Critical Rainfall Parameters : Proposed Landslide Warning System for the Metropolitan Region of Recife , PE , Brazil

In the Metropolitan Region of Recife, Northeast Brazil, landslides caused a total of 214 deaths between 1984 and 2012. Efforts have been made, but there is still need to improve the risk area management by discovering the correlations between the rainfall and landslides and by implementing a disaster warning and prevention system in order to reduce the number of accidents and fatalities. The purpose of this study is to propose rainfall parameters likely to trigger mass movements, as a contribution to risk management in this region. It specifically addresses the study undertaken in three municipalities: Recife, Camaragibe and Jaboatão dos Guararapes, which have disorganized occupation in high and very high risk areas. In order to achieve the prime objective, rainfall and landslides were tracked during 2009 to obtain correlations between them. After the data was logged, the rainfall accumulation associated with landslides was checked. According to each area with civil defense action of the municipalities involved in the study, critical values of the cumulative rainfall in 72 h (I72 h), in the long term (Pac) and the parameter Rcrit resulting from the product of I72 h by Pac, are suggested herein to be recommended for operations of the Warning and Alert states. By achieving these critical values, they increase the probability of landslide events, and are important for taking decisions and instructions on Civil Defense


Introduction
Natural disasters are increasingly frequent in large cities both at home and abroad.Recently, a number of catastrophes worldwide were due to landslides and flooding, these processes being associated with severe atmospheric instabilities.In recent decades research has evidenced that there has been a sharp increase not only in the frequency of natural disasters but also in intensity, due to the occurrence of extreme rainfall, causing serious damage and socioeconomic losses.
The world's worst landslide-related disaster ever was in the Soviet Union in 1949, when around 12,000 people died.In the last ten years, other serious disasters occurred, for example in China in August 2010, when more than 1,700 people lost their lives; in the Philippines in February 2006, with around 1,126 fatal victims (EM-DAT, 2013); and in third place, the landslides in the mountain region in the state of Rio de Janeiro, Brazil, in January 2011, when around 1000 people died.
The main factor for the disaster in the mountains in Rio de Janeiro State was the extreme rainfall events.Accumulated rainfall of 280.8 mm in 48 h, the highest recorded in the last 45 years in the region, caused several landslides and flooding.This event was classified as one of the ten worst disasters caused by landslides in the world in the last 10 years according to the UN, and the worst ever in the history of Brazil.This catastrophe exceeded the number of victims recorded in 1967 in the municipality of Caragua-tatuba, São Paulo State, when 436 people lost their lives in landslides and around 3,000 were made homeless.
EM-DAT (2012) has calculated in Brazil, for the period 1900-2012, the occurrence of 202 major natural disasters, with 12,235 fatal victims and damage worth approximately 14.6 billion dollars.These disasters include flooding, landslides, drought, epidemics and so on.According to the database of the São Paulo Technological Research Institute (IPT-SP) for the period 1988-2012, a total of 3,288 people were victims of landslides in the country, 1,240 in 2010 and 2011 alone, including records from the mountainous region outside Rio de Janeiro.
Although the Recife Metropolitan Region in Northeast Brazil has not recorded a large number of fatalities from a single rainfall event, the region has a history of disasters over the years.A total of 214 deaths from landslides in this region were calculated between 1984 and 2012.In 2011, nine fatal victims were registered after rainfall of 120.3 mm in 24 h in July (within the region's wet season).In the region, that month had a total rainfall of 535 mm, corresponding to 39% above historic average for the month, which is 386 mm according to the National Meteorology Institute.A total of 2,749 mm of cumulative rainfall had fallen between January and July, exceeding the historic average of 2,243 mm.In those cases, good risk management by implementing a disaster prevention and alert system would reduce the consequences and number of fatalities.
The aforementioned data show that the management activities in risk areas in Brazil need to be upgraded (structural and non-structural actions).Many people are living in a risk situation on hillsides and it is impossible to eliminate this risk in the short term.In this case, it is also necessary to take non-structural preventive actions by means of a good alert system and to restrict further occupation.An effective risk area management program is the main tool that the municipal administration should have at hand.The concept of a preventive plan is to take steps prior to the outbreak of accidents, based on preventing conditions potentially susceptible to their occurrence.In the case, for example, of an alert system accompanied by some parameters, such as: rainfall, weather forecast and knowledge of the critical rainfall accumulation, together with onsite follow-up, it would increase identifying possible landslide events.Fell et al. (2008) provide a guide to prepare risk maps, and comment that landslide frequency can be related to rainfall, in which the correlations indicate areas of possible landslide processes.
An early warning system is defined by UNISDR (2009) as "the set of capacities needed to generate and disseminate timely and meaningful warning information to enable individuals, communities and organizations threatened by a hazard to prepare and to act appropriately and in sufficient time to reduce the possibility of harm or loss".A people-centered early warning system necessarily comprises four key elements: (i) knowledge of the risks; (ii) monitoring, analysis and forecasting of the hazards; (iii) communication or dissemination of alerts and warnings; (iv) local capabilities to respond to the warnings received.(UNISDR, 2009).Intrieri et al. (2013) define and present a practical guideline in the design of a landslide Early Warning System.The authors describe the EWS as the balanced combination of four main activities: design (geological knowledge, risk scenarios, design criteria, and choice of geo-indicators), monitoring (instruments, installation, data collection, data transmission and data elaboration), forecasting (data interpretation, comparison with thresholds, foresting methods, and warning) and education (risk perception, safe behaviors, response to warning and population involvement).A toolbox is presented that can help end-users (such as civil protection agencies and administrations) to create an Early Warning System for every landslide requirement.Calvello (2014) presents "components" of a regional rainfall-induced landslide Early Warning System: a correlation model between rainfall events and landslide events; warning levels and alert phases; decision-making to activate the alerts; monitoring and warning strategy; emergency plan and communication strategy; all steps are connected to the risk perception.An efficient model (regional warning) and an efficient system are necessary, with three players: people, scientist and managers.Table 1 shows world examples of an alert system with some information, including rainfall thresholds.
The knowledge that cumulative rainfall could cause large numbers of slope instability processes associated with the weather forecast is key information for civil defense actions, such as, for example, triggering the state of alert.Although alerts would not prevent the occurrence of the processes, they may reduce the number of fatal victims caused by disasters, since the population living in risk areas can be informed beforehand about the need or not to leave their homes before landslides occur.
In this sense the purpose of this study is to submit a proposal of technical parameters of cumulative rainfall, which indicate the probability of landslides occurring in the areas of the Metropolitan Region of Recife, thereby providing support for the civil defense teams to raise Warning and Alert states.The subject of this article is part of the research line "Analysis and Risk Management for Erosion and Landslides", included in the CNPq PRONEX Project, which was developed by GEGEP (Geotechnical Engineering Group on Hillsides, Plains and Disasters) of the Civil Engineering Department of the Federal University of Pernambuco (UFPE), coordinated by Prof. Roberto Quental Coutinho. Bandeira (2010) did his PhD research on this project and some of his results are presented herein below.

Risk Area Management
Risk area management is a decision-making process involving the definition of requirements, recognizing acceptable options and choosing appropriate strategies to reduce risks.In the 1990s, the Office of the United Nations Disaster Relief Coordinator (UNDRO) contributed significantly to the risk area management process at an international level and suggested four stages of management: a) Risk identification and analysis; b) Structural preventive measures; c) Non-structural measures; d) Public information and capacity building for self-defense.Several countries have adopted these stages, namely: Hong Kong (China), Australia, USA and Brazil (Coutinho & Bandeira, 2012a).Internationally, a framework has also been set up consisting of three steps presented by Fell & Hartford (1997) and Fell et al. (2005), which are: i) Risk Analysis; ii) Risk Assessment; iii) Risk Management.
In Brazil, a risk area management system takes into account international suggestions, involving joint actions at three government levels (federal, state and municipal); it also receives support from higher education and research institutions.As an example of work done by those institutions, it is worth mentioning Coutinho (2013), who presents a technical report from a national technical commission study of processes involving mass movements, guidelines for intervention, urban planning and parameters for geotechnical mapping for the areas susceptible to these natural disasters.
For accident prevention municipal governments take structural actions (slope retention work) and non-structural actions (follow-up system, community talks and creating volunteer civil defense centers), the latter being directly adopted in the communities, called Proximity Management in Recife.This grassroots educational and interactive work is essential, since potential landslides are directly influenced by unsuitable anthropic actions, such as: badly executed cuts and fills, precarious drainage systems, wastewater discharge and removal of the slope surface protection (Coutinho & Bandeira, 2012a).The population's wastewater discharge and inflow throughout the year increase the moisture in the ground, weakening it, even on dry days, thereby increasing the probability of landslides during short showers (Coutinho & Severo, 2009).The population is often given instructions on this kind of problem but stricter structural actions are necessary to reduce these problems.
At a state level, specific programs have managed the work of the municipal governments, seeking the best way to counter the practices of emergency and specific actions, offering an interdisciplinary and differentiated methodology, and giving communities better living and housing conditions.
At a federal level, the government has created risk area management programs, providing financial conditions to prepare risk maps and downsizing plans, as well as offering capacity-building courses on risk area management and preparing engineering designs (based on Coutinho & Bandeira, 2012b).An important improvement in the disaster legislation in Brazil was the Federal Law -12,608 of April 10, 2012.
Although the actions are being taken at the three government levels, every year rainfall causes dozens of landslides, some fatal, evidencing the need to have more in-depth knowledge of the correlations between rainfall and slope instability processes, as well as to improve forecasts of rainfall indices in the risk areas.
In relation to civil defense actions in the Recife Metropolitan Region (MR) and the other Brazilian cities, the main cities adopt three levels of operation, as follows: Observation, Warning and Alert.The Recife-MR civil defense teams take actions based on the follow-up of weather forecasts and site inspections.Hence, this study proposes values of rainfall parameters for civil defense actions, contributing to the local government alert system.
By analyzing the frequencies of mass movements and flooding in a number of areas in Brazil, some researchers have presented critical rainfall accumulations based on which they increase the number of occurrences of these processes.Tavares et al. (2004), on analyzing the variability of time and space of the rainfall associated with the movements along the coastline of the State of São Paulo, concluded that there is a predominance of mass movements when there is 72-h cumulative rainfall of over 120 mm.These authors state that landslides with rainfall accumulation below this figure may be correlated to anthropic actions or heavy rain in a 24-h period.Gonçalves (2003) found that 60 mm or more of heavy rain falls in 24 h in Salvador, Bahia State, with major spatial repercussion; with this intensity, flooding in the municipality as well as more serious slope failures occur from an intensity of 70 mm/24 h.Macedo et al. (2004), studying the correlations between the landslides and 72-h cumulative rainfall for some regions in the State of São Paulo, confirmed that when the accumulated rainfall reaches 80 mm in 72 h, it is likely that mass movements would occur in the following places: Campos do Jordão, Campinas, ABC and Sorocaba.Now, in places in the Baixada Santista and Paraíba Valley the authors found rainfall accumulation of 100 mm in 72 h; and 120 mm for the northern coast of São Paulo.
Gusmão Filho et al. (1987), when examining the frequencies of mass movements on the hills of Olinda, Northeast Brazil, found that the instability of the slopes in this location, as a result of the rise in water level in the area, is the result of the joint action between the intensity of accumulated rainfall (P ac ), from January to the date under consideration, with occurrence of a daily rainfall of minimum intensity (I).So the authors defined a parameter R crít , to the value of 60,000 mm 2 , as the product of 24-hr rainfall (I) by the rainfall accumulation (P ac ) to the date of the event (R crít = P ac x I), and countless situations could trigger landslides (Gusmão Filho, 2001).For example, when the accumulated rain is 800 mm, probably the ground is heavily saturated and a daily 75 mm rainfall would loosen the mass.This study shows how important the weather forecast is to identify the probable value of I and, consequently, the value of R crít , which would help decide on accident prevention measures.
This kind of study is a valuable contribution to improving risk area management.The technical parameters suggested for the Warning and Alert states, associated with good weather forecasts and site inspections to identify the features of instability, can provide effective preventive measures and risk mitigation, also reducing loss of human lives, by adopting an alert system.The purpose of this system is to expedite the departure of people living in risk areas and prevent them from becoming victims of landslides, but this system is only effective if the technical parameters of critical cumulative rainfall are known.In addition to those parameters, it is necessary to have a good monitoring system of the climate and weather conditions, by means of, for example, weather radars, which can warn about the arrival of heavy rain.Radar can warn hours beforehand about the occurrence of heavy rain, and over a range of hundreds of miles.It is also essential to have an automatic permanent rain gauge network installed in risk areas (high and very high risk) with ongoing analysis of the results.In short, to ensure fewer fatalities caused by landslides it is necessary that the civil defense system draw up its preventive plan based on identifying the risk areas and on the critical rainfall parameters, It should also implement technologies and support of partner agencies in their taking preventive and emergency measures of risk management.The people in the communities need to be involved, trained and to feel co-responsible.
Some examples of alert systems could be the issue of property alerts and sirens, sending SMS messages or similar devices.In the city of Rio de Janeiro, for example, the Rio Alert System issues warnings based on weather reports, namely, the intensity of expected rainfall.When heavy rain is forecast, which may cause isolated landslides, property alerts to the population are issued (via the press and Alert Rio website: www.sistemaalerta-rio.com.br).The city of Rio de Janeiro also has Community Heavy Rainfall Warning, where the local inhabitants receive SMS alert warnings of rainfall.These warnings occur when heavy rain is forecast for the next few hours that may cause isolated landslides.The sirens installed in the most critical areas also issue warnings, and are sounded when very heavy rain is expected in the next few hours and might cause generalized landslides, in which case it would be a maximum alert situation.

Study Area Characteristics
The study area in this article covers the municipalities of Recife, Camaragibe and Jaboatão dos Guararapes, in the State of Pernambuco, Northeast Brazil.Together they occupy an area of around 530 km 2 , and are between projections 265000 to 300000 East and 9085000 to 9125000 North, of the geographic coordinate system WGS-1984 zone 25S, according to the Universal Transverse Mercator (UTM) projection (Fig. 1).
In terms of climate, the study areas within the climate range type As', as wet tropical climate according to the W. Köppen classification, with a dry summer and a wet season, which begins in the fall.Figure 2 shows a historic series of the average rainfall logged in the city of Recife between 1910 and 1985, considered as a benchmark for the entire Metropolitan Region.Figure 2 shows that between March and August there is a concentrated wet season, with monthly averages of over 150 mm.This period is considered critical for the civil defense units in the Recife Metropolitan Region, where several landslides have been recorded.The monthly maximum rainfall is logged in May, June and July, with monthly averages of more than 300 mm rainfall.On average, the September-February period has a low monthly rainfall (less than 150 mm).
Specifically in the municipality of Recife, the annual average total rainfall is 2,243 mm.In recent years the maximum monthly rainfall recorded in June 2005 in the municipalities of the Metropolitan Region was 681.3 mm in Recife; 728.8 mm in Camaragibe, and 609.9 mm in Jaboatão dos Guararapes.In 2011, the rainfall volumes logged for some months in Recife were also high, for example in April (635.7 mm), May (685.1 mm) and July (550.6 mm).The annual total rainfall recorded in the municipality of Recife in 2011 was 3,097.9mm, higher than the annual average according to the Pernambuco State Water and Climate Agency.
The local climate is closely related to the soil formation of the soils and the latter with slope failure.The high humidity rates and temperature of Recife-MR affect the chemical weathering processes of the soils, breaking down weaker minerals, such as feldspars and micas.These minerals are quite common in the granitic rocks of the crystalline substratum and in the sedimentary overburden of the Barreiras Formation, present in large areas of Recife-MR (Alheiros et al., 2003).According to Bandeira (2003), the action of chemical weathering in the region strongly impacts processes of mass movements, since the feldspar particles, initially in size of sand particles in the Barreiras Formation sediments, contribute to increasing the clay content, leading to slope failure.
The Barreiras Formation is commonly found in the hillside areas of Recife-MR.This deposit associated with Cenozoic events of a climate and/or tectonic nature, towards the end of the Tertiary period (Pliocene) widely covered the exposed surfaces of the substratum, filling a fairly disturbed relief.In general, this formation consists of claysand sediments, cream to reddish in color, depending on the intensity of the oxidation of the iron, and occurs through a vast sedimentary overburden (Alheiros et al., 1988).This geological unit extends along the Brazilian coast from Rio de Janeiro (Southeast) to the State of Amapá (North), covering Mesozoic sedimentary deposits in several coastal basins (Bezerra et al., 2006).In the study area herein, the Barreiras Formation sediments are located in two areas (Fig. 3): the first is farther North, covering the table l and to the North of Camaragibe and Recife; and the second, farther South, consists of table l and remnants and hills between the northern and southern municipal boundaries of Jaboatão dos Guararapes and Recife, respectively.The locations of the Barreiras Formation sediments coincide with the hillside areas, where landslides are frequent.
In some hillside areas of the municipality of Camaragibe and Jaboatão dos Guararapes residual soil of the crystalline substratum occurs, more widespread on the hillsides of the municipality of Jaboatão dos Guararapes (Bandeira, 2010).Here the residual soil of the gneiss-migmatite complex is predominant, with a clayey-sand texture, and consequently susceptible to a larger number of landslides in this type of soil.
The relief of the study area herein is characterized by a sharp division between the plain and the hills.On the  plain several places are below sea level, and flooding often occurs.In the hilly areas altitudes of up to 200 m above sea level can be encountered.Hilltop occupation is normally planned, but a low-income population in precarious conditions occupies the hillsides.Between 1996 and 2002 there was a sharp increase of 251,600 inhabitants in the hillside areas, to cause higher demographic density, covering a wider area in the municipalities.In 2004 the CONDEPE/FIDEM State Agency found that there were 345,714 homes built on hillsides and in areas prone to flooding in Recife-MR, representing 38% of all households existing in the region (FIDEM, 2006).

Methodological Aspects
Seven main work stages described below were set up to identify and develop a proposal for cumulative rainfall parameters, likely to cause landslides in the study area of this paper.a) Expansion of the rain gauge network: the municipalities considered the rainfall to be similar in the areas; this is the reason for few rain gauges throughout the study area (530 km 2 ) -only five instruments.Under the GEGEP -CNPq/PRONEX and CNPq-Universal Project, another 17 instruments were installed, 11 of which were manual gauges (Ville de Paris) and six rainfall data loggers, in order to give a good correlation between the rainfall and landslides (see Figs. 4 and 5).The data loggers provide information about the intensity and duration of the rain, helping to understand the landslide events caused by rainfall concentrated in less than 24 h.b) Defining the rain gauge collection points: the location of the risk zones (high or very high risk), the geological characteristics of interest and the areas of civil defense action determined the sites where the new rain gauges were to be installed.These risk zones, were obtained from preexisting maps subsidized by federal government funds (Fig. 5). Figure 5 -Location of rain gauges in the 'High' and 'Very High Risk' areas, and the local rainfall indices during 24 h registered on April 13 th , 2009 (Coutinho et al., 2010).
c) With regard to geological aspects, the rain gauges were installed on sites where the soil had sedimentary characteristics of the Barreiras Formation and residual characteristics from the crystalline substratum, and where historically a larger number of landslides had occurred in the study area.In their working areas the civil defense teams operate on a decentralized basis (Coutinho & Bandeira, 2012b).In the municipality of Recife, for example, the territory is divided into Regionals: North, Northeast, West, Northwest and South.In Camaragibe the division is by Area: Area I, II, III, IV-Tabatinga and IV-Vera Cruz.In the municipality of Jaboatão dos Guararapes the Regionals are: Cavaleiro, Jaboatão Centro, Curado, Muribeca, Prazeres and Praias.So, for each working area of the civil defense teams, rain gauges were installed to accompany the rainfall and record the landslides.d) Logging rainfall and landslides: throughout 2009 rainfall in the study was monitored on a daily basis using the rain gauges.Landslide data were obtained first in the civil defense units of the municipalities and from news coverage.These data were logged on electronic spreadsheet.Landslide information was logged as follows: place and date of occurrence; type of movement and geological-geotechnical characteristics.
e) Identification of accumulated rainfall associated with landslides: For information about landslide-related cumulative rainfall, the rainfall accumulation was recorded in 72 h or less before the date of the landslide and in the long term (from early in the year), collected from the data logged by the rain gauge closest to the event.From these data, histograms were drawn, showing the number of landslides as a result of certain intervals of cumulative rainfall for each working area of the civil defense teams.
f) Proposed rainfall parameters for civil defense actions: at this stage rainfall accumulations were proposed for the Warning and Alert states.The critical rainfall accumu- lated in 72 h (I 72 h ), in the long term (P ac ) and R crit values were proposed in each study area, the last obtained using the products of I 72 h by P ac .These values were based on data from the graphs drawn during the previous stage.
g) Validation of the proposed alert systematics: through ROC analysis (Fawcett, 2006) the performance of the recommended rainfall parameters was assessed for civil defense actions.This stage was developed in all study areas in the municipality of Camaragibe.

Results Obtained and Discussions
Extending the rain gauge network in this study was necessary to obtain the rainfall accumulations that had caused the landslides in the areas.Positioning the instruments close to the most critical areas of risk helped to achieve a better correlation between the rainfall and processes of slope failure.
An example of this key step in the work is the rainfall logged on April 13, 2009, in the municipalities involved in the study.In Camaragibe, for example, rainfall was recorded as follows: 99.7 mm in Vera Cruz (using the instrument installed in this study) and 54.2 mm in Timbi (using the instrument previously existing in the municipal station) for the same 24 h.It should be mentioned that the latter instrument is used by the civil defense as a benchmark for its work; in other words, in Vera Cruz the civil defense would only act if there was an urgent call from the population, since the rainfall logged in the rain gauge that the civil defense team uses as a base (in Timbi) was around 50% of that logged in Vera Cruz.In the municipality of Recife, 123.6 mm of rain fell in 24 h in the South Regional, while in the North Regional the rainfall was 67.25 mm for the same period.In Jaboatão dos Guararapes 160 mm of rainfall was logged in the district of Socorro, while in Prazeres district the rain gauge used by the civil defense as a benchmark for its actions recorded rainfall of 80.75 mm during the same 24 h.This information shows how important it is to extend the rain gauge network for this study and for civil defense actions.Figure 5 illustrates the location of the rain gauges in the areas of high and very high risk and the rainfall data logged on April 13, 2009.
Throughout the base year of this study ( 2009), 1,367 cases of slope failure and 10 fatalities were recorded.Of this total, 827 occurred in the municipality of Recife, registering six deaths; 160 landslides in Camaragibe, with one death; and 380 landslides in Jaboatão dos Guararapes, with three fatalities.The majority of the landslides recorded in Recife occurred in the North Regional (32%) and South Regional (28%), which include areas of high landslide risks.In Camaragibe, 45% of landslides in 2009 occurred in Area II, concentrated in Tabatinga and Bairro dos Estados, which are considered to be the most problematic in the municipality.In Jaboatão dos Guararapes most of the landslides occurred in the Regionals of Jaboatão Centro and Cavaleiro (77% logs).In general the landslides in the mu-nicipalities involved in this study were closely related to high rainfall indices in February, April, June and July (Fig. 6).
After registering the 1,367 landslides, the accumulated rainfall was investigated to find which had triggered the processes.The following items describe the results found in each place.

Rainfall and landslides in the municipality of Recife
For each civil defense working area in the municipality of Recife (Regionals: South, Northeast, Northwest, West and North) the number of landslide events associated with cumulative rainfall in 72 h (Fig. 7) and in the long term, from January 1 st (Fig. 8) were checked.Of the 827 landslide events in this municipality, 619 (75%) were directly related to rainfall and 208 (25%) landslides were not caused by rain as the main factor, with evidence of low rainfall indices.These landslides may have been aggravated by anthropic action and were not considered in the correlation analysis between the accumulated rainfall and slope failure processes.
Figure 7a shows that the rainfall caused 190 landslides in the South Regional.Of this total, 131 (69%) occurred when the 72-h cumulative rainfall was over 60 mm, with a higher concentration of rainfall accumulation of 120-150 mm.The 30 landslides (16%) occurring due to cumulative rainfall of 40-60 mm were caused by heavy rainfall in 24 h (> 40 mm).On the other hand, landslides occurring with accumulated rainfall of less than 40 mm/72 h (29 landslides) were related to the already saturated ground conditions, since slope failures were recorded after April, with accumulated indices higher than 600 mm from January 1 st .Seventy-five percent (142) of all landslides in this Regional occurred in Lagoa Encantada.This Regional had the highest number of landslide processes in 2009.
In the Northeast Regional (Fig. 7b) 92 landslides occurred, 66 of them (72%) when the 72 h cumulative rainfall was more than 40 mm.In this Regional, the landslide was more evenly distributed in relation to the cumulative rainfall.When the accumulated rainfall was under 40 mm 26 (28%) landslides were registered; most of which occurred in May onward, when the ground was heavily saturated and the rain was concentrated during 24 h.In the Northwest Regional 101 landslides were recorded in 2009.Sixty-eight cases (67%) of this total occurred when the 72 hour accumulated rainfall was above 40 mm (Fig. 7c).In this area most cases were practically evenly distributed, with cumulative rainfall of 80-100 mm or less.When the accumulated rainfall was less than 20 mm/72 h 14 landslides were recorded, but all occurring after April, when the ground was heavily saturated.Around 84% (16 cases) of the 19 processes of slope failure with accumulated rain of 20-40 mm in 72 h also occurred after the high rate of ground saturation.The West Regional suffered 75 landslides, 50 of which (67%) with cumulative rainfall of 80 mm or more in 72 h (Fig. 7d), with a slightly higher concentration for the 120-150 mm rainfall.The majority of landslide events, with accumulated rainfall below 80 mm in 72 h, occurred after the high degree of ground saturation.
In the North Regional with 161 recorded landslides ranked second with the highest number of processes than anywhere else in the study area.This area had a relatively even distribution of landslides between the ranges of cumulative rainfall.Of all the landslides, 118 (63%) were recorded when the 72-h rainfall accumulation was above 40 mm (Fig. 7e).Fourteen (45%) of the 31 landslides as a result of accumulated rainfall of 20-40 mm in 72 h occurred in May (after soil saturation); ten (32%) occurred in January and February, most of them related to the concentrated rainfall during 24 h; and seven (23%) in April, with the majority of landslides related to the average rainfall of 35 mm in 72 h .Eight (67%) of the 12 landslides with rainfall under 20 mm/72 h happened after the high saturation rate of the ground; and four (33%) in February and March showed signs of being more closely related to anthropic actions.
By analyzing the long days of accumulated rainfall in the municipality of Recife (Fig. 8), it is found that after 750 mm of cumulative rainfall several landslides occurred in all the Regionals, representing 51%-81% of the total, with major events in the South, Northeast and West Regionals.In the South, Northeast, Northwest and West Regionals there is a clear upward trend in the number of landslides with cumulative rainfall values.The North Regional had a significant number of landslides with cumulative rainfall below 450 mm.The registered landslides with accumulated rainfall below 450 mm were closely related to the rainfall concentrated in 72 h, as occurred in the Northeast (11 cases -Fig.8b), Northwest (11 cases -Fig.8c) and North (34 cases -Fig.8e) Regionals.It should be stressed that in the Northwest and North Regionals, the landslides with cumulative rainfall of 450-600 mm were also related to the rainfall concentrated in 72 h, with logs above 70 mm.
These landslide records show how important it is to know the product between long days of accumulated rainfall and the 72 h rainfall.Considering these data, the critical accumulation in 72 h could be considered 40 mm for the Northeast, Northwest and North Regionals; 60 mm for the South Regional; 80 mm for the West Regional; and the long-term critical accumulation (P ac ) for all the Recife Regionals could be considered 750 mm.
In relation to the geological-geotechnical aspects, the sediments in the Barreiras Formation can be said to be characteristic of Recife hillsides studied, only differing in the predominant sizes of particles in each Regional.In the South Regional, for example, sandy sediments prevail; however, the soils in the Northeast and Northwest Regionals are sandy-clay.The West Regional presents pockets of soils interspersed with sands and clays, which could provide a preferential landslide route on the hillsides; and the North Regional has more clayey soils in this formation.

Rainfall and landslides in the municipality of Camaragibe
One hundred and forty-six (91%) of the 160 landslides that occurred in the municipality of Camaragibe were directly related to rainfall, and since the other 14 (9%) landslides were not triggered by rainfall, they were excluded from the critical rainfall analyses.
Figure 9 shows the distributions of landslide events due to cumulative rainfall in 72 h for the municipality of Camaragibe.This figure shows that in Areas I, III and IV (Vera Cruz), most of the processes had cumulative rainfall of more than 80 mm in 72 h.However, the four landslide events in Area I, recorded with accumulated rainfall of 80-100 mm/72 h, occurred after the ground was heavily saturated, as in the two logs below 80 mm/72 h.In Areas II and IV (Tabatinga), the majority of processes occurred with accumulated rainfall above 60 mm/72 h, recording in this condition 55 and 18 landslides, respectively.Two of the nine landslides registered in Area II, with cumulative rainfall below 60 mm/72 h, happened in March, due to heavy rainfall of 42-58 mm in 48 h; and six occurred after April when the ground was already saturated, with 30 mm of rainfall concentrated in 24 h.
The landslides associated with the long days of accumulated rainfall in the municipality of Camaragibe are illustrated in Fig. 10. Figure 10a, corresponding to Area I, shows that 81% of the landslides (25 cases) were recorded when the rainfall accumulation exceeded 750 mm.In Area II, when the accumulated rainfall exceeded 600 mm, 41 landslide events (62%) were logged in separate places in this area (Fig. 10b).The 22 landslides with accumulated rainfall below 450 mm were related to the previous days' heavy rainfall (60 mm/72 h).Area III also showed a number of landslides with accumulated rainfall of more than 600 mm (Fig. 10c), the main driving factor of 75% of the landslides (15 events), that is, the saturated state of the ground.In Area IV-Tab, when the long-term accumulated rainfall reached 750 mm, 14 landslides (74%) were registered of the total 19 events in the area (Fig. 10d).In this Area, the three slope failures, with a recorded rainfall of 600-750 mm, were due to the heavy rainfall the previous days (136 mm/72 h); and the two registered landslides with cumulative rainfall of less than 450 mm were due to the rainfall of 53 mm/24 h, on February 12, and of 98.5 mm/24 h, on February 27.
Area IV (Vera Cruz) had fewer landslides (10 cases), with eight (80%) registered with cumulative rainfall of less than 450 mm (Fig. 10e), in contrast to the other areas.These eight landslides occurred due to heavy rainfall every day during February; four of the landslides occurred with rainfall of 98.5 mm/24 h and four due to the concentrated rainfall of 88.5 mm/72 h, with no relation to the long days of rainfall.The other landslides in this Area (2 cases) that were correlated to the long-term cumulative rainfall occurred with rainfall of 600 mm or more.In short, in this area 80% of the cases were linked to rainfall concentrated in 24 h.This behavior may be due to the small size of the existing risk areas in that location.In general, except for Area IV (Vera Cruz), there was an increase in the number of landslides in Camaragibe with increased cumulative rainfall.
These data show that the 72 h-critical accumulation (I 72 h ) for Camaragibe could be considered as follows: 100 mm for Area I; 60 mm for Areas II and IV (Tabatinga), and 80 mm for Areas III and IV (Vera Cruz).On the other hand, the long-term critical accumulation (P ac ) can be considered to be 750 mm for Areas I and IV (Tabatinga) and 600 mm for Areas II, III and IV (Vera Cruz).
Concerning the geological-geotechnical characteristics, Area I consists of alternating sand and clay layers from the Barreiras Fm.Area II consists of sandy soils from this formation and residual silty soils of granite.The predominant characteristic of Area III, on the other hand, is clayeysand residual soil of granite (Bandeira et al., 2011).
Tabatinga and Vera Cruz in Area IV reveal clayey-sand characteristics with pebble, of the Barreiras Formation.

Rainfall and landslides in the municipality of Jaboatão dos Guararapes
Figure 11 illustrates the distribution of landslide events due to 72 h-accumulated rainfall in each Regional of the municipality of Jaboatão dos Guararapes (Jaboatão Centro, Cavaleiro, Curado, Muribeca, Prazeres).In this municipality 380 landslides occurred, 324 (85%) of which were directly correlated with the rainfall.There are signs that the main factor for slope failure in the other landslides was anthropic action.Figure 11a shows that 80% of the landslide events (114) in Regional I (Jaboatão Centro) were due to accumulated rainfall of more than 80 mm/72 h, with a greater concentration of landslides with rainfall accumulation of more than 120 mm.The occurrence of 19 of the 22 landslides after rainfall accumulation of less than 80 mm/72 h was due to the high long-term rainfall accumulation (more than 600 mm).Most of the landslides (106) in Regional II (Cavaleiro), which were rainfall-related (130 cases), occurred with cumulative rainfall of 80 mm/72 h or more (Fig. 11b); and also with a larger concentration of landslides with accumulated rainfall of more than 120 mm.Ninety percent (90%) of the 24 cases that occurred with accumulation of less than 80 mm were correlated with the high degree of ground saturation, since they occurred after April when the long-day rainfall accumulation was already high.
In Regional III (Curado) there were 35 processes of slope failure concentrated in the District of Curado IV. Figure 11c shows that most of the landslides occurred after a rainfall accumulation of 100 mm/72 h (32 cases), with   higher concentration of landslides with cumulative rainfall of more than 150 mm.Around 50% of the landslide events (18 cases) in this Regional were in February, due to the heavy rainfall (115-137 mm/24 h) each day from the 22 nd to 25 th of that month.Regional IV (Muribeca) had six landslide events, four recorded with rainfall of more than 100 mm/72 h (Fig. 11d).All the landslides occurred after the long days of accumulated rainfall exceeding 600 mm.In this Regional there was a relatively even landslide distribution in the bands of accumulated rainfall in 72 h.It is worth mentioning that between January and April 2009 there was at least one record of rainfall above 100 mm/72 h, but there were no landslides due to these concentrated rains.All recorded landslides in this Regional occurred after the ground was saturated; in other words, after the beginning of April when the long-term accumulations exceeded 600 mm.Practical experiments in the Recife Metropolitan Region show that rainfall of 100 mm/72 h or more cause various landslides in the Region.The fact that no landslides occurred in the first four months of the year in this Regional can be explained by the small hillside areas to be found there, associated with low demographic density and, consequently, little influence of anthropic factors and interference in the original conditions of the relief.
In Regional V (Prazeres) 39 rainfall-related landslide occurred (Fig. 11e).When the rainfall accumulated above 60 mm/72 h, 31 landslides were recorded, corresponding to around 80% of all landslides in Regional V; generally, however, the number of landslides was relatively uniform in the distribution bands, except for the 120-150 mm band, which had the highest concentration.The two slope failures with rainfall of less than 20 mm/72 h were logged in February and July, the latter when the ground was already heavily saturated.Five landslides were recorded when the 72 h accumulation was 20-40 mm, after a large accumulation of long days of rain.When the accumulation was 40-60 mm/72 h, only one landslide in June was logged, after the long days of accumulation had exceeded the 600 mm.It is found that in Regionals II and V when more than 150 mm was accumulated there were fewer landslides compared to the processes associated with the 120-150 mm accumulation.This fact can be explained by the period of occurrence of landslides with higher accumulations between February and April, while the landslide events with accumulations of 120-150 mm occurred in June to August, when the ground had a higher saturation level.In Regional IV the same case was observed for a 120-150 mm accumulation compared to the landslides occurring with cumulative rainfall of 100-120 mm/72 h.
The correlations between the landslides and long days of cumulative rainfall are shown in Fig. 12.In Regional I, 90% of the cases had cumulative rainfall over 600 mm, corresponding to 103 of a total 114 landslides (Fig. 12a).Regional II, had around 70% (90 cases) of a total of 130 landslides, with accumulated rainfall above 750 mm (Fig. 12b).The slope instabilities had rainfall accumulations of less than 450 mm (23 cases) due to the concentrated rainfall in 72 h or less in February (14 landslides with rainfall of 113 mm/24 h; two with rainfall of 94 mm/48 h; and seven with cumulative rainfall of 164 mm/72 h).In Regional III, the same number of landslides (17) occurred with cumulative rainfall under 450 mm and over 750 mm (Fig. 12c).However, similar to the events in Regional II, slope failures with rainfall accumulation of less than 450 mm occurred due to the high rainfall during 72 h or less in February (15 landslides with rainfall of 113 mm/24 h; and two with accumulated rainfall of 164 mm/72 h), and were not directly related to the long days of accumulated rain.Figure 12d, referring to Regional IV shows that 100% of the landslides (6 cases) occurred when the cumulative rainfall was already over 600 mm.Regional V had a total of 39 landslides events (Fig. 12e), mostly (62%) with accumulated rainfall above 750 mm.The majority of logs of landslides with accumulated rainfall below 450 mm were due to the concentrated volumes of rain in 24 h, six occurring in January with rainfall of 118 mm/24 h; and three in February with concentrated rainfall of 91.3 mm/24 h.
With this information, estimated values of critical accumulations of rainfall in 72 h (I 72 h ) for the Regionals of Jaboatão dos Guararapes, are as follows: 80 mm for Regionals I and II -80 mm; Regionals III and IV -100 mm; and Regional V -60 mm.The values relating to the critical accumulations of long-term rainfall (P ac ) are: Regionals I and IV -600 mm, and Regionals II, III and V -750 mm.
In relation to the geological formation, the hillsides of Regionals I, II and III have a predominantly clayey soil residual of mylonite.However, the Regionals IV and V hillsides are predominantly of sandy sedimentary soils of the Barreiras Formation.

Proposed Rainfall Parameters for Civil Defense Actions
In this section technical rainfall parameters are proposed, which increase the probability of triggering landslides in the study areas, creating support for the municipal civil defense teams to properly establish the levels of operation in four states: Observation, Warning, Alert and Maximum Alert.Tables 2, 3 and 4 summarize the data (rainfall parameters -I 72 h , P ac and R crit ) for the municipalities of Recife, Camaragibe and Jaboatão dos Guararapes, respectively.
It is recommended in all study areas to begin the state of Observation on December 1 st , since the Region's monthly historic average is over 50 mm, which is a key value to start this operation.For the states of Warning, Alert and Maximum Alert parameters were proposed that are characteristic of each area in the municipalities.The references values were in general different for each area and between municipalities confirming the importance to have specific rainfall information.
It is worth mentioning that in any study area when the parameters P ac and R crit are reached during the rainy season (February to September) they will be valid for the Warning level; on the other hand, I 72 h is valid for any period of the year.It should be said that the Warning level could mean the occurrence of isolated landslides in the areas.
When establishing the Alert level in all study areas, it should be announced when I 72 h and R crit are expected to be reached simultaneously and continuing rainfall is forecast.The Maximum Alert level should be announced when there is an Alert level in several areas in the municipality.Tables 2, 3 and 4 summarize the aforementioned data for the municipalities of Recife, Camaragibe and Jaboatão dos Guararapes, respectively.
The cumulative rainfall parameters suggested herein as support for Civil Defense actions in the study areas are limit values.At the start of the rain the Observation operating level is adopted (from December 1 st ), after what is considered the dry season (October-November).At this level there is no need for site visits, unless at the request of the local inhabitants.When the cumulative rainfall is predicted to reach the critical values within 72 h (I 72 h ), particularly in 24 h, or to reach the long-term critical rainfall figures (P ac ) or the R crit , then the Warning level begins.This is the level in which the civil defense technical team begins its most active field inspections, to check the need for temporary or definitive removal of families living in the high risk and very high risk areas, considering the possibility of landslide events.When it is expected that R crit and I 72 h reached at the same time, the Alert level is given, remaining at this level when weather forecasts indicate continuing heavy rainfall.At this operating level the civil defense may inform the population about the possible total evacuation from the highest risk areas; and this could be helped by implementing a sound system or other form of alert communication, informing the population about the strong possibility of generalized landslide events.At this level the need is assessed to evacuate the population living in the most critical risk areas and to set up a shift system for the civil defense teams, with technical personnel on call 24 h to provide assistance.The field inspections are also carried out at this stage.
Figure 13 presents the proposal in a follow-up graph of the rainfall parameters (I 72 h and R crit ) to establish the Observation, Warning and Alert levels for Area II of Camaragibe.The civil defense teams recommend drawing up a graph for the other study areas in the municipalities.Figure 13 shows that when the I 72 h and R crit values were considerably higher, there was a major increase in the number of landslides and consequences.Continuing this study, different levels of alert associated with the gravity of the consequences could be established.R crit must be monitored by following up the cumulative rainfall product since December 1 st for the expected cumulative rainfall in 72 h.Continuing at the Alert level itis suggested that the rainfall indices are followed up daily and, if possible, at 60-min intervals.It is worth mentioning that the R crit value proposed herein would only be valid when it is achieved during the wet season, which in Recife-MR is concentrated from February to September.If the location has some dry days, the critical cumulative rainfall may be reached after the wet season, when it will no longer be necessary to establish the state of Alert, but it is recommended to establish the Warning state, whenever necessary.Table 5 shows the criteria for the operating levels and their actions.

Validation of Rainfall Parameters Suggested for Actions in the Study Areas
In order to assess the performance of the rainfall parameters suggested herein for Alert level the rates of successfully predicted landslides were calculated using the ROC analysis (Fawcett, 2006).To do this the chosen pilot area was the municipality of Camaragibe.Table 6 gives the data used in this analysis and Tables 7 and 8 present a summary of the results.Figure 14 illustrates a ROC graph of the Camaragibe Areas.
Table 7 shows that the true positive classified rate of Alert days was higher than the number of days classified as false negative, in most areas, except in Area I and Area IV-VC.Nevertheless, the number of true positive days in Area I represents 68% of the landslide events in that area.In relation to Area IV-Vera Cruz, it should be mentioned that it has an atypical characteristic, showing relief in the form of a single valley, with disorganized occupation.The landslide events in this area are due to its peculiar characteristics, closely related to anthropic actions.On analyzing the entire municipality, the number of days said to be true positive (29 days) is 66% of total landslide occurrences (97 landslides of a total 146).
In Table 8 (ROC Analysis) the true positive rates of Areas II, III and IV-TAB are higher than 0.50, signifying a good result.In relation to the item Precision, Areas I, II and III are seen to have values of 0.37-0.47.Area IV-TAB had a precision of 0.28.It should again be mentioned that the atypical case is Area IV-Vera Cruz, which had the lowest value (0.08).
Figure 14 presents the ratio between the true and false positive rates.In this figure, when the points have ordered pairs close to 0-false positive rate, 1-true positive rate, they are more accurate as well as those located above the diagonal line.Figure 14 shows that points B, C and D, corresponding to Areas II, III and IV-TAB, respectively, have more satisfactory values, but no area has a point below the diagonal line (high false positive rate values), which demonstrates that the suggested parameters are satisfactory for alert system recommendations.Level/regional Regional I (Jab.Centro) Regional II (Cavaleiro) Regional III (Curado) Regional IV (Muribeca) Regional V     With regard to the isolated values of the critical cumulative rainfall in 72 h (I 72 h ), the 100 mm value suggested for the warning state in Area I of Camaragibe and Regionals III (Curado) and IV (Muribeca) of Jaboatão dos Guararapes (Tables 2 and 3), is similar to those proposed by Macedo et al. (2004) for the Baixada Santista and Paraíba Valley in São Paulo.The 80 mm value in 72 h or less for the warning state in the West Regional in the city of Recife (Table 1), for Areas III and IV (Vera Cruz) in Camaragibe (Ta-ble 2); and for Regionals I (Jaboatão Centro) and II (Cavaleiro) in Jaboatão dos Guararapes (Table 3), is similar to that proposed for the São Paulo Regions of Campos do Jordão, Campinas, ABC and Sorocaba.The critical cumulative rainfall value of 60 mm in 72 h recommended for the warning state in the South Regional of the city of Recife, and for Areas II and IV (Tabatinga) of Camaragibe and Regional V (Prazeres) of Jaboatão dos Guararapes, are lower than those found by Macedo et al. (1999) for the São Paulo  Regions.The critical rainfall accumulated in 72 h for the Northeast, Northwest and North Regionals of Recife was 40 mm.This is the lowest critical value.
Figure 15 illustrates a landslide event on April 13, 2009, in the South Regional of Recife.This landslide, which destroyed a property, occurred after a concentrated rainfall of 123.6 mm/24 h and 137 mm/72 h, with long-term accumulation of 1,200 mm and an R parameter of 164,400 mm 2 .It is thereby found that this landslide occurred with higher rainfall parameters than those suggested for the state of Warning and Alert in Recife-MR.Other landslides events recorded years after this study confirmed the critical rainfall parameters proposed herein, such as, for example, the landslides on March 23, 2010, where the cumulative rainfall of 104.2 mm/72 h and particularly 80 mm/24 h caused slope instabilities in a number of locations in the study area.
In 2009, the year when the rainfall and landslides were followed up in this study, 10 fatal victims were registered due to the landslides on hillsides: six deaths in Recife, one in Camaragibe and three in Jaboatão dos Guararapes.Table 9 shows the rainfall accumulation in 72 h, long days

Figure 6 -
Figure 6 -Landslides and monthly rainfall during 2009 in the studied area.

Figure 8 -
Figure 8 -Landslides due to accumulated rainfall from January 1 st in Recife Regionals, Pernambuco -Total of 827 landslides.

Figure 10 -
Figure 10 -Landslides due to accumulated rainfall from January 1 st in Camaragibe Regionals, Pernambuco -Total of 160 landslides.

Figure 12 -
Figure 12 -Landslides due to accumulated rainfall from January 1 st in Jaboatão dos Guararapes Regionals, Pernambuco -Total of 380 landslides.
= 80 mm or P ac = 600 mm or R crit = 48,000 mm 2 I 72 h = 80 mm or P ac = 750 mm or R crit = 60,000 mm 2 I 72 h = 100 mm or P ac = 750 mm or R crit = 75,000 mm 2 I 72 h = 100 mm or P ac = 600 mm or R crit = 60,000 mm 2 I 72 h = 60 mm or P ac = 750 mm or R crit = 45,in all areas of the municipality *At this level, isolated landslides may occur, particularly with heavy rainfall concentrated in 24 h.

Figure 13 -
Figure 13 -Follow-up of rainfall parameters suggested for civil defense actions in Area II of Camaragibe.

Table 4 -
Proposed accumulated rainfall parameters for the Civil Defense operating levels in Jaboatão dos Guararapes.

Table 5 -
Civil Defense operating levels and recommended actions for Recife-MR.
st Follow up of rainfall indices and meteorologyWarningWhen I 72 h is expected to occur at any time of the year; or when P ac or R crit is expected to occur within the wet season (Feb-Sep)

Table 3 -
Proposed accumulated rainfall parameters for the Civil Defense operating levels in Camaragibe.

Table 6 -
Landslides and Operating Levels in in Camaragibe.

Table 7 -
Performances of Alerts using ROC analysis in Camaragibe.³ 100 mm and R crit ³ 75,000 mm ³ 60 mm and R crit ³ 45,000 mm

Table 8 -
Performances of alerts using ROC analysis in Camaragibe.