World’s fast-growing urban areas, especially in developing countries, will likely suffer from the consequences of climate change, a major issue facing human societies. The projected impacts of climate change are likely to require some adaptation, which in turn is based on future projections of whether adaptive capacity assets will be drawn upon in times of need [1] . Adaptive response, a component of vulnerability, is dependent on a range of variables that operate at diverse scales [1] ; among them, the frequency and intensity of extreme events which are in connection to population vulnerability, may favor natural disasters.

Although part of the natural system, extreme weather and climate events can cause impacts and major disruptions in natural and human systems. The level of impacts depends on the intensity of the change and also upon the capacity of the social groupings to avoid and/or to overcome their negative effects. Therefore, aspects such as poor infrastructure facilities, inexistent or weak institutional me- chanisms to cope with extremes of climate variability and change such as lack of financial resources as well as socio-spatial disparities, are potential contributors to more severe impacts of extreme weather and climate events. Because such extreme events can reduce the economic growth and development and obstruct competitiveness; their evaluation must be part of a more complex task of assessing extremes to provide knowledge and to inform societal risk relevant to adaptation [2] . Components of vulnerability include exposure, sensitivity risk and resilience [3] . Adaptation measures can contribute to reducing risks making a society less vulnerable and more resilient to extremes.

Precipitation extremes over São Paulo state have been discussed in several stu- dies [4] [5] [6] [7] [8] . Studies on observed precipitation extremes over South- eastern Brazil, where the São Paulo state is located, are numerous [9] [10] [11] , among others. Model simulations of extremes over South America, including São Paulo state were presented in [12] and [13] . Most of these studies have shown that rainfall extremes have been increasing during the last five decades, and future projections show the similar trend. Uncertainties in future of precipitation projections over South America were discussed in [14] [15] [16] . The main source of uncertainty has been identified as the differences in physical parameterizations among different models and also the possible role of urbanization, something that is not yet well represented in climate models.

In the present study, the analyses are focused on two metropolitan regions of São Paulo state using a regional model and future projections results to provide information that can be useful to adaptation measures in face of the vulnerabilities of the sites. The Metropolitan Region of Campinas (MRC) and the Metropolitan Region of Baixada Santista (MRBS) are dynamic economic centers with rich ecosystems, but also low environmental threshold¹, aggravated by the population increase and occupation of more risk-prone locations, such as hilly regions, which are affected by landslides or areas at risk of flooding. These actions led to relevant environmental changes during the last decades and made population more vulnerable to the impacts of extreme weather and climate.

São Paulo state is located in southeastern South America, in a region affected by the South Atlantic Convergence Zone (SACZ) [11] [17] [18] [19] , a meteorological system that causes heavy and persistent precipitation. Heavy rainfall events over Southeastern Brazil have been associated with the SACZ and cold fronts [9] . In addition, these events can have the contribution of humidity flux from Amazonia, which is one of the sources of moisture to southeastern Brazil [20] . A review of large scale and synoptic systems influence on extreme precipitation over Southeastern Brazil is given in [21] , where some cases during the first decade of 21^st century were discussed. Positive precipitation trend identified in São Paulo city, the largest city in Brazil, during the wet season from 1933 to 2010, was attributed in part to natural climate variability and also to the urbanization effects [7] . Besides the heat island effect, the influence of sea breeze intensification due to warmer nights on the positive precipitation trend was suggested by [6] .

The Eta Regional model has been used in operational weather forecasting at the Center for Weather Forecasting and Climate Studies/National Institute for Space Research (CPTEC/INPE). This model has also been used in climate simulations and projections of climate change [6] [22] [23] [24] [25] . These projections were obtained from the Eta model downscaling of the HadCM3 with four integrations using the A1B scenario, and HadGEM2-ES, MIROC5 and BESM from CMIP5, using the RCPs 4.5 and 8.5 scenarios for 1961-1990, 2010-2040, 2041-2070 and 2071-2100. Precipitation, temperature and atmospheric circulation climatology of the base period were validated against observations in [22] . The four members simulated the seasonal difference between DJF and JJA, re- presenting the typical SACZ pattern over tropical and subtropical South America.

The model results and observed data information are presented in Section 2. In Section 3, the geographic and socio-economic features of the two metropolitan regions are informed. Section 4 indicates the precipitation changes projected by the model in the regions, the member spread and uncertainties. Extreme precipitation indices are shown in Section 5, and precipitation patterns simulated by the model in cases of extreme monthly precipitation are discussed in Section 6. Conclusions are shown in Section 7.

The observed precipitation data were obtained from the São Paulo state data set archive: http://www.sigrh.sp.gov.br/cgi-bin/bdhm.exe/plu. Simulation and projections were analyzed from results of the Regional Eta model, with 40 km horizontal resolution and 38 vertical levels. The lateral boundary conditions were provided by the HadCM3 coupled climate model [26] with the A1B scenario for the future projections. Four members of simulations and projections, based on climate sensitivity were analyzed. Details of these integrations can be found in [22] . Maps of precipitation difference between two future periods (2011-2040) and (2070-2099) and the base period (1961-1990) are shown in percentages compared to the base period for the ensemble, ensemble plus and ensemble minus standard deviation among the members. Results from the four members were analyzed in the two locations, MRC and MRBS. Uncertainties are analyzed from the maps and considering members spread in the graphs.

Daily data from 1961 to 1990 (base period) and for three periods of the 21^st century (2011-2040, 2041-2070 and 2071-2098) were used to calculate the precipitation indices shown in Table 1. The climatological monthly observed data and model results were used to obtain the annual cycle during the base period. Daily rain intensity was analyzed regarding class intervals of precipitation <1 mm to >51 mm. Monthly data was used to calculate the Standardized Precipitation Index (SPI). This index identifies extreme wet/dry (SPI ≥ 2.0/SPI ≤ −2.0), severe wet/dry (1.5 ≤ SPI < 2.0)/(−2 < SPI ≤ −1.5) and moderate wet/dry (1.0 ≤ SPI < 1.5)/(−1.5 < SPI ≤ −1.0) conditions. The anomalous precipitation patterns over South America associated with extreme wet monthly precipitation obtained from the 4 members in the 2 locations were analyzed in composites of precipitation anomalies for each season.

The vulnerability of each municipality is presented by means of the Social Vulnerability Index (SVI), developed by [27] which combines several variables of the 2010 Brazil census to indicate the vulnerability of each municipality. The SVI seeks to highlight the social inequality scenario in each metropolitan region, indicative of exclusion and social vulnerability, and was constructed using sixteen indicators grouped into three main dimensions: urban infrastructure, human capital and income and employment. Each dimension corresponds to a set of assets, resources or structures whose access, absence or failure indicate that the pattern of family life is low, suggesting, in extreme cases, the lack of access and non-compliance of social rights. Each variable has a relative weight, revealing their importance for the whole index.

The first dimension, the urban infrastructure vulnerability, reflects the conditions of access to basic sanitation and urban mobility and has three variables, each one with a weight in the index: percentage of people in households with inadequate water and sewage (0.3); percentage of population living in households without regular waste collection services (0.3); percentage of people living in households below poverty estimate and who spend more than one hour a day com- muting (0.4). The second dimension, human capital, involves health and education

and has 8 indicators, each one weighted 0.125. The third and last dimension, income and employment, is composed of five indicators, each one weighted 0.2. The combination of the variables allows a unique value of exclusion and social vulnerability to the cities of the metropolitan regions. The SVI ranges between 0 and 1, the closer to 1, the greater the social vulnerability of a given municipality. Classes of social vulnerability are: 0 - 0.2 very low; 0.2 - 0.3 low; 0.3 - 0.4 medium; 0.4 - 0.5 high; and above 0.5 very high.

Results from the Eta-HadCM3 model simulations from the base period (1961-1990) and model projections for the future are discussed in this section. The annual cycle of observed and model ensemble precipitation in MRC and MRBS are shown in Figure 4. The annual cycles are reasonably well simulated, although with overestimated values in MRC and underestimated values in MRBS, where there is larger underestimation in the austral summer and autumn. The differences between the two annual cycles are likely related to the different precipitation regime between the two regions. In Figure 4, it is seen that at MRC, located relatively far from the ocean, the precipitation regime presents a seasonal contrast, with large amounts during summer, when a series of processes operate in the area bringing more moisture, and when there is development of the SACZ, while precipitation in winter, when the atmospheric circulation is different from summer, is primarily linked to the cold fronts. On the other hand, although MRBS is also affected by the SACZ, it presents lower inter- and intra seasonal variability and larger precipitation than MRC during the whole year as it is influenced by the humidity advection from the sea. In addition, the reduction from the summer to autumn and winter is gradual, while in MRC there is a greater reduction. The monthly differences between the future periods and the base period are indicated by the columns in Figure 4. There is a dominance of rainfall increase from late spring to early winter, in the three future periods, in

MRC. Reduced rainfall is projected from middle winter to middle spring. At MRBS small reductions are projected for August and September. From middle spring to early winter the projections show increased precipitation except in February and July, when there is a very small reduction. Then, in both locations, the rainy season is projected to be wetter and the dry season to be drier than in the base period.

Spatial changes in two future periods: near future (2011-2040) and distant future (2070-2099) compared to the base period (1961-1990) of MRC and MRBS are shown in Figures 5-8. The uncertainties are discussed in terms of the standard deviation among the members. The ensemble mean, ensemble mean plus or minus one standard deviation indicate large or small uncertainties and the members spread. In DJF (2011-2040) the ensemble mean indicates an increase of precipitation between 5% and 10% at MRC (Figure 5(a)). The increase interval is between 0% and 20%, from the ensemble minus to the ensemble plus. During DJF (2070-2099) the range of increase is between 10% to 20% in the ensemble and between 0% to 30% from the ensemble minus to the ensemble plus (Figure 5(b)). At MRBS the projections also show precipitation increase for the ensemble between 0% and 5% in the majority of municipalities (Figure 5(c)). Compared to the MRC, the uncertainty is larger at MRBS, since the ensemble minus standard deviation indicates a reduction between 0% and 5% and the ensemble plus standard deviation indicates an increase between 5% and 10% (Figure 5(c)). The uncertainty at MRBS is reduced during DJF (2070-2099) since the ensemble, ensemble minus and ensemble plus indicate increases between 0% and 20% (Figure 5(d)). Figure 9 shows the spread among the four members of precipitation differences between three future periods and the base period. At MRC all members show increased precipitation in the future periods during DJF, indicating a convergence in the model response and then more confidence on the results (Figure 9(a)). At MRBS there is convergence among the members on precipitation increase during the last two future time slices (Figure 9(b)).

In MAM (Figure 6(a)), the uncertainties range from −10% to 20% at MRC during the period (2011-2040) and between −5% and 30% during (2070-2099),

(Figure 6(b)). At MRBS the interval is between −5% to 10% (Figure 6(c)) during (2011-2040) and the uncertainties reduce during (2070-2099) as the changes show increase between 10% and 30% (Figure 6(d)). The spread in Figure 9 shows precipitation increase in the majority of members in both locations. JJA (2011-2040) and (2070-2099) present the larger uncertainties regarding the standard deviation among the members in both regions (Figure 7). The ensemble minus the standard deviation shows reduced precipitation in both locations and the ensemble plus the standard deviation shows increased precipitation. The interval ranges from −40% to 10% in MRC and from −20% to 30% in MRBS during (2011-2040). During (2070-2099) the intervals change from −40% to 20% in MRC and from −10% to 30% in MRBS. The uncertainties are also seen in the spread (Figure 9), although the ensemble shows reduction during two future periods in MRC and increase during three future periods in MRBS. In SON, MRC displays changes between −10% and 5% and MRBS changes between −5% and 20% during (2011-2040) (Figure 8(a), Figure 8(c)). The interval of changes increases in (2070-2099) from −20% to 30% in MRC and between 0% and 40% in MRBS (Figure 8(b), Figure 8(d)). Therefore the uncertainties in SON are larger in MRC than in MRBS, which is also seen in Figure 9 that shows precipitation increase in MRBS projected by the ensemble and the majority of members, while in MRC there is large spread among the members.

The indices of Table 1 are shown in Figure 10(a) and Figure 10(b) for MRC and MRBS, respectively. SDII and SDII10 are well represented by the model, while R1 and R10 are overestimated in MRC. In this location, the projections show a gradual increase in the mean precipitation above 1mm (SDII) and 10 mm (SDII10) from the base period (1961-1990) to the last period of 21^st century (2071-2098), with low spread among the members, represented by the minimum and maximum values. The number of rainy days (R1) reduces along the period, while there are small changes in the number of days with heavier precipitation (R10). The concurrent reduction of rainy days and precipitation increase, in the future, indicate more intense rains concentrated in a few days. In MRBS the model underestimates the four indices, but there is also an increase of SDII and SDII10 and reduction of R1.

The changes in the distributions of rain intensity are presented in Figure 11 (MRC) and Figure 12 (MRBS). In MRC there is an increase of days without rain in two future periods (2011-2040) and (2041-2070), but in (2071-2100) the uncertainty is high, since there are two members with opposite changes. There are less days with light rains in the three future periods, and the reduction is larger in the last period. The number of heavy rain increases, mainly in the last two periods (Figure 11(b)). In MRBS (Figure 12) the behavior of the days without rain is similar, with increase of dry days in the first two future periods and more uncertainty in the last. There is also a reduction of days with light rains and increase of heavy rains in the future projections. This result is consistent with other studies over South America that show an increase in the frequency of daily precipitation extremes in future periods [6] [24] [34] .

Besides the heavy daily precipitation, which can cause damages in the two metropolitan regions, as discussed in session 3, monthly extremes are also important to agriculture and water resources. These are calculated from the Standard Precipitation Index (SPI), which identifies extreme, severe and moderate monthly events. As the SPI is calculated from anomalies relative to the climatology of each period, the results do not show a systematic change in the frequency of extreme, severe and moderate cases from the base to the future periods.

In order to see what is the simulated precipitation pattern related to the extremes in the two locations, seasonal composites from extreme wet monthly events during the base period were analyzed. Composites for each region showed similar patterns; therefore we show the composites regarding the selected cases of precipitation extremes in both regions of MRC and MRBS (Figure 13). In DJF the pattern resembles the dipole associated with the SACZ variability [11] [18] [19] [35] and indicates, in this case, the weakness or southward displacement of the SACZ (Figure 13(a)). Previous studies show that this dipole is related to the

Pacific South America (PSA) wave-train pattern [10] [19] [35] and humidity flux from the Amazon region [36] . In MAM the pattern may show the influence of frequent or/and persistent frontal systems. Dry conditions to the north, mainly related to the ITCZ displaced northwards may indicate a large scale subsidence which traps the frontal systems over part of southeast and south regions (Figure 13(b)). In JJA the pattern indicates again the presence of frontal systems and dry conditions in the tropical region (Figure 13(c)). The precipitation dipole returns in SON, season that marks the beginning of the rainy season in the South America Monsoon region [37] [38] [39] , although weaker than in DJF (Figure 13(d)).

It is also noted a precipitation dipole over Chile, which indicates the displacement of synoptic systems northward of the normal track. Therefore, the regional ETA model represents well the patterns associated with extreme monthly precipitation in the two regions of southeastern Brazil. It identifies the pattern of frontal systems, in MAM and JJA, and the influence of the SACZ in DJF and SON. The impact of these systems on observed extreme precipitation over Southeastern Brazil have been discussed in several studies, such as [9] [11] [22] . The ability of the ETA model in identifying seasonal patterns of extreme precipitation related to the influence of observed systems can increase the confidence on model results, reducing uncertainties, and also can be useful for seasonal predictions.

The two metropolitan regions (MRC and MRBS) are vulnerable in face of climate change. The model underestimates precipitation in MRBS, but represents well the annual precipitation cycle in MRC, although with overestimated values. In MRC, the projections in the near and distant future periods show precipitation increase during DJF, while there are more uncertainties in the other seasons. In MRBS, precipitation increase is projected during DJF, MAM and SON of the last period of the 21^st century. The uncertainties are larger in JJA in both locations. The ensemble mean shows increased precipitation from late spring to middle summer and reduced precipitation from middle winter to early spring months, indicating drier conditions in the dry season and wetter conditions in the rainy season. The increase in the summer season and the decrease in the spring are consistent with [40] which discusses the delay in the South American Monsoon System due to drier conditions in this later season in future projections. In MRBS, the reduction from middle winter to early spring is very small, likely due to the location close to the sea. The projections show a precipitation increase during the rainy days and a reduction of the rainy days’ number in both areas, indicating intensity increase in future projections.

The changes may impact the economy and may increase social problems due to floods and droughts in MRC and floods and landslides in MRBS. The vulnerabilities and the impacts of climate change need to be considered in adaptation actions in both regions. Even disregarding the effects of climate change, adaptation to the existing level of changes in both study areas is already critical. Due to the rapid transformation of these lively metropolitan regions, which place population at greater risk for natural disasters, including those expected to worsen with climate change, new challenges are posed, among them to understand and respond to the complex interactions between urban development and the climate system in view of building more resilient and adapted cities.

To IVA-FAPESP Project (Proc. 08/58161-1), FAPESP-Belmont Grant 2012/ 51876-0 and FAPESP RCUK PULSE Grant 2011/51843-2. I. F. A. C. and J. A. M. are grateful also to CNPq for research support.

Cavalcanti, I.F.A., Nunes, L.H., Marengo, J.A., Gomes, J.L., Silveira, V.P. and Castellano, M.S. (2017) Projections of Precipitation Changes in Two Vulnerable Regions of São Paulo State, Brazil. American Journal of Climate Change, 6, 268-293. https://doi.org/10.4236/ajcc.2017.62014