Hydrological Sciences Journal

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/thsj20

Nonstationary extreme precipitation in Brazil

Bianca Nespoli Cortez, Gabrielle Ferreira Pires, Alvaro Avila-Diaz, Humberto

Paiva Fonseca & Lais Rosa Oliveira

To cite this article: Bianca Nespoli Cortez, Gabrielle Ferreira Pires, Alvaro Avila-Diaz, Humberto
Paiva Fonseca & Lais Rosa Oliveira (2022): Nonstationary extreme precipitation in Brazil,
Hydrological Sciences Journal, DOI: 10.1080/02626667.2022.2075267

To link to this article: https://doi.org/10.1080/02626667.2022.2075267

View supplementary material

Accepted author version posted online: 11

May 2022.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=thsj20
Publisher: Taylor & Francis & IAHS

Journal: Hydrological Sciences Journal

DOI: 10.1080/02626667.2022.2075267
Nonstationary extreme precipitation in Brazil

Bianca Nespoli Corteza, Gabrielle Ferreira Piresa*, Alvaro Avila-Diazb,c,

Humberto Paiva Fonsecaa, Lais Rosa Oliveiraa
a
Department of Agricultural Engineering, Federal University of Viçosa, Viçosa , Brazil;
b
Universidad de Ciencias Aplicadas y Ambientales - UDCA, Bogota 111166, Colombia
c
Natural Resources Institute, Federal University of Itajubá, Itajubá 37500-903, Brazil

*P. H. Rolfs Av., s/n, Viçosa 36570-900, MG, Brazil, gabrielle.pires@ufv.br

Abstract

This study provides the first estimates of the frequency distribution of

observational and future maximum daily precipitation considering nonstationarity
for the entire Brazilian territory. We assess observational data from 1980 to 2015
and projected data from 2020 to 2099 for two climate change scenarios and four
downscaled climate models. We modeled extreme precipitation according to the
extreme values theory and calculated the precipitation intensity associated with
the return periods of 5, 10, 25, 50, and 100 years in nonstationary conditions. The
results indicate that nonstationarity is identified in 17.5% of the study area during
1980-2015. The analysis of future climate projections indicated an increase in the
return levels of extreme precipitation compared to the historical period in at least
90% of the national territory. It shows that engineering design must urgently
consider the nonstationarity of extreme precipitation under the risk of
increasingly unsafe infrastructure.

Keywords: extreme events, climate change, CMIP5, nonstationarity, NASA,

INPE
 1. Introduction

Recent studies show that extreme precipitation has changed over the past decades and

increased in intensity and/or frequency on the local, continental and global scales

(Alexander et al., 2006; McPhillips et al., 2018; Westra et al., 2013; Xavier et al., 2019;

Eyring et al., 2021). Although extreme events are associated with natural climate

variability, it is virtually certain that anthropogenic activity contributes to the

intensification of extreme precipitation on a global scale (Eyring et al., 2021; Donat et

al., 2016; Lehmann et al., 2015; Porto de Carvalho et al., 2014). In addition, increases in

extreme heavy precipitation are more frequent than decreasing trends (Westra et al.,

2013; Donat et al., 2016; Lehtonen and Jylha 2019), particularly related with the

increase in atmospheric moisture content associated with warming (Eyring et al., 2021) .

These findings suggest a need to better understand extreme precipitation events

behavior and how these changes can impact social and ecological systems. In urban

environments, the impacts may be related to urban mobility, water, energy supply, and

health, especially waterborne diseases. Furthermore, these events are associated with

urban floods and landslides, which are also correlated with accelerated and disorderly

urbanization (McPhillips et al., 2018). Most urban centers in Brazil are the result of a

disorderly expansion, lack of infrastructure, and the occupation of areas at geological

and geotechnical risk, leaving the population exposed and vulnerable to such impacts.

Densely urbanized areas might be even more vulnerable to extreme events disasters in

the future (Ávila et al., 2016; de Loyola Hummell et al., 2016; Debortoli et al., 2017;

Lyra et al., 2018).

In Brazil, numerous cases of urban floods and landslides are recorded every

year. Recent episodes were recorded in the states of Bahia (in 2021), Minas Gerais and

São Paulo (in 2020 and 2022), Espírito Santo and Rio de Janeiro (2020), Santa Catarina
in 2008, and in some municipalities in Rio de Janeiro (Angra dos Reis in 2009, Niterói

in 2010, and mountaintop cities in 2011) (IBGE, 2019a). All these events resulted in

many deaths and damages to local infrastructure and left thousands of homeless people.

With the intensification of climate change, the population might become even

more exposed, more vulnerable, and less adaptable to extreme events (Marengo et al.,

2011, Debortoli et al., 2017). Adaptation is urgent in this context and directly depends

on managing natural risks through technical solutions of geological-geotechnical

infrastructure and water resources (Marengo et al., 2021). The most common approach

to engineering design relies on defining an acceptable risk associated with the

probability of occurrence of an extreme event. These estimates are based on the

statistical distribution of stationary annual maximums time series (Lang et al., 1999;

Ouarda and Charron, 2019).

The notion of a stationary climate, which guides many technical solutions, is

based on the dynamics of the natural system that float within an immutable envelope of

climatic variability (Milly et al., 2008). However, human interference may introduce

nonstationarity to many meteorological and hydrological time series, especially in the

future (Field et al., 2012; Milly et al., 2008). Therefore, for technical solutions to be

effective and safe, they must be designed and operated considering climate change, not

based only on climate observations as projects are currently developed.

However, designing technical solutions that enable adaptation to climate change

depends on updated and accessible extreme events data. Frequently, the available data is

not generated in a format that managers and design engineers can easily use to mitigate

and adapt to future conditions (Markus et al., 2018). Thus, multidisciplinary studies

involving climate and engineering disciplines can play an essential role in assisting

infrastructure development to adapt the population to climate change.

 A few studies have identified trends in extreme precipitation for some regions of

Brazil (e.g., Porto de Carvalho et al., 2014; Silva Dias et al., 2013; Xavier et al., 2019,

Avila-Diaz et al., 2020b). More specific studies that address nonstationary extreme

precipitation series modeling are even more scarce (Assis et al., 2018; Pedron et al.,

2017; Zilli et al., 2017), and are focused on some municipalities in the south and

southeastern regions of the country. There is no study that assesses or estimates

nonstationary extreme rainfall events for the entire national territory in any time period.

Thus, this study aims to assess the occurrence of extreme rainfall in

nonstationary conditions in Brazil and its projections for climate change scenarios. In

this sense, the study focused on the observational period during the last four decades

(1980–2015). Furthermore, four downscaled resources from the National Aeronautics

and Space Administration (NASA) Earth Exchange Global Daily Downscaled

Projections (NEX-GDDP) and Brazilian National Institute for Space Research (INPE)

were used to assess nonstationary extreme rainfall for the 2020-2099 period. We expect

that the information provided in this research will be relevant to studies of impact,

vulnerability, and adaptation to climate change in Brazil's different socioeconomic

sectors.

2. Materials and methods

2.1.Study area and precipitation data

Annual extreme daily precipitation for the entire Brazilian territory (Figure 1) was

analyzed in this study. Much of the extreme rainfall occurs due to mesoscale systems

that can often be incorporated into synoptic systems that are influenced by large-scale

conditions, which in turn are also influenced by climatic variability (Cavalcanti, 2012).
Extreme precipitation in Brazil has been associated with anomalies both in the Pacific

and Atlantic Oceans, frontal systems, persistent systems such as the Intertropical

Convergence Zone (ITCZ) and South Atlantic Convergence Zone (SACZ), and

mesoscale convective systems (Lima et al., 2020; Cavalcanti, 2012; Luiz-Silva et al.,

2021). In addition, climate change increases the mean surface air temperature and sea

surface temperature, as well as changes the atmospheric water vapor content and

transport (Liu et al, 2020), with the potential to change the spatial patterns, intensity,

and frequency of extreme precipitation (O’Gorman and Schnieder, 2009; O’Gorman,

2015; Lehtonen and Jylhä, 2019)..

The maximum annual daily precipitation (mm.day-1) for each pixel of the study

area for the observational period (1980 to 2015) were obtained from the database of

Xavier et al. (Xavier et al., 2016), with the spatial resolution of 0.25°× 0.25° (latitude ×

longitude). This climate dataset was assumed as the reference because it is the most

complete gridded dataset, with 9,259 rain gauges to generate the daily fields of

precipitation over Brazil. In addition, this dataset has been used to analyze extreme

precipitation in many previous studies (Almagro et al., 2020; Avila-Diaz et al., 2020a,

2020b; Costa et al., 2020; Lucas et al., 2021).

Figure 1(about here)

The future projection of extreme precipitation was derived from RCP4.5

(Thomson et al., 2011) and RCP8.5 (Riahi et al., 2011) scenarios from 2020 to 2099

(subdivided into two periods: 2020-2059 and 2060-2099).

We chose the climate model data based on Ávila-Diaz et al. (2020a), considering

the best response to reproduce the annual maximum 1-day precipitation (RX1day)

between downscaled models and gridded observational datasets in the historical period
(1980-2005). Therefore, the RX1day index followed the same definition and

nomenclature in the climate community established by the Expert Team on Climate

Change Detection and Indices - ETCCDI (Seneviratne et al., 2021; Zhang et al., 2011).

Thus, we assessed downscaled climatic data from four climate models: the Hadley

Center Global Environmental Model, version 2 (HADGEM2-ES; Collins et al., 2011),

the Model for Interdisciplinary Research on Climate (MIROC5) (Watanabe et al.,

2010), Meteorological Research Institute Coupled Atmosphere-Ocean General

Circulation Model, version 3, MRI-CGCM3 (Yukimoto et al., 2012) and the

Community Climate System Model, version 4 CCSM4 (Lawrence et al., 2012). This

study has focused on statistically and dynamically downscaled approaches (Table 1). In

this sense, HADGEM2-ES and MIROC5 models were downscaled using a dynamical

approach by the Weather Forecast and Climate Studies Center (CPTEC) of the National

Institute for Space Research (INPE) (Table 1). Furthermore, MRI-CGCM3 and CCSM4

models were statistically downscaled by the National Aeronautics and Space

Administration (NASA) Earth Exchange Global Daily Downscaled Projections (NEX-

GDDP) (Table 1) using the Global Meteorological Forcing Dataset (GMFD; Sheffield

et al., 2006) and the Bias-Correction Spatial Disaggregation algorithm (Thrasher et al.,

2012 ).

Model outputs were interpolated to the same spatial resolution of the Xavier et

al.’s data. Finally, the main results presented here were determined by the ensemble

average (individual model estimates are shown in the Supplementary Material, Figures

S2 and S5, as well as the uncertainty in simulated precipitation, Figures S10 to S13).

Table 1 (about here)

 2.2. Extreme precipitation modeling

The generalized function of extreme values (GEV) provides a robust framework for

analyzing climatic extremes associated with return periods. This analysis structure

allows the incorporation of distribution parameters conditioned to covariables

representing climatic variability over time (Katz et al., 2002; Ouarda and Charron,

2019). According to extreme values theory, annual maximums' behavior can generally

be represented by one of the three GEV families: Gumbel, Fréchet, and Reversed

Weibull (Jenkinson, 1955).

GEV represents these three distributions combined, in a single parametric

family, through three parameters: location (µ), scale (σ), and shape (ξ) that governs the

tail behavior of the GEV distribution (Equations 1 and 2) (Coles, 2001). We adopted the

classic and stationary model, namely GEV0 (µ, σ, ξ), considering all parameters as

constants throughout time. However, as statistical properties (distribution type and

parameters) are expected to differ from the historical/observational to future/projected

data, µ, σ and ξ are calculated for each model, dataset, and time-period in all pixels.

≠0

( |μ, , ) = − 1+ (1)

=0

( |μ, , ) = − − (2)
 2.2.1. Nonstationary analysis

The inclusion of nonstationarity in GEV has been implemented in several studies in

order to represent time variation in climate variables (e.g., Agilan and Umamahesh,

2017; Cheng et al., 2014; Almeida et al., 2016; El Adlouni et al., 2007; Tramblay et al.,

2013, 2012; Vu and Mishra, 2019; Westra et al., 2013; Zhou et al., 2012; Xavier at al.,

2019; Liu et al., 2021). To integrate the effects of nonstationarity in extreme

precipitation modeling, the GEV parameters must change over time (Coles, 2001). This

solution of representing nonstationarity through time-dependent functions or covariates

intricately linked to temporal variation has been widely adopted in recent studies

(Cheng et al., 2014; Sarhadi and Soulis, 2017; Vu and Mishra, 2019).

In the present study, to account for nonstationarity, the location (µ) and scale (σ)

parameters assume a time-dependent trend, and only the shape parameter (ξ) remains

constant, as the accurate estimation of this parameter is very challenging (Coles, 2001).

Thus, the nonstationary and time varying GEV function is defined as GEVt (µt, σt, ξ).

We assume the location and scale parameters as a linear function of time to account for

nonstationarity, and thus, µt and σt are given by Equations 3 and 4, respectively. Several

authors show that the linear variation of the parameters with time (e.g., Sarhadi and

Soulis, 2017; Vu and Mishra, 2019 Xavier at al., 2019; Liu et al., 2021) is a plausible

approximation of nonstationary extremes, and, with our data, the quadratic variation did

not show any improvement in nonstationarity modeling, according to the deviation

statistic (D, Equations 5 and 6).

μ = + (3)

= + (4)
 The parameters vector (θ) is estimated for each GEV model: three parameters

for GEV0, θ = (µ, σ, ξ); four parameters for GEV1, θ1 = ( , , , ), and five

parameters for GEV2, θ2 = ( , , , , ) (Table 2). Importantly, distribution type

and parameters are calculated for each model, dataset, and time-period in all pixels.

Table 2 (about here)

2.2.2. Parameters estimation

The maximum likelihood method (Smith, 1985) was used to calculate and obtain the

best estimates of all the statistical parameters of the GEV models. The method consists

of a system of equations formed by the partial derivatives of the likelihood probability

function (ℓ) concerning each parameter. The solution of this system was determined

with the quasi-Newton Broyden-Fletcher-Goldfarb-Shanno (BFGS) optimization

method, as widely used in similar studies (Vasiliades et al., 2015). The algorithm is

available in gev.fit function in the ismev free package running on R (Heffernan et al.,

2018). The initial guess of the parameters, when not set by the user, is calculated based

on the mean and variance of the precipitation data, as described in the Supplementary

Material (section 1) and Heffernan et al., 2018.

2.2.3. Stationarity analysis and extreme precipitation model choice

The existence of trends in extreme precipitation was assessed using two tests (α=0,05):

the Mann-Kendall (Kendall, 1956; Mann, 1945) and the deviation (Coles, 2001) tests.

Mann-Kendall provides a preview of the existence and the sign of the trend in extreme
precipitation, highlighting the importance of nonstationary assumption. In addition, the

magnitude of the trend was determined by Sen’s slope estimator. The deviation test

compares the GEV models' validity based on the difference in their log of the maximum

likelihood. The deviation statistic (D, Equation 5) indicates if the nonstationary models

are better than the stationary model to represent the extreme precipitation (Coles, 2001;

El Adlouni et al., 2007; Tramblay et al., 2012; Xavier et al., 2019; Liu et al., 2021). The
2
asymptotic D distribution is given by the chi-square ( k) distribution with k degrees of

freedom. Here, k accounts for the difference in dimensionality (or complexity) of the

nonstationary and stationary models. Therefore, the extreme value distribution choice

accounts for both model performance and complexity.

= 2[ℓ ( )− ℓ ( )] (5)

Where ℓ (GEV1) and ℓ (GEV0) are the maximum likelihood logarithm of the

GEV1 and GEV0 models, respectively. Large D values indicate that the GEV1 is more

appropriate as it explains the data variation better than the GEV0, while the opposite

indicates that the increase in the model's complexity does not benefit the model's ability

to represent the data (Coles, 2001). Moreover, for nonstationary data, the use of

Equation 6 (comparing GEV1 and GEV2) identifies which nonstationary model better

represents extreme precipitation behavior. Therefore, for each pixel of the study area,

Equations 5 and 6 determined which GEV model was used to represent extreme

precipitation. Data that show adherence to GEV0 are stationary, and those that show

adherence to either GEV1 or GEV2 are nonstationary.

= 2[ℓ ( )− ℓ ( )] (6)
 2.3.Calculation of extreme precipitation depths

Following previous studies (Agilan and Umamahesh, 2017; Cheng et al., 2014; Sarhadi

and Soulis, 2017; Tramblay et al., 2012), we calculated extreme precipitation depths

associated with the return periods (r) of 5, 10, 25, 50 and 100 years for the observational

(1980 to 2015, ), model historic (1980-2005, ) and future (2020 to 2099, )

periods (Equations 7 and 8 for stationary conditions and Equations 9 and 10 for

nonstationary conditions).

for ≠0

= μ − {1 − [− log(1 − )] } (7)

for =0

= μ− [− (1 − )] (8)

where, represents the return level or extreme precipitation depth (mm.day-1),

in the period p (observational, historical or future) associated with the return period r

(years), and µ, σ, ξ, are the location, scale, and shape parameters, respectively.

As in nonstationary pixels µ and σ vary with time (Equations 9 and 10), a

different GEV (and consequently different precipitation depths for each return period) is

calculated for each year of the analysis period, implying in an increasing (or decreasing)

frequency of extreme precipitation. Here we extend the notion of return period to the

nonstationary case by communicating it as an interpretation of the expected number of

events in r years is one´, where r is 5, 10, 20, 50, or 100 years. This has been suggested
as an efficient way to communicate risks for designing engineering projects when

nonstationarity has been identified and the notion of return period as in the stationary

case becomes unprecise (Cooley, 2013; Salas and Vogel, 2018; Yang 2019). In that

sense, to ensure infrastructure design safety, here we consider that the return levels

associated with each return period in the nonstationary case are the highest estimated

value (Equations 9 and 10).

for ≠0

= max(μ − {1 − [− log(1 − )] }) (9)

for =0

= max(μ − [− (1 − )]) (10)

2.4.Changes in future extreme precipitation

Changes in future extreme precipitation depths (∆ ) were estimated comparing climate

change projections ( ) to each model historical set of simulations ( ) and not

directly to observed data (Equaqtion 11).

Extreme precipitation anomaly was determined for two scenarios (RCP4.5 and

RCP8.5), two time periods (2020-2059 and 2060-2099), and four climate models (Table

1). The analysis of projected precipitation data followed the same methodology as the

observational and historical periods: for each pixel, we identify the existence of trends,

estimate the GEV parameters for both stationary and nonstationary conditions and fit

the best GEV model according to the deviation statistic.

 ∆ = − (11)

3. Results

3.1.Observational period: stationarity analysis and GEV model estimate

Average extreme precipitation for 1980-2015 varies considerably (from 40 to

100 mm.day-1) over Brazil. For instance, Northern and Southern Brazil show the highest

(100 mm.day-1), and the interior northeastern region has the lowest (40 mm.day-1) value

(Figure 2-a).

The spatial distribution of the trend magnitude is very heterogeneous (Figure 2-

b). According to the Mann-Kendall test, approximately 7.5% of the study area showed a

significant positive trend in extreme precipitation, while 7.0% showed a significant

negative trend, corresponding to 14.5% of the area with nonstationary extreme

precipitation (dark blue areas in Figure 2-c). Positive trends are strong in the south and

northeastern coast. On the other hand, negative trends are strong in magnitude over the

central-western Amazon and central Brazil. Statistically, positive trends are as high as

1.2 mm.day-1year-1, which would translate into an increase in 43.2 mm.day-1 throughout

the observational period.

Figure 2 (about here)

The deviation test indicated that 82.5% of the study area adhered to the

stationary model (GEV0). Furthermore, 16.7% of the area adhered to the model with the
location parameter as a function of time (GEV1), and only 0.8% to the model with both

location and scale parameters varying with time (GEV2). Thus, according to the

deviation test, nonstationary models (GEV1 and GEV2) represented 17.5% of the study

area (Figure 3). The deviation statistic showed good agreement with Mann-Kendall test

to identify nonstationary precipitation (compare Figures 2-c and 3).

Figure 3 (about here)

3.2.Extreme precipitation depth

3.2.1. Differences in stationary and nonstationary extreme precipitation for

different return periods in the observational period

Extreme precipitation estimated under stationary conditions (Figure 4a-e) and estimated

by the best-selected model (either stationary or nonstationary, as defined by the

deviance test, Figure 4f-j) show similar spatial distribution for each return period

( ), with differences that intensify as the recurrence time increases (Figure 4k-o).

Even though sparsely present in the country, this intensification is more accentuated in

the following states: Amazonas, Pará, Amapá, Sergipe, Alagoas, Mato Grosso, and São

Paulo. The difference between nonstationary and stationary models can be higher than

10 mm.

Even though most areas show increased precipitation depths under nonstationary

modeling, a few pixels show decreased extreme precipitation when nonstationarity is

considered (mainly in Amazonas, Pará, Bahia, and Paraná, Figure 4-o), a pattern that

intensifies with the increase in return periods. Specifically, only 0%, 0.1%, 0.6%, 1.6%

and 2.6% of the area showed a reduction in precipitation depths under nonstationary
modeling, while 17.5%, 17.4%, 16.9%, 15.9% and 13.1% of the area showed an

increase in , for the return periods of 5, 10, 25, 50, and 100 years respectively

(Figure S1). This result strongly indicates that wrongly considering that the data is

stationary underestimates the magnitude of the precipitation extremes in regions where

nonstationarity is identified in the precipitation series.

Figure 4 (about here)

3.3.2. Extreme precipitation estimated for 2020-2059 and 2060-2099

Notably, most of the study area predicted future extreme precipitation depths are greater

than those calculated for the historical period, regardless of the time period, climate

change scenario, and associated return period (please see Figure 5-f to 5-o and Figure 6-

f to 6-o for the ensemble anomaly, and Figures S10 to S13 for the precipitation

magnitudes and uncertainty). In addition, for both periods and scenarios, there is an

increase in areas with nonstationary behavior in extreme precipitation (Figures S2 to

S7). While in the historical period great part of the area showed extreme precipitation

ranging from 50 mm.day-1 to 80 mm.day-1, the climate change scenarios indicate that

extreme precipitation in the same areas is greater than 80 mm.day-1 in all return periods

throughout the 21st century (Figures S10 to S13).

Notably, the intensification of extreme precipitation magnitudes implies a

greater frequency of specific return levels. For instance, extreme precipitation depths

that occurred once every 25 years in the historical period might occur more frequently.

In fact, they might occur every 5 years during 2020-2059 in the northeastern coast,

southern and southeastern regions of the country (Figures S10 to S13). Similarly,

according to the same projections, extreme precipitation depths that occurred once every
50 years in central Amazon, Rio de Janeiro, São Paulo, Acre, and Paraná might occur

every 10 years.

Furthermore, the greater the return period, the greater the increment in estimated

precipitation depths. In general, greater precipitation than historical period was found in

RCP 4.5 in 97.3%, 96.2%, 92.7%, 89% and 85.3% of the study area, while in RCP 8.5 it

reached 97.7%, 96.3%, 92.2%, 88% and 84.2%, for the return periods of 5, 10, 25, 50,

100 years, respectively, for 2020-2059 projections (Figure S8). Increases limited to 20

mm are prevalent (> 50% of the area) for r = 5, 10 and 25, while precipitation increases

greater than 40 mm are prevalent for r = 50 and 100.

However, spatial distribution is heterogeneous. For 2020-2059, more intense

precipitation is predicted to occur in the left and northern border of the country, as well

as from northwestern Amazon to southeastern Brazil (Figure 5 f-o). In the same period,

less intense precipitation is predicted in part of the Cerrado and the semi-arid region,

especially in RCP8.5 (Figure 5-o), where extreme precipitation depths decrease from 20

to 40 mm.day-1. In 2020-2059 future extreme precipitation might increase as little as 20

mm.day-1 in the five-year return period and as high as 150 mm.day-1 in the hundred-year

return period. The difference between the scenarios indicates that precipitation will be

more intense in RCP 8.5 in the majority of the area (Figure 5a-e).

The spatial distribution of the increase in extreme precipitation is similar in

2060-2099, even though they show an even greater increase in extreme precipitation

when compared to the historical period. In general, in 2060-2099 greater precipitation

than historical period was found in RCP 4.5 in 98.6%, 97.4%, 93.7%, 90.1% and 86%

of the study area, while in RCP 8.5 it reached 99.5%, 99%, 96.6%, 93.1% and 95.3%,

for the return periods of 5, 10, 25, 50, 100 years, respectively (Figure S9). In addition,
increases limited to 20 mm are prevalent (> 50% of the area) for r = 5 and 10, while

precipitation increases greater than 40 mm are prevalent for r = 25, 50 and 100.

Finally, the most affected regions in climate change scenarios are the states of

Pernambuco, Amapá, Rio Grande do Sul, Rio de Janeiro, Espírito Santo, Minas Gerais,

and São Paulo. The last four states are the most populous states of Brazil.

Figure 5 (about here)

Figure 6 (about here)

4. Discussion

The results indicate that 17.5% of the study area presents nonstationarity in precipitation

extremes during the observational period. Therefore, the idea that impacts of extreme

precipitation might only occur in the future is incorrect and already potentially

occurring in many areas over the country. Furthermore, these changes were detected in

several areas with potential socioeconomic impacts, indicating that an assessment and

an adequate knowledge of the changing hydroclimatic extremes is essential for

planning, project execution, and management of hydraulic infrastructures.

A wide range of statistical methods assists the design of hydraulic structures that

protect society and the environment from the impact of extreme precipitation events,

such as urban floods and landslides. Both the safety and financial cost of these

structures are related to the risk associated with the frequency of extreme precipitation,

determined by traditional methods that assume that the probability distribution functions

are stationary and governed by a single set of parameters (Milly et al., 2008; McPhillips

et al., 2018). However, our study indicates that stationary assumptions underestimate
extreme precipitation in 7.5% of the national territory, and infrastructure built in these

areas is potentially unsafe. In these areas, stationary modeling of extremes

underestimates maximum daily precipitation from 5 to 10 mm.day-1 in a 10-year return

period and up to 30 mm.day-1 in a 100-year return period.

Thus, a growing concern of water resource management experts worldwide has

been to determine if traditional methods developed for stationary regimes can still be

applicable (Salas et al., 2018). Once the existence of nonstationarity of extreme

precipitation is identified, engineering design must incorporate statistical methods that

are not based on the stationarity premise. Numerous recent studies have shown that

nonstationary GEV can be an efficient tool to explain the spatial and temporal evolution

and the dependencies between random variables and the climate (Cheng et al., 2014; El

Adlouni et al., 2007; Gonzalez-Alvarez et al., 2018; Katz et al., 2002; Xavier et al.,

2019; Liu et al., 2021).

More critically than the observational period, future extreme precipitation is

expected to increase in at least 90% of the national territory. In the 5 and 10-year return

periods, extreme precipitation increases up to 20 mm.day-1 in most areas, reaching 80

mm.day-1 in a few pixels. In 25, 50, and 100-year return periods increase from 20 to 60

mm.day-1, reaching more than 120 mm.day-1 for a few pixels. These results indicate

more intense and frequent disasters like urban floods and landslides, increasing social,

environmental, and economic risks.

Even though CMIP5 models agree that Brazil will endure more variability in

precipitation in the following decades, a comprehensive assessment of the drivers of

these changes remains a challenge. Models have shown improvements in recent years,

but they still have difficulties in simulating precipitation patterns, mainly because the

coarser resolution model cannot represent well processes related to convective

precipitation, subgrid convention, and mesoscale circulation (Alves et al., 2020).

However, studies indicate a future increase in ENOS variability (Cai et al., 2018),

changes in width and strength of ITCZ (Byrne et al., 2018); as well as changes in

position and intensity of SACZ (Zilli et al., 2021), all of which have potential to disturb

precipitation extremes in Brazil.

In addition to climate projections, urban population growth is another crucial

driver of urban floods. The accelerated urbanization inevitably creates impermeable

areas, increasing runoff. As more people live in densely populated urban areas, often

located in floodplains and low-lying coastal areas, their exposure to flood risks also

increases (Hammond et al., 2015). Urban drainage systems and infrastructure lifespan

are long (50 to 100 years), but the expected increase in runoff and the intensification of

precipitation extremes may prevent the drainage system from reaching the desired level

of service in the future (Alashan, 2018; Hammond et al., 2015; Mailhot and Duchesne,

2010; Willems et al., 2012; Xiong et al., 2019; Zahmatkesh et al., 2015; Zhou et al.,

2012). Such reality was evident in recent events of greater than average precipitation

that hit Minas Gerais, Rio de Janeiro, Espírito Santo, and São Paulo, causing many

impacts. The growing evidence of the increasing frequency of extremes urge the

development of new approaches to estimate risk and promote adaptation in urban areas,

and probabilistic models that incorporate nonstationarity are a significant part of this

challenge (Milly et al., 2008; McPhillips et al., 2018)

The sole occurrence of extreme events does not necessarily lead to disasters but

is determined by populations' exposure and vulnerability. Therefore, increasing our

resilience to extreme events and planning for recovery demands understanding risks

associated with extreme events and population vulnerabilities (McPhillips et al., 2018).

In addition, previous policy decisions on land use, zoning, and infrastructure affect the
level of risk exposure, susceptibility, and the adaptive capacity of the region to deal

with the impacts caused by extreme events (Cho and Chang, 2017).

More than 80% of the municipalities with very high vulnerability in Brazil are

located in the northern and northeastern regions, especially in the latter (Almeida et al.,

2016). The northeast region shows less exposure to natural risks than southern Brazil

but presents severe vulnerability conditions, with less adaptive and recovery capacity.

The south and southeastern regions have lower levels of vulnerability (Almeida et al.,

2016), with an increased capacity to resist, recover and adapt to socioeconomic,

cultural, and environmental challenges. However, southern and southeastern regions

concentrate the largest areas of high or very high landslide susceptibility (IBGE,

2019a), as they have high population density located in areas of steep slopes. These

regions concentrate more than 56% of the Brazilian population (IBGE, 2019b).

However, further local studies must be conducted to accurately assess Brazilian

population exposure and vulnerability to the increase in precipitation extremes

calculated in this work.

Finally, even though we provide a new and robust analysis of extreme

precipitation in Brazil. The results of this work are subjected to many uncertainties.

First, modeling precipitation is a particularly challenging task, and models still diverge,

mainly in the change's magnitude. The differences in physical parameters between the

different models and the role of urbanization are still not well represented in the climate

models and are sources of uncertainty already identified by other studies (Cavalcanti et

al., 2017). Therefore, our estimates are based on a set of models from different

downscale approaches, which is helpful to inform internal variability in the climate

system and to a better estimation of uncertainty in projection of climate change (Chen et

al., 2021). However, even in cases where the models do not agree on the magnitude,
they agree on the change signal, which indicates that the increase in the maximum

precipitation depths is a robust result. Second, the downscaled climate models

resolution (0.25ºx0.25º) is sufficient to represent large and mesoscale phenomena that

drive precipitation but may lose some local processes of convective or orographic order.

Third, uncertainties in the observed maximum precipitation reflect in the quantiles

calculated for the historical period and affect the modeling process.

The difficulty of the calculation process and the lack of dialogue between the

scientific community and the technicians who apply these estimates inhibit the

implementation of such a methodology as a usual practice. Perhaps some of the

difficulties can be surpassed by the development of software and digital platforms that

make this information clear and accessible, as well as training programs and courses to

transfer and disseminate the necessary knowledge. However, local government and

public managers make critical decisions that affect urban environments, and

communication with those actors is a crucial step to raise awareness and implement

such measures.

5. Conclusions

This study provides the first estimate of the frequency distribution of maximum daily

precipitation considering nonstationarity for the entire Brazilian territory. We show that

17.5% of the study area has nonstationary maximum daily precipitation in the historical

period and that neglecting nonstationarity, in most of the area, underestimates the

magnitudes of the daily maximum precipitation events for all return periods (e.g., 5, 10,

25, 50, and 100 years). Furthermore, considering nonstationarity, we found that

maximum daily precipitation might increase in at least 90% of the national territory in

the future, regardless of the climate scenario. Importantly, these changes are estimated
for the most populated regions of the country.

This scenario, along with accelerated urbanization, might explain the recent

increase in disasters related to urban floods and landslides over the country. In case

there are no adjustments, more intense and more frequent urban floods, with the

increase in social, environmental, and economic risks, might occur. It suggests the

urgent need to prepare and adapt urban environments to the current and future extreme

precipitation events and increase the local population's resilience.

To this end, large-scale analysis like the presented in this work should be refined

and include local specificities that may increase the accuracy of the estimated frequency

and magnitude of extreme precipitation events. For example, it provides precipitation

simulation in finer resolutions (at least 1km), evaluating the steepness of local reliefs

and population occupation of high disaster risk areas. However, the greatest challenge to

the proper adaptation of urban environments is communicating climate and social

studies to the engineers and decision-makers. The correct maximum precipitation

frequency and intensity determine urban infrastructures' risk and economic cost, and the

needy population allocation.

Finally, there is still a long way to adapt urban environments to climate change,

and local managers and decision-makers must be aware that this is an urgent matter, and

the lack of action exposes society to an imminent risk and might cause the loss of

thousands of lives in the future.

Acknowledgements

The authors thank the graduate program in Applied Meteorology at the Federal
University of Viçosa. This study was supported by The National Council for Scientific
and Technological Development (CNPq) and the Coordination of Training of Higher
Education Personnel (CAPES). AAD also thanks the financial support received from
CNPq postdoctoral Programme. The authors are thankful for the downscaled climate
simulations generated by CPTEC/INPE and made available on the PROJETA platform,
and we thank NEX-GDDP dataset, prepared by the Climate Analytics Group and NASA
Ames Research Center using the NASA Earth Exchange, and distributed by the NASA
Center for Climate Simulation (NCCS).
References

Agilan, V., Umamahesh, N. V., 2017. What are the best covariates for developing non-
stationary rainfall Intensity-Duration-Frequency relationship? Advances in Water
Resources, 101, 11–22. https://doi.org/10.1016/j.advwatres.2016.12.016

Alashan, S., 2018. Data Analysis in Nonstationary State. Water Resources

Management, 32(7), 2277–2286. https://doi.org/10.1007/s11269-018-1928-2

Alexander, L. V., et al., 2006. Global observed changes in daily climate extremes of
temperature and precipitation. Journal of Geophysical Research Atmospheres,
111(5), 1–22. https://doi.org/10.1029/2005JD006290

Almagro, A., et al., 2020. Performance evaluation of Eta/HadGEM2-ES and

Eta/MIROC5 precipitation simulations over Brazil. Atmospheric Research, 244,
105053. https://doi.org/10.1016/j.atmosres.2020.105053

Almeida, L. Q., Welle, T., Birkmann, J., 2016. Disaster risk indicators in Brazil: A
proposal based on the world risk index. International Journal of Disaster Risk
Reduction, 17, 251–272. https://doi.org/10.1016/j.ijdrr.2016.04.007

Alves, L. M., et al., 2021. Assessment of rainfall variability and future change in Brazil
across multiple timescales. International Journal of Climatology, 41(S1), E1875–
E1888. https://doi.org/10.1002/joc.6818

Assis, L. C., et al., 2018. A model-based site selection approach associated with
regional frequency analysis for modeling extreme rainfall depths in Minas Gerais
state, Southeast Brazil. Stochastic Environmental Research and Risk Assessment,
32(2), 469–484. https://doi.org/10.1007/s00477-017-1481-1

Ávila, A., et al., 2016. Recent precipitation trends, flash floods and landslides in
southern Brazil. Environmental Research Letters, 11(11).
https://doi.org/10.1088/1748-9326/11/11/114029

Avila-Diaz, A., et al., 2020a. Extreme climate indices in Brazil: evaluation of

downscaled earth system models at high horizontal resolution. Climate Dynamics,
54(11–12), 5065–5088. https://doi.org/10.1007/s00382-020-05272-9

Avila-Diaz, A., et al., 2020b. Assessing current and future trends of climate extremes
across Brazil based on reanalyses and earth system model projections. Climate
Dynamics, 55(5–6), 1403–1426. https://doi.org/10.1007/s00382-020-05333-z

Byrne, M. P., et al., 2018. Response of the Intertropical Convergence Zone to Climate
Change: Location, Width, and Strength. Current Climate Change Reports, 4(4),
355–370. https://doi.org/10.1007/s40641-018-0110-5

Cai, W., et al., 2018. Increased variability of eastern Pacific El Niño under greenhouse
warming. Nature, 564(7735), 201–206. https://doi.org/10.1038/s41586-018-0776-9
Cavalcanti, I. F. A., 2012. Large scale and synoptic features associated with extreme
precipitation over South America: A review and case studies for the first decade of
the 21st century. Atmospheric Research, 118, 27–40.
https://doi.org/10.1016/j.atmosres.2012.06.012

Cavalcanti, I. F. A., et al., (2017). Projections of Precipitation Changes in Two

Vulnerable Regions of São Paulo State, Brazil. American Journal of Climate
Change, 06(02), 268–293. https://doi.org/10.4236/ajcc.2017.62014

Chen, D., M., et al., 2021: Framing, Context, and Methods. In Climate Change 2021:
The Physical Science Basis. Contribution of Working Group I to the Sixth
Assessment Report of the Intergovernmental Panel on Climate Change [Masson-
Delmotte, V., P. Zhai, A. Pirani, S.L. Connors, C. Péan, S. Berger, N. Caud, Y.
Chen, L. Goldfarb, M.I. Gomis, M. Huang, K. Leitzell, E. Lonnoy, J.B.R.
Matthews, T.K. Maycock, T. Waterfield, O. Yelekçi, R. Yu, and B. Zhou (eds.)].
Cambridge University Press. In Press.

Cheng, L., et al., 2014. Non-stationary extreme value analysis in a changing climate.
Climatic Change, 127(2), 353–369. https://doi.org/10.1007/s10584-014-1254-5

Cho, S. Y., Chang, H., 2017. Recent research approaches to urban flood vulnerability,
2006–2016. Natural Hazards, 88(1), 633–649. https://doi.org/10.1007/s11069-017-
2869-4

Chou, S. C., et al., 2014a. Assessment of Climate Change over South America under
RCP 4.5 and 8.5 Downscaling Scenarios. American Journal of Climate Change,
03(05), 512–527. https://doi.org/10.4236/ajcc.2014.35043

Chou, S. C., et al., 2014b. Evaluation of the Eta Simulations Nested in Three Global
Climate Models. American Journal of Climate Change, 03(05), 438–454.
https://doi.org/10.4236/ajcc.2014.35039

Coles, S., 2015. An introduction to extreme values. In The effects of brief mindfulness
intervention on acute pain experience: An examination of individual difference
(Vol. 1).

Collins, W. J., et al., 2011. Development and evaluation of an Earth-system model –

HadGEM2. Geoscientific Model Development Discussions, 4(2), 997.
https://doi.org/10.5194/gmdd-4-997-2011

Cooley, D., 2013. Return Periods and Return Levels Under Climate Change. Extremes
in a Changing Climate, 25. 97-114 https://doi.org/10.1007/978-94-007-4479-0_4

Costa, R. L., et al., 2020. Analysis of climate extremes indices over northeast Brazil
from 1961 to 2014. Weather and Climate Extremes, 28.
https://doi.org/10.1016/j.wace.2020.100254

de Loyola Hummell, B. M., Cutter, S. L., & Emrich, C. T., 2016. Social Vulnerability to
Natural Hazards in Brazil. International Journal of Disaster Risk Science, 7(2),
111–122. https://doi.org/10.1007/s13753-016-0090-9
Debortoli, N. S., Camarinha, P. I. M., Marengo, J. A., & Rodrigues, R. R., 2017. An
index of Brazil’s vulnerability to expected increases in natural flash flooding and
landslide disasters in the context of climate change. Natural Hazards, 86(2), 557–
582. https://doi.org/10.1007/s11069-016-2705-2

Donat, M. G., et al., 2016. More extreme precipitation in the worldâ €TM s dry and wet
regions. Nature Climate Change, 6(5), 508–513.
https://doi.org/10.1038/nclimate2941

El Adlouni, S., et al., (2007). Generalized maximum likelihood estimators for the
nonstationary generalized extreme value model. Water Resources Research, 43(3),
1–13. https://doi.org/10.1029/2005WR004545

Eyring, V., et al., 2021. Chapter 3: Human influence on the climate system.
In: AR6 Climate Change 2021: The Physical Science Basis. IPCC, 207.
04-202. https://doi.org/10.1002/2014GL06106

Field, C.B., et al., eds., 2012. Managing the risks of extreme events and disasters to
advance climate change adaptation. Special Report of Working Groups I and II of
the Intergovernmental Panel on Climate Change (IPCC). Cambridge: Cambridge
University Press.

Gonzalez-Alvarez, A., et al., (2018). Effect of the non-stationarity of rainfall events on

the design of hydraulic structures for runoff management and its applications to a
case study at Gordo Creek watershed in Cartagena de Indias, Colombia. Fluids,
3(2). https://doi.org/10.3390/fluids3020027

Hammond, M. J., et al., 2015. Urban flood impact assessment: A state-of-the-art

review. Urban Water Journal, 12(1), 14–29.
https://doi.org/10.1080/1573062X.2013.857421

Heffernan, J. E., Stephenson, A. G., Gilleland, E. 2018. Ismev: An introduction to

statistical modeling of extreme values. R package version, 1, 42.

IBGE. 2019a. Macrocaracterização dos Recursos Naturais do Brasil: Suscetibilidade a

Deslizamentos do Brasil (Primeira Aproximação).
https://loja.ibge.gov.br/macrocaracterizac-o-dos-recursos-naturais-do-brasil-
suscetibilidade-a-deslizamentos-do-brasil.html

IBGE. 2019b. Estimativas da população residente no Brasil e Unidades da Federação

com data de referência em 1º de julho de 2019.
https://www.ibge.gov.br/estatisticas/sociais/populacao/9103-estimativas-de-
populacao.html?=&t=o-que-e

Jenkinson, A. F., 1955. The frequency distribution of the annual maximum (or
minimum) values of meteorological elements. Quarterly Journal of the Royal
Meteorological Society, 81(348), 158–171.
https://doi.org/10.1002/qj.49708134804
Katz, R. W., Parlange, M. B., Naveau, P., 2002. Statistics of extremes in hydrology.
Advances in Water Resources, 25(8–12), 1287–1304.
https://doi.org/10.1016/S0309-1708(02)00056-8

KENDALL, M.G., 1956. Rank Correlation Methods. British Journal of Statistical

Psychology. 9, 196. https://doi.org/10.1111/j.2044-8317.1956.tb00172.x

Lang, M., Ouarda, T. B. M. J., Bobée, B., 1999. Towards operational guidelines for
over-threshold modeling. Journal of Hydrology, 225(3–4), 103–117.
https://doi.org/10.1016/S0022-1694(99)00167-5

Lawrence, D. M., et al., 2012. The CCSM4 land simulation, 1850-2005: Assessment of
surface climate and new capabilities. Journal of Climate, 25(7), 2240–2260.
https://doi.org/10.1175/JCLI-D-11-00103.1

Lehmann, J., Coumou, D., Frieler, K., 2015. Increased record-breaking precipitation
events under global warming. Climatic Change, 132(4), 501–515.
https://doi.org/10.1007/s10584-015-1434-y

Lehtonen, I., Jylhä, K., 2019. Tendency towards a more extreme precipitation climate in
the Coupled Model Intercomparison Project Phase 5 models. Atmospheric Science
Letters, 20(5), 1–7. https://doi.org/10.1002/asl.895

Lucas, E. W. M., et al., 2021. Trends in climate extreme indices assessed in the Xingu
river basin - Brazilian Amazon. Weather and Climate Extremes, 31(January),
100306. https://doi.org/10.1016/j.wace.2021.100306

Liu, B., et al., 2020. Global atmospheric moisture transport associated with precipitation
extremes: Mechanisms and climate change impacts. Wiley Interdisciplinary
Reviews: Water, 7(2), 1–25. https://doi.org/10.1002/wat2.1412

Liu, M., et al., 2021. Non-stationary frequency analysis of extreme streamflow

disturbance in a typical ecological function reserve of China under a changing
climate. Ecohydrology, 1–20. https://doi.org/10.1002/eco.2323

Luiz-Silva, W., et al., 2021. An overview of precipitation climatology in Brazil: space-

time variability of frequency and intensity associated with atmospheric systems.
Hydrological Sciences Journal. https://doi.org/10.1080/02626667.2020.1863969

Lyra, A., et al., 2018. Climate change projections over three metropolitan regions in
Southeast Brazil using the non-hydrostatic Eta regional climate model at 5-km
resolution. Theoretical and Applied Climatology, 132(1–2), 663–682.
https://doi.org/10.1007/s00704-017-2067-z

Mailhot, A., Duchesne, S., 2010. Design Criteria of Urban Drainage Infrastructures
under Climate Change. Journal of Water Resources Planning and Management,
136(2), 201–208. https://doi.org/10.1061/(asce)wr.1943-5452.0000023

Mann, H. B., 1945. Non-Parametric Test Against Trend. Econometrica, 13(3), 245–259.
http://www.economist.com/node/18330371?story%7B_%7Did=18330371
Marengo, J. A., et al., 2011. Recursos hídricos em regiões áridas e semiáridas. Recursos
hídricos em regiões áridas e semiáridas.
http://plutao.sid.inpe.br/col/dpi.inpe.br/plutao/2011/06.11.02.16/doc/Marengo_Var
iabilidade.pdf?languagebutton=en

Marengo, J.A., et al., 2021. Extreme Rainfall and Hydro-Geo-Meteorological Disaster

Risk in 1.5, 2.0, and 4.0°C Global Warming Scenarios: An Analysis for Brazil.
Frontiers in Climate. 3, 1–17. https://doi.org/10.3389/fclim.2021.610433

Markus, M., et al., 2018. Communicating the Impacts of Projected Climate Change on
Heavy Rainfall Using a Weighted Ensemble Approach. Journal of Hydrologic
Engineering, 23(4), 04018004. https://doi.org/10.1061/(asce)he.1943-
5584.0001614

McPhillips, L. E., et al., 2018. Defining Extreme Events: A Cross-Disciplinary Review.

Earth’s Future, 6(3), 441–455. https://doi.org/10.1002/2017EF000686

Milly, P. C. D., et al., 2008. Climate change: Stationarity is dead: Whither water
management? Science, 319(5863), 573–574.
https://doi.org/10.1126/science.1151915

O’Gorman, P. A., & Schneider, T. 2009. The physical basis for increases in
precipitation extremes in simulations of 21st-century climate change. Proceedings
of the National Academy of Sciences of the United States of America, 106(35),
14773–14777. https://doi.org/10.1073/pnas.0907610106

O’Gorman, P. A. 2015. Precipitation Extremes Under Climate Change. Current Climate

Change Reports, 1(2), 49–59. https://doi.org/10.1007/s40641-015-0009-3

Ouarda, T. B. M. J., Charron, C., 2019. Changes in the distribution of hydro-climatic

extremes in a non-stationary framework. Scientific Reports, 9(1), 1–8.
https://doi.org/10.1038/s41598-019-44603-7

Parey, S., Hoang, T.T.H., Dacunha-Castelle, D., 2010. Different ways to compute
temperature return levels in the climate change context. Environmetrics, 21, 698–
718. https://doi.org/10.1002/env.1060

Pedron, I. T., et al., 2017. Trends and variability in extremes of precipitation in

Curitiba – Southern Brazil. International Journal of Climatology, 37(3), 1250–
1264. https://doi.org/10.1002/joc.4773

Porto de Carvalho, J. R., et al., 2014. Annual maximum daily rainfall trends in the
midwest, southeast and southern Brazil in the last 71 years. Weather and Climate
Extremes, 5(1), 7–15. https://doi.org/10.1016/j.wace.2014.10.001

Riahi, K., et al., 2011. RCP 8.5-A scenario of comparatively high greenhouse gas
emissions. Climatic Change, 109(1), 33–57. https://doi.org/10.1007/s10584-011-
0149-y
Salas, J. D., Obeysekera, J., Vogel, R. M., 2018. Techniques for assessing water
infrastructure for nonstationary extreme events: a review. Hydrological Sciences
Journal, 63(3), 325–352. https://doi.org/10.1080/02626667.2018.1426858

Sarhadi, A., Soulis, E. D., 2017. Time-varying extreme rainfall intensity-duration-

frequency curves in a changing climate. Geophysical Research Letters, 44(5),
2454–2463. https://doi.org/10.1002/2016GL072201

Seneviratne, S. I., et al., 2021. Weather and Climate Extreme Events in a Changing
Climate. In: Climate Change 2021: The Physical Science Basis. Contribution of
Working Group I to the Sixth Assessment Report of the Intergovernmental Panel
on Climate Change. In P. Z. Masson-Delmotte, V., E. A. Pirani, S. L. Connors, C.
Péan, S. Berger, N. Caud, Y. Chen, L. Goldfarb, M. I. Gomis, M. Huang, K.
Leitzell, & R. Y. and B. Z. Lonnoy, J. B. R. Matthews, T. K.

Sheffield J., Goteti G., Wood E.F., 2006. Development of a 50-year high-resolution
global dataset of meteorological forcings for land surface modeling. Journal Of
Climate.19(13), 3088-3111. doi:10.1175/JCLI3790.1

Silva Dias, M. A. F., et al., 2013. Changes in extreme daily rainfall for São Paulo,
Brazil. Climatic Change, 116(3–4), 705–722. https://doi.org/10.1007/s10584-012-
0504-7

Smith, R. L., 1985., Maximum likelihood estimation in a class of nonregular cases.

Biometrika, 72(1), 67–90. https://doi.org/10.1093/biomet/72.1.67

Thomson, A. M., et al., 2011. RCP4.5: A pathway for stabilization of radiative forcing
by 2100. Climatic Change, 109(1), 77–94. https://doi.org/10.1007/s10584-011-
0151-4

Thrasher, B., et al., 2012. Technical Note: Bias correcting climate model simulated
daily temperature extremes with quantile mapping. Hydrology and Earth System
Sciences, 16(9), 3309–3314. https://doi.org/10.5194/hess-16-3309-2012

Tramblay, Y., et al., 2012. Climate change impacts on extreme precipitation in

Morocco. Global and Planetary Change, 82–83, 104–114.
https://doi.org/10.1016/j.gloplacha.2011.12.002

Tramblay, Y., Neppel, L., Carreau, J., Najib, K., 2013. Analyse fréquentielle non-
stationnaire des pluies extrêmes dans le Sud de la France. Hydrological Sciences
Journal, 58(2), 280–294. https://doi.org/10.1080/02626667.2012.754988

Vasiliades, L., Galiatsatou, P., Loukas, A., 2015. Nonstationary Frequency Analysis of
Annual Maximum Rainfall Using Climate Covariates. Water Resources
Management, 29(2), 339–358. https://doi.org/10.1007/s11269-014-0761-5

Vu, T. M., Mishra, A. K., 2019. Nonstationary frequency analysis of the recent extreme
precipitation events in the United States. Journal of Hydrology, 575(May), 999–
1010. https://doi.org/10.1016/j.jhydrol.2019.05.090
Watanabe, M., et al., 2010. Improved climate simulation by MIROC5: Mean states,
variability, and climate sensitivity. Journal of Climate, 23(23), 6312–6335.
https://doi.org/10.1175/2010JCLI3679.1

Westra, S., Alexander, L. V., Zwiers, F. W., 2013. Global increasing trends in annual
maximum daily precipitation. Journal of Climate, 26(11), 3904–3918.
https://doi.org/10.1175/JCLI-D-12-00502.1

Willems, P., Arnbjerg-Nielsen, K., Olsson, J., Nguyen, V. T. V., 2012. Climate change
impact assessment on urban rainfall extremes and urban drainage: Methods and
shortcomings. Atmospheric Research, 103, 106–118.
https://doi.org/10.1016/j.atmosres.2011.04.003

Xavier, A. C., King, C. W., Scanlon, B. R., 2016. Daily gridded meteorological
variables in Brazil (1980–2013). International Journal of Climatology, 36(6),
2644–2659. https://doi.org/10.1002/joc.4518

Xavier, A. C. F., et al., 2019. Stationary and non-stationary detection of extreme

precipitation events and trends of average precipitation from 1980 to 2010 in the
Paraná River basin, Brazil. International Journal of Climatology, 40(2), 1197–
1212. https://doi.org/10.1002/joc.6265

Xiong, L., et al., 2019. Impacts of Climate Change on Urban Extreme Rainfall and
Drainage Infrastructure Performance: A Case Study in Wuhan City, China.
Irrigation and Drainage, 68(2), 152–164. https://doi.org/10.1002/ird.2316

Yang L., Li J., Sun H., Guo Y., Engel B.A. 2019. Calculation of nonstationary flood
return period considering historical extraordinary flood events. Journal Flood Risk
Management. 12(2), 1-10. doi:10.1111/jfr3.12463

Yukimoto, S., et al., 2012. A new global climate model of the Meteorological Research
Institute: MRI-CGCM3: -Model description and basic performance-. Journal of the
Meteorological Society of Japan, 90(A), 23–64. https://doi.org/10.2151/jmsj.2012-
A02

Zahmatkesh, Z., et al., 2015. Low-impact development practices to mitigate climate

change effects on urban stormwater runoff: Case study of New York City. Journal
of Irrigation and Drainage Engineering. 141, 1–13.
https://doi.org/10.1061/(ASCE)IR.1943-4774.0000770

Zhang, X., et al., 2011. Indices for monitoring changes in extremes based on daily
temperature and precipitation data. Wiley Interdisciplinary Reviews: Climate
Change, 2(6), 851–870. https://doi.org/10.1002/wcc.147

Zhou, Q., et al., 2012. Framework for economic pluvial flood risk assessment
considering climate change effects and adaptation benefits. Journal of Hydrology,
414–415, 539–549. https://doi.org/10.1016/j.jhydrol.2011.11.031
Zilli, M. T., et al., 2017. A comprehensive analysis of trends in extreme precipitation
over southeastern coast of Brazil. International Journal of Climatology, 37(5),
2269–2279. https://doi.org/10.1002/joc.4840
TABLES

Table 1. Global climate models used in regionalized climate change projections.

Horizontal
Model Downscaling Access link References
resolution
HADGEM2- (Brazil
ES MCTI, 2016;
20 km x 20 Chou et al.,
Dynamical https://projeta.cptec.inpe.br/
km* 2014a,
MIROC5
2014b; Lyra
et al., 2018)
MRI-
https://cds.nccs.nasa.gov/nex- (Thrasher et
CGCM3 0.25°x 0.25° Statistical
gddp/ al., 2012)
CCSM4
* Interpolated to 0.25° x 0.25°
Table 2. Models of the Generalized Function of Stationary (GEV0) and Nonstationary
Extreme Values (GEV1 e GEV2).
Time-
Model GEV Model Description Function varying
parameters
Classic model, with all
GEV0 (μ, , ) -
parameters constant.
Model with the location
GEV1 parameter as a function of ( + , , ) μ
time.
Model with location and
GEV2 scale parameters as a ( + , + , ) μ ,
function of time.
FIGURE CAPTIONS

Figure 1. Study area.

Figure 2. Extreme precipitation during the observational period (1985-2015). (a) average
maximum annual precipitation (AMAP, mm.day-1), (b) extreme precipitation trend (EPT,
mm.day-1.year-1) determined by Sen’s slope estimator and (c) p-value for Mann-Kendall test
(dark blue areas indicate statistically significant trends, α=0.05).

Figure 3. Spatial distribution of the adherence to the different GEV models according to the
deviation test for the historical period (1985-2015).

Figure 4. Historical maximum precipitation depths associated with the return periods of 5, 10,
25, 50, and 100 years ( , , , , ), in mm.day-1 (a) to (e) estimated with the
stationary model (GEV0), (f) to (j) with the model indicated by the deviation test (GEV0, GEV1,
GEV2) and (k) to (o) the difference between them.

Figure 5. Future maximum precipitation depth anomaly, from 2020 to 2059, associated
with the return periods of 5, 10, 25, 50, and 100 years ( , , , , ),
in mm.day-1. (a) to (e) between RCP 8.5 and RCP 4.5, (f) to (j) between RCP 4.5 and
the historical period, and (k) to (o) between RCP 8.5 and the historical period.

Figure 6. Future maximum precipitation depth anomaly, from 2060 to 2099, associated with the
return periods of 5, 10, 25, 50, and 100 years ( , , , , ), in mm.day-1. (a)
to (e) between RCP 8.5 and RCP 4.5, (f) to (j) between RCP 4.5 and the historical period, and
(k) to (o) between RCP 8.5 and the historical period.