Are the socioeconomic impacts associated with tropical cyclones in Mexico exacerbated by local vulnerability and ENSO conditions?

Tropical Cyclones (TCs) are among the most dangerous natural hazards because they can cause severe economic losses and high mortality. Climate risk is defined as a metric that depends on social vulnerability and the occurrence of natural hazards. A social vulnerability index was constructed for this study using two metrics: the degree of local marginalization and the local social gap. The accumulated rainfall and duration of extreme precipitation associated with TC passages are examined as a natural hazard during the period 1981–2017. TC days are depicted as days when TC‐related rainfall exceeded the 95th percentile of daily precipitation from May to November, defined as summer precipitation. In this way, changes in climate risk under El Niño‐Southern Oscillation (ENSO) conditions are explored to determine regions where both social vulnerability and TC days are high. These changes are useful to find out when disasters have more chances to occur. In the present study, climate risk was found to increase more than 80% from average in southwestern Mexico during strong El Niño years. Under neutral conditions, climate risk values rise to more than 40% than average over northwestern Mexico. Under strong La Niña conditions, climate risk increases by more than 80% from average over the eastern coast of Mexico. Our approach is validated through a comparison between anomalies in climate risk and disaster costs (socioeconomic impacts). Both local vulnerability and ENSO conditions exacerbate socioeconomic impacts associated with TCs, and an analysis of linear trends in TC rainfall and TC days reveals that most of the coastal regions in Mexico have a significant rising trend in both variables. Thus, Mexico should be prepared to face more TC extreme rainfall events. Suggestions for how Mexico can meet the objectives of international risk agendas are discussed.


| ENSO, rainfall, and tropical cyclones
The El Niño-Southern Oscillation (ENSO) is the main driver of seasonal atmospheric conditions over the tropical and subtropical Americas. It is well known that El Niño produces strong subsidence over the Caribbean region and central-southern Mexico, and this subsidence prevents deep tropical convection from organizing (Magaña, 2003;Wang and Enfield, 2003). On the other hand, La Niña induces a strong upward movement, which supports the development of convection over the tropical Americas (Magaña, 2003;Wang and Enfield, 2003). Sun et al. (2015) found that this climatic oscillation can strongly modulate the occurrence of extreme rainfall events in Mexico, mainly in September-October-November and December-January-February.
In addition to a direct effect on tropical rainfall, ENSO also modulates tropical cyclone (TC) activity (particularly trajectories and lifetimes; Kossin et al., 2010;Camargo et al., 2008). During El Niño (La Niña) conditions, below-normal (above-normal) TC activity is expected over the Caribbean Sea and Gulf of Mexico (Camargo et al., 2007). During La Niña conditions, straight-forward TC trajectories are more likely to form over the Caribbean region and make landfall in northeastern Mexico. However, this type of trajectories is not expected during El Niño (Dominguez and Magaña, 2018). Over the Eastern Pacific Ocean, only the TC trajectories that form near the western coast of Mexico and Central America and travel later to the mid-Pacific Ocean were found to be affected by El Niño conditions at 97.5% level of confidence (Camargo et al., 2008;Dominguez and Magaña, 2018). By modulating TC formation and track, ENSO also indirectly modulates rainfall in certain regions of Mexico, and it is expected to receive more (less) rainfall produced by TCs under La Niña (El Niño) conditions. However, under neutral conditions, no strong signal of the ENSO modulation on the type of tracks has been found in the North Atlantic and Eastern Pacific Ocean basins. Any type of TC trajectories can occur on both oceans under these ENSO conditions. Furthermore, interannual variability of TCs over the North Atlantic and Eastern Pacific Oceans can be influenced by the Atlantic Multi-decadal Oscillation (AMO) and the Pacific Decadal Oscillation (PDO), respectively. These decadal oscillations play a significant role in modulating atmospheric mechanisms that affect TC landfalling. For example, the combination of the cold PDO phase with La Niña conditions can steer TCs towards the western coast of Mexico (Gutzler et al., 2013;Wood and Ritchie, 2013), producing more TC-related rainfall over land. However, this combination does not mean an increase in the whole TC activity over this basin. Over the North Atlantic Ocean, it is well known that TCs are more active under the positive AMO phase (AMO+) and La Niña conditions because the sea surface temperatures are warmer than during the AMO-phase, and the vertical wind shear is decreased, mainly over the main development region. The merger of both oscillations increases the chances of TC landfalling along the eastern coast of Mexico (Goldenberg et al., 2001;Nogueira et al., 2013;Dominguez et al., 2020).
TC-related rainfall can be defined as the precipitation that is inside of a radius of 500 km from the TC center (Englehart and Douglas, 2001). Although TC size may vary due to closeness to shorelines (Cerveny et al., 2000;Rogers et al., 2000), and large-scale environmental conditions, such as vertical wind shear, humidity, and sea surface temperatures Hill and Lackmann, 2009), this threshold distance from the TC center has been widely accepted (Cavazos et al., 2008;Jiang and Zipser, 2010). Besides, Khouakhi et al. (2016) showed that TC precipitation decreases as moving inland from shorelines in North America. So, it is inferred that the extreme precipitation produced by TCs is only concentrated on the 500-km radius.
TCs play an essential role in the hydrological cycle of the arid and semi-arid regions over Mexico. They not only contribute up to 50% of seasonal precipitation but also can fill reservoirs and help to recover dam levels over these regions (Dominguez and Magaña, 2018;Sisto et al., 2016). However, TCs also produce such intense rainfall that their impact can damage local infrastructure and housing (CENAPRED, 2019). The local impacts produced by TCs may range from high to low economic losses, which mainly depend on infrastructure, social conditions of the population, economic development of the region, and the effectiveness of local strategies (World Bank, 2018;Estrada et al., 2015). What is more, a Mexican coastal city can simultaneously experience flash floods, storm surges, heavy rainfall, and severe property damage produced by intense TCs (Trepanier et al., 2017). However, even if the meteorological impacts are similar, each affected coastal city can experience different TC damage outcomes (Caetano et al., 2017;Holland et al., 2019). It is thus important to examine the natural hazards of TCs from multiple perspectives. of hydrometeorological hazards produced by TCs are flash floods, heavy rainfall, intense winds, and coastal storm surges.
Vulnerability is the social, economic, and physical conditions that increase the susceptibility of communities to the impacts of hazards (UNISDR, 2017), and it is independent of natural variability. Some examples are the degree of fragility in housing, social organization, level of preparedness to face hazards, degree of social marginalization, socioeconomic activities, institutional-political development, and social gap (Alcántara-Ayala, 2019b). These social variables determine to what extent a community is prepared to face one or several hazards at the same time. There is not a unique methodology to define vulnerability because it depends on the hazard, and several methods have been used to estimate its magnitude. For example, previous studies have quantified vulnerability to floods and droughts over Mexico (Neri and Magaña, 2016;Zúñiga and Magaña, 2018), and their findings show that the use of combined normalized variables in a single index can provide an adequate estimation of vulnerability. Besides, official social data is harder to quantify when compared to natural hazards, and it is only obtained every 10 years by the Mexican government through a national census.
Risk is defined as the degree of loss of life or damaged assets that occurs to a society or a community in a certain period of time. It is determined probabilistically as a function of hazard, vulnerability, and capacity (Aven, 2016). Thus, climate risk should be defined as an interaction of atmospheric variables with social conditions (Oliver-Smith et al., 2017;Alcántara-Ayala, 2019a;Sutton, 2019).
Any natural hazard can produce high economic losses and mortality if the vulnerability is high, and the impacts of hazards will generate a high risk. This conjunction will consequently lead to disasters (Oliver-Smith et al., 2017;Silver et al., 2019). The chances for severe disruptions in the functioning of a society at any scale, which lead to high human, material, economic, and environmental losses, are defined as a disaster (UNISDR, 2017). In this sense, disasters are not natural; they are socially constructed because they depend on social vulnerability (Alcántara-Ayala, 2019a). Disasters should not be explained only in terms of natural hazards, but also the social response and organization to face them must be considered (Alcántara-Ayala, 2019b). Both physical and social variables must be analysed to quantify risk, which is essential to identify the chances of generating disasters and, according to the magnitude of risk, develop risk management actions and define adequate prevention measures (Ward, 2020;Zúñiga and Magaña, 2018).
Hydrometeorological hazards produced 45.5% of all disasters over Mexico during the 1900-2018 period (Alcántara-Ayala, 2019b). TCs are the main natural phenomena that cause high-impact economic losses. During the 1929-2013 period, seven of the top 10 disasters in Mexico occurred mainly over southern Mexico, and they were associated with TCs (Sanchez-Rodriguez and Cavazos, 2015). For example, the disasters associated with Hurricanes Stan and Wilma, 2005 (category 1 and 4, when they made landfall over Mexico, respectively) generated economic losses of 4,136 million dollars. Hurricanes Manuel and Ingrid, 2013 (category 1 and tropical storm, respectively) produced damages of 4,200 million dollars (Alcántara-Ayala, 2019b; CENAPRED, 2019). On average, TCs represent 86.5% of the annual cost of disasters in Mexico, and the main TC hazard is the extreme rainfall they produced (CENAPRED, 2019). Although TC intensity is important to define hazards related to TCs, it does not hold a clear relationship with the heavy rainfall that a TC can produce (Yu et al., 2017), and consequently, both tropical storms and hurricanes can cause high economic losses in the country. Apart from that, 48.8% of Mexican counties, where disasters commonly occurred, are under significant poverty conditions. Most of the serious damages and high economic losses have been registered in counties located over southern Mexico, as their social marginalization is enormous, and their economic development is low (Alcántara-Ayala, 2019b).
It is important to highlight that the disasters in Mexico are not only linked to heavy rainfall produced by any hydrometeorological phenomenon, but also the rise in deterioration of watersheds, deforestation of natural areas, and environmental degradation. This rise makes communities more physically and socially vulnerable to intense rainfall (Oliver-Smith et al., 2017;Alcantara-Ayala, 2019b). For example, the duration of intense rainfall, even weaker than the 95th percentile of daily rainfall, for various days can saturate the soil moisture, and river flows can increase the transport of riverbed sediments. The latter may cause people who live on the riverside or coastal cities to experience floods and landslides, mainly over central-southern Mexico. Consequently, disruptions in life and basic services can occur (Alcantara-Ayala, 2004;Zuñiga and Magaña, 2018).
Climate risk of TC extreme precipitation over Mexico has not been quantified, and this information is fundamental for public policymaking and decision-makers since TCs are the most significant hydrometeorological hazards that are associated with disasters. Here, we build on previous studies by exploring TC extreme rainfall days under ENSO conditions and social vulnerability to estimate regional climate risk over Mexico. We hypothesize that ENSO largely modulates TC extreme rainfall events in the same way that other TC features (trajectories and regional contribution to seasonal rainfall), which along with social vulnerability, play an important role in defining the areas with a high probability of disasters. The main objective is to find out the modulation of ENSO on climate risk, considering both a vulnerability index and the probability of TC extreme rainfall day occurrence in years under strong ENSO conditions. Integrated multi-hazard early warning systems should be created to reduce disasters (UNESCO, 2019). Diminishing disaster costs involves identifying vulnerable regions and developing policies for making well-informed decisions (UNDRR, 2015;UN, 2015). National and local stakeholders need to be familiar with scientific studies, promoting science-policy interactions (Claudet et al., 2020). Therefore, Mexico needs climate risk studies that determine under which natural and social conditions disasters occur to implement prevention actions. Scientific research should be directed to understand and manage disaster risk reduction adequately (UNDRR, 2015).
The next section describes both the social and physical datasets were used to define TC extreme rainfall and social vulnerability. The methodology for determining trends in TC precipitation, as well as for quantifying vulnerability and climate risk, is also presented. Section 3 analyzes how climate risk to TC extreme rainfall days changes under ENSO conditions, and a validation of our approach is carried out through a comparison between anomalies in climate risk and economic losses. Finally, the summary and conclusions are given in section 4.

| DATA AND METHODS
Climate risk from TCs was analysed under different ENSO conditions ( Figure 1) for four regions of Mexico: northeastern, northwestern, southeastern (commonly known as Yucatan Peninsula), and southwestern Mexico.

| HURDAT, CHIRPS, and TC rainfall attribution
TC tracks were obtained from HURDAT. This database contains TCs over the North Atlantic and Eastern North Pacific Oceans from the beginning of the century, but the TC location and frequency are considered more accurate in the satellite era from 1970 to present (Vecchi and Knutson, 2008). TCs, whose wind magnitude corresponds to a tropical depression, tropical storm, or hurricane were included in this study. TC-related rainfall is defined as the precipitation that is inside a 500-km radius of influence from the TC center location designated by HURDAT. This approximation has been widely recognized to quantify adequately extreme rainfall events associated with TC passage (Cavazos et al., 2008;Jiang and Zipser, 2010;Khouakhi et al., 2016;Xu et al., 2017).
The number of rain-gauge stations in Mexico has dramatically decreased from the 1990s until present, making a station-based analysis of daily rainfall difficult. Moreover, rain-gauges have a sparse spatial cover over Mexico, and they could not provide enough information about specific local changes (Perdigón-Morales et al., 2018). Quantifying TC extreme rainfall demands the use of high-resolution datasets. CHIRPS (Climate Hazards group Infrared Precipitation with Stations) is a quasiglobal dataset (50 S-50 N) that combines satellite data with rain-gauge stations to create a grid with a high spatial resolution (0.05 ), and its temporal resolution is daily, pentadal, or monthly from 1981 to present (Funk et al., 2015). We aimed at characterizing the spatial field of TC extreme rainfall events as finely as possible, so we used CHIRPS for the 1981-2017 period. Firstly, the 95th percentile (P95) was obtained using an empirical gamma function at each grid in the domain of interest (15 N-32 N and 118 W-86 W). Only precipitation that occurred from May to November (hereafter defined as summer precipitation) was considered for our analysis. Moreover, two kinds of extreme rainfall events are defined for each year of the study period: (a) the annual number of extreme wet days, which are days whose rainfall is higher than P95, and (b) days only associated with TC passages, hereafter TC days. A day is considered a TC day if TC-related rainfall exceeded the P95. In that sense, we defined the annual probability of the TC hazard F I G U R E 1 Topography (m) and the four study regions: Northeast, southeast (commonly known as Yucatan Peninsula), northwest, and southwest of Mexico occurrence as the ratio between TC days and the total number of extreme rainfall days for a given year.
During the 1981-2017 period, 4 years were defined as strong El Niño episodes and three as strong La Niña events. The Oceanic Niño Index (obtained from https:// origin.cpc.ncep.noaa.gov/products/analysis_monitoring/ ensostuff/ONI_v5.php) is computed on a three-monthly basis, so we averaged the values from May-June-July to September-October-November per year to define intense episodes of ENSO. Strong El Niño years are years whose anomaly average is equal or higher than +1.0 C, and Strong La Niña years are years whose average is lower or equal to −1.0 C. Weak El Niño (La Niña), values from 0.5 C to 0.99 C (from −0.5 C to −0.99 C), were not considered in our study. Twenty years are defined as neutral conditions (from −0.49 C to 0.49 C) during the 1981-2017 period (Table 1). This approach is useful for determining the atmospheric response to strong ENSO conditions. However, these conditions are less common than weak El Niño and La Niña events, limiting the number of events that were considered in our study.
We explored how the ENSO conditions modulate the annual accumulated rainfall, TC rainfall, TC days, and the annual probability of TC hazard occurrence. The annual accumulated rainfall was obtained as the sum of summer precipitation in a given year. We also defined the annual TC rainfall as the sum of the precipitation produced by all TCs formed either in the Eastern Pacific or the North Atlantic Ocean and made landfall in Mexico for a given year. The annual number of TC days is determined as the sum of the days when TCs made landfall in Mexico. We classified and then averaged the annual accumulated rainfall, the annual probability of TC hazard occurrence, annual TC rainfall, and TC days, depending on the ENSO phase. Those mean values for each ENSO phase were used to calculate their anomalies, which were obtained by subtracting mean values from their climatology.

| Trends and test for TC precipitation
Statistical significance of trends in the magnitude of TC rainfall and TC days were tested using two tests: t-test and Mann-Kendall test. The t-test is a widely used test to evaluate the statistical significance, at 95% of significance level, of slopes from linear regression models.
The Mann-Kendall test is independent of linear regression models, and outliers do not influence the Mann-Kendall test because its statistic is based on the sign of differences, not directly on the values of the variable (Sharma and Babel, 2014;Wilks, 2011): In Equation (1), S is the Mann-Kendall's test statistic, and the differences for two values is defined as x j -x k . Our null hypothesis is no trend in the data.
where S is the Mann-Kendall's test statistic, and the variance is computed as follows: where n is the number of years; J indicates the number of groups of repeated values, and t j is the number of repeated values in the jth group (Wilks, 2011). The Mann-Kendall test was also used to evaluate the statistical significance at the 95% significance level.

| Methodology for quantifying vulnerability
We define the vulnerability to TC extreme rainfall days as a quantity that depends on two social variables: the degree of local marginalization and local social gap. The first one is a metric that measures the extent to which the population has access to education, suitable housing, and medium-to-high labor incomes. The lower the metric, the lower the social lacks and poverty (CONAPO, 2012 health services, low access to basic services, and lowquality housing at a local level. This local value is compared to a national frame. It is important to mention that the social gap is not a measure of poverty by itself since it does not incorporate the labor income and social security (CONEVAL, 2010). The lower the metric, the fewer the social deprivations of a population group. These two variables were chosen for this study because these indices are sensitive to the detection of resilience in regions affected by extreme rainfall events; in other words, they define social vulnerability. It is worth noting that the values of both variables are computed every 10 years at a local level by the National Commission for Knowledge and Use of Biodiversity (CONABIO, 2010). Consequently, their values vary slowly when compared to natural hazards. Vulnerability is considered a low-frequency modulator of climate risk and disasters (Neri and Magaña, 2016). The degree of local marginalization and local social gap are normalized by using Equation (4) because the two metrics have different magnitudes: Where X is the social variable and X min and X max are the lowest and the highest value of the variable in the whole country, respectively. Figure 2a show that this social variable has increased by up to 0.8 in magnitude over some regions of the country (Figure 2c). On the contrary, the local social gap has slightly decreased. This small reduction suggests that the number of people in poverty conditions has increased (Figure 2c), and fewer people have access to education, adequate housing, health, and basic services, decreasing at the same time the difference in the local social gap because more people are in the same socioeconomic condition (CONEVAL, 2010) (Figure 2f). In this way, social vulnerability is defined as an index obtained from the multiplication of the two social variables, and it may be expressed from 0 to 1. When the index is zero, it means non-vulnerable, and when the index is one, it means significantly vulnerable. Figure 2g

| Methodology for quantifying climate risk
Climate risk "R" depends on a physical phenomenon called natural hazard, and a social condition called vulnerability, as shown in Equation (5), where "H" means natural hazards, which is defined as the probability of TC hazard occurrence, and "V" represents local social vulnerability, which is depicted as an index that considers the degree of local marginalization and local social gap of the population that face such natural hazards (Figure 2h). Values range from 0 to 1, where 0 means non-risk, and 1 means the highest risk and the imminent occurrence of a disaster.
Equation (5) does not include exposure-which is considered an element of social vulnerability-because a community is not vulnerable if it is not exposed (Neri and Magaña, 2016). Guimarães-Nobre et al. (2019) propose that ENSO can change extreme rainfall probabilities over coastal regions and, consequently, environmental risk and economic damages worldwide. We build on this idea by exploring a relationship between ENSO and climate risk over Mexico. Mean climate risk for different ENSO conditions was defined as the average of climate risk for the years when ENSO phases are active. The climatological risk is the average of climate risk for the whole period. Anomalies in climate risk were also defined under ENSO conditions. These anomalies are computed as the average of climate risk under ENSO conditions minus the climatological risk. Once the subtraction is obtained, it is divided by the climatological risk and multiplied by 100%. In this way, anomalies are expressed as changes in percentage from the average. To validate the skillfulness of climate risk, its regional anomalies under different ENSO conditions were compared with regional disaster costs.

| EM-DAT: Disaster cost dataset
The Centre for Research on the Epidemiology of Disasters (CRED) of the Université Catholique of Louvain created a database called EM-DAT (EM-DAT, 2020). It has records of economic losses associated with different hazards (earthquakes, fires, convective storms, cold waves, tropical cyclones, droughts, among others) from 1900 until now. However, we only used the information associated with TCs that are recorded during the 1981-2017 period. The disaster costs per administrative state unit were also analysed under different ENSO phases to explore if this climate oscillation has a signal on the study regions and to validate our climate risk analysis.

| RESULTS
Summer precipitation in Mexico varies from a semi-arid climate over northern Mexico to a wet climate over southern Mexico. The northeastern region receives up to 600 mm per year, while the northwestern region rains less than that amount annually. The accumulated summer rainfall over southern regions is more than 1,000 mm (Figure 3a). The highest values of extreme daily precipitation (P95), more than 54 mm day −1 , are located in coastal regions of the Gulf of Mexico, and the lowest values of P95 are located in northwestern Mexico, except for the south of Baja California Sur (Figure 3b). The extreme precipitation pattern is well related to the orography (Figure 1), as the largest values of P95 are located in shorelines, and lower values are inland. This behaviour could be associated with the presence of the eastern and western Sierra Madre mountain ranges. The climatological extreme rainfall days is an average of the days that exceed P95 from May to November during the 1981-2017 period (Figure 3c), and the climatological TC days is an annual averaged amount of days when the rainfall produced by TCs exceed P95, in the same study period (Figure 3d). Overall, our results capture TC days adequately in most of the country, but they overestimate the number of TC days over the south of Baja California Sur (from 22 N to 25 N).
On average, TC produced 2-3 extreme rainfall days per year in places where there are 5-6 extreme summer precipitation days (Figure 3c,d). Thus, TCs may causẽ 50% of the P95 events, demonstrating that these tropical phenomena largely influence the extreme precipitation days, mainly over shorelines of the Gulf of Mexico, Caribbean Sea, and Eastern Pacific Ocean basin.

| ENSO as the main climatic modulator of TC days
Under strong El Niño conditions, the summer precipitation anomalies over central-southern Mexico are below normal, and these regions can experience a deficit of 200 mm in their accumulated rainfall (Figure 4a). This reduction in overall precipitation also leads to fewer extreme precipitation days ( Figure 4b) and results in negative anomalies of those days (Figure 4c). The latter is because deep convection chances diminish due to a southward displacement of the ITCZ, an enhancement of subsidence (Magaña et al., 2003), and easterly wave F I G U R E 3 (a) Mean accumulated summer rainfall (mm), (b) mean extreme summer precipitation values (mm/day), (c) mean extreme summer precipitation days per year, and (d) mean extreme rainfall days per year produced by TC passages during the 1981-2017 period. Extreme rainfall is the precipitation value higher than the 95th percentile of daily precipitation from May to November, defined as summer rainfall E3314 tracks are below normal (Dominguez et al., 2020) over these regions. Under neutral conditions, slightly positive anomalies of summer precipitation are seen over the coastal states of the Gulf of Mexico (Figure 4d), but no pattern in the anomalies of extreme days is seen (Figure 4f). TCs tracks and easterly waves have no clear signal under neutral conditions, leading to no signal in extreme precipitation days (Dominguez et al., 2020). Although the places where TCs can make landfall are highly dependent on the type of track (Dominguez and Magaña, 2018), this relationship does not hold under neutral conditions. Interestingly, under neutral ENSO conditions, the chances of extreme precipitation events only increase over northwestern Mexico when compared to strong El Niño and La Niña conditions, particularly in Baja California Sur (Figure 4f). Under strong La Niña conditions, the anomalies in summer precipitation are above-normal (around 200 mm) in central-southern Mexico (Figure 4g), which is opposite to what happens during strong El Niño years. However, strong La Niña years can also result in negative summer rainfall anomalies over the northwestern region ( Figure 4g). During this ENSO phase, more extreme precipitation days, which is indicated by positive anomalies of extreme days, occur in the southwestern region and the Yucatan Peninsula when compared to strong El Niño events (Figure 4i).
The large-scale environment is modulated by ENSO conditions, and this can also influence the number of TCs days per season and the amount of TC rainfall ( Figure 5). During El Niño years, TCs tend to produce less accumulated rainfall when compared to the other ENSO phases (up to 60 mm) over northeastern Mexico and the Yucatan Peninsula (Figure 5a). As a result, the number of TC days is also reduced by this ENSO phase over those regions (Figure 5c). Under neutral years, the signal of the increase in TC days over northwestern Mexico (Figure 5f) remains consistent when compared to total extreme rainfall days (Figure 4f), particularly in Baja California Sur. This relationship means that the extreme precipitation days over this region are highly linked to TC passages. This result is consistent with Zuñiga and . Moreover, the chances of TC days in the coastal states that are located west of southwestern Mexico also increase during neutral conditions (Figure 5f). Under La Niña years, TCs produce more precipitation over the Yucatan Peninsula, northeastern Mexico, and the coastal states of the Gulf of Mexico when compared to the other ENSO phases (Figure 5g). So, TC days also increase over these regions (Figure 5i).
TC density is computed as the number of TC tracks per unit area (1 x 1 ) from May to November for a given year. The TC density for each ENSO phase is an average per unit area of the tracks that occurred during the years when the phase was active. For example, the TC tracks that occurred in 1982, 1987, 1997, and 2015 are considered to make an average of TC density under El Niño conditions (Figure 6a). The anomalies of TC density are the difference between TC density for each ENSO conditions and the climatological TC density for the whole period. The anomalies of TC density under El Niño conditions show that few TCs tend to impact Mexico (Figure 6b). Besides, TC tracks that are more than 500-km-distant from Mexico, such as tracks that form over the main development region and recurve into the southeastern United States, can induce subsidence and moisture flux divergence, leading to an inhibition of tropical convection over the Mexican continental mass (Dominguez and Magaña, 2018). As a result, the chances of TC days over Mexico decrease (Figure 5c).
Under neutral conditions, it is noticeable that more TCs affect the regions located over the Eastern Pacific basin, especially Baja California Sur (Figure 6d). The analysis of TC days under neutral years also supports this result (Figure 5e,f). However, large-scale circulations are not strongly modulated by this ENSO phase. Our results represent a mean for neutral conditions, and each seasonal TC frequency under this phase is not necessarily on average. A particular TC season could be inactive (below-normal) or active (above-normal) in terms of the number of TCs, as part of the interannual climate variability, but TC tracks may not make landfall over the Mexican continental mass in either case (Dominguez and Magaña, 2018). That is why TC days over western Mexico depend on the types of TC tracks over the Eastern Pacific basin and not only on TC frequency (Dominguez and Magaña, 2018). Under La Niña years, more TCs tend to impact northeastern Mexico, the coastal states of the Gulf of Mexico, and the Yucatan Peninsula (Figure 6e,f). This enhances the threat of extreme rainfall events occurring over the latter regions (Figure 5i).
In summary, ENSO plays an essential role as a climate modulator of extreme precipitation events and TC days. Strong El Niño years result in less extreme precipitation events and fewer TC days over Mexico than in other ENSO phases. This relationship can be reinforced by enhanced subsidence and moisture divergence associated with distant TC tracks from the Mexican continental mass. Although the TC activity can be normal or belownormal over the Eastern Pacific Ocean under strong El Niño conditions (Figure 6b), Farfan et al. (2013) show that at least one hurricane (category 2) can occur. The latter can lead to a disaster if vulnerability is high (Figure 2h). Strong La Niña years not only result in an increment of TC days (natural hazards) over the east of Mexico but also in a slight decrease of TC days over the central-western coastal regions of Mexico. Interestingly, under neutral years, the chances of TC days increase over northwestern Mexico and the central-western coastal F I G U R E 6 An annual average of tropical cyclone density for: (a) strong El Niño years, (c) neutral years, (e) strong La Niña years, and anomalies of tropical cyclone density for: (b) strong El Niño years, (d) neutral years, (f) strong La Niña years during the 1981-2017 period regions of Mexico when compared to other ENSO phases (Figure 5f).

| Trends in tropical cyclone precipitation
Linear trends in TC rainfall and TC days were analysed for the 1981-2017 period. Both accumulated TC rainfall and TC days have increased (Figure 7), with no significant decreases over Mexico. Two tests were applied to have robustness in the results. The trends are robust, since both statistical tests revealed the same spatial distribution. For example, southeastern, northwestern, and southwestern regions have significant rising trends, both accumulated rainfall and days related to TC passages. These results are consistent with Cavazos et al. (2008) over northwestern Mexico and Perez-Morga et al. (2013) over southwestern Mexico. Knutson et al. (2020) found that TC rainfall will increase 14% worldwide under 2 C global warming at medium-to-high confidence, so Mexico should be prepared to face TCs that can produce more torrential rain in the coming years under climate change. However, no robust consensus in projected changes in TC tracks and TC sizes was obtained at a global scale. Moreover, Kossin et al. (2020) recently demonstrated that the frequency of major TCs increased by about 8% per decade for the 1979-2017 period.

| Climate risk modulated by ENSO and disaster costs
The southwestern region has a high mean probability of TC hazard occurrence (up to 0.8) under strong El Niño conditions (Figure 8a), but this likelihood diminishes by up to 0.3 during strong La Niña conditions (Figure 8g). Under neutral conditions, the probability of TC hazard occurrence over Baja California Sur is more than 0.8 (Figure 8d), which surpasses the probability during strong El Niño and La Niña years. Hence the probability of TC hazard occurrence is greatest over the Baja California Sur during neutral conditions. On the eastern coast of Mexico, the probability of TC hazard occurrence is more than 0.6 under La Niña conditions (Figure 8g).
ENSO induces changes in climate risk due to its strong relationship with the occurrence of TC days. Under strong El Niño conditions, southwestern Mexico is at a high risk of developing a disaster because its values are more than 0.5 (Figure 8b), and the climate risk enlarges more than 80% from average over the Pacific coast of southwestern Mexico (Figure 8c). Most of the cities of this region rely on touristic activities, and they are highly vulnerable to extreme precipitation because the local degree of marginalization and the local social gap are more than 0.7, which is high (Figure 2b,e). Thus, any disruption of touristic activities will likely cause a disaster. On the other hand, climate risk diminishes more than 80% from average over the eastern coast of Mexico (Figure 8c). The occurrence of TC hazards over the northeastern and northwestern regions varies from 0.1 to 0.6 (Figure 8a), and their social vulnerability is less than 0.3 (low) over those regions (Figure 2h). As a result, mean climate risk over northwest and northeast of Mexico is 0.3 or less (low) (Figure 8b). Even when the probability of TC hazard occurrence is high, mean climate risk will be close to 0 if the social vulnerability is low. For example, the vulnerability in the Peninsula of Baja California is low (Figure 2h), and consequently, this impacts the magnitude of climate risk (Figure 8b). Under strong La Niña conditions, the eastern coast of Mexico has moderate values of risk that range from 0.2 to 0.5. Specifically, the northeast of Mexico and the Yucatan Peninsula have moderate values of climate risk (0.3-0.4) even when both regions have a high probability of hazard occurrence (more than 0.6) because their social vulnerability is low (no more than 0.3). Therefore, mean climate risk is also low over those regions (Figure 8h). A strong dipole in climate risk anomalies appears between the eastern and western coasts under this phase (Figure 8i). The anomalies in climate risk are more than 80% from average over northeastern Mexico and the Yucatan Peninsula, leading to a possible disruption in their economic activities.
One way to validate our approach is to compare the positive anomalies in climate risk, which show the regions that have more chances to go through disasters, with economic losses. We used the EM-DAT dataset to explore where disasters occurred under different ENSO phases in Mexico during the study period. The socioeconomic impacts for each ENSO phase are defined as a mean cost of the disasters in years when strong ENSO conditions are present. During strong El Niño years, 18.8% of all Mexican states suffered expensive disasters. The highest climate risk and their positive anomalies are located over southwestern Mexico (Figure 8b,c). The EM-DAT shows that economic losses related to TCs are up to 60 US million dollars on average over this region (Figure 9a). However, the negative anomalies in climate risk do not indicate that the northeastern region also burdens with disaster costs (Figure 8c and 9a). During neutral conditions, 37.5% of all states were economically affected by disasters. Most of the TC hazard-prone regions over the western coast of Mexico go through up to 100 US million-dollar losses (Figure 9b). Positive anomalies in climate risk show where disasters have F I G U R E 8 Mean probability of TC hazard occurrence (days when tropical cyclones produced extreme precipitation) for: (a) strong El Niño years, (d) neutral years, (g) strong La Niña years, mean climate risk for: (b) strong El Niño years, (e) neutral years, (h) strong La Niña years, and mean anomalies in climate risk (%) during (c) strong El Niño phase, (f) neutral ENSO phase, and (i) strong La Niña phase more chances to occur (Figure 8f), but they fail over the eastern coast, mainly in the Yucatan Peninsula. During strong La Niña years, only 28.1% of Mexican states reported economic losses, but the states located over the eastern coast registered more than 100 US million dollars in socioeconomic impacts ( Figure 9c). Therefore, these ENSO conditions are the most dangerous ones for the country. Interestingly, the positive anomalies in climate risk show adequately the regions where disasters have more chances to materialize, except for the north of the Peninsula of Baja California that has negative anomalies in climate risk and disaster costs are 90 US million dollars (Figure 8i and 9c).
In summary, it is concluded that the local vulnerability and ENSO conditions largely influence changes in climate risk and exacerbate socioeconomic impacts associated with TCs. Under strong El Niño years, climate risk enhances 80% more from average over the southwestern region. While under strong La Niña years, this region diminishes its climate risk values up to 80% from average. The occurrence of TC days threatens the eastern coast of Mexico under this phase, and consequently, the climate risk amplifies its magnitude 80% more than average. Remarkably, climate risk raises more than 40% from average over the northwestern region under neutral conditions, mainly in Baja California Sur. These results under different ENSO conditions only show where disasters have more chances to occur, and they are based on the probability of TC hazard occurrence and social vulnerability.
Our definition of climate risk needs additional social information to improve its magnitude. For example, it would be ideal to have the level of local preparedness to face hazards and public perception of risk in a high spatial and temporal resolution ). As inferred from Figure 8, climate risk values could be moderate (0.2-0.4), but wrong decisions, incompetent local strategies to face hazards, and weak prevention measures can turn extreme events into disasters (Alcántara-Ayala et al. 2019).

| SUMMARY AND CONCLUSIONS
The extreme rainfall events are defined as the number of days when precipitation produced by tropical cyclones (TCs) exceeds the 95th percentile of daily rainfall distribution in a given year, and these TC days are considered natural hazards. In that sense, TCs are critical phenomena that cause~50% of the extreme precipitation events, mainly over shorelines of the Gulf of Mexico, Caribbean Sea, and Eastern Pacific basin. El Niño-Southern Oscillation (ENSO) is the main driver of atmospheric circulations over the tropical and subtropical Americas and modulates the occurrence of TC days. Under strong El Niño conditions, TC activity tends to diminish in both basins, leading to a decrease in the occurrence of TC hazards in Mexico. Under strong La Niña conditions, the eastern coast of Mexico is threatened by an enhancement of TC days, mainly over northeastern Mexico and the Yucatan Peninsula. We find a remarkable signal under neutral conditions: more TC days tend to affect Baja California Sur because more TC tracks impact this region on average. Although our results overestimate the number of TC days over the south of Baja California Sur, the behaviour of TC days over this region under neutral conditions is consistent with Martínez-Sanchez and Cavazos (2014). Moreover, the south and northwest of Mexico have experienced an increase in TC rainfall and TC days. Our results are also consistent with Dominguez and Magaña (2018), who also found a positive trend in TC rainfall for Baja California Sur using rain gauge data.
A social vulnerability index was constructed using two metrics: the degree of local marginalization and the local social gap. The two variables measure to what extent local groups of population have access to education, suitable housing, and medium-to-high labor incomes when compared to a national frame. These factors underline social vulnerability.
Climate risk is defined as the multiplication of the probability of TC hazard occurrence by social F I G U R E 9 Mean costs of disasters in US million dollars per administrative state unit reported by EM-DAT for: (a) strong El Niño years, (b) neutral years, and (c) strong La Niña years vulnerability. It was also explored under different strong ENSO conditions. During strong El Niño years, climate risk is triggered by more than 80% from average in southwestern Mexico. During strong La Niña years, climate risk is reduced up to 80% from average over this region, but its magnitude is amplified 80% more than average over the eastern coast of Mexico. Under neutral conditions, climate risk enlarges its magnitude by more than 40% than average over northwestern Mexico, mainly the Peninsula of Baja California. However, this relationship highly depends on the type of TC tracks over the Eastern North Pacific Ocean, as each type of TC track determines the place of landfalling (Dominguez and Magaña, 2018). These results only point out the regions where disasters have more chances to occur under different strong ENSO phases based on our definition of climate risk. They should not be interpreted as predictors of TC activity or TC days. Only skillful climate predictions can be provided by strong ENSO conditions. These seasonal forecasts should determine the occurrence of certain types of TC tracks and the specific impacted regions where social vulnerability is high. That could help to achieve a proper disaster risk reduction. Moreover, future early warning systems (EWSs) for TCs will be valuable if a multi-risk approach is considered (Claudet et al., 2020). New EWSs should include several TC hazards (storm surges, high speed of the wind, translational speed, TC rainfall, and TC size) and social vulnerability at the same time.
Our approach was validated through a comparison between positive anomalies in climate risk and the costs of disasters. Under El Niño conditions, the northeastern and southwestern regions can experience disasters of up to 60 US million dollars. Under neutral conditions, more Mexican states can be affected by TCs when compared to the other ENSO phases, and disasters may occur over the northwest and south of Mexico. Under La Niña conditions, the eastern coast of Mexico undergoes economic losses of up to 100 US million dollars. It is concluded that the country has more expensive socioeconomic impacts under this phase. Our results show that positive anomalies in climate risk adequately point out where disasters have more chances to occur under strong La Niña conditions. However, our results under different strong ENSO phases should be improved by adding other social information.
These findings are the first approach to quantify climate risk under ENSO conditions. Our results can help policymakers point out regions where disasters have more chances to occur under different strong ENSO phases. To meet the objectives of international risk agendas (Sendai framework, Agenda 2030, and Decade for Ocean Science), the scientific community needs to familiarize itself with vulnerability and risk and not only focus on natural hazards (Sutton, 2019). Social response and organization are also significant factors that determine where disasters materialize (Alcántara-Ayala,-2019b). In that sense, disasters will be decreased if vulnerability is aimed to be reduced, even when the frequency and intensity of TCs increase in the future.
Mexico needs further studies that define how TC hazards and vulnerability vary regionally. Future work should create a TC hazard index that adequately incorporates the main TC features over Mexico, for instance: TC size, their translational speed, extreme precipitation, and TC category, as shown in Holland et al. (2019). Apart from that, population growth, the level of local preparedness to face hazards, risk perception of the population, and the effectiveness of local decisions should be estimated to have a better understanding of how vulnerability and disasters are socially constructed.

ACKNOWLEDGMENTS
We thank Dr. Jose Santos and three anonymous reviewers for their helpful comments on the early version of this manuscript. The comments made by Dr. Bradford Barrett in the early version of the manuscript are highly appreciated, as well as the technical assistance provided by Dulce Nieto. This study was financially supported by UNAM-PAPIIT under the grant IA100620. A. Jaramillo acknowledges the fellowship from DGAPA at UNAM. ORCID Christian Dominguez https://orcid.org/0000-0002-2787-052X Alejandro Jaramillo https://orcid.org/0000-0002-4175-7726