Physically-based landslide prediction over a large region: Scaling low-resolution hydrological model results for high-resolution slope stability assessment

Highlights

• Downscaling soil moisture using topographical attributes.

• Coupling hydrological model and slope stability model with different spatial resolutions.

• Developed a useful and efficient method to conduct landslide hazard prediction.

• The coupled model is more robust than the conventional rainfall threshold method.

Abstract

Rainfall-triggered shallow landslides are widespread natural hazards around the world, causing many damages to human lives and property. In this study, we focused on predicting landslides in a large region by coupling a 1 km-resolution hydrological model and a 90 m-resolution slope stability model, where a downscaling method for soil moisture via topographic wetness index was applied. The modeled hydrological processes show generally good agreements with the observed discharges: relative biases and correlation coefficients at three validation stations are all <20% and >0.60, respectively. The derived scaling law for soil moisture allows for near-conservative downscaling of the original 1-km soil moisture to 90-m resolution for slope stability assessment. For landslide prediction, the global accuracy and true positive rate are 97.2% and 66.9%, respectively. This study provides an effective and computationally efficient coupling method to predict landslides over large regions in which fine-scale topographical information is incorporated.

Introduction

Rainfall-triggered shallow landslides are worldwide natural hazard (Hong et al., 2006; An et al., 2016), causing a large number of fatalities and property losses (Liao et al., 2010; He et al., 2016; Godt et al., 2009). Globally, landslide hazards cause approximately 1000 deaths and USD 4 billion losses per year (Pradhan and Youssef, 2010). China is severely affected by landslide hazards, which has led to 1100 fatalities and 5–10 billion US dollars since 2000 (Hong et al., 2017). Due to frequent fatalities and extensive property damages that landslide hazard may cause (Papathoma-Kohle et al., 2015), it is essential to investigate and predict landslide hazards to avoid the damages (Zhang et al., 2019). Methods for hazard assessment range from heuristic susceptibility approaches (Fookes, 1997; Guzzetti et al., 2000; Griffiths and Edwards, 2001; Griffiths, 2002) to lumped regional empirical rainfall thresholds (Bogaard and Greco, 2018; Glade et al., 2000; Caine, 1980; Guzzetti et al., 2007, 2008) to detailed physical-based coupled hydrological-slope stability models (He et al., 2016; Zhang et al., 2016; Wilkinson et al., 2002; Alvioli and Baum, 2016; Montrasio and Valentino, 2016a, b; Van Asch et al., 2007). However, most of the studies focused on a single slope or landslide event in a relatively small catchment, usually in the magnitude of 10²–10³ km² in a resolution of 3 arcsec (~90 m). This resolution is also a prerequisite to validate model results against observed landslides from inventories (Tian et al., 2008). Yet, the corresponding computational burden jeopardizes the applicability of physically-based modeling to larger area (e.g., basins≫ 10,000 km²) (Bellugi et al., 2011; Camilo et al., 2017) and to derive necessary model parameters (Bardossy and Singh, 2008; Yao et al., 2012). In order to apply physically-based models in landslide prediction with high resolution, some computational expediency is needed in large basins.

Low-resolution but large scale hydrological models (typically applied at 1 km resolution or coarser) are widely available due to increasing availability of climate data and land surface datasets (Sun et al., 2017; Huang et al., 2019; Yao et al., 2019; van Beek et al., 2011; Wada et al., 2011; Luo et al., 2018). These models can be calibrated against observed discharge (Beck et al., 2016; Chao et al., 2018) so it captures the spatiotemporal variations across the basin. However, this coarse resolution does not fit the purpose for validation of landslide prediction because the size of a rainfall-triggered shallow landslide event is usually only tens or hundreds m² (Chen et al., 2017; Zhang et al., 2018a). In this case, a downscaling method should be used in order to couple coarse-resolution hydrological models with fine-resolution slope stability models.

Soil moisture is an important component in water and energy balance, affecting subsurface flow, soil evapotranspiration, hydrological response of a catchment etc., as a linking variable between hydrological model and slope stability model (Bogaard and Greco, 2016; Zhang et al., 2019). For slope stability model, the degree of soil saturation affects ground water table response to rainfall and thereby slope stability (Talebi et al., 2007; Bogaard and Greco, 2016; Krzeminska et al., 2012). Several studies showed that topography could be a good indicator for spatial soil moisture distribution (Beaudette et al., 2013; Burt and Butcher, 1985; Sveditchnyi et al., 2003). The topographic wetness index (TWI), as one of the topographic indexes, shows a high correlation with soil moisture during wet conditions (Grayson and Western, 1998; Brocca et al., 2010). Therefore, it is reasonable to downscale the soil moisture via using TWI to link hydrological model and slope stability model under different spatial resolution.

The main objective of this study is to assess the potential of physically-based landslide hazard assessment on the scale of 100,000 km². In this study we use topographic information to downscale the results of the coarse-scale hydrological model to the finer resolution slope stability model in an efficient and expedient manner. The hydrological model we used is the Coupled Routing and Excess STorage (CREST) model (Wang et al., 2011), a distributed hydrological model, which has been applied at a resolution of 1 km to the Shaanxi Province in Northwest China for an area of more than 200,000 km² for the period of 2009–2012. We downscaled the soil moisture information via TWI and land surface conditions to a finer resolution of 90 m. For the landslide assessment, we used a physically-based landslide model, namely, the SLope-Infiltration-Distributed Equilibrium (SLIDE) model (Montrasio and Valentino, 2008; He et al., 2016), which used the result of downscaled soil moisture as input variable. The results of the landslide hazard assessment are compared to classical lumped regional rainfall thresholds to quantify the possible improvement.

This article is organized as follows. It firstly describes the study area and datasets in Section 2. Then a brief introduction of hydrological model and slope stability model along with the downscaling method are described in Section 3. Section 4 presents results of our research and hence a comparison between the results and the results of landslide prediction using classic rainfall threshold is discussed in section 5. Conclusions, limitations of the research and its potential improvements are described in the end.