SciELO - Scientific Electronic Library Online

vol.41Effects of Transplanting Time on 15-Nitrogen Utilization and Industrial Quality of Flue-Cured TobaccoTree Growth and Nutrient Dynamics in Pine Plantations in Southern Brazil author indexsubject indexarticles search
Home Pagealphabetic serial listing  

Services on Demand



  • text new page (beta)
  • English (pdf)
  • Article in xml format
  • How to cite this article
  • SciELO Analytics
  • Curriculum ScienTI
  • Automatic translation


Related links


Revista Brasileira de Ciência do Solo

On-line version ISSN 1806-9657

Rev. Bras. Ciênc. Solo vol.41  Viçosa  2017  Epub Dec 04, 2017 

Division 3 - Soil Use and Management

Multivariate Analysis of Erosivity Indices and Rainfall Physical Characteristics Associated with Rainfall Patterns in Rio de Janeiro

Roriz Luciano Machado(1)  * 

Daniel Fonseca de Carvalho(2) 

Janaina Ribeiro Costa Rouws(3) 

Wilk Sampaio de Almeida(4) 

Paulo Tarso Sanches de Oliveira(5) 

(1)Instituto Federal de Educação, Ciência e Tecnologia Goiano, Campus Ceres, Ceres, Goiás, Brasil.

(2)Universidade Federal Rural do Rio de Janeiro, Instituto de Tecnologia, Departamento de Engenharia, Seropédica, Rio Janeiro, Brasil.

(3)Empresa Brasileira de Pesquisa Agropecuária, Embrapa Agrobiologia, Seropédica, Rio Janeiro, Brasil.

(4)Instituto Federal de Educação, Ciência e Tecnologia de Rondônia, Campus Colorado do Oeste, Colorado do Oeste, Rondônia, Brasil.

(5)Universidade Federal de Mato Grosso do Sul, Faculdade de Engenharias, Arquitetura e Urbanismo e Geografias, Campo Grande, Mato Grosso do Sul, Brasil.


The identification of areas with greater erosive potential is important for planning soil and water conservation. The objective of this study was to evaluate the physical characteristics of rainfall events in the state of Rio de Janeiro, Brazil, and their interactions with rainfall patterns through multivariate statistical analysis. Rainfall depth, kinetic energy, 30-min intensity (I30), duration of rainfall events, and the erosivity indices KE >10, KE >25, and EI30 in 36 locations (stations) were subjected to principal component analysis (PCA) and canonical discriminant analysis (CDA). Based on evaluation of the respective historical series of hyetographs, it was found that the advanced pattern occurs with highest frequency (51.8 %), followed by the delayed pattern (26.1 %), and by the intermediate pattern (22.1 %). All the evaluated rainfall characteristics have high response capacity in describing localities and rainfall patterns through PCA and CDA. In CDA, the Tukey test (p<0.05) applied to the scores of the first canonical discriminant function (CDF1) allowed differentiation of the stations with respect to the rainfall and erosivity characteristics for the advanced and delayed patterns. In the delayed pattern, the localities of Angra dos Reis, Campos, Eletrobrás, Manuel Duarte, Santa Isabel do Rio Preto, Tanguá, Teresópolis, Vila Mambucaba, and Xerém had the highest CDF1 scores, indicating that they have rainfalls with higher depth, I30, and duration because the standardized canonical coefficient (SCC) and the correlation coefficient (“r”) of these characteristics were positive. The rainfall events in the state of Rio de Janeiro differ from one locality to another in relation to the advanced and delayed rainfall patterns, mainly due to the physical characteristics of rainfall depth, I30, and duration, indicating a higher risk of soil loss and runoff in the localities where rainfall events with the delayed pattern prevail.

Key words: water erosion; EI30; canonical discriminant analysis


Water erosion poses a serious threat to natural resources worldwide, with significant economic and environmental consequences (Prosdocimi et al., 2016) and evaluation of this erosion has become an important tool in soil and water conservation plans (Anache et al., 2017). Soil erosion is directly related to the physical characteristics of rainfall events (Ran et al., 2012), primarily their intensity and duration, and the diameter and final speed of raindrops, which determine their erosivity (Goebes et al., 2014).

The indices EI30 and KE >25 have been widely used to calculate erosivity (Oliveira et al., 2013) since they have provided good correlation with soil losses in different regions (Bertol et al., 2007; Bertol et al., 2008; Silva et al., 2009). These indices are calculated through the product of kinetic energy and intensity of the rainfall (Marques et al., 1997).

Rainfall events with the same average intensity do not necessarily have the same kinetic energy (Mohamadi and Kavian, 2015), which indicates that a given precipitated volume may lead to different soil losses, even under the same conditions of land use and cover. This reality may be related to the effect of the hydrological rainfall pattern, defined in accordance with the occurrence of peak intensity in relation to its duration (Flanagan et al., 1988). Various studies have shown the influence of rainfall patterns in the erosion process, especially in runoff, soil loss, and particle distribution (Flanagan et al., 1988; Eltz et al., 2001; Oliveira et al., 2010; Wang et al., 2016).

The state of Rio de Janeiro has wide climatic variability, mainly caused by the diverse topography and consequent cloud formation. Thus, the state has orographic rainfalls and convective and frontal rainfalls (André et al., 2008; Dereczynski et al., 2009). These characteristics influence the volume, intensity, and patterns of rainfalls and, consequently, soil erosion.

Evaluations of rainfall erosivity and patterns have been made for various Brazilian localities individually (Mehl et al., 2001; Evangelista et al., 2005; Cassol et al., 2008; Machado et al., 2008; Peñalva-Bazzano et al., 2010; Oliveira et al., 2012; Eltz et al., 2013). However, there are few studies investigating the use of these rainfall physical attributes in relation to the erosive processes that aim to group various localities of a same region or state with similar characteristics. This type of approach is especially useful in identifying areas with a higher risk of soil, water, and nutrient losses.

Due to the large number of characteristics (variables) related to the rainfall events, multivariate analysis statistical techniques can be better suited to evaluate the data; they have been applied to soil science and various areas of knowledge (Maluche-Baretta et al., 2006; Dai et al., 2014). Multivariate techniques include canonical discriminant analysis (CDA) (Cruz-Castillo et al., 1994) and principal component analysis (PCA) (Pearson, 1901; Raziei et al., 2009; Santos et al., 2010). These are exploratory techniques of the data to summarize the set of original variables and facilitate interpretation of the interdependence among these variables. Applied to pluviographic data, these techniques allow localities to be distinguished and grouped in accordance with the rainfall physical characteristics that affect the erosive process. Through the PCA, it is possible to visualize the variables (characteristics) and the levels of the factors evaluated (in this case, localities and rainfall patterns) on a two-dimensional plane and, therefore, evaluate relationships between variables through components (PCA) and the potential to discriminate (CDA) locations according to the rainfall characteristics.

Canonical discriminant analysis allows identification of differences among the groups evaluated (treatments or factors) based on measurements of characteristics of the observations belonging to these groups, facilitating understanding of the relationships among the variables evaluated within these groups. The CDA estimates linear functions (called canonical functions or variables) based on the quantified original variables, separating the groups of individuals to maximize the variance between groups and minimize the variance inside them (Cruz-Castillo et al., 1994).

Utilization of the PCA and CDA multivariate techniques has not been widespread in the literature on rainfall studies (Machado et al., 2008; Jin et al., 2015), and even less so with the aim of evaluating and grouping localities according to the rainfall characteristics related to erosivity.

Considering that the different regions of Rio de Janeiro do not exhibit homogeneity of characteristics related to the potential of rainfall events to cause erosion, the aim of this study was to evaluate the physical characteristics of rainfall events, erosivity indices, and rainfall patterns in the state using multivariate statistical techniques to identify homogeneous regions.


The present study was conducted using the rainfall records of 36 stations in the state of Rio de Janeiro that belonged to the Superintendência Estadual de Rios e Lagoas - Serla (currently part of the Instituto Estadual do Ambiente - INEA), the Instituto Nacional de Meteorologia (INMET), the Agência Nacional das Águas (ANA), and the Light Serviços de Eletricidade S.A. company (Figure 1). The period of analysis encompassed the years 1974 to 1996, with an average of 11 years of data, considering that it was not possible to establish a common period for all series studied. Regarding rainfall data, in Brazil it is still rare to have series of more than ten years (Montebeller et al., 2007), especially in working with areas of considerable geographic extension.

Figure 1 Location of the rainfall stations studied in the state of Rio de Janeiro, Brazil. 

The daily hyetographs, with amplitude of 10 mm of rainfall depth and lowest reading range of 0.1 mm, were converted to the digital format using a digitizing table and the system for digitalization of hyetographs HidroGraph 1.02. From the digitized data, the erosivity indices (EI30, KE >10, and KE >25) for each station were calculated using the computer program CHUVEROS (Program created by the Professor Elemar Antonino Cassol; not published). For each rainfall event, the program provides the rainfall depth and duration, kinetic energy, and I30, among other characteristics. It also characterizes the rainfall events in hydrological patterns.

Rainfall events were characterized in the following hydrological patterns: advanced (V), intermediate (I), and delayed (T), when the peak of highest intensity occurs in the first third, second third, and third third of the rainfall period, respectively (Horner and Jens, 1941). The rainfall characteristics and erosivity indices evaluated were rainfall depth (mm), kinetic energy - KE (J m-2), maximum 30-min intensity - I30 (mm h-1), rainfall duration (h), KE >10 (MJ ha-1 yr-1), KE >25 (MJ ha-1 yr-1), and EI30 (MJ mm ha-1 h-1 yr-1). The erosivity indices KE >25 and KE >10 were determined by the product of the kinetic energy and intensity of rainfall segments longer than 25 and 10 min h-1, respectively. The kinetic energy determination used in CHUVEROS is the same as EI30, i.e., from Wischmeier and Smith (1958) modified by Foster et al. (1981), and described as KE = 0.119 + 0.0873 × log I, where I is rainfall intensity in mm h-1 (Montebeller et al., 2007).

Joint evaluation of physical characteristics and erosivity indices to obtain the groups of stations/localities with similar characteristics regarding the different rainfall patterns was made through multivariate principal component analysis (Santos et al., 2010) and canonical discriminant analysis (Cruz-Castillo et al., 1994), using the programs Canoco (ter Braak and Smilauer, 1998) and SAS (SAS, 1997), respectively. For this study, the 12 months of the year were considered as replicates for each combination between station and rainfall pattern, and they were obtained from the mean of the erosive events in each month, for all years of the historical series.

The interpretation of PCA results considered the scores of the first two principal components (Y1 and Y2) and the values of the linear correlation coefficients between the original variables and the first two components. The cluster analyses of the treatments and of the above-mentioned variables through CDA used the mean of 12 months for each rainfall pattern.

Complementarily and to facilitate graphical interpretation of the CDA, multivariate analysis of variance was made through the Wilks’ Lambda test (Green and Carroll, 1978), and the Tukey test (p<0.05) was applied to the scores of the treatments.


The advanced rainfall pattern (V) was the most frequent (51.8 %), followed by the delayed pattern (T) (26.1 %), and the intermediate pattern (I) (22.1 %) in the state of Rio de Janeiro under the conditions studied (Table 1). This predominance of the advanced pattern also occurs in other regions of Brazil, such as São Borja - RS (Cassol et al., 2008), Quaraí - RS (Peñalva-Bazzano et al., 2007), Lavras - MG (Evangelista et al., 2005), Seropédica - RJ and Nova Friburgo - RJ (Carvalho et al., 2005), and São Gabriel - RS (Eltz et al., 2013). In contrast, the localities Manuel Duarte, Santa Isabel do Rio Preto, and Vila Mambucaba stood out from the others with their trend of higher frequency of rainfalls with delayed pattern, which is a rare characteristic in Brazil.

Table 1 Rainfall patterns for different stations in the state of Rio de Janeiro, Brazil 

Station code Station Altitude Advanced pattern Intermediate pattern Delayed pattern

Events % Events % Events %
Coasal Plain mesoregion
02242067 1 - Álcalis 7.4 81 55.9 26 17.9 38 26.2
02242101 25 - Rio Mole 10 120 45.6 69 26.2 74 28.1
Central Fluminense mesoregion
02142053 7 - Carmo 341 31 55.4 11 19.6 14 25.0
02242074 8 - Cordeiro 485 153 62.4 46 18.8 46 18.8
02142362 14 - Ilha dos Pombos 146 176 50.9 71 20.5 99 28.6
02242070 21 - Nova Friburgo 842 98 54.1 35 19.3 48 26.5
02142051 30 - Santa Maria Madalena 617 113 57.7 36 18.4 47 24.0
Metropolitan mesoregion
02243235 2 - Andorinhas 80 594 54.9 258 23.8 230 21.3
02242095 3 - Cachoeiras de Macacu 40 383 48.0 200 25.1 215 26.9
02243239 6 - Capela Mayrink 460 312 43.7 172 24.1 230 32.2
02243186 9 - Ecologia Agrícola 33 138 61.1 44 19.5 44 19.5
02243083 10 - Eletrobrás 40 266 44.3 131 21.8 203 33.8
02242094 11 - Escola União 10 413 43.0 232 24.2 315 32.8
02242100 12 - Fazenda Coqueiro 140 545 51.0 275 25.7 248 23.2
02242096 13 - Fazenda Santo Amaro 260 476 52.3 232 25.5 202 22.2
02344017 15 - Ilha Guaíba 64 138 59.0 52 22.2 44 18.8
02242097 17 - Japuíba 50 454 58.6 172 22.2 149 19.2
02242098 23 - Posto Garrafão 640 443 38.5 330 28.7 378 32.8
02243157 29 - São Bento 6 138 60.5 45 19.7 45 19.7
02242100 26 - Sambaetiba 10 425 62.5 134 19.7 121 17.8
02242091 31 - Tanguá 40 317 50.3 165 26.2 148 23.5
02242072 32 - Teresópolis 860 57 57.0 25 25.0 18 18.0
02243151 34 - Vassouras 437 125 64.8 33 17.1 35 18.1
02243238 36 - Xerém 138 407 52.9 196 25.5 167 21.7
North Fluminense mesoregion
02141044 5 - Campos 11 104 52.0 46 23.0 50 25.0
02241020 19 - Macaé 4 91 48.4 42 22.3 55 29.3
Northwest Fluminense mesoregion
16 - Itaperuna 115 60.8 30 15.9 44 23.3
South Fluminense mesoregion
4 - Angra dos Reis 2 161 51.9 70 22.6 79 25.5
02243008 20 - Manuel Duarte 400 172 27.6 106 17.0 345 55.4
02243165 22 - Piraí 388 141 65.9 40 18.7 33 15.4
02244092 24 - Resende 440 149 56.2 66 24.9 50 18.9
02243251 18 - Ribeirão das Lajes 462 315 50.2 145 23.1 167 26.6
02243205 27 - Santa Cecília 371 250 50.8 112 22.8 130 26.4
02244033 28 - Sta. I. do Rio Preto 567 275 36.2 179 23.6 305 40.2
02244103 33- Tocos 453 389 52.6 172 23.3 178 24.1
02344016 35 - Vila Mambucaba 6 197 29.6 140 21.0 329 49.4
Total Mean ----- 51.8 22.1 26.1

From identification of the advanced pattern as predominant in the state of Rio de Janeiro, it can be inferred that soil losses are smaller than they would be if there were predominance of rainfall events with intermediate and delayed patterns (Bertol et al., 2007). The higher water content in the soil preceding the rainfall events, common in these patterns, favors disaggregation of the particles due to the impacts of the raindrops, increases surface sealing, reduces infiltration capacity, and causes greater runoff (Peñalva-Bazzano et al., 2010). Using erosion plots under natural rainfall in Seropédica - RJ, Carvalho et al. (2009) identified that 64.6, 21.3, and 14.1 % of the rainfall events were classified as advanced (V), intermediate (I), and delayed (T), respectively, generating 35.1, 6.6, and 58.3 % of the soil losses, under different management practices in an Argissolo Vermelho Amarelo (Red Yellow Ultisol). Eltz et al. (2001) also confirmed the predominance of soil losses in the delayed pattern, working with simulated rainfall in this same soil class in Santa Maria, RS.

Through the PCA, 96.1 % of the information contained in the set of the seven original variables resulted from the first two principal components (Table 2), whose contributions were 83.0 and 13.1 % for the first (Y1) and second (Y2) axes, respectively. Higher Y1 values were obtained for the stations Manuel Duarte (man), Santa Isabel do Rio Preto (sir), and Vila Mambucaba (vil). In complement, figure 2 shows the behavior in the first axis, in which the stations of the South Fluminense mesoregion are distant from the others, situated farther from and to the right of the origin. In the second axis, the highest values of principal components occurred for the stations Angra dos Reis (ang), Campos (cam), Capela Mayrink (cap), and Xerém (xer).

Table 2 Values of the principal components (Y1 and Y2) for the rainfall stations, rainfall patterns, and percent of information retained by them (% variance and % accumulated variance) 

Factor Y1 Order Y2 Order Factor Y1 Order Y2 Order
alc - Álcalis 0.076 14 1.008 6 nov - Nova Friburgo -0.666 35 -0.791 33
and - Andorinhas -1.014 39 0.033 19 pir - Piraí -0.421 27 -0.511 28
ang - Angra dos Reis 0.696 4 2.17 1 pos - Posto Garrafão -0.753 36 -0.388 26
cac - Cach. de Macacu 0.083 13 0.13 15 res - Resende -0.608 32 -0.675 31
cam - Campos 0.557 7 1.44 2 rio - Rio da Cidade 0.188 12 0.04 18
cap - Capela Mayrink -0.637 33 1.205 4 sam - Sambaetiba -0.473 29 0.208 14
car - Carmo -0.583 31 -1.033 36 sce - Santa Cecília 0.448 8 -1.078 37
cor - Cordeiro -0.144 19 -1.024 35 smm - Santa M. Madalena -0.064 16 0.072 17
eco - Ecologia Agrícola -0.398 26 -0.751 32 sbe - São Bento -0.145 20 -0.365 25
ele - Eletrobrás 0.582 6 0.525 11 sir - Sta Izabel do R. Preto 2.115 2 -0.667 30
esc - Escola União -0.242 25 -0.942 34 tan - Tanguá 0.334 10 -0.097 21
fco - Faz. Coqueiro -0.426 28 0.798 8 ter - Teresópolis 0.31 11 0.584 10
fsa - Faz. Santo Amaro -0.778 38 0.736 9 toc - Tocos -0.097 17 -1.413 39
Ipo - Ilha dos Pombos -0.134 18 -1.326 38 vas - Vassouras -0.664 34 -0.649 29
Igu - Ilha Guaíba -0.163 21 1.065 5 vil - Vila Mambucaba 1.53 3 0.121 16
Ita - Itaperuna -0.774 37 -0.397 27 xer - Xerém -0.199 23 1.211 3
jap - Japuíba -0.167 22 0.381 12 Advanced pattern -0.57 30 0.264 13
laj - Ribeirão das Lajes 0.367 9 -0.295 24 Intermediate pattern -0.024 15 0.017 20
mac - Macaé -0.217 24 0.961 7 Delayed pattern 0.594 5 -0.281 22
man - Manuel Duarte 2.480 1 -0.284 23 --- --- --- --- ---

Figure 2 Principal component analysis of rainfall and erosivity characteristics considering the 36 rainfall stations and the three rainfall patterns (advanced - V, intermediate - I, and delayed - T). RD: rainfall depth; KE: kinetic energy; I30: maximum 30 - min intensity; EI30, KE >10, and KE >25: erosivity indices. Other abbreviations see Table 2. 

The first principal component (Y1) has high positive correlation with six of the rainfall characteristics and erosivity indices studied (r >0.91) (Table 3). Hence, all characteristics, except rainfall duration, contributed to Y1 and were considered good predictors of the effects of the treatments and, therefore, adequate for utilization in other statistical analyses. The high positive correlation with Y2 (0.75) indicates that duration is the rainfall characteristic of highest correlation with the stations Angra, Campos, Capela Mayrink, and Xerém (Figure 2). These stations have low altitudes in common (Table 1) and are more influenced by the sea breeze, according to Dereczynski et al. (2009).

Table 3 Linear correlation coefficients between the rainfall and erosivity characteristics and the first two principal components (Y1 and Y2) 

Variable Y1 Y2
RD -0.95 0.27
KE -0.98 0.14
I30 -0.91 -0.31
EI30 -0.97 -0.11
KE >10 -0.98 -0.15
KE >25 -0.91 -0.37
Duration -0.63 0.75

RD: rainfall depth; KE: kinetic energy; I30: maximum 30 - min intensity; EI30, KE >10, and KE >25: erosivity indices.

It was not possible to find the determinant effect of the rainfall and erosivity characteristics in the rainfall patterns (V, I, and T) for the stations studied, because all of them approach the origin of the axes Y1 and Y2 (Figure 2). However, rainfall depth was the characteristic closest to the advanced pattern. Regarding the localities and the respective patterns, the most correlated stations were Andorinhas, Fazenda Santo Amaro, Sambaetiba, Capela Mayrink, Fazenda Coqueiro, Posto Garrafão, Itaperuna, Vassouras, Resende, and Nova Friburgo for the advanced pattern; Santa Maria Madalena, Japuíba, São Bento, Piraí, Cachoeiras de Macacu, Álcalis, Rio Mole, Escola União, Carmo, Cordeiro, Ilha dos Pombos, and Tocos for the intermediate pattern; and Manuel Duarte, Santa Izabel do Rio Preto, Vila Mambucaba, Ribeirão das Lajes, Tanguá, Eletrobrás, and Santa Cecília for the delayed pattern (Figure 2).

Most of the stations correlated with the advanced pattern are located in the Metropolitan region. However, they are closer to the border with the Central Fluminense region, which is the Serrana region on other maps. According to Montebeller et al. (2007), this region has the highest erosivity indices of the rainfall events in the state. In contrast, the stations that have delayed pattern were more associated with the South Fluminense mesoregion. According to Brito et al. (2016), these parts of the state have total monthly rainfall events influenced by valley-mountain atmospheric circulations and also by orographic rainfall.

Through CDA, the first two functions explain 64.3 % of the total data variation, 42.2 % by the first function (CDF1) and 22.1 % by the second function (CDF2). The likelihood ratio test showed that CDF1 and CDF2 were significant at the 0.01 probability level, and the explanation of the variation between the samples given by the other CDFs is small. The canonical correlations were 0.70 and 0.58 for CDF1 and CDF2, respectively, considered as high. The results indicate high association between localities, as well as rainfall patterns with the EI30 and KE >25 indices and the physical characteristics of rainfalls (Table 4).

Table 4 Standardized canonical coefficients (SCC) and correlation coefficients (r) between the first canonical discriminant functions (CDFs) and the seven variables evaluated 

Variable Discriminant function


RD 0.77 0.76 -5.30 -0.47
KE -0.10 0.68 4.35 -0.38
I30 1.39 0.47 0.86 0.11
EI30 -0.79 0.40 -1.21 -0.24
KE >10 -0.26 0.44 0.04 -0.13
KE >25 -0.11 0.32 0.46 0.06
Duration 0.89 0.73 1.38 -0.05
% of variance explained 42.2 22.1

RD: rainfall depth; KE: kinetic energy; I30: maximum 30-min intensity; EI30, KE >10, and KE>25: erosivity indices.

The first discriminant functions (CDF1 and CDF2) differed among localities and rainfall patterns through the multivariate Wilks’ Lambda test (p<0.0001) (Green and Carroll, 1978). For the characteristics I30, duration, and EI30, there were higher values (in modulus) of the standardized canonical coefficient for CDF1, while rainfall depth (RD), followed by duration, along with their correlation coefficients, were higher for CDF2 (Table 4).

Based on I30, rainfall intensity is the physical characteristic most adequate for different localities in the state of Rio de Janeiro with respect to erosive potential, under different rainfall patterns. This index is used to calculate the EI30 utilized in the Universal Soil Loss Equation (USLE) and is considered one of the best estimators of erosivity in countries with temperate climates (Wischmeier and Smith, 1978; Oliveira et al., 2013).

Despite subjectivity in visualization of the groups due to the large number of points, three groups of stations could be identified through the CDA in analysis of the mean scores of CDF1 for the combination of stations versus rainfall patterns (108 observations) (Figure 3). The capV group (Capela Mayrink and advanced pattern) has lower scores; the sirT (Sta I. do Rio Preto) and ManT (Manuel Duarte) group of stations has higher values, both for the delayed pattern; and the third group, with intermediate scores, contained the other stations associated with each pattern (Figure 3). The results found in the delayed pattern were similar to those obtained by the PCA.

Figure 3 Standardized scores of the significant canonical discriminant functions (CDF) of rainfall and erosivity attributes for the 36 localities of the state of Rio de Janeiro, Brazil. 

In the group of the stations for each rainfall pattern that was isolated, greater dispersion was observed in the advanced pattern, and it was possible to identify three groups of stations. The Capela Mayrink station had a lower score, while Santa Isabel do Rio Preto, Manuel Duarte, Vila Mambucaba, and Cordeiro had higher values (Figure 4a). In the intermediate pattern, the Itaperuna station exhibited the lowest score, while Manuel Duarte, Teresópolis, and Santa Isabel do Rio Preto showed the highest scores (Figure 4b). The other stations had intermediate values. In the delayed pattern, three groups were identified - the stations Vassouras and São Bento had the lower scores, whereas Manuel Duarte, Sta. I. do Rio Preto, Vila Mambucaba, Teresópolis, and Angra dos Reis had higher scores (Figure 4c).

Figure 4 Standardized scores of the significant canonical discriminant functions (CDF) of rainfall and erosivity attributes in the advanced (a), intermediate (b), and delayed (c) patterns for the 36 localities of the state of Rio de Janeiro, Brazil. 

To facilitate interpretation of the group presented in figures 3 and 4, the Tukey test (p<0.05) was applied to the scores (index resulting from the linear combination of all variables) of the stations/localities within each rainfall pattern. Since CDF1 explained most of the data variation (42.2 %), only its result relative to the Tukey test (p<0.05) will be presented (Table 5).

Table 5 Means of standardized canonical scores of the canonical discriminant function (CDF1) of the physical characteristic and erosivity variables for the rainfall patterns 

Station Advanced Intermediate Delayed

CDF1 Tukey Test CDF1 Tukey Test CDF1 Tukey Test
alc-Álcalis -59.95 abcdef* -28.97 a -109.10 ab*
and-Andorinhas -87.40 ghi -86.76 a -182.42 ab
ang-Angra -62.56 abcdefg -52.69 a -57.37*** a
cac-C. de Macacu -72.52 cdefg -58.33 a -80.91 ab
cam-Campos -56.41 abcde -29.93 a -64.58 a
cap-Capela Mayrink -105.19** i -114.14 a -123.84 ab
car-Carmo -70.54 bcdefg -110.07 a -141.69 ab
cor-Cordeiro -54.53*** abcd -68.34 a -130.62 ab
eco-Ecologia Agrícola -65.61 abcdefg -63.62 a -161.74 ab
ele-Eletrobras -73.68 cdefgh -48.52 a -60.56 a
esc-Escola União -74.18 cdefgh -75.1 a -95.74 ab
fco-Fazenda Coqueiro -86.03 ghi -54.03 a -104.27 ab
fsa-F. Santo Amaro -99.01 hi -76.73 a -141.53 ab
igu-Ilha Guaíba -65.61 abcdefg -110.43 a -83.73 ab
ipo-Ilha dos Pombos -62.14 abcdefg -68.33 a -97.30 ab
ita-Itaperuna -64.82 abcdefg -120.16** a -151.99 ab
jap-Japuíba -71.82 bcdefg -55.09 a -100.00 ab
laj-Rib. das Lajes -59.18 abcde -32.16 a -82.30 ab
mac-Macaé -80.07 efghi -73.9 a -90.56 ab
man-Manuel Duarte -46.76*** ab -22.86*** a -12.98*** a
nov-Nova Friburgo -69.77 bcdefg -53.63 a -184.20 ab
pir-Piraí -59.97 abcdef -39.56 a -185.00 ab
pos-Posto Garrafão -84.53 fghi -105.79 a -129.91 ab
res-Resende -70.89 bcdefg -55.53 a -132.60 ab
rio-Rio Mole -69.54 bcdefg -38.44 a -165.32 ab
sam-sambaetiba -69.90 bcdefg -61.65 a -173.07 ab
sbe-São bento -62.95 abcdefg -33.34 a -204.09** ab
sce-Santa Cecília -59.19 abcdef -38.48 a -77.29 ab
sir-Sta I. do Rio Preto -41.95*** a -25.22*** a -13.30*** a
smm-S. Mª. Madalena -72.11 cdefg -44.25 a -105.38 ab
tan-Tanguá -68.61 bcdefg -55.97 a -64.63 a
ter-Teresópolis -78.02 defgh -25.13*** a -56.23*** a
toc-Tocos -59.69 abcdef -61.01 a -77.26 ab
vas-Vassouras -55.95 abcde -38.78 a -289.11** b
vil-Vila Mambucaba -52.14*** abc -50.78 a -36.14*** a
xer-Xerém -87.23 ghi -66.39 a -67.51 a

* Means followed by different letters in the column differ by Tukey’s T test (p≤0.05); ** stations with lowest scores; *** stations with highest scores.

In the advanced pattern, the Capela Mayrink station showed the lowest score, while Santa Izabel do Rio Preto (sir), Manuel Duarte, Vila Mambucaba, and Cordeiro showed the highest values (Table 5). The other stations did not differ, and some of them were similar to sir (man, vil, cor, vas, cam, laj, sce, toc, alc, pir, ipo, ang, sbe, ita, igu, and eco stations) and others similar to cap (mac, pos, fco, xer, and fsa stations) with respect to the rainfall and erosivity characteristics. For the intermediate pattern, the localities did not differ regarding the physical characteristics of rainfall events and erosivity evaluated as a single group, considering CDF1 (Tukey <0.05). However, a lower score was observed for the Itaperuna station and higher scores for the Manuel Duarte, Teresópolis, and Santa Izabel do Rio Preto stations (Table 5).

For the delayed pattern, three groups of stations were identified: the group with the lowest values was formed by the stations Vassouras and São Bento; the group with the highest scores was composed of the stations Angra dos Reis, Campos, Eletrobrás, Manuel Duarte, Santa Isabel do Rio Preto, Tanguá, Teresópolis, Vila Mambucaba, and Xerém; the other stations showed intermediate scores (Table 5).

Considering that most characteristics are directly related to erosive potential, the more negative the mean of the scores, the greater the possibility of lower erosive potential of the rainfall events under the influence of the stations of the respective groups and rainfall patterns (Table 5). However, the stations with the highest scores, regardless of the rainfall patterns (except for Santa Isabel do Rio Preto), are related to higher values of rainfall erosivity, corroborating the study of Montebeller et al. (2007). According to these authors, the sites with the highest erosivity indices in the state occur in the Sul Fluminense region (Baía de Ilha Grande) and also in the direction from the Baixadas Litorâneas to Central Fluminense (Serrana region). This demonstrates the influence of the relief, notably the Serra do Mar mountain range, in this behavior of rainfall events. The stations with the highest scores (Table 5) coincide with most of those correlated with the delayed pattern obtained in the PCA (Figure 2), showing a certain consistency of this analysis in discriminating stations and rainfall patterns. A possible explanation for this is the procedure common to these techniques of making linear combinations of the original variables.

The stations belonging to the group with the highest significant score for CDF1, in relation to the others in the delayed pattern, indicate that these localities (Figure 5) are similar regarding the expected high risk of soil, water, and nutrient losses due to the physical characteristics of rainfall events in them. This may occur especially in unprotected soils due to higher antecedent water content, which leads to lower infiltration capacity, lower cohesion of aggregates, and greater surface sealing and runoff. In addition, higher erosive potential is expected in this group because of the rainfall events with physical characteristics of high rainfall depth, high I30, and long duration, because the standardized canonical coefficient (SCC) and the correlation coefficient (r) of these characteristics were positive. Except for the Campos station, the highest scores in the delayed pattern were obtained for stations located in the regions South Fluminense, Central Fluminense, and Metropolitan, which had the highest monthly and annual mean values of rainfall (André et al., 2008; Brito et al., 2016) and the highest values of rainfall erosivity in the state (Montebeller et al., 2007; Machado et al., 2013). These results are highly influenced by the topography and higher air humidity coming from the ocean in the southeast direction (André et al., 2008).

Figure 5 Location of the stations of delayed pattern with respect to the variation in standardized canonical scores of the first canonical discriminant function (CDF1). 

The results obtained suggest conducting studies in erosion plots using simulated rainfall of a delayed pattern, because this would allow evaluation of the effect of different soil management systems and conservation practices on water erosion. The aim would be to control this type of soil degradation through systematization of techniques that can be extrapolated to the localities of this group.


Principal component analysis proved to be adequate for the data. In addition, characteristics such as rainfall depth, erosivity index EI30, kinetic energy (KE), maximum 30-min intensity (I30), the erosivity indices KE >10 and KE >25, and duration of the rainfall events showed significant correlation with the localities and rainfall patterns studied.

Canonical discriminant analysis differentiates the stations/localities in the rainfall patterns. Through this technique, three groups of homogeneous stations are formed, regardless of the rainfall pattern, as well as three groups of similar stations for the advanced and delayed patterns for physical characteristics of rainfall events and erosivity in the state.


Our thanks to UFRRJ/CPGA-CS, Capes (financial support to the first author), Serla (currently INEA), INMET, ANA, and Light.


Anache JAA, Wendland EC, Oliveira PTS, Flanagan DC, Nearing MA. Runoff and soil erosion plot-scale studies under natural rainfall: a meta-analysis of the Brazilian experience. Catena. 2017;152:29-39. ]

André RGB, Marques VS, Pinheiro FMA, Ferraudo AS. Identificação de regiões pluviometricamente homogêneas no estado do Rio de Janeiro, utilizando-se valores mensais. Rev Bras Meteorol. 2008;23:501-9. ]

Bazzano MGP, Eltz FLF, Cassol EA. Erosividade e características hidrológicas das chuvas de Rio Grande (RS). Rev Bras Cienc Solo. 2010;34:235-44. ]

Bazzano MGP, Eltz FLF, Cassol EA. Erosividade, coeficiente de chuva, padrões e período de retorno das chuvas de Quaraí, RS. Rev Bras Cienc Solo. 2007;31:1205-17. ]

Bertol I, Leite D, Engel FL, Cogo NP, González AP. Erodibilidade de um Nitossolo Háplico alumínico determinada em condições de campo. Rev Bras Cienc Solo. 2007;31:541-9. ]

Bertol I, Zoldan Junior WA, Fabian EL, Zavaschi E, Pegoraro R, González AP. Efeito de escarificação e da erosividade de chuvas sobre algumas variáveis de valores de erosão hídrica em sistemas de manejo de um Nitossolo Háplico. Rev Bras Cienc Solo. 2008;32:747-57. ]

Brito TT, Oliveira-Júnior JF, Lyra GB, Gois G, Zeri M. Multivariate analysis applied to monthly rainfall over Rio de Janeiro state, Brazil. Meteorol Atmos Phys. 2016;128:1-10. ]

Carvalho DF, Cruz ES, Pinto MF, Silva LDB, Guerra JGM. Características da chuva e perdas por erosão sob diferentes práticas de manejo do solo. Rev Bras Eng Agr Amb. 2009;13:3-9. ]

Carvalho DF, Montebeller CA, Franco EM, Valcarcel R, Bertol I. Padrões de precipitação e índices de erosividade para as chuvas de Seropédica e Nova Friburgo, RJ. Rev Bras Eng Agr Amb. 2005;9:7-14. ]

Cassol EA, Eltz FLF, Martins D, Lemos AM, Lima VS, Bueno AC. Erosividade, padrões hidrológicos, período de retorno e probabilidade de ocorrência das chuvas em São Borja, RS. Rev Bras Cienc Solo. 2008;32:1239-51. ]

Cruz-Castillo JG, Ganeshanandam S, Mackay BR, Lawes GS, Lawoko CRO, Woolley DJ. Applications of canonical discriminant analysis in horticultural research. HortScience. 1994;29:1115-9. [ Links ]

Dai Z, Li R, Muhammad N, Brookes PC, Wang H, Liu X, Xu J. Principle component and hierarchical cluster analysis of soil properties following biochar incorporation. Soil Sci Soc Am J. 2014;78:205-13. ]

Dereczynski CP, Oliveira JS, Machado CO. Climatologia da precipitação no município do Rio de Janeiro. Rev Bras Meteorol. 2009;24:24-38. ]

Eltz FLF, Cassol EA, Pascotini PB, Amorim RSS. Potencial erosivo e características das chuvas de São Gabriel, RS, de 1963 a 1993. Rev Bras Eng Agr Amb. 2013;17:647-54. ]

Eltz FLF, Mehl HU, Reichert JM. Perdas de solo e água em entressulcos em um Argissolo Vermelho-Amarelo submetido a quatro padrões de chuva. Rev Bras Cienc Solo. 2001;25:485-93. ]

Evangelista AWP, Carvalho LG, Bernardino DT. Caracterização do padrão das chuvas ocorrentes em Lavras, MG. Irriga. 2005;10:306-17. [ Links ]

Flanagan DC, Foster GR, Moldenhauer WC. Storm pattern effect on infiltration, runoff, and erosion. Am Soc Agr Biol Eng. 1988;31:414-20. ]

Foster GR, McCool DK, Renard KG, Moldenhauer WC. Conversion of the universal soil loss equation to SI metric units. J. Soil Water Conserv. 1981;36:355-59. [ Links ]

Goebes P, Seitz S, Geißler C, Lassu T, Peters P, Seeger M, Nadrowski K, Scholten T. Momentum or kinetic energy - How do substrate properties influence the calculation of rainfall erosivity? J Hydrol. 2014;517:310-6. ]

Green PE, Carroll JD. Analysing multivariate data. Illinois: Dryden Press; 1978. [ Links ]

Horner WW, Jens SW. Surface runoff determination from rainfall without using coefficients. Trans Am Soc Agr Eng. 1941;107:1039-117. [ Links ]

Jin L, Zhu J, Huang Y, Zhao HS, Lin KP, Jin J. A nonlinear statistical ensemble model for short-range rainfall prediction. Theor Appl Climatol. 2015;119:791-807. ]

Machado RL, Carvalho DF, Rouws JRC, Gomes DP, Eduardo EN. Erosividade das chuvas associada a períodos de retorno e probabilidade de ocorrência no Estado do Rio de Janeiro. Rev Bras Cienc Solo. 2013;37:529-47. ]

Machado RL, Carvalho DF, Costa JR, Oliveira Neto DH, Pinto MF. Análise da erosividade das chuvas associada aos padrões de precipitação pluvial na região de Ribeirão das Lajes (RJ). Rev Bras Cienc Solo. 2008;32:2113-23. ]

Maluche-Baretta CRD, Amarante CVT, Klauberg Filho O. Análise multivariada de atributos do solo em sistemas convencional e orgânico de produção de maçãs. Pesq Agropec Bras. 2006;41:1531-9. ]

Marques JJGSM, Alvarenga RC, Curi N, Santana DP, Silva MLN. Índices de erosividade da chuva, perdas de solo e fator erodibilidade para dois solos da região dos cerrados - primeira aproximação. Rev Bras Cienc Solo. 1997;21:427-34. ]

Mehl HU, Eltz FLF, Reichert JM, Didoné IA. Caracterização de padrões de chuvas ocorrentes em Santa Maria (RS). Rev Bras Cienc Solo. 2001;25:475-83. ]

Mohamadi MA, Kavian A. Effects of rainfall patterns on runoff and soil erosion in field plots. Int Soil Water Conserv Res. 2015;3:273-81. ]

Montebeller CA, Ceddia MB, Carvalho DF, Vieira SR, Franco EM. Variabilidade espacial do potencial erosivo das chuvas no estado do Rio de Janeiro. Eng Agric. 2007;27:426-35. ]

Oliveira JR, Pinto MF, Souza WJ, Guerra JGM, Carvalho DF. Erosão hídrica em um Argissolo Vermelho-Amarelo, sob diferentes padrões de chuva simulada. Rev Bras Eng Agr Amb. 2010;14:140-7. ]

Oliveira PTS, Rodrigues DBB, Sobrinho TA, Carvalho DF, Panachuki E. Spatial variability of the rainfall erosive potential in the state of Mato Grosso do Sul, Brazil. Eng Agric. 2012;32:69-79. ]

Oliveira PTS, Wendland E, Nearing MA. Rainfall erosivity in Brazil: a review. Catena. 2013;100:139-47. ]

Pearson K. On lines and planes of closest fit to systems of points in space. Philos Mag. 1901;2:559-72. ]

Prosdocimi M, Cerdà A, Tarolli P. Soil water erosion on Mediterranean vineyards: a review. Catena. 2016;141:1-21. ]

Ran Q, Su D, Li P, He Z. Experimental study of the impact of rainfall characteristics on runoff generation and soil erosion. J Hydrol. 2012;424-425:99-111. ]

Raziei T, Saghafian B, Paulo AA, Pereira LS, Bordi I. Spatial patterns and temporal variability of drought in western Iran. Water Resour Manag. 2009;23:439-55. ]

Santos JF, Pulido-Calvo I, Portela MM. Spatial and temporal variability of droughts in Portugal. Water Resour Res. 2010;46:1-13. ]

Statistical Analysis Systems Institute Inc. SAS/STAT software: changes and enhancements through release 6.12. Cary: Statistical Analysis Systems Institute Inc.; 1997. [ Links ]

Silva RB, Iori P, Silva AFM. Proposição e validações de equações para estimativa da erosividade de dois municípios do estado de São Paulo. Irriga. 2009;14:533-47. [ Links ]

ter Braak CJF, Smilauer P. Canoco reference manual and user’s guide to canoco for windows: Software for canonical Community Ordination (Version 4). Microcomputer Power; 1998. [ Links ]

Wang W, Yin S, Xie Y, Liu B, Liu Y. Effects of four storm patterns on soil loss from five soils under natural rainfall. Catena. 2016;141:56-65. ]

Wischmeier WH, Smith DD. Predicting rainfall erosion losses: a guide to conservation planning. Washington, DC: USDA; 1978. (Agricultural handbook, 537) [ Links ]

Wischmeier WH, Smith DD. Rainfall energy and its relationships to soil loss. Trans. Am. Geophys. Union. 1958;39:285-91. [ Links ]

Received: September 9, 2016; Accepted: May 8, 2017

* Corresponding author E-mail:

Creative Commons License  This is an Open Access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.