Long-term variation of precipitation indices in Ceará State, Northeast Brazil



The state of Ceará is a semiarid region in north-eastern Brazil, having a high spatial and temporal variability of precipitation, which poses challenges for water resources management. The objective of this study is to describe and analyse the long-term variation of monthly precipitation indices and their relation to sea surface temperature (SST) anomalies. Data from 55 weather stations in the state of Ceará from 1974 to 2009 was analysed. In general, a decreasing tendency in monthly precipitation was observed over almost all the state of Ceará. The results point to a tendency for dry months to become dryer and to a decrease in precipitation intensity. SST anomalies from October to March correlate with precipitation indices from January to April, showing decreasing lag times towards the end of the wet season. The most influential SST anomalies locations are Niño 1+2, Niño 3, Niño 3.4 and Global Tropics.

1. Introduction

Water resources are of great importance to many economic activities, being essential to agriculture, water supply, industry and energy production. Impacts from either an increase or a decrease in mean and extreme precipitation may include increased floods, droughts, surface and groundwater contamination, and soil erosion, which in turn may raise the risk of diseases, property damage, possible crop failure and compromise livestock rearing.

In a worldwide study conducted by Alexander et al. (2006), a significant annual tendency towards wetter conditions was observed between 1901 and 2003 suggesting that global wetting is likely to be part of a longer-term trend. Magrin et al. (2007) report a similar tendency in South America. The studies conducted by Santos and Brito (2007) and Santos et al. (2009) in a semiarid region in Brazil (Rio Grande do Norte, Paraíba and Ceará states), report a tendency towards more humid conditions on a yearly scale, especially in continuous wet days, total annual precipitation and extreme precipitation values. However, Haylock et al. (2006) verified drier conditions for Northeast Brazil, together with a significant variability in total annual precipitation. López-Moreno et al. (2009) suggest a need for denser monitoring networks, because of large differences in both the sign and magnitude of trends at nearby stations, emphasizing the high spatial variability of precipitation indices and the need for local studies.

The state of Ceará is in the semiarid region of north-eastern part of Brazil, with high spatial and temporal variability of precipitation (Chu, 1983; Haylock et al., 2006; Alves et al., 2009). In some regions, it is not uncommon that 70% of the annual precipitation occurs in only 1 month (Andrade et al., 2010).

The annual and monthly rainfall variability is related to different atmospheric systems which are connected to Tropical Pacific and Atlantic sea surface temperature (SST) (Hastenrath, 1986; Uvo et al., 1998; Ferreira and Mello, 2005; Sun et al., 2006). At the beginning of the rainy season (January), precipitation over the state can be attributed to frontal systems (Ferreira and Mello, 2005). From February to May, precipitation is related to the displacement of the Intertropical Convergence Zone (ITCZ) to the South. The ITCZ location is controlled, to a large extent, by the delayed feedback of the Atlantic SST variability (Hastenrath and Greischar, 1993; Moncunill, 2006; Santos and Brito, 2007; Santos and Manzi, 2011). When Atlantic SST favours northeast trade winds with a weakening of the southeast trade winds, ITCZ migrates to the Southern Hemisphere with a resulting increase in precipitation in March and April (Hastenrath, 2012). Precipitation in these months accounts for approximately 50% of total annual precipitation (Andrade et al., 2010). The end of the wet season is determined by the return of the ITCZ to the Northern Hemisphere. Unusually, the rainfall season may be extended until June/July due to easterly wave disturbances (EWD) (Ferreira et al., 1990; Kayano, 2003, Ferreira and Mello, 2005; Torres and Ferreira, 2011).

The effect of decreased wetness on agricultural production depends more on the timing of drought conditions than on their severity or duration (Mishra et al., 2010), emphasizing the need for analysis of evolution of precipitation indices. In this region, farming is mostly rainfed, and knowledge of rainfall variability is of extreme importance for water resources management (Santos et al., 2009). The growing season begins with the wet season, searching for suitable soil water content. A disruption in the rainfall season with dry spells, locally called ‘veranicos’, will result in a negative impact on rainfed agriculture. Changes in extreme rather than mean values lead to greater impact, and its analysis becomes very important in water resources management (López-Moreno et al., 2009; Santos et al., 2009).

Local climate variability may be assessed by analysis of time series of meteorological indices calculated from precipitation records (Moncunill, 2006; Santos and Brito, 2007; Durão et al., 2009; Santos et al., 2009; Özger et al., 2010; EEA, 2011; Villarini et al., 2011). The Climate Variability and Predictability (CLIVAR) expert team on climate change developed a series of extreme climate indices from daily temperature and precipitation data to study climate changes (Alexander et al., 2006) and has been used by several researchers for analysis of trends in precipitation indices.

Considering the spatial and temporal variability of precipitation regimes in Ceará, it is essential to try to identify long-term patterns and tendencies in precipitation and associated indices. The objective of this study is to describe and analyse, on a monthly scale, long-term variations precipitation indices – total precipitation, precipitation intensity and frequency, duration of wet and dry spells – in the state of Ceará, Brazil and relate them to SST anomalies.

2. Data and methods

2.1. Study area

The state of Ceará lies in the north-eastern part of Brazil with a total area of approximately 146 000 km2, with 560 km of coastline. Total population in the state of Ceará is around 8.5 million, of which 2.4 million live in the state capital, Fortaleza, with a population density of 7815 hab km−2, contrasting with a 6.7 hab km−2 in the Aiuaba municipality (IPECE homepage2011). These statistics highlight the spatial variability of population, and respective different needs regarding water resources management.

Ceará has a quite diverse climate, with distinct climate regions, most of the state being classified as BSh', according to the Koppen climate classification, with a potential evapotranspiraton that exceeds precipitation. The territory may be divided into four major climatic regions (Figure 1): Litorânea (Ocean), Serrana (the Sierras), Cariri and Semiarid (Andrade et al., 2010), according to the water resources plan of the state of Ceará. Annual precipitation varies from 600 mm in the semiarid region to 1400 mm in the ocean region (Santos et al., 2009) with a potential evapotranspiration of 1850–1900 mm.

Figure 1.

Climatic regions, water bodies, desertification areas and precipitation gauges used in the study in the state of Ceará, Brazil. (Adapted from http://www2.ipece.ce.gov.br/atlas/capitulo1/12/index.htm).

Geologically, the state is formed, mostly, of crystalline nappes, characterized by shallow soils, low infiltration capacity and high surface runoff (Andrade et al., 2010). This lack of storage capacity and increased surface runoff which translates into mostly ephemeral streams, has led to development of small-scale water storage systems as can be seen in Figure 1, which shows natural and artificial water bodies with a surface area above 5 ha. The two major artificial lakes account for 21% of lake surface area (51 414 ha of 240 000 ha) and all artificial lakes account for over 91% of lake surface area. A need for adequate management of surface storage is evident, because this water is used for human consumption in urban and/or settlement areas, for animal husbandry and agricultural development, although most agriculture in Ceará is rainfed. Changes in precipitation regimes will affect current management practices and should be taken into account for development of this region.

Land use and vegetation in the state of Ceará is predominantly mangrove caatinga, which is a type of scrubby forest, and evergreen vegetation on the hilly areas of the state. Rainfed agriculture and irrigation districts play a significant role in land use, which are affected by the overall water deficit in the state that runs for approximately 10–11 months in the semiarid region, and for 6–8 months (June–July to January–February) in the Ocean area, the sierras and Cariri (Figure 1), which is a tropical hot sub-humid climatic region (Andrade et al., 2010).

A reduction of anthropogenic activity and restoration of the vegetation cover is being implemented in the State of Ceará, with positive results. Nonetheless, federal programmes have been necessary for preservation of these ecosystems (e.g. Chapada do Araripe and Ceará). The areas most prone to desertification have been identified in the state of Ceará (Figure 1), which are extremely vulnerable to over-exploitation and inappropriate land use as well as climate variability.

2.2. Data

Precipitation data was obtained from Funceme (2008). From a total of 669 stations with daily precipitation data, 141 stations had over 30 years of data and 55 had no missing data. The analysis was performed using daily precipitation data from those 55 weather stations from 1974 to 2009 (36 years), the longest period with no missing data for a greater number of stations scattered throughout the state of Ceará, Brazil (Figure 1). No stations were excluded based on proximity to urban areas and the normally implicit urban heat island effect, because these areas do not seem to have an effect on long-term trends (Peterson and Owen, 2005).

A homogeneity test was performed on the annual data using the Shapiro–Wilk parametric hypothesis test of composite normality at a significance level of 5% for rejection of data series. Only homogenous data was used in this study.

2.3. Precipitation indices

The precipitation indices herein analysed were based on the indices developed by the CLIVAR Expert Team on Climate Change Detection and evaluated on a monthly basis. These indices are based on percentiles, absolute values, threshold values and duration (Alexander et al., 2006). Computation of these indices was performed according to the equations described in Table 1.

Table 1. Precipitation indices and respective equations (Adapted from http://cccma.seos.uvic.ca/ETCCDMI/list_27_indices.shtml)
PWD – monthly precipitation on wet days – RR ≥ 1 mmmath formula(1)
CDD – monthly maximum length of dry spell, maximum number of consecutive days with RR < 1mmCDDj = max(days(RRij) < 1 mm)(2)
CWD – monthly maximum length of wet spell, maximum number of consecutive days with RR ≥ 1mmCWDj = max(days(RRij) ≥ 1 mm)(3)
NDD – monthly count of days with RR < 1 mmNDDj = days(RRij < 1 mm)(4)
NWD – monthly count of days with RR ≥ 1 mmNWDj = days(RRij ≥ 1 mm)(5)
NP75/NP90/NP95 – monthly number of days with RR above the 75/90/95 percentileNPpj = days(RRij ≥ Pp)(6)
Rx1day – monthly maximum 1-d precipitationRx1dayj = max(RRij)(7)
Rx5day – monthly maximum consecutive 5-d precipitationRx5dayj = max(RRkj)(8)
SDII – simple daily intensity indexmath formula(9)

j = month in analysis.


RRij = daily precipitation amount on day i in month j.


RRkj = precipitation amount for the 5-d interval ending day k, in month j.


RRwj = daily precipitation amount on wet days, w (RR ≥ 1 mm) in month j.


W = number of wet days.


p = user defined threshold (75, 90 and 95 percentiles).


Pp = precipitation at 75, 90 and 95 percentiles.

Extreme frequency indices were calculated at a monthly scale at different cut-offs –75, 90 and 95 percentiles rather than a fixed threshold (Nicholls and Haylock, 2000; Bates et al., 2008; Guerreiro et al., 2010). Percentiles were evaluated using all data set for each and all stations (Silva et al., 2009).

2.4. Data analysis methods

To identify similarities among the precipitation indices under study, an agglomerative hierarchical cluster analysis was performed. A combination of the Ward's linkage method and the squared Euclidean distances, as a measure of similarity, was carried out using the statistical package, SPSS version 19. Data was standardized by the z-score method, in order to minimize the influence of different units of measurement.

A correlation between standardized values of SST anomalies in the Niño 1+2, Niño 3, Niño 4, Niño 3.4, Tropical North Atlantic (NATL), Tropical South Atlantic (SATL), Global Tropics (TROP) regions (Table 2) and the standardized monthly indices was performed for a lag-time of 1–6 months (lag-time of 1 month means that January SST is correlated to the February indices). This analysis was performed for the wet season of January to June.

Table 2. Sea surface temperature regions (Adapted from http://www.esrl.noaa.gov/psd/data/climateindices/list)
 Latitude limitLongitude limit
Niño 1+2 (Extreme Eastern Tropical Pacific)0–10°South90°West–80°West
Niño 3 (Eastern Tropical Pacific)5°North–5°South150°West–90°West
Niño 4 (Central Tropical Pacific)5°North–5°South160°East–150°West
Niño 3.4 (East Central Tropical Pacific)5°North–5°South170°West–120°West
Tropical North Atlantic5°–20°North60°West–30°West
Tropical South Atlantic0°–20°South30°West–10°East
Global Tropics10°South–10°North0°–360°

A monthly linear tendency of each precipitation index was adjusted for the selected time period (1974–2009) through regression analysis of the index on time (years). Regression slopes were estimated and tested for the hypothesis of being equal to zero (no tendency) at the 5% significance level.

A spatial interpolation of the slopes of calculated indices was performed using the inverse distance weighted (IDW) interpolation algorithm, assuming that data from stations that are closer together tend to have more similar characteristics than stations that are farther apart.

3. Results and discussion

According to the rescaled distance, the clustering procedure generated three distinct groups: Group 1 is composed of precipitation indices associated with counts of dry days (NDD and CDD); Group 2 is composed of precipitation indices related to counts of wet days (NWD, CWD, N_P75, N_P90 and N_P95) and Group 3 is associated to precipitation values (Rx1day, Rx5day, PWD and SDII). These clusters are shown in Figure 2. The precipitation indices used to characterize tendencies of precipitation in the studied region were chosen based on the similarity within each group and on the work of Santos and Manzi (2011), the Rx5day, CWD, N_P75 and CDD.

Figure 2.

Dendogram of studied precipitation indices.

The number of rain gauges that had a correlation above 0.5 among the selected precipitation indices and SST anomalies is shown in Figures 3-5 for a lag time of 1–3 months. For lag times of 4–6 months, little correlation was detected and therefore not presented in this work.

Figure 3.

Number of stations with correlation between SST and precipitation indices greater than 0.5 for 1 month lag time.

Figure 4.

Number of stations with correlation between SST and precipitation indices greater than 0.5 for 2 months lag time.

Figure 5.

Number of stations with correlation between SST and precipitation indices greater than 0.5 for 3 months lag time.

SST anomalies verified in October affect indices in January, whereas SST anomalies in November and December affect the indices in February, SST anomalies in January and February affect the indices in March and SST anomalies in March affect the indices in April (Table 3). The SST regions that mostly affect precipitation indices are Niño 3.4, Niño 3, TROP and Niño 1 + 2. The index continuous dry days is the mostly affected by SST anomalies.

Table 3. Months with correlation between SST and precipitation indices greater than 0.5 and respective SST anomalies
Month in analysisMonths of SST anomaliesLocation of SST
JanuaryPrevious October – lag 3 monthsNiño 1+2, Niño 3, Niño 3.4, Global Tropics
FebruaryPrevious November (little) – lag 3 months

Previous December (more) – lag 2 months

Niño 1+2, Niño 3, Niño 3.4

Niño 1+2, Niño 3, Niño 3.4, Global Tropics

MarchPrevious January (little) – lag 2 months

Previous February (more) – lag 1 month

Niño 1+2, Niño 3, Niño 3.4

Niño 1+2, Niño 3, Niño 3.4, Global Tropics

AprilPrevious March – lag 1 monthNiño 1+2, Niño 3, Niño 3.4, Global Tropics

Average monthly precipitation in the state of Ceará shows a seasonal pattern with a dry period beginning in July and ending in December, and a wet period occurring between the months of January and June. March and April are the wettest months (Figure 6) being both influenced by SST anomalies with a 1-month lag time (Table 2) and mostly related to Niño 1+2, Niño 3, Niño 3.4 and TROP (Figures 3-5). During the wet period, the rainfall is mainly distributed in the climatic Ocean, Sierras and Cariri regions as characterized in Figure 1.

Figure 6.

Total monthly precipitation (PWD).

No significant variability was observed for annual precipitation, either positive or negative. Nonetheless, significant variability was observed in some stations for monthly precipitation. Table 4 shows that 60% of the calculated monthly precipitation variability (55 stations × 12 months) was negative, being 90% of the calculated significant monthly precipitation variability also negative.

Table 4. Number of stations with positive and negative variability in monthly precipitation
Significant positive000110010100
Significant negative012121308932
Total positive501417201928134314111322
Total negative54138353627421241444233

Figures 7-10 show long-term variations in precipitation indices – CDD, CWD, NP75, Rx5day. Long-term variations, even if not statistically significant, are identified with a positive sign (+) whenever the variation is positive, and a negative sign (−) for negative variation. Shades of grey are more intense at the extreme values for both tendencies, and white for a slope of zero (no changes over time). The stations that showed significant positive or negative tendencies (5% significance level) are identified with bigger positive signs (+) and bigger negative signs (−).

Figure 7.

Tendencies in monthly number of continuous dry days (CDD).

Figure 8.

Tendencies in monthly number of continuous wet days (CWD).

Figure 9.

Tendencies in monthly number of wet days with precipitation above 75 percentiles (NP75).

Figure 10.

Tendencies in monthly maximum consecutive 5-d precipitation (Rx5day).

3.1. CDD – Continuous Dry Days – monthly maximum length of dry spell

CDD is the index that is mostly affected by SST anomalies (Figures 3-5) with effects at 1, 2 and 3 months lag time (Table 2).

The monthly maximum length of dry spells shows an increasing and significant tendency almost over the entire state of Ceará from September to November (Figure 7), which are also the drier months, implying that dry spells are intensifying in the dryer months. As for the other months, an increasing tendency is also verified, sparsely distributed throughout the region, with some exceptions particularly in August. The drier conditions observed from February to May, considering that crops plantation begins in December for the Cariri region and between January and March for the rest of Ceará, may affect germination of rainfed crops that are very important for economic development of the region. Crops may also be affected by the increase in dry spells in January and February.

The increased length of dry spells may lead to reduced plant growth or even compromise the harvest itself, as they are more affected by duration of dry spells than total amount of rain in the season (Alves et al., 2009, Menezes et al., 2010). Rainfed crops which have their initial stage at the beginning of the wet season may be affected by the increase in CDD of January and February. However, rainfed crops like corn and beans, which are planted in the middle of the wet season (March), may benefit from the decrease in dry spells. Sun et al. (2006) identified a statistical significant correlation between corn yield and Ninõ 3.4 SST anomalies.

3.2. CWD – Continuous Wet Days–(monthly maximum length of wet spell)

CWD is the precipitation index least affected by SST anomalies (Figures 3-5). CWD values tend to increase in January and August throughout all Ceará state. A tendency to decrease is verified in March and April (Figure 8), which may imply a decrease in overall soil moisture during the wet season. This may compromise agricultural production, especially subsistence agriculture (self-sufficiency farming) where farmers are totally dependent on rainfall.

The increase in CWD in June and August expresses a lengthening of the rainy season, which is related to a greater action of the eastern waves over the precipitation regime in the state of Ceará (Ferreira et al., 1990; Torres and Ferreira, 2011). For the other months, an increasing tendency also occurs in some restricted areas over the region. CWD is related to SST in TROP and the Niños (Figures 3-5). ITCZ location is dependent on Tropical Pacific and Atlantic SST (Hastenrath and Greischar, 1993; Uvo et al., 1998; Santos and Brito, 2007) suggesting that CWD is also related to ITCZ location.

3.3. N_P75–Number of Days with Precipitation above the 75 percentile

There is an increasing tendency in January for the number of days with precipitation above the 75th percentiles. From February to May (Figure 9) there is a decreasing tendency starting on the Ocean and east and north-east Sierra region of Ceará and rotating clockwise for Cariri, in April, and again to Ocean and western Sierra, in May. This tendency is related to SST anomalies, which promotes the shift of the ITCZ in the Southern Hemisphere. When SATL and NATL SST have lower than normal values, they lead to an increase in the southeast trade winds and a weakening of the northeast trade winds. This anomaly results in a delay of the ITCZ movement on the Southern Hemisphere with a resulting decrease in precipitation over the state of Ceará (Uvo et al., 1998; Alves et al., 2009).

From June to August, a positive tendency may be observed, expressing an increase in the number of very wet days, although not statistically significant at the 5% level. This lengthening of the rainy season and the increase in N_P75 are related to the eastern waves as discussed by Kayano (2003) and Ferreira and Mello (2005).

3.4. Rx5day–Monthly Maximum Consecutive 5-d precipitation

In general, an increase in the Rx5day tendency was verified for the whole state of Ceará from January to March, except of the sierra region in February, where a significant decrease occurs (Figure 10). Significant decreases are evident from September to December, which are in the dry season. Like N_P75, Rx5day shows a positive tendency in time over the state in the months of January, June and August. The increase in extreme precipitation events at the beginning of the wet season (January) may lead to an increase in soil erosion, in an already susceptible semiarid region, since vegetation is leafless and soil vegetation cover is scarce.

4. Conclusions

In general, a decreasing tendency in monthly precipitation was observed over almost all the state of Ceará, which may affect water storage and rainfed agriculture.

The wetter January shows an increase in CDD, but also an increase in continuous wet days: it rains more for longer continuous periods and the dry spells in between are also longer.

The precipitation indices point to a tendency for the dry months to become dryer with a dry season starting with a tendency towards a slight increase in monthly total precipitation and ending with a decrease in total precipitation. These results indicate that water should be stored at the beginning of the wet season rather than at the end, for irrigation purposes, animal water supply and for human consumption.

Extreme precipitation tends to increase at the beginning of the wet season, making watersheds more prone to flooding and to erosion phenomena in an already susceptible semiarid region.

SST anomalies from October to March seem to affect precipitation indices from January to April with decreasing lag times as it approaches the end of the wet season. The most influential SST anomalies locations are Niño 1+2, Niño 3, Niño 3.4 and TROP.