Nat. Hazards Earth Syst. Sci., 18, 613–631, 2018 https://doi.org/10.5194/nhess-18-613-2018 © Author(s) 2018. This work is distributed under the Creative Commons Attribution 4.0 License. Physically based approaches incorporating evaporation for early warning predictions of rainfall-induced landslides Alfredo Reder1,2, Guido Rianna2, and Luca Pagano1 1Department of Civil, Architectural and Environmental Engineering, University of Naples Federico II, 80125 Naples, Italy 2Regional Models and geo-Hydrological Impacts Division, CMCC Foundation, 81043 Capua, Italy Correspondence: Guido Rianna ([email protected]) Received: 3 August 2017 – Discussion started: 7 September 2017 Revised: 30 January 2018 – Accepted: 31 January 2018 – Published: 28 February 2018 Abstract. In the field of rainfall-induced landslides on slop- conditions permit them to be stable for slope angles exceed- ing covers, models for early warning predictions require an ing friction angles (Pagano et al., 2008b) thanks to additional adequate trade-off between two aspects: prediction accuracy strength provided by suction (usually, they are characterized and timeliness. When a cover’s initial hydrological state is by low or null true cohesion values). The sequence of rainfall a determining factor in triggering landslides, taking evapo- events occurring over the wet season induces a general re- rative losses into account (or not) could significantly affect duction in suction levels, increasing the cover’s susceptibil- both aspects. This study evaluates the performance of three ity to an exceptional rainfall event. In contrast, evaporative physically based predictive models, converting precipitation fluxes reduce susceptibility to sliding by increasing suction and evaporative fluxes into hydrological variables useful in levels. The antecedent period, during which the contrast be- assessing slope safety conditions. Two of the models in- tween rainfall and evaporation affects suction levels, may last corporate evaporation, with one representing evaporation as weeks or months depending on the hydraulic properties of both a boundary and internal phenomenon, and the other only the soils involved and the climate regime of the area (Rahimi a boundary phenomenon. The third model totally disregards et al., 2011; Rahardjo et al., 2001). evaporation. Model performances are assessed by analysing The 2005 Nocera Inferiore landslide (hereinafter a well-documented case study involving a 2 m thick slop- “2005NIL”) was interpreted (Pagano et al., 2010) by merely ing volcanic cover. The large amount of monitoring data col- referring to precipitation recorded by a meteorological sta- lected for the soil involved in the case study, reconstituted in tion placed near the landslide area (Fig. 1). Richards’ (1931) a suitably equipped lysimeter, makes it possible to propose equation in 1-D flow conditions was adopted to convert procedures for calibrating and validating the parameters of hourly precipitation records into the evolution of soil suction the models. All predictions indicate a hydrological singular- at various depths. This simple approach highlighted the ity at the landslide time (alarm). A comparison of the mod- crucial role of antecedent rainfall (945 mm of rainfall over els’ predictions also indicates that the greater the complex- 4.5 months), which had reduced soil suction to very low ity and completeness of the model, the lower the number of values before the occurrence of the major event (143 mm of predicted hydrological singularities when no landslides oc- rainfall over 16 h). Numerical analyses indicate that suction cur (false alarms). vanishing throughout the entire cover depth can induce the attainment of slope failure conditions. Virtual scenarios built with modified antecedent rainfalls were analysed, and they indicated that the phenomenon would have not occurred if 1 Introduction the antecedent periods had been drier. The crucial factor affecting soil suction at triggering time were the weather In Italy, many sloping deposits of sand or silty sand, consti- conditions over the previous 4 months. tuting covers not exceeding few metres, experience unsatu- rated conditions throughout the hydrological year. Such state Published by Copernicus Publications on behalf of the European Geosciences Union. 614 A. Reder et al.: Physically based approaches incorporating evaporation for early warning predictions Figure 1. The 2005 Nocera Inferiore Landslide (2005NIL): (a) map indicating the landslide location and zones homogeneous with that of the landslide for soils, cover thicknesses and slope gradients; (b) plan view (DigitalGlobe 2012, http://www.earth.google.com), indicating the landslide area and the location of the weather station; (c) frontal view of the landslide area (Pagano et al., 2010, modified). A meteorological window of such long influence implies and accuracy of prediction. In tens of centimetres thin and/or that it would not be reasonable to neglect evaporative fluxes, coarse-grained soil covers, prediction accuracy does not re- as their persistency could result in significant drying pro- quire an account for evaporation because the latter plays a cesses even during the cold season, when evaporation is at its minor role in the hydrological balance, which might result lowest (about 1–2 mm day−1 in winter). Rianna et al. (2014a) in a landslide (Brand et al., 1984; Chatterjea, 1989; Morgen- measured infiltrating precipitation and actual evaporation stern, 1992). (AE) induced by the actual weather conditions on a layer Thick silty covers may instead experience a hydrologi- placed in a lysimeter, made using the same soil that was in- cal behaviour strongly affected by evaporation. In princi- volved in the 2005NIL. Monitoring showed that over a hy- ple, modelling evaporation requires coupling water and heat drological year the amount of AE occurs to the same order flows. Geotechnical engineers and geologists are still too un- of magnitude as that of infiltrating precipitation (hundreds of familiar with heat flow modelling in particular, as it entails a millimetres). number of thermal parameters and boundary conditions that Increasing efforts are being made to develop early warn- are difficult to calibrate and validate. In addition, governing ing systems to mitigate the risks of rainfall-induced land- equations need non-widespread numerical codes and their slides. Their success strongly depends on the performance high non-linearity involves difficulties in achieving numer- of the predictive models they implement in terms of timing ical solutions. This implies that in several applications evap- Nat. Hazards Earth Syst. Sci., 18, 613–631, 2018 www.nat-hazards-earth-syst-sci.net/18/613/2018/ A. Reder et al.: Physically based approaches incorporating evaporation for early warning predictions 615 oration is neglected because it is considered less important than rainfall intensity during a highly intense event that trig- gered landslides (e.g. Baum et al., 2008; Pagano et al., 2010; Formetta et al., 2016). In other applications evaporation is taken into account by following approaches with different degree of complexity (Casadei et al., 2003; Rosso et al., 2006; Šimunek et al., 2006; Ebel et al., 2010; Formetta et al., 2014; Capparelli and Versace, 2011; Arnone et al., 2011; En- drizzi et al., 2014). Complete approaches, modelling internal and boundary evaporation through hydrothermal approaches, were taken into account in studies referred to slopes in fine- grained soils differing substantially from those involved in the case at hand (Cui et al., 2005; An et al., 2017; Song et al., 2016). Concerning the hydrological behaviour of silty vol- Figure 2. Grain-size distribution of the Nocera Inferiore volcanic canic sloping covers, several authors adopted approaches in- ash (Pagano et al., 2010). corporating evaporation for the interpretation of monitoring results (Pirone et al., 2015a) and/or back analysis of previ- ous events (Greco et al., 2013; Napolitano et al., 2016). In validation procedure, of the findings retrieved by the above- such studies, however, evaporative fluxes were modelled as a mentioned lysimeter, involving 1 m3 of material forced by boundary phenomenon only. realistic boundary conditions provided by actual meteorolog- The question naturally arises whether, for silty volcanic ical evolution, instead of the traditional procedures based on sloping covers, the accuracy of the early warning prediction small specimen subject to artificial boundary conditions. A will be significantly reduced if evaporation were neglected, previous comparison among laboratory, lysimeter and field resulting in too many false alarms. conditions, restricted to retention properties of the same soil This study attempts to address this question by compar- involved in the work (Pirone et al., 2016), resulted in a sat- ing results yielded by three different mathematical-numerical isfactory agreement. This encourages throughout the work models, either taking evaporation into account as only su- considering representative lysimeter of field conditions also perficial or an internal phenomenon (Wilson et al., 1994) or for quantifying parameters needed to describe soil hydraulic neglecting it (Richards, 1931), in the interpretation of the conductivity and thermal behaviour. 2005NIL case study. The paper begins with a description of the case study Two of the models account for evaporation: one based on and presents the lysimeter data. After describing the selected a coupled (heat–water flow) approach and the other based