1 values. On the other hand, AMO phases do not influence relationships. Results highlight that processes involved in the variability of rainfall intensity are partly related to global variability modes, opening opportunities to enhance predictions in relation to the flood hazard.

extreme rainfall

ENSO

SAM

AMO

PDO

Rainfall is a climatic variable widely studied due to its major socio-economics implications. The maximum-daily rainfall (Rx1) is analyzed in this work, which is usually used as an indicator of rainfall intensity (de la Casa et al., 2019). The occurrence of local flooding or flash flooding events occur when intense rainfall occurs over a small area in a short period of time (Avashia and Garg, 2020), for which Rx1 can be a good proxy. This study focuses on San Miguel de Tucumán (TUC), the most populated city of northern Argentina (Ministerio del interior, Argentina, data available on: https://www.argentina.gob.ar/sites/default/files/poblacion_urbana_dnp.pptx_.pdf), located in subtropical South America (SSA), east of the Andes Mountain Range. The flood hazard in this city is significant due to the presence of lowlands and urban stream channels with poor maintenance (Fernández and Lutz, 2010). In fact, hundreds of people have been evacuated in the past decades due to flooding. Thus, it is essential to comprehend the variability of heavy rainfall events to improve their prediction accuracy and minimize their impact.

Our study focuses on interannual to multidecadal scales of rainfall variability to improve long-term Rx1 prediction. The possibility of non-stationary associations between rainfall and large-scale modes of variability is explored using a century-long daily-rainfall record from TUC. Rainfall variability in North-Western Argentina (NWA), the region where TUC is located, is largely influenced by forcings related to the Tropical Pacific Ocean. This has two dominant modes of variability: El Niño Southern Oscillation (ENSO) and the Pacific Decadal Oscillation (PDO).

In the case of ENSO, this is a coupled atmosphere-ocean phenomenon (Wang et al., 2017) and is characterized by two phases denoting sea-surface temperature (SST) conditions: warmer (positive phase, called “El Niño”) and colder (negative phase, called “La Niña”) (Cai et al., 2020). ENSO has periodicities of 3 to 6 years, which have significant impacts on rainfall in Southern South America (Wang et al., 2017; Krepper and Garcia, 2004). Long-term variations in ENSO are also present (Hurtado et al., 2020), but they have received less attention in the literature. PDO, on the other hand, consists of multidecadal fluctuations of the SST in the tropical Pacific, extending to the central North Pacific and tropical Indian Oceans (Newman et al., 2016; Wang et al., 2017). In the past century, relatively cold conditions of PDO were observed in 1910–1925 and 1947–1976 periods (negative PDO phase), and relatively warm conditions 1926–1945 and 1977–1998 (positive PDO phase) (Wang et al., 2017).

Previous studies on rainfall variability have mainly focused on the association with ENSO. In a recent review, Cai et al. (2020) analyzed the impacts of ENSO, which drives a pattern of stationary Rossby wave-trains that promotes rainfall anomalies. During El Niño years, the induced Rossby wave-train promotes heavy rainfall and flooding across southern Brazil, Uruguay, and northern Argentina. In northern Argentina, the relationship between rainfall and ENSO events is particularly strong from September to January, which is the late dry season and early rainy season (Hurtado and Agosta, 2021; Montecinos et al., 2000). However, few studies have examined the long-term relationship between ENSO and rainfall. In a recent study by Hurtado et al. (2020), with a multi-breakpoint detection analysis jumps in monthly series of total rainfall in Argentina were identified in the mid 1950s, 1976/77 and early 1980s, but TUC data series was not analyzed there. This study associated the mid-1950s and early 1980s breakpoints with ENSO, while the 1976/77 case was linked to a PDO phase shift, from negative to positive. Most of the existing research on the ENSO-rainfall relationship in this region has focused on monthly/seasonal total rainfall and on central and eastern Argentina, where the linkage is stronger and sustained over time. There is then a gap in our understanding of the relationship between ENSO and rainfall in the NWA region that this study aims to address.

TUC daily rainfall data is not publicly available and, in consequence, has been less studied than other series in the literature. Scardilli et al. (2017) analyzes total rainfall, wet days and wet spells for TUC from the daily record and identifies interdecadal variations which result in a positive trend but without analyzing the causes. They also detect jumps or breakpoints between 1954 and 1956, in agreement with findings for annual total rainfall in northern Argentina (Minetti and Vargas, 1998; Hurtado et al., 2020). For NWA, Ferrero and Villalba (2019) analyzed centennial monthly records of rainfall in NWA, finding positive jumps between 1956 and 1960, and between 1973 and 1979, and negative jumps between 2008 and 2013. These breakpoints seem to be linked to phase shifts in ENSO and PDO.

ENSO influence on the interannual variability of rainfall in the NWA region is controversial. Rivera and Penalba (2015), using the Standardized Precipitation Index for the 1961–2008 period concluded that NWA do not show any relationship with ENSO. Marwan et al. (2003) found that the annual total precipitation in TUC is weakly influenced by ENSO. For the 1979–1999 period, they showed that rainfall anomalies in NWA associated with ENSO have a dipolar structure, with positive anomalies over the east and negative on the west in El Niño-years, and vice versa in La Niña-years. TUC happens to be in the transition zone and thus present weak anomalies. However, Sierra and Perez (2000) show that ENSO is a predictor of corn yield production for the 1965–1997 period in Tucuman. They found that Niño 3.4 is the best ENSO-related predictor, with a significant-positive correlation since May and peaking in November. Trauth et al. (2003) detected that during El Niño events 1965–1966, 1969–1970 and 1982–1983 the TUC annual rainfall was higher than its mean. Hence, there is evidence suggesting that ENSO may serve as a predictor of rainfall variability in TUC, although the outcomes vary depending on the time period and index used for the analysis. The precipitation patterns induced by ENSO may be missed in shorter-term studies. Therefore, the use of a centennial rainfall record proves valuable in identifying distinct periods when ENSO exhibits a stronger association.

It should be noted that the onset of summer rainfall in TUC is influenced by the Subtropical Oscillation, which is a low-frequency (20–26 year) variability in land-ocean moisture transport. This oscillation can modulate, or mask, the ENSO effects (Vargas et al., 2002). This signal is observed in total summer rainfall, but not in Rx1 (Medina et al., 2021), implying that different processes may drive the variability in each rainfall metric. It is plausible then, that the connection between ENSO and Rx1 in TUC could be more pronounced compared to the total seasonal or annual rainfall.

In contrast to monthly and annual total rainfall, much less works exist about the association between ENSO and extreme-daily rainfall. Among them, Grimm and Tedeschi (2009) analyzed a large set of daily station rainfall data and found that ENSO-related changes in the frequency of extreme-daily rainfall events are generally coherent with total monthly rainfall changes. However, since significant changes in daily-extremes were much more extensive, they argue that the highest sensitivity to ENSO seems to be in the extreme-daily precipitation instead of monthly totals. Li et al. (2020) in a global study examines hourly rainfall detecting that ENSO modulates rainfall mainly through the number of rainy hours. Thus, unlike the total rainfall, there may be significant associations between ENSO and Rx1 in TUC.

In the multidecadal scale, extreme-rainfall changes in Argentina in the last decades are significantly influenced by the combination of the multi-decadal variability and long-term trends associated with the tropical oceans (Robledo et al. 2020). The interaction between the above mentioned PDO and the Atlantic Multidecadal Oscillation (AMO) drives the rainfall variability in a way that positive PDO and negative AMO enhances the moisture transport from tropics to Argentina (Barreiro et al, 2014). AMO is a near-global scale mode of multidecadal climate variability with alternating warm and cool phases over large parts of the Atlantic Ocean in the Northern Hemisphere (Knight et al., 2006). Namely, AMO is the equivalent to PDO, but in the Atlantic North Ocean.

Regarding other rainfall forcings, there is the Southern Annular Mode (SAM) which is known to affect the Southern Hemisphere (Thompson and Wallace, 2000; Fogt and Marshall, 2020). SAM positive phase is associated with anomalously low-pressure over Antarctica and high-pressure over the mid-latitudes of the Southern Hemisphere, and vice versa during its negative phase. The study of SAM is relatively recent, with the initial investigations dating back to the late 1970s and early 1980s (Fogt and Marshall, 2020). SAM effects have been identified in NWA neighboring regions. It alters the strength and position of cold fronts and mid-latitude storm systems (González et al., 2017), with positive (negative) SAM phases inhibiting (favoring) the passage of cold-fronts over Argentina. According to Garbarini et al. (2021) spring rainfall in Central and Northern Argentina is enhanced during negative SAM phases, while in summer this signal decreases. SAM effects in rainfall were also detected in Uruguay, Brazil and Central Chile (Vasconcellos and Cavalcanti, 2010; Garreaud et al., 2020; Reboita et al., 2021).

In this research, the variability of maximum daily rainfall and its connection with ENSO and SAM is explored using an approach taking into account their non-stationary nature. We identify periods where the correlation is not statistically significant, as well as periods where either ENSO or SAM exhibits a more significant correlation with rainfall variability. We test if this alternating association may be linked to PDO and AMO favoring ENSO or SAM connection depending on their phases: positive or negative. There could even be an effect of PDO and AMO phase combination (positive PDO with negative AMO, for example), which would favor or inhibit an ENSO-Rx1 or a SAM-Rx1 association. This work is structured as follows: Section 2 provides a description of the study region, Sections 3 and 4 detail the data and methodology used for the analysis, Section 5 presents the results, and Section 6 summarizes the main findings and draws conclusions.

The location of TUC and NWA regional topography are shown in Fig. 1. TUC is located ~ 500 m above sea level, in the transition zone between the Chaco Plains and the Puna Plateau. The region has a quasi-monsoonal precipitation cycle with a maximum in summer months, accounting for more than 50% of the annual precipitation (Poblete et al., 1989; Medina et al., 2021). In the austral summer, the deepening of the continental low pressure center is the main factor responsible for the occurrence of the precipitation maximum in NWA due to an easterly/northeasterly moisture flux (Vargas et al., 2002). Rainfall is maximum in TUC and nearby regions as a result of the Andes Mountain Range slope (Bookhagen & Strecker, 2008). The annual rainfall at this station serves as an indicator of the NWA’s precipitation variability (Vargas et al., 2002; Ferrero and Villalba, 2019), highlighting the importance of analyzing this extended record.

3.1 Daily rainfall record

We analyze the series of Rx1 values for the December-January-February-March season (DJFM) season, obtained from a long daily-rainfall record. Data provided by Estación Experimental Agroindustrial Obispo Colombres (EEAOC), a government agency of Tucumán province, was used in this work as representative of TUC rainfall. The daily record spans the 1911–2019 period and was obtained from a conventional station located in El Colmenar (26.8°S, 65.2°W), in the north of Tucuman's capital city. As we analyze the DJFM season, and the years were named as the calendar-year of the January month, the period analyzed is 1912–2019.

Initially, the meteorological station was operated by the Servicio Meteorologico Nacional (SMN) which collected the data until the 1970s. Then, the station was moved to Tucumán airport, but the EEAOC continued operating in El Colmenar. Therefore, there has been no change in the station's location. In addition, this time series has been used and has successfully passed quality controls in previous studies (Penalba and Robledo, 2010; Scardilli et al., 2017; Saurral et al., 2017).

To ensure the quality of our data, we checked some aspects of the daily rainfall record. We detect missing data only in the period of July-December 1943, which represents 0.5% of the entire record. As our analysis focuses solely on the DJFM, the missing data lowers to 0.2%. The time series was also checked for obvious erroneous data (e.g., negative rainfall values or values excessively high) and we did not find any. Further, for each month we compute monthly total rainfall, maximum daily rainfall and maximum consecutive days without rainfall. All values that deviated four standard deviations or more respect to the monthly mean were flagged as suspected of mistakes, a criterion used in other works based on daily rainfall series (Cavalcante et al., 2020; Medina et al., 2023). Flagged data were compared with neighboring stations, when available, to decide about its acceptance. No outliers were detected in the total rainfall for DJFM season. For maximum daily rainfall only one outlier was detected on date 3/3/1973 but we confirmed its value with a neighboring station (at ~ 40 km distance). In the case of consecutive dry days, outliers were detected in January of 1917 and March of 1933. In the first case, we were unable to compare with neighboring stations' values, while in the second case we compared them with one station at ~ 140 km distance. In both cases, data was rejected for caution. However, even with these considerations, the proportion of missing data increased to only 0.6%, and therefore should not have a substantial impact on our results. From this final daily rainfall record, Rx1 for the DJFM season was assessed.

Figure 2a displays the yearly count of annual Rx1 occurrence in each month, using the whole daily TUC record. January has the highest count, with 35 out of the 108 years of Rx1 occurring in this month. Other months with high counts include December, February, and March. In total, the period from December to March (DJFM) covers 88% of annual Rx1 occurrence, and this is the reason for selecting it for our study.

Figure 2b shows Rx1 series. High variability is observed in the raw series with a mean value of 89 mm, and extreme values of 189 mm (in March of 1973) and 36.5 mm (in February of 1916). The relative variation, estimated as the ratio between the standard deviation and the mean, is 31%. On the other hand, the series of anomalies are shown in Fig. 2c, showing the alternancy of years with higher and lower Rx1 values. Exists a major predominance of higher values from the late 1950s. Anomalies with a 10-year running mean (Fig. 2d) indicates multidecadal variability consisting of higher Rx1 values since 1960, especially in the 1980s and 1990s decades, while from 2010’s series seems to change again to lower Rx1 values.

3.2 Climate indices

Four large-scale modes of variability are considered in this study: ENSO, SAM, PDO, and AMO. Time series were averaged for the DJFM season in coincidence with the season selected for Rx1. A brief description of each mode and its availability is provided below.

- ENSO indices: temperature anomalies in regions Niño 1.2 and Niño 3.4 (available on the NOAA’s webpage: https://psl.noaa.gov/gcos_wgsp/Timeseries/), computed using the HadISST1 dataset (Rayner et al., 2003) referred to the 1981–2010 period;

- SAM index: three time series were considered: (1) SAM-20 from the The Twentieth Century Reanalysis Project (20CRV2c) (Gong and Wang (1999) methodology) encompassing the 1912–2011 period (available on: https://psl.noaa.gov/data/20thC_Rean/timeseries/monthly/SAM/), (2) SAM-M from the British Antarctic Survey which is a station-based index (Marshall (2003) methodology) encompassing the 1957–2019 period (available on: https://legacy.bas.ac.uk/met/gjma/sam.html), and (3) a concatenated time series consisting in SAM-20 for the 1912–1956 period jointly with the SAM M for the 1957–2019 period. Both SAM-20 and SAM-M are computed as the difference of the zonal means at 40°S and 65°S, but in the first case gridded data are used, while in the other case six unevenly distributed stations are considered. We performed the analysis of this work with the 3 time series and results were similar. However we chose the concatenated time series because it showed comparatively better results and had a longer temporal cover that only SAM-M;.

- PDO index (Mantua, 1999) is defined as the leading principal component of mean SST anomalies for the Pacific Ocean to the north of 20°N latitude, with the global mean SST anomaly first removed for each month in order to reduce the influence of the long-term trends (available on: http://research.jisao.washington.edu/data_sets/pdo/);

- AMO index: area weighted average of the North Atlantic Ocean (0–70°N) temperature calculated from the Kaplan SST dataset (Kaplan et al., 1998) (available on: https://www.psl.noaa.gov/data/timeseries/AMO/).

A correlation analysis is performed to explore whether there is a linear association, which may be not stable in time, between Rx1 and two of the large-scale variability modes: ENSO and SAM. This is complemented with the analysis of PDO and AMO phase shifts as forcings of the possible non-stationary association. For the correlation analysis, a new approach is proposed to a known method, which is based on the triangle-representations used in Fink et al. (2010) and Bahaga et al. (2019). This representation depicts the correlation coefficient over all possible combinations of starting dates and time window lengths. Further performing the triangle-representations, the novelty in this work is to focus on identifying the period within the Rx1 time-series with the most significant correlation and, in addition, the longest possible period when a significant correlation can be obtained. The significance threshold in this work is the 95% of confidence level. The sub-periods with highest significance in the correlation for Rx1-ENSO and Rx1-SAM are then analyzed in the context of phase shifts detected on PDO and AMO. Next sections describe the correlation analysis for ENSO and SAM, and subsequently the analysis of PDO and AMO roles in the non-stationary character of the relationships.

4.1 Correlation analysis and identification of the periods with the most significant relationship

To detect significant associations within sub-periods, a common method is employing a temporal moving-window of a fixed length, shorter than the whole period. This fixed window length (WL) is usually selected based on an identified periodicity or some other criteria based on an a-priori knowledge of the system under study. Instead of using this conventional method, we analyzed all possible WL values within the length of our time series. This methodology is the triangle-representations used in Fink et al. (2010) and Bahaga et al. (2019). We do a different focus: searching for the lowest p-value, rather than only observing the correlation values. In this case, the Pearson correlation coefficient (RHO) is assessed, together with the corresponding p-value from the Student t-test as its significance measure. For each WL, the correlated time series were first linearly detrended.

As a first step, a matrix containing four rows is constructed based on the correlation between Rx1 and each ENSO and SAM time series. The 1st row contains the start-year of the window, the 2nd WL, being 20 years the shortest value used for WL, the 3rd the corresponding RHO value, and the 4th the corresponding p-value.

Within the matrix, the lowest p-values correspond to a certain WL with given start- and end-years, which determine the sub-period of minimum p-value (PmPV). For the PmPV, for definition, the RHO value has the greatest significance between all RHO values of all possible sub-periods. This means that PmPV is the longest period when an association between Rx1 and either ENSO or SAM is more likely to exist. The advantage of this criterion over detecting the highest RHO, is avoiding the identification of short periods which pass the significance test, but that do not correspond to the period with the higher significance for the linear association.

The period with maximum WL (PMWL) for which a statistically significant RHO value can be obtained is also identified. In general, PMWL contains PmPV with identical RHO sign, implying that the correlation can be detected along a longer time than indicated by PmPV, but in general a weakened outside the PmPV period can be observed. The difference in the length between PmPV and PMWL is an indicator of the speed of weakening in the relation between Rx1 and the climate index (ENSO or SAM), as the PmPV is enlarged up to the PMWL. On the other hand, it can occur that the PmPV is not contained in the PMWL and that RHO signs result opposite. In these cases, it can be concluded that the relationship between Rx1 and the climate index is chaotic and could even not exist. In addition to the PmPV and PMWL, the triangle-representations for Rx1-ENSO and Rx1-SAM are plotted to show all results of RHO, p-value and their temporal variations.

4.2 Two methods to determine PDO and AMO role on a non-stationary association

If it is observed that the correlation becomes evident only during certain sub-periods, the association between Rx1 and ENSO or SAM could be nonstationary. In this case, an additional forcing can drive the changing associations of Rx1. The easiest evaluation of the possible role of an additional forcing would be to consider a multilinear correlation instead. However, if it has a non-linear interaction with Rx1, this method would fail in revealing its role. In our case, there are two possible additional forcings which could interact with Rx1 in a non-linear way, by favoring or inhibiting the effects of ENSO and SAM: PDO and AMO. As was mentioned in the introduction, PDO and AMO drive the multidecadal variability of the rainfall in northern Argentina. Thus, it could be thought that drives changes in the teleconnections patterns associated with ENSO and SAM.

One possibility is to check if the running correlations between Rx1 and ENSO or SAM become significant and with the same sign along a certain PDO and/or AMO phase (positive or negative). This is done qualitatively, by plotting the running correlation coefficient simultaneously with PDO and AMO smoothed series.

A second possibility is to assess the correlations between Rx1 and ENSO or SAM, by filtering first the time series according to the phases of PDO and AMO. Thus, we split the series by PDO phase (+ PDO: positive PDO phase, and –PDO: negative PDO phase), by AMO (+ AMO: positive AMO phase, and –AMO: negative AMO phase), and by their combinations (+ PDO and + AMO, +PDO and –AMO, -PDO and –AMO, -PDO and + AMO). After splitting the time series, RHO is assessed for each case. If the PDO and/or AMO phase favors or inhibits ENSO effect on Rx1, for example, then a significant RHO should be obtained for one of the phases, and a null or non-significant one in the other phase. This analysis is similar to that used by Labitzke (1987) and Labitzke and van Loon (1988) to detect the non-linear QBO role on the association between the polar stratospheric temperature during winter and solar activity.

For the analysis above proposed, a 10-year running mean of PDO and AMO averaged over DJMF was performed to focus on the multidecadal component of the series. Thus, the years which correspond to each PDO and AMO phases for the splitting process are:

+PDO: 1922–1944, 1978–2006 and 2011–2014

-PDO: 1916–1921, 1945–1977, and 2007–2010

+AMO: 1930–1964 and 1999–2014

-AMO: 1916–1929 and 1965–1998

-PDO and –AMO: 1916–1921 and 1965–1977

-PDO and + AMO: 1945–1964 and 2007–2010

+PDO and –AMO: 1922–1929 and 1978–1998

+PDO and + AMO: 1930–1944, 1999–2006 and 2011–2014

The initial and final years are 1916 and 2014, instead of 1912 and 2019 due to in the running mean process (with centered window) the half of the WL is lost at the beginning and the other half at the end of the time series.

5.1 Correlation analysis of Rx1 with ENSO and SAM

Figures 3 show the triangle-representations in terms of start-year and WL, for Rx1 vs Niño 1.2, Niño 3.4 and SAM. In each figure, PmPV and PMWL values are identified (black-filled and black-empty stars, respectively). This representation of the correlations can be interpreted as follows: moving from the start-year horizontally to the right in the WL-axis, colors represent the RHO values for start-year as increasing the WL. On the other hand, moving from the bottom vertically upward yields RHO for different start-years with the fixed WL denoted on the abscissa. Only correlations at the 95% of significance level (p < 0.05) are shown in the color scale.

For Niño 1.2 and Niño 3.4 (Fig. 3) in the WL = 20, the start-years with positive significant RHO values, between 0.6–0.8, are present around the 1950s. As the WL is increasing, positive RHO values are observed for start-years before the 1950s, due to the fact these windows encompass 1950s when a noticeable correlation exists. For WL > 30 years approximately, RHO values tend to steadily decay under 0.6 up to loss of statistical significance. On the other hand, smaller periods with negative RHO are observed both for Niño 1.2 and Niño 3.4. In both cases is observed a RHO ~ -0.4, in the Niño 1.2 start in 1987 for WL = 20 (this is the 1987–2006 period), while for Niño 3.4 start in 1923 for WL = 21 (this is the 1923–1942 period). Thus, although ENSO indices show mainly a positive RHO, negative RHO values are also observed, but with a minor temporal extent and lower absolute values.

For SAM index (Fig. 3) there are only positive RHO values. In the WL = 20, the start-years with significant RHO between 0.4–0.6 are present mainly around the 1970s and 1980s. As WL increases up to the maximum possible, RHO values remain above 0.4 only for the start-years corresponding to 1970’s. This indicates that from around the 1970s up to almost the end of the series, a significant and steady positive RHO exists between Rx1 and SAM.

To summarize the information provided by the triangle-representations of Fig. 3, in Fig. 4 are shown the PmPV and PMWL for comparing the periods with more noticeable influence of ENSO and SAM in Rx1. The PmPv results clearly show a change in the association between Rx1 and climate indices. The PmPV for ENSO indices (Niño 1.2 and Niño 3.4) is 1945–1974, while for SAM is 1974–2007, in both cases with RHO ~ 0.6. Thus, around 1974 a change in the coupling of Rx1 occurred, from a positive correlation with ENSO to a positive correlation with SAM. On the other hand, in concordance with the triangles-representations, the PMWL of Fig. 4 shows that for WL greater than PmPV, the RHO values decrease abruptly. This is due to, as WL increases, years when the correlation is lower and even non-significant are incorporated to the window. This comparison between the PmPv and PMWL confirms that the main influences of ENSO and SAM are confined specially to their corresponding PmPV.

The Rx1 and Niño 3.4 series for the PmPV, i.e. 1945–1974, are shown in Fig. 5a, and for the whole period 1912–2019 in Fig. 5b. The corresponding scatter plots (Rx1 vs Niño 3.4) are shown in Figs. 5c-d. Similarly for SAM this is shown in Fig. 6, for the 1974–2007 period and for the whole period 1912–2019. It can be clearly noticed that the association along the PmPV occurs on interannual timescales (Figs. 4a and 5a), and that this co-variability decays when the whole period is considered (Figs. 4b and 5b). The scatter plots (Fig. 4c-d, and 5c-d) also clearly show the linear association observed during the PmPV.

PmPV in Fig. 4, together with the triangle representations in Fig. 3 demonstrates a non-stationary statistical association between Rx1 and either ENSO or SAM. Also, the figures show that the sub-periods with positive association between Rx1 and each of them are non coincident. In the case of Niño 1.2, a little period with a negative RHO is observed inside the PmPV of SAM, however it shows less extent than SAM and lower RHO magnitude.

5.2 PDO and AMO roles in the non-stationary relationships of Rx1 with ENSO and SAM

Figure 7 shows the RHO values of a running correlation (WL = 20) for Rx1 vs ENSO and Rx1 vs SAM indices. WL is selected in 20 years because it is closer to the approximate length of PDO phases (see periods of + PDO and -PDO). Also in Fig. 7, the phases of PDO index are shown, which were obtained applying a 10-year running mean to PDO series. It can be noticed that, during the extended period 1945–1977 of the negative PDO phase, a significant positive RHO is only observed between Rx1 and ENSO indices (blue and red points). Contrary, during the positive PDO phase along 1978–2007, RHO becomes notably significant and positive for Rx1 vs SAM (green points), while a few points with negative RHO and just above the significance threshold appear for Rx1 vs Niño 1.2 (blue points). For the other time interval of negative PDO phase, this is 1922–1944, also a few points with negative RHO values between Rx1 and Niño 3.4 are observed, while non-significant correlation with SAM is observed. No clear associations are noticed if AMO is considered instead of PDO (not shown).

To further analyze for a possible influence of the PDO and AMO phases over the correlation of Rx1 with ENSO and SAM, Table 1 shows the RHO values when the series are separated according to the phases of PDO and AMO. These results corroborate that the PDO negative phase favors the Rx1-ENSO positive correlation. On the other hand, the PDO positive phase inhibits this positive correlation and seems to favor a negative correlation of lower magnitude in comparison with the other case. In addition, in the positive PDO phase appears a positive correlation between Rx1 and SAM. Contrasting with results of Fig. 7, in this case the RHO values are lower, which is because all years with either + PDO or -PDO were joined and some of these years can not show a relation Rx1-ENSO or Rx1-SAM. When Rx1, ENSO and SAM are separated according to AMO phases and then correlated, the results do not support a consistent role of their different phases. On the other hand, when the combination of PDO and AMO phases is used to split the data series, only ENSO indices show significant RHO values for negative PDO, but regardless of the AMO phase. Thus, AMO seems to have no influence in the association between Rx1 and ENSO or SAM.

The analysis of Fig. 7 and Table 1 shows that PDO is possibly the forcing that can explain the non-stationarity of the correlation for Rx1 vs ENSO and Rx1 vs SAM. Negative PDO phase favors the positive Rx1-ENSO association, while the positive PDO phase favors a positive Rx1-SAM and a weaker negative Rx1-ENSO association. However, in the case of SAM, significant correlation was not found for the first negative PDO phase (1922–1944), which can be due to deficiencies in the SAM index used or it can be due to the absence of a relationship in this period. In relation to the former hypothesis, the relationship is clearly observed in the positive PDO phase when the station-based Marshall’s SAM is available (after 1957), while it is not observed for the period corresponding to the 20CRV2c SAM (prior to 1957). It is known that the sensitivity of the relationship between SAM and rainfall depends on the choice of SAM index and the data set used to calculate it (Ho et al., 2012).

Table 1

Correlation coefficients between Rx1 and Niño 1.2, Niño 3.4 and SAM considering the complete period (1912–2019), and the periods split according to: the PDO positive (+ PDO) and negative (-PDO) phases, the AMO positive (+ AMO) and negative (-AMO) phases, and the four possibilities of the PDO and AMO combined positive and negative phases. The number of years used in each correlation is indicated for each phase or phase combination. Only significant values (95% confidence level, p-value < 0.05) are shown.
	Rx1 vs. Niño 1.2	Rx1 vs. Niño 3.4	Rx1 vs. SAM
Complete series	-	-	-
+ PDO (56 years)	-0.29	-0.30	0.28
-PDO (43 years)	0.48	0.43	-
+AMO (51 years)	-	-	-
-AMO (48 years)	-	-	-
+PDO & -AMO (29 years)	-	-	-
-PDO & +AMO (24 years)	0.65	0.45	-
+PDO & +AMO (27 years)	-	-	-
-PDO & -AMO (19 years)	0.44	0.46	-

In this work, we contribute with the analysis of a centennial time series of northern Argentina, that has been less analyzed in the literature. The proposition of determining the PmPV as a new approach in the correlation analysis, allowed us to detect a time-varying coupling of Rx1 with ENSO and SAM. Specifically, we detect a shift from a significant positive correlation between Rx1 and ENSO in the period 1945–1974, and between Rx1 and SAM in the period 1974–2007. A priori, it could be said that the change in the coupling of Rx1 in 1974 can be related with a phase change, from negative to positive in the PDO (which occurred in 1976–1977) and whose impact was detected in various rainfall series of Argentina (Hurtado et al., 2020). Our new contribution in this sense, is the finding of the impacts of PDO in the associations of Rx1 with ENSO and SAM. In our knowledge, Rx1 is a variable that has not yet been analyzed with this focus in the NWA region.

Specifically, the period 1945–1974 coincides with an extended negative phase of the PDO from 1945 to 1977. In addition, the PDO positive phase in the consecutive period, lasting from 1978 to 2007, is almost coincident with the PmPV of the correlation Rx1-SAM, suggesting that it could favor the association between Rx1 and SAM, while favor a weaker and sign-inverted relation Rx1-ENSO. This non-stationary association is the reason for which most works analyzing rainfall variability in NWA (Rivera and Penalba, 2015; Lauro et al., 2019; Medina et al., 2021), have not found significant correlations with ENSO or SAM.

The decay in the association between rainfall and ENSO since the 1970s was also observed in other regions of the world, e.g. in the Indian monsoon (Kucharski et al., 2007). In contrast, according to Torralba et al. (2015), in Northeastern Brazil the association between rainfall and ENSO was enhanced from the 1970s onwards. A global climate shift in the 1970s would have changed the teleconnection patterns of ENSO around the globe affecting the link with the rainfall. Results of Barreiro et al. (2014) indicate that the Atlantic and Pacific influence can explain a large transition from dry to wet decades over Northern Argentina during the beginning of the 1970’s. This global shift also is observed in TUC rainfall data, but we focus on the changes in the relation Rx1-ENSO and Rx1-SAM rather than in the rainfall values. We show that with only the Pacific’s influence is enough to explain the change in the association between Rx1 and either ENSO or SAM in TUC. More investigation in this sense is needed to determine the physical processes that favor the changes in these associations by the PDO phases.

RHO values obtained for Rx1 vs ENSO in the negative PDO phase are in agreement with the observed response of the rainfall to ENSO events in Eastern Argentina. That is, Niño (Niña) years implies higher (lower) rainfall values. In our particular case focusing on Rx1, when PDO is in the negative phase, the Niño (Niña) years implies heavier (lighter) daily-rainfall. The relationship is inverse and weaker in the positive phase of PDO.

On the other hand, RHO values between Rx1 and SAM are in contrast with the negative association observed in other works about Argentina and Uruguay rainfall (Garbarini et al., 2021). In our case, when PDO is in the positive phase, SAM implies heavier (lighter) daily rainfall. Positive RHO values were also detected for Southeastern Brazil, where extreme precipitation is associated with anomalous circulation over the region that is indirectly affected by SAM (Vasconcellos and Cavalcanti, 2010; Reboita et al., 2021). A deeper analysis is needed to determine the mechanism of association in the case of TUC. We contribute with our results to the knowledge of SAM influence in the rainfall in a location of Argentina, a region of the world where the principal focus of the existing works is the link to the ENSO variability.

Results of this work show that Rx1 has certain predictability in relation to large-scale variability modes, despite having more temporal-spatial variability in comparison with total seasonal or annual rainfall. Rx1 shows strong interannual and multidecadal variability, linked in the first case with ENSO and SAM in a non-stationary relationship conditioned by the PDO phases. About the multidecadal variability, is clearly observed a shift to years with higher Rx1 values from late 1950s, and specifically in the 1980s and 1990s, with an apparent return to lower Rx1 values from 2010s. This result is in agreement with Ferrero and Villalba (2019) that analyze seasonal total rainfall in the NWA region, detecting a a non-significant wet shift in the 1950s, a noticeably wet period starting in the 1970s and a dry period beginning between 2008 and 2013, almost coincident to our detection. Although we do not focus on this multidecadal behavior, the wetter decades of 1980s and 1990s are encompassed in the positive PDO phase. In these years, Rx1 values show a positive correlation with SAM in conjunction with a weak negative correlation with ENSO (Niño 1.2). The posterior return of lower Rx1 values coincide with a change from positive to negative PDO phase in the late 2000s.

We remark the importance of analyzing the link between rainfall variability and climate forcings taking in account the possible non-stationarity in the relationships. This can be clearly observed in the case of TUC series analyzed here, in which when the whole series is correlated without split in sub-periods, non-significant or low correlations are found. Finally, results highlight that processes involved in the variability of rainfall intensity in this city are partly related to global variability modes, opening opportunities to enhance predictions in relation to the flood hazard.

Acknowledgments

The authors acknowledge research project PIP 2957 (CONICET). We thank Estación Experimental Agroindustrial Obispo Colombres (EEAOC) for providing Tucuman rainfall data.

Statements & Declarations

Funding

This work was supported by project PIP 2957 (CONICET).

Competing Interests

The authors have no relevant financial or non-financial interests to disclose.

Author Contributions

Conceptualization: all authors; Methodology: all authors; Formal analysis and investigation: Franco D. Medina; Writing - original draft preparation: Franco D. Medina; Writing - review and editing: all authors; Supervision: Ana G. Elias.

Data Availability

Data of climate indices are available on the followings webpages: ENSO (https://psl.noaa.gov/gcos_wgsp/Timeseries/), SAM (https://psl.noaa.gov/data/20thC_Rean/timeseries/monthly/SAM/ and https://legacy.bas.ac.uk/met/gjma/sam.html), PDO (http://research.jisao.washington.edu/data_sets/pdo/) and AMO (https://www.psl.noaa.gov/data/timeseries/AMO/). Rainfall data is not publicly available due to authors have not permission to share it, but can be request to Estación Experimental Agroindustrial Obispo Colombres (EEAOC, Tucumán, Argentina, website: https://www.eeaoc.gob.ar/).

Avashia V, Garg A (2020) Implications of land use transitions and climate change on local flooding in urban areas: An assessment of 42 Indian cities. Land use policy 95:104571. https://doi.org/10.1016/j.landusepol.2020.104571
Bahaga TK, Fink AH, Knippertz P (2019) Revisiting interannual to decadal teleconnections influencing seasonal rainfall in the Greater Horn of Africa during the 20th century. Int J Climatol 39:2765–2785. https://doi.org/10.1002/joc.5986
Barreiro M, Díaz N, Renom M (2014) Role of the global oceans and land–atmosphere interaction on summertime interdecadal variability over northern Argentina. Clim. Dyn., 42 (7) (2014), pp. 1733–1753, https://doi.org/10.1007/s00382-014-2088-6
Cai W, McPhaden MJ, Grimm AM et al (2020) Climate impacts of the El Niño–Southern Oscillation on South America. Nat Rev Earth Environ 1, 215–231 (2020). https://doi.org/10.1038/s43017-020-0040-3
Cavalcante RBL, da Silva Ferreira DB, Pontes PRM, Tedeschi RG, da Costa CPW, de Souza EB (2020) Evaluation of extreme rainfall indices from CHIRPS precipitation estimates over the Brazilian Amazonia. Atmos Res 238:104879. https://doi.org/10.1016/j.atmosres.2020.104879
de la Casa AC, Ovando GG, Díaz GJ (2019) Interannual variability of seasonal rainfall in Cordoba, Argentina, evaluated from ENSO and ENSO Modoki signals and verified with MODIS NDVI data. SN Appl. Sci. 1, 1624 (2019). https://doi.org/10.1007/s42452-019-1650-6
Fernández D, Lutz M (2010) Urban flood hazard zoning in Tucumán Province, Argentina, using GIS and multicriteria decision analysis. Engineering Geology, Volume 111, Issues 1–4, 2010, Pages 90–98, ISSN 0013-7952. https://doi.org/10.1016/j.enggeo.2009.12.006
Ferrero M, Villalba R (2019) Interannual and Long-Term Precipitation Variability Along the Subtropical Mountains and Adjacent Chaco (22–29° S) in Argentina. Front Earth Sci 7:148. https://doi.org/10.3389/feart.2019.00148
Fink AH, Schrage JM, Kotthaus S (2010) On the potential causes of the nonstationary correlations between West African precipitation and Atlantic hurricane activity. J Clim 23:5437–5456. https://doi.org/10.1175/2010JCLI3356.1
Fogt RL, Marshall GJ (2020) The Southern Annular Mode: Variability, trends, and climate impacts across the Southern Hemisphere. WIREs Clim Change. 2020; 11:e652. https://doi.org/10.1002/wcc.652
Garbarini EM, González MH, Rolla AL (2021) Modulation of seasonal precipitation in Argentina by the South Pacific high. Int J Climatol 41 (Suppl 1E3279–E3297. https://doi.org/10.1002/joc.6924
Garreaud RD, Boisier JP, Rondanelli R, Montecinos A, Sepúlveda HH, Veloso-Aguila D (2020) The Central Chile Mega Drought (2010–2018): A climate dynamics perspective. Int J Climatol 40:421–439. https://doi.org/10.1002/joc.6219
Gong D, Wang S (1999) Definition of Antarctic oscillation index. Geophys Res Lett 26(4):459–462. https://doi.org/10.1029/1999GL900003
González M, Garbarini E, Rolla A, Eslamian S (2017) Meteorological drought indices: rainfall prediction in argentina. In: Eslamian S (ed) Handbook of Drought and Water Scarcity. Principle of Drought and Water Scarcity, vol 1. Taylor & Francis Publishing (CRC Group, Abingdon, England, pp 540–567. Chapter 29
Grimm AM, Tedeschi RG (2009) ENSO and Extreme Rainfall Events in South America. J Clim 22(7):1589–1609. https://doi.org/10.1175/2008JCLI2429.1
Ho M, Kiem AS, Verdon-Kidd DC (2012) Hydrol Earth Syst Sci 16:967–982. https://doi.org/10.5194/hess-16-967-2012. The Southern Annular Mode: a comparison of indices
Hurtado S, Zaninelli P, Agosta E (2020) A multi-breakpoint methodology to detect changes in climatic time series. An application to wet season precipitation in subtropical Argentina, Atmospheric Research, Volume 241, 2020, 104955, ISSN 0169–8095, https://doi.org/10.1016/j.atmosres.2020.104955
Hurtado SI, Agosta EA (2021) El Niño Southern Oscillation-related precipitation anomaly variability over eastern subtropical South America: Atypical precipitation seasons. Int J Climatol. 2021; 41: 3793– 3812. https://doi.org/10.1002/joc.6559
Kaplan A, Cane M, Kushnir Y, Clement A, Blumenthal M, Rajagopalan B (1998) Analyses of global sea surface temperature 1856–1991. J Phys Res, 103, 18,567 – 18, 589.
Knight JR, Folland CK, Scaife A (2006) Climate impacts of the Atlantic Multidecadal Oscillation. Geophys Res Lett 33:L17706. 10.1029/2006GL026242
Krepper CM, García NO (2004) Spatial and temporal structure of trends and interannual variability of precipitation over La Plata Basin. Quatern Int 114:11–21. https://doi.org/10.1016/S1040-6182(03)00038-7
Kucharski F, Bracco A, Yoo JH, Molteni F (2007) Low-frequency variability of the Indian monsoon–ENSO relationship and the tropical Atlantic: The “weakening” of the 1980s and 1990s. J Clim 20:4255–4266. https://doi.org/10.1175/JCLI4254.1
Labitzke K (1987) Sunspots, the QBO, and the stratospheric temperature in the north polar region. Geophys Res Lett 14:135–137. https://doi.org/10.1029/GL014i005p00535
Labitzke K, van Loon H (1988) Associations between the 11-year solar cycle, the QBO and the atmosphere. Part I: the troposphere and stratosphere in the northern hemisphere in winter. J Atmos Terr Phys 50:197–206. https://doi.org/10.1016/0021-9169(88)90068-2
Lauro C, Vich A, Moreiras S (2019) Streamflow variability and its relationship with climate indices in western rivers of Argentina. Hydrol Sci J 64(5):607–619. https://doi.org/10.1080/02626667.2019.1594820
Li XF, Blenkinsop S, Barbero R et al (2020) Global distribution of the intensity and frequency of hourly precipitation and their responses to ENSO. Clim Dyn 54:4823–4839. https://doi.org/10.1007/s00382-020-05258-7
Mantua NJ (1999) The Pacific Decadal Oscillation. A brief overview for non-specialists. Encyclopedia of Environmental Change
Marwan N, Trauth MH, Vuille M et al (2003) Comparing modern and Pleistocene ENSO-like influences in NW Argentina using nonlinear time series analysis methods. Clim Dyn 21:317–326. https://doi.org/10.1007/s00382-003-0335-3
Marshall GJ (2003) Trends in the Southern Annular Mode from observations and reanalyses. J Clim 16:4134–4143. https://doi.org/10.1175/1520-0442%282003%29016<4134%3ATITSAM>2.0.CO%3B2
Medina F, Bazzano F, Heredia T, Elias A (2021) Posibles forzantes de variaciones de largo plazo de la precipitación de verano en Tucumán, Argentina. Meteorológica; 46; 1;26–47. https://doi.org/10.24215/1850468Xe003
Medina F, Zossi B, Bossolasco A, Elias A (2023) Performance of CHIRPS dataset for monthly and annual rainfall-indices in Northern Argentina. Atmospheric Research, Volume 283, 2023, 106545, ISSN 0169–8095. https://doi.org/10.1016/j.atmosres.2022.106545
Minetti JL, Vargas WM, Atmósfera (1998) 11(4), 205–221
Montecinos A, Díaz A, Aceituno P (2000) Seasonal Diagnostic and Predictability of Rainfall in Subtropical South America Based on Tropical Pacific SST. J Clim 13(4):746–758. https://doi.org/10.1175/1520-0442(2000)013<0746:SDAPOR>2.0.CO;2
Newman M, Alexander MA, Ault TR, Cobb KM et al (2016) The Pacific Decadal Oscillation, Revisited. J Clim 29(12):4399–4427. https://doi.org/10.1175/JCLI-D-15-0508.1
Penalba O, Robledo F (2010) Spatial and temporal variability of the frequency of extreme daily rainfall regime in the La Plata Basin during the 20th century. Clim Change 98:531–550. https://doi.org/10.1007/s10584-009-9744-6
Poblete AG, Minetti JL, Sierra AM (1989) La Influencia del Régimen Hidrico Andino-Puneño en el Oasis del Noroeste Argentino y Cuyo. Rev de geofisica 30:137–149
Rayner NA, Parker DE, Horton EB, Folland CK, Alexander LV, Rowell DP, Kent EC, A (2003) Kaplan, Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century. J Geophys Res 108(D14):4407. https://doi.org/10.1029/2002JD002670
Reboita MS, Ambrizzi T, Crespo NM, Dutra LMM, Ferreira GWdS et al (2021) Impacts of teleconnection patterns on South America climate. Ann N Y Acad Sci 1504:116–153. https://doi.org/10.1111/nyas.14592
Rivera JA, Penalba OC (2015) El Niño/La Niña events as a tool for regional drought monitoring in Southern South America. Drought: research and science-Policy interfacing, 293–299. https://doi.org/10.1201/b18077-50
Robledo F, Vera C, Penalba O (2020) Multi-scale features of the co-variability between global sea surface temperature anomalies and daily extreme rainfall in Argentina. Int J Climatol 40:4289–4299. https://doi.org/10.1002/joc.6462
Saurral RI, Camilloni IA, Barros VR (2017) Low-frequency variability and trends in centennial precipitation stations in southern South America. Int J Climatol 37:1774–1793. https://doi.org/10.1002/joc.4810
Scardilli AS, Llano MP, Vargas WM (2017) Temporal analysis of precipitation and rain spells in Argentinian centenary reference stations. Theor Appl Climatol 127:339–360. https://doi.org/10.1007/s00704-015-1631-7
Sierra EM, Pérez SP (2000) Relación entre el fenómeno Enos y los rendimientos de maíz 1965/66-1997/98 en la provincia de Tucumán. Revista de la Facultad de Agronomía‚ 20‚ (2) p.201–206. http://ri.agro.uba.ar/cgi-bin/library.cgi?a=d&c=rfa&d=2000sierraem
Thompson D, Wallace J (2000) Annular Modes in the Extratropical Circulation. Part I: Month-to-Month Variability. J Clim 13(5):1000–1016. https://doi.org/10.1175/1520-0442(2000)013<1000:AMITEC>2.0.CO;2
Trauth M, Bookhagen B, Marwan N, Strecker M (2003) Multiple landslide clusters record Quaternary climate changes in the northwestern Argentine Andes. Palaeogeography, Palaeoclimatology, Palaeoecology, Volume 194, Issues 1–3, 109–121. 0031–0182. https://doi.org/10.1016/S0031-0182(03)00273-6
Torralba V, Rodríguez-Fonseca B, Mohino E, Losada T (2015) The non-stationary influence of the Atlantic and Pacific Niños on North Eastern South American rainfall. Front Earth Sci 3:55. https://doi.org/10.3389/feart.2015.00055
Vargas W, Minetti J, Poblete (2002) A. Low-frequency oscillations in climatic and hydrological variables in southern South America’s tropical-subtropical regions. Theor Appl Climatol 72, 29–40. https://doi.org/10.1007/s007040200010
Vasconcellos FC, Cavalcanti IF (2010) Extreme precipitation over Southeastern Brazil in the austral summer and relations with the Southern Hemisphere annular mode. Atmosph Sci Lett 11:21–26. https://doi.org/10.1002/asl.247
Wang C, Deser C, Yu JY, DiNezio P, Clement A (2017) El Niño and southern oscillation (ENSO): a review. Coral Reefs of the Eastern Tropical Pacific. Springer, Dordrecht, pp 85–106. https://doi.org/10.1007/978-94-017-7499-4_4

Download PDF

Version 1

posted

You are reading this latest preprint version

Non-stationary associations between rainfall intensity and large-scale variability in an east-Andean city over the last 108-years

Status:

Version 1

Abstract

Figures

1. Introduction

2. Study region

3. Data

3.1 Daily rainfall record

3.2 Climate indices

4. Methodology

4.1 Correlation analysis and identification of the periods with the most significant relationship

4.2 Two methods to determine PDO and AMO role on a non-stationary association

5. Results

5.1 Correlation analysis of Rx1 with ENSO and SAM

5.2 PDO and AMO roles in the non-stationary relationships of Rx1 with ENSO and SAM

6. Discussion and conclusions

Declarations

References

Status:

Version 1