1 Introduction

“El Niño” was first introduced to the scientific community in reference to the anomalous climatic event that took place in 1891 along the coast of Peru, described as an abnormal intrusion of warm oceanic water from the north, replacing the normally cold coastal-upwelled water and favoring the occurrence of strong rainfall and flooding in the otherwise arid northern coast of Peru (Carranza 1891). The warm southward ocean flow was named “Corriente del Niño” (Child’s current) in reference to the weaker climatological version of this current that is normally present after Christmas time (Carrillo 1893).

Nowadays, the term “El Niño” (EN) is used as shorthand for referring to the warm phase of the large-scale El Niño–Southern Oscillation (ENSO) coupled ocean–atmosphere phenemenon, characterized by anomalously high SST in the central–eastern equatorial Pacific (El Niño) and the reduction of the zonal gradient in sea level pressure across the basin (Southern Oscillation; SO). However, the relation between EN and the SO is not always strong (Deser and Wallace 1987), while the SO has been shown to exist even without the ocean dynamics generally associated with EN (Clement et al. 2011). In practice, the definition of “El Niño” is therefore more a matter of convenience of its users than a strict scientific result (Trenberth 1997).

Although the essential physics of ENSO have been largely identified (e.g. Neelin et al. 1998), recent research is focused on understanding the diversity among the individual events, for which a popular procedure is to classify these based on whether the maximum SST anomalies are predominantly found in the central or eastern Pacific (see review by Capotondi et al. 2015), although this classification is somewhat arbitrary (Takahashi et al. 2011). One possibly “true” distinct type of EN could consist of the extreme EN of 1982–83 and 1997–98, as they appear to correspond to a different dynamical regime from the rest of EN due to the nonlinear activation of deep convection in the cold eastern Pacific (Takahashi et al. 2011; Takahashi and Dewitte 2016). These extreme events have been associated with intense warming in the FEP and disproportionally large rainfall anomalies in the arid western coast of South America (e.g. Woodman 1985, 1999; CAF 2000).

From the perspective of the FEP impacts, the only other “very strong” event in the last century was the 1925–26 EN (Quinn et al. 1987; Fig. 1a) and by several other measures in this region, this EN can be considered among the three strongest, with 1982–83 and 1997–98 (Fig. 1b–f). It was due to the detailed report of this event by Murphy (1926) that “El Niño burst onto the international scientific scene” as a legitimate research topic (Cushman 2004), leading to the discovery by Berlage in 1929 (Cushman 2004) of the statistical relation between EN variability, using an index of rainfall in northern Peru (Eguiguren 1894) as a proxy, and the large-scale atmospheric Southern Oscillation (Walker 1924), culminating with the concept of ENSO, an essentially coupled ocean–atmosphere phemonenon (Bjerknes 1969).

Fig. 1
figure 1

El Niño-related indices for the twentieth century: a El Niño magnitudes estimated according to its coastal manifestations (Quinn et al. 1987, updated by Quinn 1992; the 1997–98 event was added), b annual rainfall (cm) in Guayaquil, coastal Ecuador [Sep(−1)–Aug(0)], annual mean discharge (m\(^3\)/s) of the Zaña (c) and d Viru rivers, eProsopis pallida annual growth ring width (rainfall proxy) from Casma, coastal Peru (Rodriguez et al. 2005), f SST anomaly in Puerto Chicama (°C), g monthly detrended sea level anomaly (cm) at Balboa, h the equatorial SST “cold tongue index” (°C), and i the Southern Oscillation Index (reversed axis). The vertical grey lines correspond to January 1925, 1983, and 1998

Although the 1925–26 EN was relatively well documented at the time, some of the ideas that appeared well justified at the time, particularly the role of northerly winds (Schott 1931), have been discarded in the subsequent years with the establishment of the ENSO paradigm in the 1970s and 1980s (Wyrtki 1975; Wallace et al. 1998; Neelin et al. 1998), but without taking a close look at the 1925 EN. Thus, it is timely to revisit the 1925–26 EN in an integrated way, under the light of modern theory and expanded datasets, to recover potentially valuable information and insights on the nature of EN and its diversity.

2 Data sources and processing

Monthly series for Puerto Chicama SST (7.7°S, 79.4°W, 1925–2002), the Piura river discharge (1925–1998), Piura rainfall (1932–2008), and estimated EN magnitudes of Quinn (1992) were obtained from the JISAO data archive (http://jisao.washington.edu/data). The monthly Piura discharge data for March and April 1925 were absent, but these values were estimated as discussed in "Appendix B".

Monthly precipitation for Milagro (1921–1981) and Guayaquil (1915–2000) and mean air temperature for Iquique (1900–1988) from the NOAA NCDC GHCN v2 database were obtained from the IRI Data Library (http://iridl.ldeo.columbia.edu/). Annual precipitation values were calculated based on the hydrological year from September of the previous year to August. We also used the annual tree-ring width series for an individual of Prosopis pallida (locally known as “algarrobo”) in San Rafael, Casma (near the coast at 9.5°S), for 1908–2002, a proxy for annual precipitation (Rodriguez et al. 2005).

Daily “research quality” mean sea surface height (SSH) for Balboa (8.97°N, 79.57°W, 1907–2012), on the Pacific side of the Panama Canal, was obtained from the University of Hawaii Sea Level Center (http://uhslc.soest.hawaii.edu/). Monthly means were calculated from this data. The climatology for daily anomalies was calculated using six harmonics of the annual period fitted to the daily data. Anomalies were linearly detrended over the full period.

We used ship-based ocean surface data from the ICOADS database (Worley et al. 2005; Woodruff et al. 2011) in two formats. Firstly, we used the gridded ICOADS 2-degree Enhanced v. 2.5.2 monthly summaries, specifically the mean SST, wind, and cloudiness, which we obtained from the NOAA ESRL website (http://www.esrl.noaa.gov/psd/data/gridded). This dataset consists averages of individual observations over the corresponding 2° grid boxes, with no spatial interpolation. For broad-scale mapping, we constructed 3 month-mean SST and surface wind anomalies during the 1925–26 EN from the gridded monthly ICOADS dataset, with no smoothing or interpolation, limiting the results to those grid cells that contained at least 3 and 10 observations for SST and wind, respectively (wind has stronger high frequency variability). The climatology was constructed from the same data by first averaging temporally and then filling the spatial gaps with a first guess and 9-point smoothing ten times to merge the filled values with the averaged ones. Additionally, we constructed monthly time-series for equatorial segments along four shiptracks that had good data coverage (Fig. 3a).

We also used the individual ship observations from ICOADS Release 2.5 (dataset ds540.0 at NCAR CISL RDA) to produce monthly averages every degree latitude from 9° to 30°S along a well-transited shipping route from Panama to the coasts of Ecuador, Peru and Chile, that started after the opening of the Panama Canal in 1914. Because of this and the onset of World War II in the Pacific, there is good data availability along this track between the years 1920 and 1942. Therefore, unless explicitly indicated, our base period for climatologies for all variables is 1920–1939.

We used a database of approximately 2500 news articles from the Peruvian newspaper “El Comercio” for the period from January 1925 to December 1926, focusing particularly on mentions of meteorological or hydrological phenomena in the coast of Peru (Chang 2014), as well as a database of articles from the newspaper “El Tiempo” from the Piura region (Rojas-Rosas 2014). Numerical meteorological and river discharge data was also retrieved from the newspapers, particularly from the Hipólito Unanue meteorological station in Lima and the daily raingauge data from the Harvard Observatory in Arequipa. Since El Comercio is based in Lima and communications with the northern coast were not in real-time, the news articles often lag the actual events and date and time were seldom reported precisely (Chang 2014), so we indicate the publication date and page number for the relevant newspaper articles in footnotes.

We obtained 18 temperature profiles from the Arcturus expedition (Beebe 1926), which were made in a broad region between Panama and the Galapagos islands between March 30 and June 9, 1925 (Table S1). Four oceanographic profiles of temperature and salinity made in the upper 200 m by a British cruise (NODC code GB012817) along the coast between Ecuador and Panama between April 17–19, 1925 (Table S1) were retrieved from the World Ocean Database 2013 (WOD13; http://www.nodc.noaa.gov/OC5/SELECT/dbsearch/dbsearch.html). We also digitized selected meteorological, hydrological, and oceanographic data from tables and graphs in Murphy (1926), Zegarra (1926), Berry (1927), Zorell (1929), Bailey (1930), Schott (1931), Sheppard (1930, 1933), Petersen (1935), Schaeffer et al. (1958), Woodman (1985), and Reparaz (2013).

Two long-term indices used to represent the basin-scale ENSO variability are the cold tongue index (CTI; SST anomaly in 6°S–6°N, 180°–90°W, 1845–2011) from the JISAO data archive (http://jisao.washington.edu/enso/), and the Southern Oscillation Index (SOI; normalized pressure difference between Darwin and Tahiti for 1866–2013) from the CRU website (http://www.cru.uea.ac.uk/cru/data/soi/, based on Ropelewski and Jones 1987).

We complemented the analysis with the following reconstructed observational products, which were trusted only to the extent that they were consistent with actual in situ data: The Hadley Centre Global Sea Ice and Sea Surface Temperature (HadISST) v1.1 (Rayner et al. 2003) and the NOAA Extended Reconstructed SST (ERSST) v3b, Smith et al. 2008) SST products, the SODA 2.2.4 ocean reanalysis (Giese and Ray 2011; http://apdrc.soest.hawaii.edu/), and the NCEP twentieth century Reanalysis v2 (hereafter 20CRv2; Compo et al. (2006, 2011); http://www.esrl.noaa.gov/psd/data/gridded).

3 Ocean–atmosphere evolution and processes

The large-scale evolution of the 1925–26 EN followed approximately the development phases of the “canonical El Niño” (Rasmusson and Carpenter 1982, hereafter RC82; Harrison and Larkin 1998). The “onset phase” in November 1924–January 1925 was characterized by anomalously cool conditions and easterly equatorial wind anomalies in the central and eastern equatorial Pacific (Figs. 2a, 3b–e), followed by strong anomalous warming near the coast of South America peaking in March 1925 (Figs. 2b, 3b).

Fig. 2
figure 2

Seasonal mean SST (°C) from ICOADS (colors, shown only for at least three observations per grid cell) and surface wind vectors anomalies (shown only for at least ten observations per grid cell and a minimum magnitude of 1 m/s). Also shown is the SST anomaly reconstruction HadISST 1.1 (contours; interval: 0.5 °C, slight smoothing). The averaging periods are indicated in eachpanel

Important departures from the RC82 and HL98 composites are that (1) the warming off Peru took place a couple of months earlier than the corresponding “peak phase” of RC82, so that it coincided with the warmest months (Takahashi 2005) that is most favorable for deep convection (e.g. Takahashi 2004; Huaman and Takahashi 2016), while (2) cool conditions remained in the central-eastern Pacific, expanding to the west, until June 1925 (Figs. 2b, c, 3e), which we later argue is also important for the convective dynamics in the FEP (Sects. 3.2, 3.4).

Fig. 3
figure 3

a Segments of the four ship tracks with the best data coverage in the central and eastern equatorial Pacific during the period 1920–39 (numbered 14 westward from the coast; standard Niño 1+2 and 3.4 regions indicated) and raw (thin) and 1-2-1-smoothed (colors, thick) monthly be SST (°C) and fi zonal wind (m/s) anomalies along each of the ship tracks based on the gridded ICOADS data

The warming was confined to the coast in February–March 1925 but then spread westwards, reaching the central Pacific (~170°W) around August 1925 and peaking by the end of the calendar year (Fig. 3b–e), corresponding to the “mature phase” of the canonical EN (RC82). Westerly wind anomalies started developing in the central Pacific in March 1925 (Fig. 3h–i), indicating the onset of the Bjerknes feedback associated with the FEP warming (Dewitte and Takahashi 2016) and leading to the establishment of the warm ENSO phase.

In the next subsections we address specific mechanisms relevant to the development of this EN event.

3.1 Lack of initial Kelvin wave forcing

In the ENSO paradigm, the coastal warming is associated with downwelling equatorial Kelvin waves (e.g. Chiodi et al. 2014) that depress the thermocline and raise the sea level. However, in contrast to other EN events, the sea level data at Balboa, Panama, does not show positive anomalies in early 1925, and in fact they are negative in February (Fig. 4). Furthermore, we find that large sea level anomalies (e.g. >10 cm) in other EN events are associated with substantially weaker monthly SST anomalies than the one observed at Puerto Chicama in March 1925 (Fig. 4), with the exception of the peaks of the extreme 1982–83 and 1997–98 EN (Fig. 4d, f). Thus, a sea level anomaly associated with a downwelling Kelvin wave sufficiently strong to account for the observed warming in early 1925 could be expected to have been greater than 20 cm.

Fig. 4
figure 4

Daily detrended sea surface height anomaly at Balboa (cm; colors) and monthly sea surface temperature anomaly (°C; grey lines) for Puerto Chicama for selected El Niño events

On the other hand, the second warming peak in 1925, starting in November, coincides with a 20 cm sea level anomaly, indicating that a downwelling equatorial Kelvin wave pulse was responsible (Fig. 4a). Downwelling Kelvin waves at this time are consistent with the westerly wind anomalies in the central Pacific (Fig. 3h–i), although this data cannot resolve specific westerly wind events.

Predominantly positive sea level anomalies were observed starting in May 1925 (Fig. 4a), after which the positive SST anomalies extended southward to 30°S (Fig. 5b). Particularly, the warming pulses in August–September 1925 and December 1925–February 1926 coincide with the positive sea level pulses around those times (Fig. 4a), providing further support to their interpretation as downwelling Kelvin wave pulses that were able to propagate the warming signal polewards into Chile, whereas the warming peak in March 1925 was restricted to north of 20°S (Fig. 5b).

Fig. 5
figure 5

a Monthly sea surface temperature (shading, °C) and surface wind (m/s), and b the anomalies (with respect to 1920–39) from ICOADS along the near-coastal track. The zero meridional wind contour is solid white in a

Additional evidence of the absence of downwelling equatorial Kelvin wave forcing is provided by the hydrographic measurements from the Arcturus expedition east of the Galapagos (March 30–June 9, 1925) and the ship GB012817 along the coast of Colombia and Ecuador (April 17–19, 1925). The data from both indicate SST anomalously above 27 °C near the equator, sharply decreasing to 19–20 °C at the 50 m depth (Fig. 6a, c). The equatorial anomalies calculated from the SODA climatology transition from generally positive at the surface to negative at 100 m (Fig. 6b, d, e). Conversely, strong downwelling Kelvin waves would result in a deep (>100 m) warm layer (Cucalon 1987; Garcés-Vargas et al. 2005; see profiles for 1983 and 1998 in Fig 7b, d, f, h).

Fig. 6
figure 6

ac Temperature and df its anomalies at a, d 0 m, b, e 50 m, and c, f 100 m based on hydrographic measurements from the Arcturus (circles; March 30–June 9, 1925, starting on the east) and ship GB012817 (times symbols; April 17–19, 1925, starting on the south). The anomalies are calculated with respect to the SODA 1920–1939 climatology, except around the Galapagos were SODA is less reliable

Fig. 7
figure 7

a Location of four oceanographic stations from the GB012817 ship (labeled 14) in April 17–19, 1925 (starting with station 4). The surface temperature (°C) and salinity (pps) are indicated in red and blue, respectively. Observed profiles of b, d, f, h) temperature and c, e, g, i) salinity (solid) are shown for stations b, c 1, d, e 2, f, g 3, and h, i 4. Also included are the climatology for April (1920–39, thick short dashed), and the data for April 1983 (long dashed) and 1998 (dot-dashed) for the nearest grid cell from SODA 2.2.4

The zonal wind showed weak positive anomalies (<1 m/s) in the FEP prior to the warming (Fig. 3f, g), which could have helped with the warming.

3.2 Northerly winds and the ITCZ

The most outstanding aspect of the atmospheric circulation in early 1925 was the extreme southward extension of the Panama wind jet, which climatologically reaches the equator in February and March (Fig. 8b, c) but in 1925 extended to 8–9°S and fanned out to the Galapagos (Fig. 8f, g). Similarly, the Papagayo jet further west extended almost to the equator near 95°W (Fig. 8f, g). The northerly anomalies were on the order of 2–3 m/s (Figs. 3j, k, 5b, 9), substantially stronger than in the EN composite of (Harrison and Larkin 1998) for February (~0.6 m/s around 3°N in their Fig. 6) and the RC82 composite for March–May (Fig. 19b in RC82). The onset of these northerly winds from January to February was abrupt, as was their retreat from March to April (Figs. 3j, k, 5a, b, 8a, b).

Fig. 8
figure 8

ICOADS sea surface temperature (shading, °C) and surface winds (m/s) for January through April for the 1920–39 climatology (top row), and year 1925 (bottom row). The 26 °C isotherm (solid and light purple) and the zero meridional wind contour (dashed) are included

The enhancement of the northerly winds could have been the result of external atmospheric forcing. In the case of the gap jets through Central America, the forcing could be associated with the SLP difference between the Atlantic and the Pacific (Karnauskas et al. 2008). The ICOADS data shows positive SLP anomalies around Central America in both February and March 1925, with negative anomalies off Ecuador only clearly in March (Fig. 9a, b). However, although these (absolute) northerly winds were connected to the northeasterlies in the Caribbean, the northerly anomalies themselves were limited to the Pacific (Figs. 5b, 9). Furthermore, in situ observations of the integrated January–April northerly wind at Balboa and Cristobal, at both ends of the Panama Canal, indicate that 1925 did not have unusually high northerly winds (Schaeffer et al. 1958; Fig. S1). On the other hand, in February–March 1925, the Southern Oscillation was in a positive state (Fig. 1i), suggesting an enhanced South Pacific anticyclone, while the actual wind data does not show substantial subtropical wind anomalies (Fig. 5b), indicating that the northerly wind anomalies were not forced from the south Pacific. Therefore, atmospheric forcing from the Caribbean or the south Pacific does not appear to have had a key role in driving the northerly wind anomalies.

Fig. 9
figure 9

Monthly anomalies of surface wind (m/s; vectors) overlayed on a, b cloud cover (octas; shading) and c, d) sea level pressure (hPa; shading anomalies from ICOADS for a, c February and b, d March 1925. Only grid cells with at least 15 observations are shown (slight spatial smoothing). Anomaly vectors with magnitude larger than 1 m/s are darker

Another possibility is that local air–sea interaction amplified and maintained the northerly wind anomalies that could have been initiated by weak external atmospheric or oceanic forcing. Particularly, the meridional gradient in absolute SST near the coast was reversed, with the warmest/coolest waters found south/north of the equator (Figs. 5a, 8f, g). The SST anomalies presented a meridional dipole pattern with maxima at °S and 5°N and the northerly wind anomalies in-between (Fig. 5b). This anomalous SST gradient could directly reinforce the northerly winds via thermally-induced pressure gradients (Lindzen and Nigam 1987; Battisti et al. 1999) and, perhaps more importantly, the anomalously high SST south of the equator could strengthen the southern hemisphere ITCZ that climatologically is present in February–April around 5°S (Huaman and Takahashi 2016). The latter is suggested by the positive cloudiness anomaly in the ICOADS data around 2°S in February and, most notably, around 7°S in March (Fig. 9c, d). This enhancement of the SH ITCZ is consistent with the surface wind convergence off the coast of northern Peru (Fig. S2), as well as with the heavy rainfall observed in the otherwise arid northern-central coast of Peru (Sect. 3.4, "Appendix A"). Additionally, the barometric pressure measured at Chicama (7.7°S) at 7 a.m. (to reduce the diurnal land heating effect; the 4 pm data has similar variability but with lower values) indicates a large 9–10 hPa drop from around 1017 hPa in the beginning of January to an average of 1008 hPa in the second half of March (Fig. 10b), consistent with the establishment of the equatorial trough off the coast of Peru, with lower pressure than the March mean value of 1010 hPa according to the TAO buoy at 5°S, 95°W for the years 2001–2003, probably an indication of the intensity of the anomalous SH ITCZ. On the other hand, the ICOADS data shows that the fanning of the Panama jet is associated with net surface wind divergence in the eastern Pacific north of the equator in February and March (but not in the Caribbean) instead of the climatological convergence (Fig. S2), consistent with the reduced cloudiness in the NH in March (Fig. 9d). The above suggests an anomalous local meridional overturning cell in the FEP in February–March 1925 with ascent/descent in the southern/northern hemisphere.

Fig. 10
figure 10

Daily series for January–April 1925: a sea surface temperature in Puerto Chicama and Callao (°C; open circles and dots, respectively), b barometric pressure in Chicama (hPa; 1-2-1 smoothed), c precipitation in Zorritos (mm; Petersen 1935), d Piura river discharge (m\(^3\)/s; thick circles are discharge reconstructed from water height; small and large black triangles indicate days with moderate and heavy rainfall, respectively, in the city of Piura, Table S2), e Virú (black) and Chicama (grey) river discharge (m\(^3\)/s; Zegarra, 1926), f precipitation in Trujillo (mm; *the accumulated value for March 7–9 was reported on March 9), and g precipitation in Lima (mm; El Comercio, 1925). Data was digitized from Murphy (1926) unless explicitly indicated. Days with missing data are shaded (except for the gaps in the Chicama discharge), and, in the case of precipitation, they are assigned a value of zero (following Petersen 1935). In bf, thick black lines indicate when the near-coastal SST analysis of Schott (1931) exceeded 26 °C at the corresponding locations (for Zorritos, dashed indicates extrapolation)

The existence of an approximate threshold SST for the activation of deep convection (Graham and Barnett 1987; Johnson and Xie 2010; Takahashi and Dewitte 2016; Jáuregui and Takahashi 2017) introduces a nonlinearity that can explain the abruptness of the onset of the SH ITCZ and northerly wind (Xie and Philander 1994, hereafter XP94; Wang and Wang 1999), which would take place more easily in the warm seasonal peak driven by insolation (Takahashi 2005). In this sense, the coastal EN could be described as an amplification of the seasonal cycle.

The wind speed anomalies associated with the northerly wind anomalies present a dipole pattern (not shown), with reduced/enhanced speed in the southern/northern hemisphere, contributing to reduced/enhanced surface evaporation and, therefore, enhanced/reduced SST. This positive wind speed-evaporation-SST (WES) feedback (XP94) probably was key in establishing this coastal EN event, which would imply that it would correspond to shallow solar warming. Another possible mechanism is meridional warm advection, which is discussed in the next subsection.

3.3 The “Corriente del Niño”

Meridional warm advection associated with an anomalously strong southward “Corriente del Niño” (El Niño Current; Carrillo 1893) was the first EN mechanism identified (Carranza 1891; Schott 1931). In 1891, in addition to warm water along the northern coast of Peru, carcasses of crocodiles and tree debris from north of 3°S were found at 8°S (Carranza 1891). In 1925, the report at 4.6°S of a lizard not previously found in Peru but abundant off Ecuador at 3.2°S (Murphy 1926) and, in 1926, of seeds at 4.6°S of mangroves that are found to the north of 3.6°S (Berry 1927), are also indications of southward advection.

From the end of January through February 1925, ship drift data along the coastal track indicate southward flow north of the equator, with a speed on the order of 30–50 cm/s (Zorell 1929). Murphy (1926) also reported coastal measurements of southward flow of around 50 cm/s further south (5 and 2°S) within the same period. Schott (1931) discussed the southward progression of warm SST fronts in early 1925 using cruise data and coastal stations, with the warmest front reaching Puerto Chicama in February 27, Callao in March 12 (Fig. 10a) and as far south as Pisco in March 16 (14°S), from which he inferred a southward propagation speed of 40–50 cm/s, although onshore advection, as observed in the 1982–83 and 1997–98 events (Morón 2011), is another possibility.

Having discarded the possibility of strong downwelling Kelvin waves (Sect. 3.1), the strong northerly winds are the most likely forcing of the warm countercurrent [e.g. Philander and Pacanowski (1981)]. We can produce a rough estimate of the wind-driven surface current u based on the wind observations by neglecting the Coriolis force near the equator and considering the frictional balance \(r_s \mathbf{{u}}={\tau }/\rho H\), where \(r_s\approx\) (2 days)\(^{-1}\) is a frictional dissipation rate (Zebiak and Cane 1987; Dewitte 2000), \(\rho \approx 10^3\) kg/m3 is the water density, \({\tau }=\rho _a C_D|\mathbf{u}_a|\mathbf{u}_a\) is the surface wind stress with \(|{u}_a|{u}_a\) indicating the surface wind pseudo-stress, \(\rho _a=1.2\) kg m\(^{-3}\) the air density and \(C_D=2\times 10^{-3}\) the drag coefficient (Perigaud et al. 2000). Based on the individual ICOADS observations, the near-coastal pseudostress had a mean northerly component of \(15.5\pm 4.7\) m\(^2\)/s\(^2\) around the equator (1°S–1°N) in February–March 1925. Assuming an Ekman layer depth of \(H=30\) m (Oerder et al. 2015), we calculate the equatorial southward current speeds as 18 ± 5 cm/s. This is a lower bound, since the high SST implies reduced atmospheric stability, i.e. larger \(C_D\) and wind stress, while the ocean stratification associated with the shallow fresh warm surface (Fig. 7h, i) suggests a shallower H. Additionally, changing observational practices (i.e. Beaufort scale vs anemometers) and anemometer heights due to increasing ship sizes (Cardone et al. 1990) introduces further underestimation of the wind stress. Thus, the estimated wind-driven current speed provides a lower bound that is consistent with the other observational estimates for the “Corriente del Niño” speed.

Additionally, the northerly wind anomaly enhanced the upwelling in the Panama bight (Alory et al. 2012), with cold water extending almost to the surface, except for a very shallow warm layer (Fig. 7b), and salinities typical of the 100 m depth (Fig. 7c). This is similar to what was observed in 1891 (Schott 1931) but contrasts sharply with the extreme EN conditions in April 1983 and 1998, which featured a deep fresh warm layer of around 80 m depth (Fig. 7a–g).

3.4 Local and remote sea surface temperature control on eastern Pacific precipitation

The dependence of coastal precipitation of SST approximately follows the threshold model of XP94 (see Sect. 3.2), but with a threshold SST of around 26 °C (Woodman 1999; Takahashi 2004; Ramos 2014). This model provides an adequate qualitative explanation for the temporal evolution of the periods of strong rainfall and river discharge events at different latitudes along the Peruvian coast (Fig. 10c–g), as they approximately coincide with the periods in which the local near-coastal SST [based on the ship analysis of Schott (1931)] was greater than 26 °C (indicated with lines in Fig. 10c–g). These conditions were progressively established from north to south and ended in the reverse order. The details of the severe impacts at various sites along western Peru and Ecuador associated with heavy rainfall and flooding in 1925, which can provide important guidance for paleoclimatic and historical EN reconstructions as well as for risk management, are presented in Appendix A.

On the other hand, the cool central Pacific SST is also known to enhance the precipitation in the mid and upper basins along the western Andes of Peru (Lavado-Casimiro and Espinoza 2014). This is explained by the connection of SST in the western Pacific warm pool connects to the whole tropical free troposphere via deep convection, cooling it during basin-scale La Niña conditions (Yulaeva and Wallace 1994; Chiang and Sobel 2002), which reduces the tropospheric stability and facilitates convection (Vecchi and Soden 2007; Jáuregui and Takahashi 2017), while also producing easterly near-equatorial upper-air wind anomalies over South America that are also favorable for convection (Kousky and Kayano 1994; Vuille et al. 2000).

The role of local and remote forcing is verified for the Piura river by the relatively high positive correlation between its annual mean discharge (reconstructed as described in Appendix B) with February–March mean SST in the FEP (\(r=0.69\) in the Niño 1+2 region; Fig. 11a, b), and negative values in the western–central equatorial Pacific (\(r=-0.17\) in the region we denote as \(T_w\) (155°E–175°W, 5°S–5°N; Fig. 11a, c). On the other hand, we see in the scatter plots that the two relationships are not linear and that the correlation with Niño 1+2 is strongly dominated by the 1982–83 and 1997–98 events, which are outliers in the correlation with \(T_w\) (Fig. 11b, c). However, if we simply subtract \(T_w\) from Niño 1+2, providing a rough index of tropospheric stability in the eastern Pacific, we not only find an enhanced correlation (\(r=0.72\)), but generally a more monotonic nonlinear relationship (Fig. 11d). The high SST in \(T_w\) during the warm ENSO phases in 1926 and 2016 explain why the river discharge was not as high in those years despite the high Niño 1+2 SST, whereas 1925 and 2008 had high discharges due to the low SST in \(T_w\) and high Niño 1+2 SST. The 2008 case is interesting because it was primarily regarded as a basin-scale La Niña (cool ENSO) event (Bendix et al 2011). The nonlinearity of the relation between Niño 1+2 \(-T_w\) and the discharge, with a sharp increase in slope above −1.5 °C, can also explain why the strong EN (warm ENSO) failed to produce rainfall in northern Peru as strong as in 1983 or 1998 despite having similar Niño 3.4 SST (L’Heureux et al. 2016).

Fig. 11
figure 11

a Linear correlation between the annual discharge averaged for the Piura river with the February–March SST from HadISST 1.1 (1925–2016). Scatter plots between the same discharge and the SST averaged over b the Niño 1+2 region, c the Tw region (155°E–175°W, 5°S–5°N), as well as with d the difference between the two (Niño 1 + 2 minus Tw)

We hypothesize that a similar process reduced the stability of the seasonal southern hemisphere ITCZ in 1925, which in the context of the XP94 model could have been through a combination of lowering of the threshold SST and/or enhancing the rate of the precipitation increase with SST due to basin-scale La Niña (cool ENSO) conditions, which cools the free troposphere and produces easterly upper-air anomalies (both of which are indicated by the 20CRv2 in 1925, not shown). In this view, this type of coastal EN would be the result of the interannual destabilization of the seasonal cycle in the FEP by the cool ENSO phase.

4 Discussion

The ENSO paradigm is based on the interaction between equatorial ocean dynamics and zonal winds through SST. Historically, however, the association of northerly winds in the FEP with EN had been noted by Eguiguren (1894) and Murphy (1926), while Schott (1931) went further to propose that these winds were the forcing of the coastal EN. This hypothesis subsequently was countered by the finding that the coastal winds tend to strengthen during EN (Wyrtki 1975; Enfield 1981; Rasmusson and Carpenter 1982) and Wooster (1980) argued that Schott failed by “underestimating the magnitude of the time and space scales involved”. But none of this later studies explicitly analyzed the 1925 and their failure was in implicitly assuming that the same mechanisms act in the same way in every event, despite Wyrtki’s (1975) conclusion that “El Niño certainly does not have only a single cause”.

Nevertheless, the RC82 and the Harrison and Larkin (1998) EN composites do show northerly wind anomalies during the coastal warming phase, but weak compared to 1925. The seasonality of these anomalies appears to be critical for the strong feedback between SST, the ITCZ and the northerly wind, as the SST and the ITCZ off Peru peak climatologically around March (Takahashi 2005; Huaman and Takahashi 2016). We argue that this feedback was made more effective in 1925 by the dominant cold conditions in the rest of the equatorial Pacific, which promoted convection in the FEP by the destabilition of the troposphere and with moist easterly advection from the Amazon. However, strong northerly anomalies were also observed in early 1926, around the peak of the warm ENSO phase (Fig. 12a). This suggests that perhaps longer-term changes, like a lower convective threshold for convection (Johnson and Xie 2010), could have made this mechanism more effective in the past. In fact, the latitude of the trade-wind confluence in these two years has been the lowest in the 1920–2012 period, including the extreme 1982–83 and 1997–98 events and there is a slight (but not significant) northward trend in this latitude, perhaps a response to the stabilization associated with the long-term tropical tropospheric warming (Johnson and Xie 2010; Jáuregui and Takahashi 2017). Consistent with this, the low tropospheric stability estimated as the difference between the 700 hPa potential temperature from the 20CRv2 and SST in the Niño 1+2 region (not shown) also has a small albeit not significant positive trend. On the other hand, the SST difference between the Niño 1+2 and the \(T_w\) region, a stability proxy for the FEP (see Sect. 3.4) does not show a clear trend (Fig. 12b), although uncertainty in SST reconstructions is an issue for trends in the zonal SST gradient in the tropical Pacific (Deser et al. 2010). On the other hand, many climate future climate change projections with global climate models indicate a joint trend in increasing rainfall and northerly wind anomalies off northern Peru similar to the proposed for the 1925 EN (Belmadani et al. 2014), but the models in general continue to have strong biases in this region, particularly the double ITCZ syndrome (Zhang et al. 2015).

Fig. 12
figure 12

a Latitude of zero meridional wind along the near-coastal track based on the February–March mean wind from ICOADS, the b February–March mean Niño 1+2 minus Tw SST (°C) from HadISST 1.1 (black, circles) and ERSST v3b (grey, crosses) and c the annual (Jan–Dec) mean Piura river discharge (m3/s). Linear fits are shown dashed. Vertical lines indicate the years 1925, 1983, and 1998

If we focus on the extreme EN impacts in western tropical South America (e.g. Figs. 1a–f, 12c), both the large-scale version (1982–83, 1997–98) or this coastal version have similar signatures. So not only is the record too sparse to identify trends that differentiate between the two types, but the two trends would be responding to different processes. This is important for long-term reconstructions of ENSO diversity. The interpretation of paleoclimatic and historical records of extreme EN impacts in the FEP (high temperatures, flooding, impacts on marine ecosystems) would need to carefully consider these two types, ideally contrasting with proxies from other regions that help with the discrimination of the two.

5 Conclusions

In this study we revisited the very strong 1925 El Niño (EN), the third strongest, after 1982–83 and 1997–98, in terms of its impacts on the far-eastern Pacific (FEP) associated with heavy rainfall on the coasts of southern Ecuador and Peru in February–April 1925. In situ instrumental records and extensive newspaper reports allowed us to reconstruct the ocean–atmosphere evolution in the tropical Pacific, particularly in the FEP, as well as the large extent of the impacts associated with heavy rainfall and flooding in coastal Peru and Ecuador, which extended as far south as 12°S.

The 1925 EN event was the one that introduced the concept of El Niño to the scientific community. However, this very strong and archetypical EN event in terms of its FEP signature was restricted to this region and coincided with anomalously cold conditions in the rest of the equatorial Pacific. This “coastal” EN took place in February–April 1925, in the transition from a cool to a warm ENSO state. However, the coastal warming was not associated with the equatorial zonal dynamics that are the essence of ENSO. Instead, in situ hydrographic data in the FEP indicate that the warming was shallow and the tide-gauge data at Balboa indicates that sea level was lower than normal, so downwelling Kelvin waves were unlikely to have played a significant role in producing the coastal warming. On the other hand, in situ ship-based wind data indicate strong northerly winds that reached 7°S on average in February−March 1925, which climatologically only extend to the equator near the coast. The meridional asymmetry in SST was also reversed, i.e. with a warmer southern than northern hemisphere. The ship data also indicates reduced cloudiness, increased sea level pressure, and net wind divergence, north of the equator, and the opposite off northern Peru, indicative of the weakening/strengthening of the ITCZ to the north/south of the equator. The above then indicates a reversal of the north–south asymmetry relative to the equator in the coupled ocean–atmosphere system that includes the ITCZ, meridional wind, and SST.

The abrupt onset of this coastal EN, with the southern ITCZ and northerly anomalies, suggests strong external forcing and/or strong nonlinear coupled feedbacks. Regarding the former, the in situ data indicates that the Panama gap jet was not anomalously stronger and that the south Pacific anticyclone was not weaker in this period, indicating that the associated external forcing was not strong. Thus, it seems more likely that the nonlinearity associated with a threshold in SST for the activation of the ITCZ south of the equator was responsible for the abrupt transition, as in the model of Xie and Philander (1994, XP94). The mechanism for the ocean warming probably involved a combination of the wind-evaporation-SST mechanism (XP94) and oceanic southward warm advection, i.e. the enhanced “Corriente del Niño”. The existence of this warm countercurrent was shown with ship drift data and is consistent with a calculation of a frictional current driven by the observed northerly winds. To the extent that a nonlinear coupled feedback underlies the onset and retreat of the southern ITCZ, the very strong coastal EN can be described as an enhancement of the insolation-driven seasonal cycle, particularly through the destabilization of the seasonal SST-ITCZ-meridional wind dynamics.

We propose that an important mechanism for destabilizing the ITCZ involves the zonal equatorial SST gradient in the Pacific, particularly the contrast between the SST in the FEP and in the western–central Pacific, as the latter affects the temperature in the free tropical troposphere and, therefore its static stability, as well as the zonal upper air moisture flow in the FEP that affects convection. This is shown to be the case using a river discharge record on the coast of Peru at 5°S, the climatological latitude of the southern ITCZ band, which shows a strong nonlinear relation with the difference in SST between the FEP and western-central Pacific, with more strongly enhanced discharge when the former is less than ~1.5 °C colder than the latter. In early 1925, the cold conditions in the central-western Pacific and the warm eastern Pacific both contributed to high river discharges, whereas during the warm ENSO phase the effects of the two regions oppose each other as in 1926 and notably during the strong 2015–16 EN (L’Heureux et al. 2016).

In the context of ENSO diversity, there have been several studies that have classified EN events according to their spatial pattern. But if we focus on those EN events that have very strong impacts on coastal Peru and Ecuador with very high SST and coastal rainfall, we can identify two major types:

  • Very strong warm ENSO events, e.g. 1982–83 and 1997–98, that are associated with the zonal dynamics in the equatorial Pacific and a nonlinear Bjerknes feedback that enhances their growth (Takahashi and Dewitte 2016), which are potentially predictable several months in advance,

  • Very strong “coastal” EN, e.g. 1891 and 1925, with cold to neutral conditions in the rest of the equatorial Pacific and associated with meridional dynamics in the FEP involving the abrupt enhancement of the ITCZ and warming south of the equator and strong northerly winds. Based on our current knowledge, this type of event is not as predictable as the warm ENSO events.

Further studies will be needed to verify the proposed mechanisms, but this will prove challenging as climate models continue to have strong biases in this region, particularly associated with a warm coastal bias and an excessively strong southern ITCZ, perhaps similar to this very strong coastal EN. In situ data in this region, particularly observations of the atmospheric circulation associated with the southern ITCZ, would be very valuable for the validation of both the mechanisms and the models themselves.