Shallow and deep landslides induced by rainfall in the Lisbon region (Portugal): assessment of relationships with the North Atlantic oscillation

Natural Hazards and Earth System Sciences, 5, 331–344, 2005 SRef-ID: 1684-9981/nhess/2005-5-331 European Geosciences Union © 2005 Author(s). This work is licensed under a Creative Commons License. Natural Hazards and Earth System Sciences Shallow and deep landslides induced by rainfall in the Lisbon region (Portugal): assessment of relationships with the North Atlantic Oscillation J. L. Zˆ ezere1, R. M. Trigo2,3, and I. F. Trigo4 1Centro de Estudos Geogr´ aficos, Universidade de Lisboa, Portugal 2CGUL, Faculdade de Ciˆ encias, Universidade de Lisboa, Portugal 3Departamento de Engenharias, Universidade Lus´ ofona, Lisboa, Portugal 4Instituto de Meteorologia, Lisboa, Portugal Received: 26 October 2004 – Revised: 25 February 2005 – Accepted: 7 April 2005 – Published: 18 April 2005 Part of Special Issue “Landslides and debris flows: analysis, monitoring, modeling and hazard” Abstract. The aim of this study is to assess the impact of the North Atlantic Oscillation (NAO) on both the winter precipitation and the temporal occurrence of different landslide types in Portugal. The analysis is applied to five sample areas located just north of Lisbon, the capital of Portugal. These sites are particularly relevant because actual dates of most of the recent landslide events are known but also because the landslides occurred in a suburban area with growing urbanization pressure. Results show that the large inter-annual variability of winter precipitation observed in western Iberia, i.e. Portugal and parts of Spain, is largely modulated by the NAO mode. In particular, precipitation falling in Portugal between November and March presents a correlation coefficient of R=−0.66 with the NAO index. Precipitation distribution for the reference rain gauge in the study area reveals that the probability of a wet month to occur is much higher for low NAO index composites than for the corresponding high NAO index composite. It is shown that this control, exerted by NAO on the precipitation regime, is related to corresponding changes in the associated activity of North-Atlantic storm tracks that affect the western Iberia. Landslide activity in the study area is related to both intense, short duration precipitation events (1–15 days) and long-lasting rainfall episodes (1–3 months). The former events trigger shallow translational slides while the later episodes are usually associated with deeper and larger slope movements. This second group of landslides is shown to be statistically associated with the 3-month average of the NAO index. Correspondence to: J. L. Zˆ ezere ([email protected]) 1 Introduction Portugal is prone to slope instability due to geological, geomorphological and climatic factors (Rodrigues and Coelho, 1989; Ferreira and Zˆ ezere, 1997). In the area north of Lisbon studied in this paper, landslides have caused considerable economic losses during the last 50 years. Landslide consequences include damages on property, houses and particularly roads, and are strongly related with the increasing humanpressurerelated to urban developmentaroundthePortuguese capital. In fact, human activity within the region has introduced significant land use changes, which are partially responsible for the acceleration of natural geomorphologic activity during the previous half century (Ferreira et al., 1987; Zˆ ezere, 2002). Slope movements in the Lisbon area are induced by rainfall, as are most landslides worldwide (Fukuoka, 1980; Crozier, 1986; Gostelow, 1991; Corominas, 2001). The role of precipitation on landslide activity near Lisbon has been discussed and analysed in previous works (Zˆ ezere et al., 1999a, 1999b; Zˆ ezere, 2000; Zˆ ezere and Rodrigues, 2002), by applying cumulative rainfall methods. Previous results obtained using empirical relationships between rainfall amount and duration, and slope instability show that critical rainfall conditions for failure are not the same for different types of landslide movement (Zˆ ezere, 2000; Zˆ ezere and Rodrigues, 2002; Trigo et al., 2005). Shallow translational soil slips are most commonly activated by intense precipitation that falls within the 1 to 15 days long range. On the contrary, activity of the more deeply-seated landslides of rotational, translational and complex types is related to successive weeks of nearly constant rainfall, over periods of 30 to 90 days. The physical explanation of the different behaviour of shallow and deep landslides is probably related 332 J. L. Zˆ ezere et al.: Shallow and deep landslides induced by rainfall Table 1. Main slope movements and basic morphometrical parameters of the landslides identified in 5 study areas (61.6 km2)in the region north of Lisbon. Type of number % of total slope angle depth area per landslide estimated volume movements of cases landslides (◦) (m) (m2) per landslide (m3) events mean std. dev. mean std. dev. mean std. dev. mean std. dev. shallow translational slide 212 58.9 23 8 1.0 0.5 552 693 286 376 translational slide 55 15.3 17 7 3.1 1.7 4059 5470 5232 8822 rotational slide 39 10.8 19 7 5.2 4.3 9407 21664 34843 116434 complex and composite movements 54 15.0 20 8 4.0 3.4 9747 27011 42998 202801 total landslides 360 100.0 21 8 2.1 2.5 4481 18241 12224 82489 to infiltration processes, namely the different pressure head responses to rainfall controlled by soil characteristics, slip surface depths and effective hydraulic diffusivities (Iverson, 2000). Here we improve the empirical discrimination betweenprecipitation eventsthattriggershallow and deep slope movements in the study area, taking in account the different landslide mechanisms. The present study aims at characterizing the atmospheric conditions that favor the occurrence of landslides in the study area, particularly those slope movements associated with long-lasting rainfall episodes. Therefore, a comprehensive description of the precipitation regime at the regional (Portugal) and local (Lisbon area) scales, particularly the interannual variability of precipitation, is examined here to understand the temporal variation of landslide activity. TheNorthAtlantic Oscillation (NAO)hasbeenrecognized for more than 70 years as one of the major patterns of atmospheric variability in the Northern Hemisphere (Walker, 1924). However, only in recent years has this important atmospheric circulation mode become the subject of a wider interest (e.g. van Loon and Rogers, 1978; Rogers, 1984; Barnston and Livezey, 1987). More recently, the study by Hurrell (1995) had significant impact on the climatological community and has been followed by an increasing number of studies. It is within this context, that several studies have established links between the NAO index and winter season precipitation in Western Europe and, in particular, over the Mediterranean basin (Hurrell, 1995; Qian et al., 2000; Trigo et al., 2004a). This control exerted by NAO on the precipitation field over Europe is likely related to corresponding changes in the associated activity of North Atlantic storm tracks (Serreze et al., 1997; Osborn et al., 1999). In fact, this strong NAO-precipitation link has been used to develop rainfall forecast models that predict precipitation in Iberia several months in advance (e.g. G´ amiz-Fortis et al., 2002; Rodriguez-Fonseca and Castro, 2002). The output of these models may provide useful seasonal forecasting tools for waterresource managers and risk assessment teams (Trigo etal., 2004a). The relationship between landslide activity and low frequency atmospheric circulation patterns (associated with anomalous precipitation at the seasonal scale), has been undertaken before using the Southern Oscillation, better known as El Ni˜ no. Recent El Ni˜ no events, such as those occurred in 1982/1983 and 1997/1998 were directly responsible for anomalously high precipitation in different parts of the Globe. The unusual rainfall produced widespread landslide activity in unrelated regions, such as California (Coe et al., 1998; Godt, 1999) and Kenya (Ngecu and Mathu, 1999). To the best of our knowledge no similar study has been published for any European region, despite the fact that the NAO exerts a significant control on the precipitation regime in southern Europe. Therefore, the main objectives of this study are: (i) to describe the main characteristics of episodes of landslide activity observed during the previous 50 years, in several locations north of Lisbon; (ii) to discriminate precipitation events responsible for shallow and deep landslides using empirical analytical tools; (iii) to characterize the impact of NAO on the location of cyclones that strike Iberia, as well as the impact of NAO on the precipitation regime at the regional (Portugal) and local (north of Lisbon) scales; (iv) to assess the influence of the NAO on the occurrence of landslides near Lisbon, in particular the deeper slope movements triggered by long duration rainfall events. 2 Geomorphologic setting The region north of Lisbon (Fig. 1) is located in the Portuguese Meso-Cenozoic sedimentary basin, and has a general monocline structure dipping 5◦to 25◦towards south and southeast. Geological formations include very heterogeneous lithological units (e.g. limestones, marly limestones, J. L. Zˆ ezere et al.: Shallow and deep landslides induced by rainfall 333 Fig. 1. Geomorphological map of the Area North of Lisbon. 1. Front of “cuesta” on Cretaceous rocks; 2. Front of “cuesta” on Tertiary rocks; 3. Other cliffs; 4. River and gorge; 5. Terrace; 6. Alluvial plain; 7. Geological boundary; 8. Elevation in meters; 9. Sample areas; 10. Landslides: shallow translational slides (red), translational slides (green), rotational slides (cyan), complex and composite slope movements (purple); 11. Location of S. Juli˜ ao do Tojal rain-gauge. J. Jurassic rocks (clays, marls, limestones, sandstones); C. Cretaceous rocks (sandstones, marls and limestones); B. Upper Cretaceous Volcanic Complex of Lisbon; O. Paleogene detritic complex; M. Miocene (sandtones, limestones and clays). 1 – Pinheiro de Loures Sample Area; 2 – Lousa Sample Area; 3 – Fanh˜ oes Sample Area; 4 – Tranc˜ ao Sample Area; 5 – Calhandriz Sample Area. marls and clays, sandstones, basalts and volcanic tufs) dating from the upper Jurassic to Miocene. The region north of Lisbon contacts the Tagus basin south-eastward along the Low Tagus fault zone (NE–SW direction). The tectonic activity of this structure has been responsible for the regional uplift of the study area during the previous 5 M years (Ferreira et al., 1987; Zˆ ezere et al., 1999b). This positive tectonic deformation explains the strong fluvial erosion and development of steep slopes, namely on those valleys running in the same direction of the dip of strata. Outcropping rocks were differentially eroded during the Quaternary, allowing the development of “cuestas” and large depressions like the Loures basin, where quaternary age sediments are preserved. 3 Landslide identification and classification Landslides were mapped and inventoried in 5 study areas with a total area of 61.6km2(Fig. 1). Within these five areas 360 slope movements were identified, excluding rockfalls and landslides triggered by bank erosion of stream channels. Therefore, the landslide density is 5.8 per km2and the total affected area is more than 1.6km2(2.6% of the total Fig. 2. Shallow translational slide triggered in November 1983, in the East slope of the Tranc˜ ao valley (Tranc˜ ao sample area). study area). Landslides were classified in four main groups (Table 1), taking into account the type of movement and the affected material (Zˆ ezere et al., 1999a, 1999b). 3.1 Shallow translational slides Shallow translational slides are the most frequent slope movements within the study area (59% of total landslide events). They are single slope movements with planar slip surfaces, and generally have small lateral dimension (mean area, 552m2; mean volume, 286m3). In most of cases, shallow landslides occur on steep valley hill slopes (mean slope angle=23◦)and affect soil material (colluvium deposits) covering impermeable rocks, such as volcanic tuffs, marls and clays (Fig. 2). 3.2 Translational slides Translational slides are generally deeper seated and larger than shallow movements (mean area 4059m2; mean volume 5232m3).These landslides typically involve the bedrock (marls and clays with limestone and marly limestone intercalations) and activate on slopes that follow the dip of the 334 J. L. Zˆ ezere et al.: Shallow and deep landslides induced by rainfall Fig. 3. Translational slide occurred in February 1979, on a cataclinal slope where basalts overlay volcanic tuffs (Pinheiro de Loures sample area). Fig. 4. Backward scarp of a confined rotational slide triggered in December 1989. The landslide took place on a gentle slope cutting clays and marls of Jurassic age (Calhandriz sample area). strata, along shear surfaces controlled by impermeable bedding planes (Fig. 3). Translational slides are strongly conditioned by geometrical factors, as it is confirmed by the gradient of the affected slopes (average = 17◦), which exceeds the dip of the strata in more than 85% of cases. Although topographic slope has a relatively minor effect on the infiltration capacity of soils (Philip, 1991), moderate slope gradients do not favour rapid drainage of water by both overland flow and sub-surface flow, thus contributing to slope instability. 3.3 Rotational slides Rotational slides are not particularly common in the study area (11% of the total slope movements) due to the inhomogeneous and anisotropic characteristic of most geological formations. The presence of zones of weakness (e.g. soft bedding planes) is a condition facilitating the development of planar slip surfaces rather than circular rupture zones. ThereFig. 5. General overview of the “Quebradas” complex slope movement (Fanh˜ oes sample area). The landslide developed in February 1979 as a deep translational slide, evolving downslope to an earthflow. 0 200 400 600 800 1000 1200 1400 1956-57 1959-60 1962-63 1965-66 1968-69 1971-72 1974-75 1977-78 1980-81 1983-84 1986-87 1989-90 1992-93 1995-96 1998-99 R (mm) MAP Fig. 6. Annual precipitation (climatological year) distribution at S. Juli˜ ao do Tojal (reference rain-gauge) from 1956/1957 to 2000/2001. The horizontal line indicates the mean annual precipitation (MAP). Diamonds indicate episodes of shallow landslides activity and triangles indicate episodes of deep landslides activity. fore, rotational slides develop on the mostly homogeneous lithological units (clays and marls) within the Calhandriz study area, where it is possible to observe rotational slides of single, confined and multiple retrogressive sub-types (Fig. 4). Rotational slides have the highest average depth (5m) and are typically larger than the translational movements (mean area 9407m2; mean volume 34843m3). 3.4 Complex and composite slope movements Landsides included in this category show at least two different types of mechanisms, in sequence (complex movement) or simultaneously (composite movement). The most significant examples combine slide and flow mechanisms as well as translational and rotational movements (Fig. 5). Most of the larger landslides that occurred in the study area during the previous 50 years belong to this type. Together with the rotational slides, this landslide group exhibits the largest average area and volume (27011m2; 42998m3). J. L. Zˆ ezere et al.: Shallow and deep landslides induced by rainfall 335 Archive investigation Field work Interviews Reconstruction of past landslide activity dates Rainfall analysis (daily data) Critical rainfall combinations (quantity-duration) responsible for landslide events Reconstruction of absolute antecedent rainfall from 1 to 90 days (Pxn= P1 + P2 +…Pn) Return periods (Gumbel law) Types of landslides Shallow slope movements Deep slope movements Rainfall triggering thresholds (intensity/duration) Fig. 7. Methodology for the statistical analysis of rainfall triggering of landslides based on the reconstruction of absolute antecedent precipitations. 4 Temporal occurrence of rainfall triggered landslides 4.1 Rainfall regime The Portuguese precipitation regime is dominated by high variability at both the inter-annual and inter-seasonal scales. The strong seasonal cycle, typical of most Mediterranean climates, is characterized by the concentration of precipitation during the months between October and March, and with little or no rainfall during the dry summer months (Trigo and DaCamara, 2000). The mean annual precipitation (MAP) at the reference rain gauge of S. Juli˜ ao do Tojal (SJT, Fig. 1) is roughly 730mm, with the rainfall occurring mostly from October to March (78% of the total amount; 72% of the total rainy days). Even at the annual scale, the rainfall regime is very irregular (Fig. 6), alternating large periods of drought with prolonged rainy periods. Very intense rainfall episodes occur episodically (e.g. in November 1967 and November 1983; return period ≥60 years), causing catastrophic flash floods on small catchments and widespread shallow slope movements on steep slopes. 4.2 Reconstruction of absolute antecedent rainfall Recent landslides in the region north of Lisbon were induced by rainfall, and landslide activity has been confined to very wet periods (Ferreira et al., 1987). Figure 7 shows the general methodology used to assess the relationship between absolute antecedent rainfall and landslide occurrence, based on previous work carried out for the studied area (Zˆ ezere et al., 1999a, 1999b; Zˆ ezere, 2000; Zˆ ezere and Rodrigues, 2002). As a first step, the dates of landslide activity were reconstructed from field work, archive investigation and interviews with the population living in the study area. This reconstruction also includes the particular type(s) of landslide(s) activated in each event. Rainfall analysis was carried out using 45 years (1956–2001) of daily precipitation at SJT, which is considered representative for the study area (Zˆ ezere, 2000; Zˆ ezere and Rodrigues, 2002). The analysis consists of calculating the cumulative absolute antecedent rainfall for 1, 5, 10, 0 20 40 60 80 100 120 140 160 180 0 20406080100 Critical Rainfall Duration (days) Rainfall intensity (mm/day) 1 2 3 4 56 7 8 9 10 11 12 13 14 15 16 17 18 19 1 2 Fig. 8. Regression line between critical rainfall intensity and corresponding event duration (period 1956–2001). Green diamonds: values associated to shallow landslide episodes. Blue triangles: values associated to deep landslide episodes. Small dots: values obtained from the yearly maximum rainfall intensity for all duration intervals (computed for years without reported landslides). 15, 30, 45, 60, 75 and 90 consecutive days prior to the dates of confirmed landslide activity during the 45 years period. The return period of the rainfall amount-duration combinations was computed using a Gumbel distribution (Gumbel, 1958). The critical rainfall combination (quantity-duration) responsible for each landslide event was assessed, assuming as critical pair the combination with the higher return period. This assumption has not a physical basis, but provides a maximum discrimination between rainfall periods characterized by landslide activity and rainfall periods not related with slope instability. Nineteen major rainfall episodes that triggered landslides were identified (Fig. 6), some of which were from the same climatological year (e.g. December 1958, March 1959). Nine of these episodes were dominated by shallow landslides, while the remaining where marked by occurrence of deepseated slope movements. The most relevant features of each landslide event are shown in Table 2, including date, rainfall amount-duration, and rainfall intensity-duration relationships. To analyse the overall effect of precipitation on landslide activity wecombine the rainfall intensity and the critical rainfall duration for the 19 reported landslide events (large diamonds and triangles in Fig. 8). A regression analysis shows that rainfall intensity increases exponentially as duration decreases, following the equation Ri=84.3D−0.57, where Ri is the rainfall intensity in mm/day and Dis the duration of rainfall in days. The maximum yearly value for durations of 1, 5, 10, 15, 30, 40, 60, 75 and 90 days were also computed for the 34 years for which no landslide activity was reported. These values are also plotted in Fig. 8 (small dots) and the vast majority of them lie below the fitted curve. Therefore, the regression curve is a reliable rainfall intensity – duration threshold for the study area. Nonetheless, we emphasize that this rain- 336 J. L. Zˆ ezere et al.: Shallow and deep landslides induced by rainfall Table 2. Temporal occurrence of rainfall triggered landslides in the region north of Lisbon from 1956 to 2001. (a) Shallow landslide episodes; (b) Deep landslide episodes. Landslide Date Critical rainfall Rainfall Return episode # (dd/mm/yyyy) amount duration intensity duration period mm (dd) mm/day (dd) (years) (a) 1 19/12/1958 149 (10) 14.9 (10) 2.5 2 09/03/1959 175 (10) 17.5 (10) 4 3 15/11/1967 137 (1) 137.0 (1) 60 5 04/03/1978 204 (15) 13.6 (15) 3.5 7 30/12/1981 174 (5) 34.7 (5) 13 8 18/11/1983 164 (1) 164.0 (1) 200 9 25/02/1987 52 (1) 52.0 (1) 2 10 22/11/1989 164 (15) 11.0 (15) 2 11 25/11/1989 217 (15) 14.4 (15) 4.5 (b) 4 15/11/1968 350 (30) 11.7 (30) 6.5 6 10/02/1979 694 (75) 9.2 (75) 20 12 05/12/1989 333 (30) 11.1 (30) 5.5 13 21/12/1989 495 (40) 12.4 (40) 20 14 09/01/1996 544 (60) 9.1 (60) 10 15 23/01/1996 686 (75) 9.1 (75) 18 16 28/01/1996 495 (40) 12.4 (40) 20 17 01/02/1996 793 (90) 8.8 (90) 24 18 06/01/2001 447 (60) 7.4 (60) 5 19 09/01/2001 467 (60) 7.8 (60) 5.5 fall threshold is statistically based and was not defined considering the hydrological response to rainfall of the soil and local slope stability conditions. For shorter duration rainfall events (Group 1, Fig. 8), the rainfall intensity required to trigger landslides is highly dependent on the duration of the precipitation event (e.g. 23mm/day for 10 days; 84mm/day for 1 day). In contrast, for long duration precipitation periods (above 30 days consecutive rainfall, Group 2, Fig. 8) the daily rainfall intensity tends to stabilize around the 8–12mm/day. 4.3 Reconstruction of calibrated antecedent rainfall According to Canuti et al. (1985) and Crozier (1986), the impact of a particular rainy event decreases in time due to drainageprocesses. Inordertoconsiderthateffect in rainfalllandslide analysis, Canuti et al. (1985) developed an index for sites in Italy that accounts for the calibrated cumulative rainfall, where the most recent data are given a higher weight. A similar approach was developed by Crozier (1986) and applied by Glade et al. (2000) to sites in New Zealand, which accounts for the draining of early precipitation and accumulation of late rainfall. In this work, antecedent rainfall was weighted using the formula proposed by Crozier (1986): CARx=K P1+K2P2+...KnPn(1) where CARxis the calibrated antecedent rainfall for day x; P1is the daily rainfall for the day before day x;Pnis the daily rainfall for the n-th day before day x. The constant Kis an empirical parameter usually considered between 0.8 and 0.9, depending on the draining capacity of the material andthehydrologicalcharacteristicsofthearea (Capecchi and Focardi, 1988). Following the work developed previously, in the present study it was assumed K=0.9 (Zˆ ezere, 2000). This constant makes rainfall occurring more than 30 days before a landslide event to become negligible (Capecchi and Focardi, 1988). Therefore, the reconstitution of calibrated antecedent rainfall was only performed for durations of 3, 5, 10, 15 and 30 days. Table 3 summarizes results of calibrated antecedent rainfall for both shallow landslide episodes (Table 3a) and deep landslide episodes (Table 3b). This distinction between landslide types is also considered in Figs. 9 and 10, where the evaluation of triggering rainfall conditions was made combining the calibrated antecedent rainfall for 5 and 30 days and the daily rainfall data. In order to define the number of days relevant to the antecedent rainfall, we combine the daily rainfall relative to the 19 reported landslide events with the corresponding calibrated antecedent rainfall for different periods (green diamonds and blue triangles for shallow landslides and deep landslides, respectively, in Fig. 11). Besides shallow and deep landslide events, Fig. 11 includes also the maximum J. L. Zˆ ezere et al.: Shallow and deep landslides induced by rainfall 337 Table 3. Calibrated antecedent rainfall (CAR) for periods of landslide activity in the region north of Lisbon from 1956 to 2001. (a) Shallow landslide episodes; (b) Deep landslide episodes. Calibrated antecedent rainfall (CAR) # of previous Landslide Date Daily 3 5 10 15 30 consecutive days episode # (dd/mm/yyyy) rainfall days days days days days with 30-days CAR (mm) (mm) (mm) (mm) (mm) (mm) >50mm (a) 1 19/12/1958 22.0 54.2 73.6 91.7 99.2 108.5 6 2 09/03/1959 45.5 27.5 55.4 82.2 82.2 83.6 4 3 25/11/1967 137.0 4.1 4.1 10.3 16.3 25.2 0 5 04/03/1978 14.5 43.9 67.1 93.8 105.0 109.8 8 7 30/12/1981 78.0 75.5 75.5 102.6 109.5 111.3 3 8 18/11/1983 163.7 42.1 59.8 88.2 96.7 103.8 10 9 25/02/1987 51.8 22.2 35.4 36.0 46.5 52.1 1 10 22/11/1989 33.5 19.9 50.7 62.9 70.9 72.1 4 11 25/11/1989 38.0 36.9 42.7 75.7 88.5 89.1 7 (b) 4 15/11/1968 45.5 43.4 47.3 74.3 92.3 107.7 18 6 10/02/1979 26.0 89.3 93.6 110.1 119.7 133.6 17 12 05/12/1989 30.1 58.1 58.1 78.8 93.7 109.7 17 13 21/12/1989 26.7 29.1 40.5 65.3 78.2 100.9 33 14 09/01/1996 66.5 17.8 29.5 54.2 79.5 88.4 17 15 23/01/1996 23.6 36.5 38.7 53.2 78.9 96.8 31 16 28/01/1996 25.5 9.2 30.9 53.7 62.3 85.6 36 17 01/02/1996 24.5 34.5 48.6 69.1 79.9 97.9 40 18 06/01/2001 24.5 22.7 39.1 53.7 79.1 87.4 17 19 09/01/2001 16.0 22.1 38.6 51.9 67.1 85.3 20 yearly precipitation in one day (pink dots) and the maximum yearly CAR (black dots) for the 34 years without reported landslide activity. The best results obtained for shallow landslide episodes correspond to the 5 days CAR (Fig. 11b). The envelope curve follows the exponential rule Dr=167.28e−0.0355CAR, where Dr is the daily rainfall. There are only 5 no-landslide events lying above the curve (pink dots), confirming this rule as a possible threshold for the shallow slope movements. The best discrimination of deep landslide events is observed in the graph corresponding to the 30 days CAR (Fig. 11d). The combined threshold of daily rainfall=16mm and 30 days CAR=85mm includes all the deep landslide episodes observed, and, equally important, it is exceeded by a single no-landslide event. Differences on the relevance of antecedent rainfall for shallow and deep landslide episodes are also impressive when we compare the number of previous consecutive days with 30 days CAR>50mm (Table 3). This number ranges from 0 to 10 days for dates of shallow movements occurrence, and increases up to a minimum of 17–18 days for dates characterised by deep landslide activity. Moreover, maximum values (20 to 40 days) are observed in late event dates in those years characterized by several landslide episodes (e.g. 1989, 1996 and 2001). 4.4 Hydrological triggering conditions of shallow and deep landslides Assessment of empirical rainfall thresholds provides no causal mechanisms for slope failures. Nevertheless, results from the statistical rainfall analysis are consistent with the different hydrological triggering conditions related to different types of landslides. Shallow translational soil slips are triggered by the rapid infiltration of water into the thin soil material (colluvium deposits) which overlies impermeable rocks. The temporary rise of the pore water pressure as well as the loss of soil apparent cohesion resulting from soil saturation are responsible for a critical reduction of the soil shear resistance and resultant failure (Gostelow, 1991; Iverson, 2000). Translational slides, rotational slides and complex and composite slope movements have deeper slip surfaces than soil slips and involve rocks and soils that yield small effective hydraulic diffusivities. They are triggered by the reduction of shear strength of affected soils and rocks, linked with the steady rising of the groundwater level resulting from long-term precipitation periods (Gostelow, 1991; van Asch et al., 1999; Iverson, 2000). In the following section we describe the large-scale atmospheric circulation responsible for the high inter-annual precipitation variability and its effects on deep-seated landsliding (Fig. 8). 338 J. L. Zˆ ezere et al.: Shallow and deep landslides induced by rainfall 0 50 100 150 01-10-58 15-10-58 29-10-58 12-11-58 26-11-58 10-12-58 24-12-58 07-01-59 21-01-59 04-02-59 18-02-59 04-03-59 18-03-59 R (mm) ` ` 0 50 100 150 01-10-67 15-10-67 29-10-67 12-11-67 26-11-67 10-12-67 24-12-67 07-01-68 21-01-68 04-02-68 18-02-68 03-03-68 17-03-68 31-03-68 R (mm) ` 0 50 100 150 01-10-77 15-10-77 29-10-77 12-11-77 26-11-77 10-12-77 24-12-77 07-01-78 21-01-78 04-02-78 18-02-78 04-03-78 18-03-78 R (mm) ` 0 50 100 150 200 01-10-81 15-10-81 29-10-81 12-11-81 26-11-81 10-12-81 24-12-81 07-01-82 21-01-82 04-02-82 18-02-82 04-03-82 18-03-82 R (mm) ` 0 50 100 150 200 250 01-10-83 15-10-83 29-10-83 12-11-83 26-11-83 10-12-83 24-12-83 07-01-84 21-01-84 04-02-84 18-02-84 03-03-84 17-03-84 31-03-84 R (mm) ` 0 50 100 150 01-10-1986 15-10-1986 29-10-1986 12-11-1986 26-11-1986 10-12-1986 24-12-1986 07-01-1987 21-01-1987 04-02-1987 18-02-1987 04-03-1987 18-03-1987 R (mm) ` 0 50 100 150 01-10-89 15-10-89 29-10-89 12-11-89 26-11-89 10-12-89 24-12-89 07-01-90 21-01-90 04-02-90 18-02-90 04-03-90 18-03-90 R (mm) ` ` Fig. 9. Daily rainfall and calibrated antecedent rainfall (CAR) in the last 5 and 30 days for years of shallow landslide activity in the region north of Lisbon from 1956 to 2001. Blue bars: daily rainfall; red curve: 5-days CAR; grey curve: 30-days CAR; green diamonds: dates of shallow landslide activity. 5 The impact of NAO on Portuguese precipitation and landslide occurrence It is now widely accepted that a large portion of winter precipitation in Iberia can be explained in terms of a relatively small number of large-scale atmospheric circulation modes that vary at monthly time scales (Trigo and Palutikof, 2001). In particular, the impact of the North Atlantic Oscillation (NAO) to model winter precipitation over Iberia has been documented in literature (e.g. Rod´ o et al., 1997; Corte-Real et al., 1998; Trigo et al., 2002; Trigo et al., 2004a, 2005). As previously referred at the beginning, the NAO corresponds to the major pattern of extratropical atmospheric variability, accounting for roughly one third of the sea level pressure variability in the Northern Hemisphere (Trigo et al., 2002; Hurrell et al., 2003). The NAO index used in this study was developed by the Climatic Research Unit (University of East Anglia, UK) and is defined, on a monthly basis, as the difference between the normalized surface pressure at Gibraltar (southern tip of Iberian Peninsula) and Stykkisholmur in Iceland (Jones et al., 1997). The NAO index for winter months presents a positive trend over the last 30 years, as a consequence its distribution is dominated by positive values, with monthly averages above zero (Jones et al., 1997). Therefore we decided to normalize the entire winter NAO index (average of NDJFM values) so it has zero mean and standard deviation 0 50 100 150 01-10-68 15-10-68 29-10-68 12-11-68 26-11-68 10-12-68 24-12-68 07-01-69 21-01-69 04-02-69 18-02-69 04-03-69 18-03-69 R (mm) ▲ 0 50 100 150 01-10-00 15-10-00 29-10-00 12-11-00 26-11-00 10-12-00 24-12-00 07-01-01 21-01-01 04-02-01 18-02-01 04-03-01 18-03-01 R (mm) ▲ ▲ 0 50 100 150 200 01-10-78 15-10-78 29-10-78 12-11-78 26-11-78 10-12-78 24-12-78 07-01-79 21-01-79 04-02-79 18-02-79 04-03-79 18-03-79 R (mm) ▲ 0 50 100 150 01-10-95 15-10-95 29-10-95 12-11-95 26-11-95 10-12-95 24-12-95 07-01-96 21-01-96 04-02-96 18-02-96 03-03-96 17-03-96 31-03-96 R (mm) ▲ ▲ ▲ ▲ 0 50 100 150 01-10-89 15-10-89 29-10-89 12-11-89 26-11-89 10-12-89 24-12-89 07-01-90 21-01-90 04-02-90 18-02-90 04-03-90 18-03-90 R (mm) ▲ ▲ Fig. 10. Daily rainfall and calibrated antecedent rainfall (CAR) in the last 5 and 30 days for years of deep landslide activity in the region north of Lisbon from 1956 to 2001. Blue bars: daily rainfall; red curve: 5-days CAR; grey curve: 30-days CAR; blue triangles: dates of deep landslide activity. 0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 5-days calibrated antecedent rainfall (mm) Daily rainfall (mm) B 0 20 40 60 80 100 120 140 160 180 0 20406080100120140 3-days calibrated antecedent rainfall (mm) daily rainfall (mm) A 0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 15-days calibrated antecedent rainfall (mm) Daily rainfall (mm) C 0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 30-days calibrated antecedent rainfall (mm) Daily rainfall (mm) 85 16 D Fig. 11. Relationship between daily rainfall and calibrated antecedent rainfall (CAR) in 3 days (a), 5 days (b), 15 days (c) and 30 days (d) (period 1956–2001). Green diamonds: values associated to shallow landslides occurrence; blue triangles: values associated to deep landslides occurrence; small dots: values obtained from the yearly maximum daily rainfall (pink colour) and from the yearly maximum CAR (black colour), computed for years without reported landslides. one. Finally, we defined the seasonal high NAO composite (low NAO composite) to be a combination of all winters with NAO index greater than 0.5 (less than −0.5). Between 1932 and 2001 (70 winters), the number of winter seasons with a high NAO index (20) is similar to the number characterized by a low NAO index (24). The remaining winters (26) are characterized by near normal NAO index values. The spatial analysis on the impact of NAO on atmospheric circulation and precipitation is performed using monthly average values of sea level pressure (SLP), and precipitation rate (mm/day) over a 2.5◦×2.5◦grid. The atmospheric data used in this study are large-scale gridded data retrieved from J. L. Zˆ ezere et al.: Shallow and deep landslides induced by rainfall 339 Fig. 12. Precipitation rate anomaly fields (mm/day) from the NCEP/NCAR reanalysis for winter months with (a) high NAO index >0.5, (b) low NAO index <−0.5 and, and, (c) their difference (represented only if significant at the 5% level). Positive (solid) and negative (dashed) isolines of the sea level pressure anomaly field (hPa) are also represented (period 1958–1997). the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis dataset (Kalnay et al., 1996). Advantages and cautions on the use of this dataset for Iberia are beyond the scope in this work and have been previously addressed in greater detail by Trigo et al. (2002, 2004a, b). The gridded data cover an area from 30◦N to 80◦N and 60◦W to 70◦E, for the period 1958–1997. In fact, NCEP/NCAR Reanalyses does not cover the 70 years period available for the NAO index and precipitation in Portugal, therefore the spatial analysis presented in Fig. 12, as well as the storm track discussion in Sect. 6 (Fig. 16) is performed for a shorter 40 year long period. SLP and precipitation rate anomaly fields for winter months characterized by high and low NAO index values were computed and are shown in Figs. 12a and 12b, respectively. Differences of the SLP between winter months (NDJFM) with high and low NAO index (solid contour lines) are Fig. 13. Inter-annual variability of the mean winter (NDJFM) Portuguese average precipitation (solid curve), and the contemporaneous winter NAO index, multiplied by −1 to facilitate analysis (dashed curve); both curves have been normalised and so are dimensionless (period 1932–2001). shown in Fig. 12c. Corresponding differences in precipitation rate, between high and low NAO composites, are also represented in Fig. 12c, wherever those differences are statistically significant at the 5% level (colour scale). This figure highlights two important issues: (i) The impact of the NAO on the northern hemisphere winter precipitation field, for both phases of the NAO index, is not restricted to the European continent but extends over large sectors of the North Atlantic, confirming results from previous works (Hurrell, 1995; Osborn et al., 1999; Trigo et al., 2004a); (ii) Figure 12c shows quasi-zonal bands of opposite anomaly signs, with impressive positive differences concentrated in the northern latitudes, extending from eastern Greenland to Finland, with maximum values south of Iceland. At lower latitudes, a strong band of negative differences extends from the Azores archipelago in the mid Atlantic Ocean, to the Balkanic Peninsula, with larger differences located west of Iberia, and particularly over Portugal. The inter-annual variability of Portuguese average winter (NDJFM) precipitation and contemporaneous winter NAO index can be observed, for the 1932-2001 period in Fig. 13. The average precipitation in Portugal was computed using 18 stations that cover the whole period and are representative of the entire territory (Trigo and DaCamara, 2000). Both curves (precipitation and NAO index) were normalized and the NAO index multiplied by minus one to facilitate visual comparisons. The correlation coefficient between both curves is R=−0.66 (statistically significant at the 1% level). Having shown that the winter precipitation in Portugal is associated with the NAO mode we now focus our attention on the precipitation that falls on SJT. For this pur-