The Oxford Research Encyclopedia of Natural Hazard Science will be available via subscription on April 26. Visit About to learn more, meet the editorial board, learn about subscriber services.

Show Summary Details

Page of

PRINTED FROM the OXFORD RESEARCH ENCYCLOPEDIA, NATURAL HAZARD SCIENCE ( (c) Oxford University Press USA, 2019. All Rights Reserved. Personal use only; commercial use is strictly prohibited (for details see Privacy Policy and Legal Notice).

date: 24 March 2019

Remote Sensing and Physical Modeling of Fires, Floods, and Landslides

Summary and Keywords

Remotely sensed data for the observation and analysis of natural hazards is becoming increasingly commonplace and accessible. Furthermore, the accuracy and coverage of such data is rapidly improving. In parallel with this growth are ongoing developments in computational methods to store, process, and analyze these data for a variety of geospatial needs. One such use of this geospatial data is for input and calibration for the modeling of natural hazards, such as the spread of wildfires, flooding, tidal inundation, and landslides. Computational models for natural hazards show increasing real-world applicability, and it is only recently that the full potential of using remotely sensed data in these models is being understood and investigated. Some examples of geospatial data required for natural hazard modeling include:

• elevation models derived from RADAR and Light Detection and Ranging (LIDAR) techniques for flooding, landslide, and wildfire spread models

• accurate vertical datum calculations from geodetic measurements for flooding and tidal inundation models

• multispectral imaging techniques to provide land cover information for fuel types in wildfire models or roughness maps for flood inundation studies

Accurate modeling of such natural hazards allows a qualitative and quantitative estimate of risks associated with such events. With increasing spatial and temporal resolution, there is also an opportunity to investigate further value-added usage of remotely sensed data in the disaster modeling context. Improving spatial data resolution allows greater fidelity in models allowing, for example, the impact of fires or flooding on individual households to be determined. Improving temporal data allows short and long-term trends to be incorporated into models, such as the changing conditions through a fire season or the changing depth and meander of a water channel.

Keywords: remote sensing, computational modeling, model-data fusion, floods, wildfires, landslides


There are more than 1,400 man-made operational satellites in orbit as of December 31, 2016 (Union of Concerned Scientists, 2016). A significant number of these collect data that can be directly used for various applications including disaster management. The data is also valuable as input into physical models that predict natural hazards. This article is focused on the applications of remotely sensed data from the perspective of their use as input into natural hazard predictive models. The sophistication of physically based predictive models in disaster management applications has increased significantly, with several of these being able to provide real-time prediction of natural hazard events since early 2000. For example, Spark (Figure 1), a bushfire prediction capability developed by CSIRO (Miller, Hilton, Sullivan, & Prakash, 2014) is able to predict the spread of the bushfire perimeter significantly faster than real time using modern Graphical Processor Unit (GPU) based computational techniques.

Remote Sensing and Physical Modeling of Fires, Floods, and LandslidesClick to view larger

Figure 1. Wildfire simulation using Spark, CSIRO’s wildfire simulation toolkit. Color shading indicates the arrival time of the fire. The simulated fire is based on the Forcett bushfire in Tasmania that occurred on January 4, 2013. Backing map uses OpenStreetMap tile images.

However, the accuracy of these predictions depends significantly on the quality of the inputs provided to these physical models. One of the key inputs into natural hazard predictive models is data related to the topography of the region being assessed. This includes aspects such as the slope of the terrain, vegetation coverage, terrain roughness, water bodies, and man-made infrastructure. The ability to capture high resolution remotely sensed data including satellite and airborne sources has significantly improved the ability to make such predictions more accurate.

The three key areas of application that form the focus of this review article are fires, floods, and landslides since these require significant information gleaned from remotely sensed data to make the predictions reliable.

Some examples of geospatial data required for natural hazard modeling include:

  • elevation models derived from RADAR and Light Detection and Ranging (LiDAR) techniques for flooding, landslide and wildfire spread models

  • accurate vertical datum calculations from geodetic measurements for flooding and tidal inundation models

  • multispectral imaging techniques to provide land cover information for fuel types in wildfire models or roughness maps for flood inundation studies

The article has been organized according to the specific requirements of remotely sensed inputs for the three important classes of disasters being reviewed here: namely fires, floods, and landslides.


This entry is focused on exploring model data fusion approaches that integrate inputs from remotely sensed sources into physical models of natural hazards leading to future predictability of these natural hazards. In this context the earliest such studies from a flooding perspective can be traced back to Moore and Grayson (1991), in which the authors describe a technique to reduce an essentially complex three-dimensional flow problem into a series of coupled simpler one-dimensional problems in areas possessing complex terrain. Although these authors did not use any remotely sensed data sets in their study it could be regarded as an early version of investigating the use of such data as input into hydrodynamic and hydrological models. An early example of the integration of physics-based three-dimensional hydrodynamic modeling with moderate resolution remotely sensed data can be found in Debroux, Prakash, and Cleary (2001), where the authors demonstrate the use of such a technique for predicting the inundation of a hypothetical tsunami. The remote sensed data input for this modeling was obtained from the U.S. Geological Survey at a 30-meter resolution. The use of high-resolution LiDAR-based data as input into hydrodynamic modeling of floods is fairly recent, with relevant studies only being reported in the early to mid-2000s (Haile, 2005), where the spatial resolution for these studies was as high as 1.5 meters. For applications related to bushfire hazards one can trace the origins of the use of satellite- based remote-sensed sources as input into models in the early to mid-2000s. Keane, Burgan, and Wagtendonk (2001) is one of the early examples on the use of satellite-based data as an input for deriving spatial fire hazard risks. In the landslide risk assessment area early examples of integrating modeling with remote-sensed data can be traced back to the early to mid-1990s. An early example of such work can be found in Terlien, Van Westen, and van Asch (1995), which combines the use of physics-based deterministic models with satellite-based remotely sensed data for landslide hazard estimation in Costa Rica and Colombia. The following sections delve deeper into the three natural hazard categories of fires, floods, and landslides with a view to gaining insights into the latest developments from a model data fusion perspective.


Bushfires and wildfires are a significant threat to human populations, infrastructure, and livestock in several parts of the world including Australia, North America, South America, and some parts of Asia and Europe. Some of the most severe incidents such as the Black Saturday bushfires that occurred in Victoria, Australia, in 2009 (Teague, McLeod, & Pascoe, 2010) are a clear reminder of the significant threat associated with these natural hazards for life and property. There are several techniques used to understand and predict bushfire behavior. One of the key methodologies used is based on physical modeling of bushfire behavior to predict the propagation of the fire front. Several such models exist and are used for operational risk prediction and fire research purposes. For example, WRF-Fire (Mandel et al., 2014) is a combined weather research and forecasting (WRF) model coupled with a surface fire behavior model, SFIRE was developed jointly by the University of Colorado-Denver, the University of Utah, and the National Center for Atmospheric Research in the United States. PHOENIX Rapidfire (Pugnet, Chong, Duff, & Tolhurst, 2013) developed principally by Melbourne University, is a wildland fire simulator designed to characterize large fast moving fires mainly in the Australian context. Spark is a bushfire spread modeling framework developed by CSIRO (Miller, Hilton, Sullivan, & Prakash, 2014) designed for intense and open use by the fire research community as well as potential operational use by emergency managers and the insurance industry. All of the above predictive capabilities require remotely sensed data as input for various requirements, including accurately describing terrain slope, vegetation, and fuel type as well as presence of infrastructure that can affect fire behavior. The key focus of this section is the use of remotely sensed data for determining vegetation and fuel type since the prevailing fuel properties are fundamental to bushfire propagation speed (Sullivan, McCaw, Cruz, Matthews, & Ellis, 2012). Consequently, in order to accurately predict bushfire spread, it is necessary to have high-quality fuel information. However, such information is difficult and costly to collect, as it generally requires in situ physical assessment of fuels (see Gould, McCaw, & Cheney, 2011). However, such information is hard to collect given the large areas at risk from bushfires combined with the ever-changing nature of the environment and the land use. Where spatially explicit fuel information does exist, it is often collected sporadically due to the expense; thus it becomes out of date when needed. Furthermore, continued advances in bushfire behavior models can mean that previously collected fuel information is no longer sufficient or appropriate to run a new or revised bushfire behavior model (Cruz et al., 2015).

As satellite data becomes more accessible at higher qualities and resolutions, acquiring up-to-date detailed fuel information quickly and efficiently is becoming increasingly feasible for example (see Keane, Burgan, & Wagtendonk, 2001). This is chiefly due to the rapid development in sensing platforms and high-speed delivery of raw data. However, converting this raw data into usable information requires a high level of processing. In the field of bushfire management, much work has been done, for example, in developing specific data processing methods for determining grassland curing state (Newnham, Verbesselt, Grant, & Anderson, 2011). Frequently the products of these methods are static maps of fuel state applicable for a given period, but often the map is much older than the most recently available data due to the time and effort required to process it. Consequently, predictions of likely fire spread can be erroneous because they are not using up-to-date information. A method of providing rapid processing of raw remotely sensed data into timely fuel information for provision to bushfire spread prediction is therefore needed.

Before specific fuel attributes can be calculated it is necessary to know vegetation type, as the attributes required are vegetation-dependent. The first step toward providing fuel information directly through remote sensing is therefore the classification of the data into vegetation types. Image classification has recently started allowing the expansion of study areas for hazard research for both pre-event risk and post-event impact analysis. For example, high resolution satellite imagery allowed the authors of Thouret et al. (2014) to obtain infrastructure information for a large area in Peru in order to assess the city’s vulnerability to flash flooding and earthquakes. The use of image classification to provide fuel-type classifications has been tried for specific regions in Spain (Riaño, Chuvieco, Salas, Palacios-Orueta, & Bastarrika, 2002) and Hawaii (Varga & Asner, 2008). However, this has not yet been directly incorporated into a fire model.

Various methods can be used to perform image classification, with the optimal method being highly application dependent. Features to consider when choosing a method include allowable computation time, available training samples, contrast between classes, data resolution, available spectral bands, and existing auxiliary data. Following Lu and Weng (2007), we categorize the methods as per-pixel, contextual-based, knowledge-based, sub-pixel or per-field methods. It is also possible to combine methods in what is termed a “combinative” approach.

The most straightforward and frequently used classification methods are those in the per-pixel category. Per-pixel algorithms assume no spatial relationship between pixels. The more commonly used algorithms in the per-pixel category include maximum likelihood (ML), neural network (NN), decision tree (DT), and support vector machines (SVM). Of these four, ML is the simplest to implement; however, a study comparing ML, NN, and DT algorithms for land cover classification in England by Pal and Mather (2003) clearly found ML to be the worst performing and DT comparable to NN in quality but better in terms of speed. Alonso-Benito, Arroyo, Arbelo, Hernndez-Leal, and Gonzlez-Calvo (2013) performed a similar comparison between SVM, ML, and NN for mapping forest fuels on Tenerife Island, Spain, and found SVM methods produced the highest percentage of accuracies.

The above per-pixel methods can be referred to as “supervised algorithms.” This means the algorithm is trained on a set of known points and then run using the set of parameters determined by this training set. It is also possible to have unsupervised classification algorithms that create classes purely based on the distributions of the input data. Two such algorithms are ISODATA and K-means clustering. These methods are provided with a set of input parameters and iteratively partition data into clusters by determining appropriate cluster centroids. Although these algorithms do not tend to appear in the fuel classification literature, they are well-established methods (see Wu et al., 2007) for situations with insufficient training data.

Contextual- and knowledge-based methods use additional levels of information to the spectral data. In the case of contextual-based methods, this is further information about the data itself, such as local variation in values surrounding the pixel. In the case of knowledge-based methods, it is taken from auxiliary data that could influence classes, such as topography. Riaño, Chuvieco, Salas, Palacios-Orueta, and Bastarrika (2002) found the addition of a texture map and topography improved their supervised ML classifications when generating fuel maps for Spain from Landsat.

The last categories are sub-pixel and per-field approaches. Sub-pixel algorithms allow pixels to be split into multiple classifications. This is useful in situations where, for example, the data resolution is too large to capture small land features. Per-field approaches consider the image as a series of objects or fields. The image is divided into objects, sets of spatially connected pixels, and classification is done on the objects rather than individual pixels. An example of a sub-pixel classification method utilized to transform soft land cover classification into hard land cover classes can be found in Aplin and Atkinson (2001).

There are currently several limitations in the application of remote sensing data for fuel state classification. The satellite cannot see through forest canopy and hence cannot directly provide information on forest surface fuels. Additionally, current resolutions are still too low to provide information on small scale features, such as roads. However, satellite resolution is constantly increasing, and in the near future it is possible this latter issue will become obsolete. Regardless of these limitations, the benefits of automation as shown through the concept in Figure 2 and the accompanying ability to capture recent data sets (more frequent than every 16 days in some instances) means that there is enormous potential for remote sensing data for calculation of fuel and coverage data and changes in these features in open terrain.

Remote Sensing and Physical Modeling of Fires, Floods, and LandslidesClick to view larger

Figure 2. A concept showing the classification of remotely sensed data for the purposes of input into a wildfire modeling capability.


The severity and unpredictability of flood events is increasing at least in part due to human induced climate change (see Min, Zhang, Zweirs, & Heger, 2011; Pall et al., 2011) in many places across the globe and accurate models that are able to predict the impact of these floods are becoming increasingly important from a land-use planning, emergency management, and climate adaptation perspective. In coastal areas there is an additional imperative due to the potential combined effects of coastal and catchment flooding (see Prakash, Hilton, & Ramachandran, 2015). Accurate and high-quality inputs from remotely sensed data are crucial for predicting the effects of flooding in these locations including bathymetry, topography, man-made infrastructure locations, vegetation, and other attributes that can affect flood behavior.

Remotely sensed data has been used to derive roughness estimates for flood modeling, which in turn is important from a flood inundation accuracy perspective. For example, Forzeiri, Degetto, Righetti, Castelli, and Preti (2011) utilized multispectral data obtained from satellite sources to retrieve hydrodynamic parameters for vegetation classification as an input into a hydrodynamic model. The method was tested along a 3 km stretch of the Avisio River in Italy. Baugh, Bates, Schumann, and Trigg (2013) were successful in improving the accuracy of hydrodynamic model results through better quality Satellite Radar Topography Mission (SRTM) derived vegetation removal algorithms. Similarly, Forzieri, Guarnieri, Vivoni, and Castelli (2011) developed an automated roughness parametrization methodology for simulating forested flood plains. They used a combination of multispectral imagery and airborne laser scan data as input into the model.

In several studies remotely sensed data and physical flood modeling have been combined through a model data fusion approach to either better inform flood scenarios or reduce the uncertainty related to modeling and remotely sensed outputs. Masood and Takeuchi (2012) used a combination of Digital Surface Model (DSM) outputs and 1D hydrodynamic simulations to inform the effect that rapid land filling by land developers have on the flood outputs in mid-eastern Dhaka, Bangladesh. Chini et al. (2014) developed a new method for flood hazard mapping that combines coarse scale global flood inundation modeling approaches with microwave remote sensing. The use of the global inundation model allows an understanding of the flood scenario from a spatio-temporal continuity perspective, whereas the satellite based data provides the spatial resolution required for more accurately determining the flood hazard. Ning, Liu, and Bao (2013) used remotely sensed data for determining a spatially distributed manning roughness as an input into the hydrodynamic model, as well as to estimate the extent of inundation for validating the outputs obtained from the hydrodynamic model. Acuna-Pedrozo, Marino-Tapia, Enriquez, Mayoral, and Villareal (2012) used a combination of remotely sensed data, a 2D hydrodynamic model output and field measurements to evaluate inundation areas resulting from the diversion of an extreme river discharge toward the sea. The combined model data fusion approach was employed here primarily to reduce the uncertainty associated with model parametrization and outputs. Zhang et al. (2015) utilized Moderate Resolution Imaging Spectro-radiometer–based remote sensing data to calibrate the wetting and drying parameter in the hydrodynamic model used for the inundation study in China’s largest freshwater lake: Poyang Lake. The calibrated model was used for prediction and management of the freshwater resource as well as for water-quality-related issues.

Remotely sensed data has also been used in a flooding context specifically to improve the quality of the hydrodynamic model. Garcia-Pintado, Neal, Mason, Dance, and Bates (2013) used Satellite based Synthetic Aperture Radar (SAR) data to provide water-level observations then assimilated into a hydrodynamic model to decrease forecast uncertainty of the hydrodynamic model. Pasquale, Perona, Wombacher, and Burlando (2014) calibrated a hydrodynamic model using pattern recognition techniques applied to non-orthorectified terrestrial photographs. The authors claim that this methodology is a relatively cheaper way of accessing riverbed information when a direct measurement of flow variables is difficult to achieve. Akif Al, Billa, Biswajeet, Ahmed, and Samih (2011) coupled hydrodynamic modeling with photogrammetry derived digital surface models to simulate flood scenarios in Malaysia. Reid and Williams (2014) developed a methodology to apply remote sensing as a means to evaluate hydrodynamic model performance in predicting coastal inundation. Jung and Jasinski (2015) explored the sensitivity of a hydrodynamic model to satellite-based Digital Surface Model (DSM) scale and accuracy. Ticehurst et al. (2015) used a reverse approach of utilizing hydrodynamic modeling outputs to improve the accuracy of daily MODIS Open Water Likelihood (OWL) flood inundation estimations.

High-resolution LiDAR-based data for flood adaptation analysis is being used by CSIRO especially for coastal cities around Australia using its City Based Flood Adaptation Solutions Tool (C-FAST). An example comparing “before” and “after” adaptation scenarios for the low-lying Elwood canal region in the City of Port Phillip, Victoria is shown in Figure 3.

Remote Sensing and Physical Modeling of Fires, Floods, and LandslidesClick to view larger

Figure 3. The use of high-resolution LiDAR-based terrain and bathymetry data to evaluate “Before” and “After” adaptation flood scenarios. The case study shown here is for the Elwood canal region in the city of Port Phillip, Victoria, Australia. The flood scenario uses a 0.4-meter Sea Level Rise assumption. Flooding is colored by retention time here with blue indicating two hours of flooding or lesser and red indicating flood water retained for around 24 hours. The dark red region close to the shore is water retained in the bay.

For this case study, one-meter resolution terrain and bathymetry data were “stitched” together to obtain a seamless three-dimensional map of the region. This was used as input into the hydrodynamic simulation. The “before” scenario used existing infrastructure whereas the “after” scenario also included infrastructure that does not currently exist in the region but will be required if a level of flood adaptation is needed. Such high resolution data allows the prediction of flooding at a home scale or street-level detail. This level of detail is important from an infrastructure-planning and household-level risk- assessment perspective. Airborne LiDAR data is, however, still expensive to acquire, and moves are afoot to replace the use of such acquisitions with satellite-based sources. For example, Digital Globe (DigitalGlobe, 2016) has a suite of high resolution three-dimensional elevation model products called VRICON that is well suited to be used as an input into high resolution flood simulators. The challenge, however, lies in the acquisition and “stitching” together of high resolution bathymetric data with these three-dimensional land-based terrain models.

Satellites are still not capable of acquiring quality bathymetric data due to key issues related to the occlusion of underwater terrain features. Jagalingam, Akshaya, and Arkal (2015) demonstrated the use of freely available Landsat 8 data to obtain bathymetric data up to a depth of 20 meters. It remains to be seen, however, if such data sets can be usefully deployed as input into high resolution flood inundation models.

An example showing the series of steps required to clean up a high resolution one-meter LiDAR acquired terrain data with bathymetric data is demonstrated in Figure 4.

Remote Sensing and Physical Modeling of Fires, Floods, and LandslidesClick to view larger

Figure 4. Steps involved in cleaning up high resolution land based LiDAR and bathymetry data. Data shown is for the city of Greater Geelong, Victoria, Australia: (a) original unprocessed LiDAR for the Bellarine Peninsula overlaid on a color aerial map. The shade of gray indicates the terrain height (AHD) with dark being low and white being high. Lakes and rivers with erroneous heights of –100 meters are circled, (b) LiDAR data for regions including Swan Island identified as missing from the data set, (c) erroneous lake and river data removed from LiDAR data and Swan Island data composited together, and (d) Bathymetry data composited with the LiDAR data.

The region shown is the City of Greater Geelong in Victoria, Australia. In the first step, the high resolution LiDAR-based DSM is clipped using an outline shapefile that follows the coastline. The terrain data has several issues at this juncture, stage (a), with inland lakes, dams, and rivers included (marked with a red circle) but with erroneous heights associated with them of 100 meters, based on an Australian Height Datum (AHD). The solution to this issue as shown in stage (b) in Figure 4, is to set up a height threshold such that these inland bodies of water are removed from the DSM. There are, however, remaining issues related to significant islands (such as Swan Island) missing from the DSM. The solution here, as shown in stage (c) in Figure 4, was to combine the one-meter DSM with an additional LiDAR scan taken at one meter for the Swan Island and nearby bathymetry. This is relatively easy to achieve since the two datasets have the same resolution. Finally, the bathymetry data as shown in stage (d) in Figure 4 has a resolution of 2.5 meters and has significant missing data in it. Figure 5 shows how these remaining issues are resolved and the bathymetry and terrain data combined as input into a “street scale” flood inundation model.

Remote Sensing and Physical Modeling of Fires, Floods, and LandslidesClick to view larger

Figure 5. Steps involved in combining terrain and bathymetry data before input into a “street scale” flood inundation model. Data shown is for the city of Greater Geelong, Victoria, Australia: (a) terrain (green) and bathymetry (red-yellow) data at resolutions of one meter and 2.5 meters respectively, (b) data resampled at a single resolution (2.5 meters) onto a single raster grid, and (c) missing data at the interface is reconstructed smoothly using a minimal surface Laplacian operation.

The three key issues that need to be dealt with as shown in stage (a) in Figure 5 are the difference in resolution between the terrain (1 meter) and bathymetry (2.5 meter) data, missing data and overlaps in the two data sets. As shown in stage (b) to resolve these issues we firstly assume that the terrain takes precedence at the overlaps and resample both data sets at the same resolution. Finally, as shown in stage (c), we fill holes in the data sets using a “minimal” surface curve fitting algorithm. The combined DSM in stage (c) is now ready as input into a high resolution flood simulation solver. The demonstration shown here is just an example of the requirements for high resolution flood inundation models. There are several other cases where the issues related to bathymetry and terrain might need to be resolved slightly differently for example in case of inland lakes with missing bathymetry.


Many remote-sensing-based solutions have been developed for landslide hazard zonation, and early models were discussed and reviewed, for example, in Van Westen (2000) and Dietrich, Bellugi, and Real De Asua (2001), where such an approach was also provided in the context of forest management. Other early attempts at combining remote sensing and deterministic modeling were proposed in Xie, Esaki, Zhou, and Mitani (2003) and also in Venkatachalam, Nagesha, and Dodagoudar (2002) where remote sensing was combined with an infinite slope method to assign factor of safety values on a DSM.

The study of landslides requires the knowledge of a relatively large number of parameters such as slope, soil type, vegetation, soil depth and local conditions such as rainfall data. It is generally necessary to acquire such a diverse range of data from a variety of sources. A recent survey conducted by Tofani, Segoni, Agostini, Catani, and Casagli (2013) investigated the current use of remote sensing technologies in landslide studies in Europe. Among the most commonly used were aerial photography, satellite radar and optical data, meteorological sensors, LiDAR (airborne and terrestrial). Obviously which technique to use depends on the particular type of landslide under assessment as well as the spatial and temporal scales of interest. Often more than one method must be used to achieve meaningful results.

Jaboyedoff et al. (2012) discussed LiDAR technologies for landslide studies, and applications were classified in four categories: (1) detection and characterization of mass movements, (2) hazard assessment and susceptibility mapping, (3) modeling, and (4) monitoring. For modeling, LiDAR is mostly used to create accurate High Resolution Digital Surface Models (HRDSM), and Airborne Laser Scanning (ALS)-DSM has been applied to the problem of shallow landslide susceptibility mapping and modeling using infinite slope stability models. As highlighted in Jaboyedoff et al. (2012), higher resolution data is critical in improving the accuracy of a model. Similarly, detailed modeling of landslides and debris flow can be greatly improved by the use of HRDSM.

Twenty-first-century models have been proposed to produce landslide hazard maps from remote sensing data as discussed in Pardeshi, Autade, and Pardeshi (2013). For example, Pradhan and Lee (2009) linked a Geographic Information System (GIS) database and satellite images to a back-propagation neural network algorithm for the generation of a landslide map in Penang Island, Malaysia, while Lee (2007) applied a similar method to the Gangneung area in South Korea. In general, machine learning algorithms are gaining popularity for landslide susceptibility (LS) mapping, and several such methods were tested in Micheletti et al. (2014). It was found that these methods produced accurate LS results for shallow landslides but that the uncertainty associated with deep seated landslides was higher. Using a fuzzy logic model, Pradhan (2010) considered 10 landslide inducing factors and combined sources from satellite data, aerial photographs, and field surveys. Different fuzzy logic operators were tested, with the best results obtained at 91% for a gamma operator.

With the increased availability of accurate remotely sensed data, much effort has been devoted to the development of early warning systems both at global and regional scales. Hong, Adler, and Huffman (2007) proposed a global scale landslide susceptibility map based on satellite data. The tool uses a simple approach to assign susceptibility values based on a weighted linear combination method of six layers known to play a role in landslide risk (e.g., slope). This allows for an estimate of the zones at higher risk, which could then be subjected to more detailed analysis if required. Rainfall induced landslides account for a large number of landslides occurrence worldwide therefore acquisition of precipitation data is of paramount importance in this context. The same research group of Hong, Adler, Negri, and Huffman (2007) also proposed a framework to develop an early warning system for rainfall-induced landslides. The system was based on satellite data and in particular microwave imagers provided by the Tropical Rainfall Measuring Mission (TRMM) multi-satellite for rainfall data. The framework is an integrated system where satellite data for the topography, land cover, etc. are overlain with the rainfall data to estimate the risk of landslide occurrence through an empirical rainfall intensity-duration threshold. Similar conceptual frameworks have been proposed in other references that differ by the details of the hydro-mechanical model developed. For example, in Liao, Hong, Kirschbaum, and Liu (2012) the authors proposed a physically based model based on infinite-slope equilibrium equations. The model was in good agreement with observations on shallow landslides for the case study presented however the model was limited by a number of simplifying assumptions discussed by the authors, such as the uniformity of geological structures of slopes. Rossi, Catani, Leoni, Segoni, and Tofani (2013) also discussed the development of a physically based hydro-geological model for shallow landslides, also incorporating Monte Carlo simulations. The authors put an emphasis on parallel implementation and runtime performance as it was acknowledged that such models can only be operationally useful if the computational cost is not prohibitive. Kirschbaum et al. (2012) combined static landslide susceptibility with satellite-based rainfall data and applied the model to a region affected by Hurricane Mitch as a case study. This study highlighted the current challenges in the development of these models, such as improvement of Intensity-Duration thresholds, inclusion of soil moisture in models, and a better mechanistic (physically based) descriptions.

In fact, despite significant advances in recent decades, several aspects of the physics of landslide remain poorly understood. For example, the run-out distance of large landslides is frequently observed to be unexpectedly large and require low friction coefficient in order for numerical simulations to be in agreement with field data (see Lucas, Mangeney, & Ampuero, 2014). The advent of accurate remotely sensed data provides an opportunity to investigate the physics of landslides beyond laboratory experiments thus eliminating scaling-related difficulties (see Ekström & Stark, 2013; Lucas, Mangeney, & Ampuero, 2014). Figure 6a shows an example of a large-scale controlled experiment and associated two-dimensional model developed at CSIRO using an advanced particle based computational technique (see Lemiale, Karantgis, & Broadbridge, 2014). Although such studies are useful to validate and calibrate a model and to undertake systematic analysis of individual parameters, they ignore important aspects observed on real landslides such as the effect of local topography as illustrated in Figure 6b.

Remote Sensing and Physical Modeling of Fires, Floods, and LandslidesClick to view larger

Figure 6. (a) Two-dimensional model of controlled landslide experiment located in Japan using a particle-based method where particles have been colored by their velocity (Lemiale, Karantgis, & Broadbridge, 2014); (b) post-failure topographic profile after the 2014 Jure Landslide in Nepal (Nepal Government Ministry of Irrigation, 2014).

The use of remote sensing in such cases is essential to develop realistic three-dimensional models.

Indeed, at the scale of individual slopes, remote sensing is finding applications in the accurate determination of essential characteristics of a landslide that can then be used in detailed modeling of a specific site. Booth, Lamb, Avouac, and Delacourt (2013) proposed an inversion method that allowed the thickness of a slow-moving landslide to be determined from one-meter resolution digitized aerial photography and calculation of local elevation. They were also able to estimate the frictional strength of the failure plane from simple slope stability calculation. This type of approach can lead to entire three-dimensional reconstruction of a site from remote sensing data acquisition.

Remote sensing is also useful to allow for an accurate representation of local conditions for risk assessment purposes. For example, Delaney and Evans (2015) analyzed the dynamics of a large landslide dam in China through a combination of freely available 90-meter resolution Shuttle Radar Topography Mission (SRTM-3) topographic data and two- and three- dimensional dynamic landslide modeling.

Miller et al. (2012) an embankment at a railway site in the United Kingdom was analyzed for stability using a hydro-mechanical modeling approach and various remotely sensed data such as Airborne Laser Scanning (ALS) and Compact Airborne Spectrographic Imaging (CASI). The ultimate aim of this research was to minimize the need for periodic on-site inspection by pre-determining areas most at risk of failure, thereby reducing the cost of infrastructure maintenance.

Remote sensing has been proposed as a way to provide a rapid assessment of landslide dams. In Dong et al. (2014) a pre-landslide Digital Terrain Model (DTM) was combined with orthorectified remote sensing images to determine the geometry of a landslide dam. It was found that a 10-meter resolution or greater for the post-landslide images (which they were able to achieve using the two-meter resolution FORMOSAT-2 satellite images) combined with 40-meter resolution DTMs yielded satisfactory results. The determined geometry of the dam was then used to determine the risk of dam failure and potential spatial impact of a flood resulting from a dam break. However, the authors pointed out that their method was difficult to implement for short lived landslide dams formed due to heavy rainfall because the weather conditions usually prevent from capturing usable images of the site immediately.

Another application of remote sensing to the study of impact flooding from landslide dam breaks was presented in Butt, Umar, and Qamar (2013). Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) data were used to estimate the volume of water stored by the natural dam and also to create a Triangulated Irregular Network (TIN) of the region. Both were subsequently used as an input in a flood analysis conducted with the one-dimensional Hydrologic Engineering Center River Analysis System (HEC-RAS) model. Other satellite data such as Landsat-7 ETM+ were also used in this study. Different scenarios were investigated with the model.


There have been significant advances in remote sensing technologies in recent years leading to the availability of geospatial data sets at higher resolutions and with more content than ever before. At the same time computational modeling capabilities have also significantly advanced especially in relation to availability of relatively cheap and desktop- or laptop-based high-performance computing platforms with significant improvements in Graphical Processing Unit (GPU)–based computing power. The confluence of these technologies is now leading to an explosion in model data fusion approaches that can lead to significant improvements in our ability to predict, manage, and mitigate against various forms of natural disasters. A review of the literature encompassing fire-, flood-, and landslide-related natural hazards has demonstrated that there have so far been few instances of explicitly attempting to utilize computational modeling as the underlying need to further remote sensing technology. Instead research has focused on (for good reasons) either furthering remote sensing technology to derive data sets and analytics that are themselves of significant value or attempting to “force fit” data sets derived from remotely sensed sources as input into computational models. Going forward the ability to optimally and successfully assimilate the advances made in these two areas will significantly depend on researchers working in these parallel domains collaborating closely with each other to ensure the model data fusion process occurs that can be deemed to be “fit-for-purpose.” As an example in the “fire modelling” domain it is now essential to perform research in the remote sensing domain that is focused on developing sensors that are able to automatically derive land classification information at a sufficient resolution and quality that is then able to be used in wildfire spread models. This would then lead to a step change in our ability to predict, assess, and mitigate natural disasters.


Acuna-Pedrozo, A., Marino-Tapia, I., Enriquez, C., Mayoral, G. M., & Villareal, F. J. (2012). Evaluation of inundation areas resulting from the diversion of an extreme discharge towards the sea: Case study in Tabasco, Mexico. Hyrdological Processes, 26, 687–704.Find this resource:

Akif Al, F., Billa, L., Biswajeet, P., Ahmed, M. T., & Samih, R. (2011). Coupling of hydrodynamic modeling and aerial photogrammetry-derived digital surface model for flood simulation scenarios using GIS: Kuala Lumpur flood, Malaysia. Disaster Advances, 4, 20–28.Find this resource:

Alonso-Benito, A., Arroyo, L. A., Arbelo, M., Hernndez-Leal, P., & Gonzlez-Calvo, A. (2013). Pixel and object-based classification approaches for mapping forest fuel types in Tenerife Island from ASTER data. International Journal of Wildland Fire, 22(3), 306–317.Find this resource:

Aplin, P., & Atkinson, P. M. (2001). Sub-pixel land cover mapping for per-field classification. International Journal of Remote Sensing, 22(14), 2853–2858.Find this resource:

Baugh, C. A., Bates, P. D., Schumann, G., & Trigg, M. A. (2013). SRTM vegetation removal and hydrodynamic modeling accuracy. Water Resources Research, 49, 5276–5289.Find this resource:

Booth, A. M., Lamb, M. P., Avouac, J.-P., & Delacourt, C. (2013). Landslide velocity, thickness, and rheology from remote sensing: La Clapière landslide, France. Geophysical Research Letters, 40, 4299–4304.Find this resource:

Butt, M. J., Umar, M., & Qamar, R. (2013). Landslide dam and subsequent dam-break flood estimation using HEC-RAS model in Northern Pakistan. Natural Hazards, 65, 241–254.Find this resource:

Chini, M., Giustarini, L., Matgen, P., Hostache, R., Pappenberger, F., & Bally, P. (2014). Flood hazard mapping combining high resolution multi-temporal SAR data and coarse resolution global hydrodynamic modelling. 2014 IEEE Geoscience and Remote Sensing Symposium (pp. 2994–2996). Quebec City, QC: IEEE.Find this resource:

Cruz, M. G., Gould, J. S., Alexander, M. E., Sullivan, A. L., McCaw, W. L., & Matthews, S. (2015). Empirical based models for predicting head-fire rate of spread in Australian fuel types. Australian Forestry, 78(3), 118–158.Find this resource:

Debroux, F., Prakash, M., & Cleary P, W. (2001). Three-dimensional modelling of a tsunami interacting with real topographical coastline using Snoothed Particle Hydrodynamics. 14th Australasian Fluid Mechanics Conference (pp. 311–314). Adelaide.Find this resource:

Delaney, K. B., & Evans, S. G. (2015). The 2000 Yigong landslide (Tibetan Plateau), rockslide-dammed lake and outburst flood: Review, remote sensing analysis, and process modelling. Geomorphology, 246, 377–393.Find this resource:

Dietrich, W. E., Bellugi, D., & Real De Asua, R. (2001). Validation of the Shallow Landslide Model, SHALSTAB, for forest management. In M. S. Wigmosta & S. J. Burges (Eds.), Land use and watersheds: Human influence on hydrology and geomorphology in urban and forest areas (pp. 195–227). Washington, DC: American Geophysical Union.Find this resource:

DigitalGlobe. (2016, November 8). Vricon 3D and elevation products. Retrieved from

Dong, J.-J., Lai, P.-J., Chang, C.-P., Yang, S.-H., Yeh, K.-C., Liao, J.-J., & Y-W, P. (2014). Deriving landslide dam geometry from remote sensing images for the rapid assessment of critical parameters related to dam-breach hazards. Landslides, 11, 93–105.Find this resource:

Ekström, G., & Stark, C. P. (2013). Simple scaling of catastrophic landslide dynamics. Science, 339, 1416–1419.Find this resource:

Forzeiri, G., Degetto, M., Righetti, M., Castelli, F., & Preti, F. (2011). Satellite multispectral data for improved floodplain roughness modelling. Journal of Hydrology, 407(1–4), 41–57.Find this resource:

Forzieri, G., Guarnieri, L., Vivoni, E. R., Castelli, F., & Preti, F. (2011). Spectral-ALS data fusion for different roughness parameterizations of forested floodplains. River Research and Applications, 27, 826–840.Find this resource:

Garcia-Pintado, J., Neal, J. C., Mason, D. C., Dance, S. L., & Bates, P. D. (2013). Scheduling satellite-based SAR acquisition for sequential assimilation of water level observations into flood modelling. Journal of Hydrology, 495, 252–266.Find this resource:

Gould, J. S., McCaw, W. L., & Cheney, N. P. (2011). Quantifying fine fuel dynamics and structure in dry eucalypt forest (Eucalyptus marginata) in Western Australia for fire management. Forest Ecology and Management, 262(3), 531–546.Find this resource:

Haile, A. T. (2005). Integrating hydrodynamic models and high resolution DEM (LIDAR) for flood modelling. Enschede, Netherlands: International Institute for GeoInformation Science and Earth Observation.Find this resource:

Hong, Y., Adler, R., & Huffman, G. (2007). Use of satellite remote sensing data in the mapping of global landslide susceptibility. Natural Hazards, 43, 245–256.Find this resource:

Hong, Y., Adler, R., Negri, A., & Huffman, G. (2007). Flood and landslide applications of near real-time satellite rainfall products. Natural Hazards, 43, 285–294.Find this resource:

Jaboyedoff, M., Oppikofer, T., Abellán, A., Derron, M.-H., Loye, A., Metzger, R., & Pedrazzini, A. (2012). Use of LIDAR in landslide investigations: A review. Natural Hazards, 61, 5–28.Find this resource:

Jagalingam, P., Akshaya, B. J., & Arkal, V. H. (2015). Bathymetry mapping using landsat 8 satellite imagery. Procedia Engineering, 116, 560–566.Find this resource:

Jung, H. C., & Jasinski, M. F. (2015). Sensitivity of a floodplain hydrodynamic model to satellite-based DEM scale and accuracy: Case study—the Atchafalaya Basin. Remote Sensing, 7, 7938–7958.Find this resource:

Keane, R. F., Burgan, R., & Wagtendonk, J. V. (2001). Mapping wildland fuels for fire management across multiple scales: Integrating remote sensing, GIS and biophysical modelling. International Journal of Wildland Fire, 10(4), 301–319.Find this resource:

Kirschbaum, D. B., Adler, R., Hong, Y., Kumar, S., Peters-Lidard, C., & Lerner-Lam, A. (2012). Advances in landslide nowcasting: evaluation of a global and regional modeling approach. Environmental Earth Sciences, 66, 1683–1696.Find this resource:

Lee, S. (2007). Landslide susceptibility mapping using an artificial neural network in the Gangneung area, Korea. International Journal of Remote Sensing, 28, 4763–4783.Find this resource:

Lemiale, V., Karantgis, L., & Broadbridge, P. (2014). Smoothed particle hydrodynamics applied to the modelling of landslides. Applied Mechanics and Materials, 553, 519–524.Find this resource:

Liao, Z., Hong, Y., Kirschbaum, D., & Liu, C. (2012). Assessment of shallow landslides from Hurricane Mitch in central America using a physically based model. Environmental Earth Sciences, 66, 1697–1705.Find this resource:

Lu, D., & Weng, Q. (2007). A survey of image classification methods and techniques for improving classification performance. International Journal of Remote Sensing, 28, 823–870.Find this resource:

Lucas, A., Mangeney, A. P., & Ampuero, J. (2014). Frictional velocity-weakening in landslides on earth and on other planetary bodies. Nature Communications, 5, 3417.Find this resource:

Mandel, J., Amram, S., Beezley, J. D., Kelman, G., Kochanski, A. K., & Kondratenko, V. Y., et al. (2014). Recent advances and applications of WRF–SFIRE. Natural Hazards and Earth System Science, 14, 2829–2845.Find this resource:

Masood, M., & Takeuchi, K. (2012). Assessment of flood hazard, vulnerability and risk of mid-eastern Dhaka using DEM and 1D hydrodynamic model. Natural Hazards, 61, 757–770.Find this resource:

Micheletti, N., Foresti, L., Robert, S., Leuenberger, M., Pedrazzini, A., Jaboyedoff, M., & Kanevski, M. (2014). Machine learning feature selection methods for landslide susceptibility mapping. Mathematical Geosciences, 46, 33–57.Find this resource:

Miller, C., Hilton, J., Sullivan, A., & Prakash, M. (2014). SPARK—A bushfire spread prediction tool. IFIP Advances in Information and Communication Technology (pp. 262–271). Melbourne, Australia: Springer.Find this resource:

Miller, P. E., Mills, J. P., Barr, S. L., Birkinshaw, S. J., Hardy, A. J., & Parkin, G., et al. (2012). A remote sensing approach for Landslide Hazard Assessment on engineered slopes. Geoscience and Remote Sensing, 50, 1048–1056.Find this resource:

Min, S.-K., Zhang, X., Zweirs, F. W., & Heger, G. C. (2011). Human contribution to more-intense precipitation extremes. Nature, 470, 378–381.Find this resource:

Moore, I. D., & Grayson, R. (1991). Terrain based catchment partitioning and run off prediction using vector elevation data. Water Resources Research, 27, 1177–1191.Find this resource:

Nepal Government Ministry of Irrigation. (2014). Report on Jure Landslide, Mankha VDC, Sindhupalchowk District. Report Number: 2071/05/08 BS. Available online.

Newnham, G. J., Verbesselt, J., Grant, I. F., & Anderson, S. A. (2011). Relative greenness index for assessing curing of grassland fuel. Remote Sensing of Environment, 115(6), 1456–1463.Find this resource:

Ning, L., Liu, H., & Bao, A. (2013). Identification of Inundation Hazard Zones in Manas Basin, China, using hydrodynamic modelling and remote sensing. Journal of Water Resource and Protection, 5, 469–473.Find this resource:

Pal, M., & Mather, P. M. (2003). An assessment of the effectiveness of decision tree methods for land cover classification. Remote Sensing of Environment, 86, 554–565.Find this resource:

Pall, P., Aina, T., Stone, D. A., Stott, P. A., Nozawa, T., & Hilberts, A. G., et al. (2011). Anthropogenic greenhouse gas contribution to flood risk in England and Wales in autumn 2000. Nature, 470, 382–385.Find this resource:

Pardeshi, S., Autade, S., & Pardeshi, S. (2013). Landslide hazard assessment: Recent trends and techniques. SpringerPlus, 2, 1–11.Find this resource:

Pasquale, N., Perona, P., Wombacher, A., & Burlando, P. (2014). Hydrodynamic model calibration from pattern recognition of non-orthorectified terrestrial photographs. Computers and Geosciences, 62, 160–167.Find this resource:

Pradhan, B. (2010). Application of an advanced fuzzy logic model for landslide susceptibility analysis. International Journal of Computational Intelligence Systems, 3, 370–381.Find this resource:

Pradhan, B., & Lee, S. (2009). Landslide risk analysis using artificial neural network model focussing on different training sites. International Journal of Physical Sciences, 4, 1–15.Find this resource:

Prakash, M., Hilton, J., & Ramachandran, L. (2015). Integrating hydrodynamic and hydraulic modeling for evaluating future flood mitigation in urban environments. In R. Denzer, R. M. Argent, G. Schimak, & J. Hrebicek, Environmental software systems, infrastructure, services and applications (pp. 282–292). Melbourne, Australia: Springer.Find this resource:

Pugnet, L., Chong, D. M., Duff, T. J., & Tolhurst, K. G. (2013). Wildland–urban interface (WUI) fire modelling using PHOENIX Rapdifire: A case study in Cavaillon France. 20th International Congress on Modelling and Simulation (pp. 228–234). Adelaide, Australia. Retrieved from this resource:

Reid, S. K., & Williams, D. D. (2014). Methodology for applying GIS to evaluate hydrodynamic model performance in predicting coastal inundation. Journal of Coastal Research, 30, 1055–1065.Find this resource:

Riaño, D., Chuvieco, E., Salas, J., Palacios-Orueta, A., & Bastarrika, A. (2002). Generation of fuel type maps from landsat TM images and ancillary data in Mediterranean ecosystms. Canadian Journal of Forest Research, 32(8), 1301–1315.Find this resource:

Rossi, G., Catani, F., Leoni, L., Segoni, S., & Tofani, V. (2013). HIRESSS: A physically based slope stability simulator for HPC applications. Natural Hazards and Earth System Sciences, 13, 151–166.Find this resource:

Sullivan, A. L., McCaw, W. L., Cruz, M. G., Matthews, S., & Ellis, P. F. (2012). Fuel, fire weather and fire behaviour in Australian ecosystems. In R. A. Bradstock, A. M. Gill, & R. D. Williams (Eds.), Flammable Australia: Fire regimes, biodiversity and ecosystems in a changing world (pp. 51–77). Melbourne, Australia: CSIRO.Find this resource:

Teague, B., McLeod, R., & Pascoe, S. (2010). 2009 Victorian Bushfires Royal Commission, Final Report Summary. Melbourne, Australia. Retrieved from this resource:

Terlien, M. T., Van Westen, C. J., & van Asch, T. W. (1995). Deterministic modelling in GIS based landslide hazard assessment. New York: Springer.Find this resource:

Thouret, J. C., Ettinger, S., Cuitton, M., Santoni, O., Magill, C., & Martelli, K., et al. (2014). Assessing physical vulnerability in large cities exposed to flash floods and debris flows: The case of Arequipa (Peru). Natural Hazards, 73(3), 1771–1815.Find this resource:

Ticehurst, C., Dutta, D., Karim, F., Petheram, C., & Guerschman, J. P. (2015). Improving the accuracy of daily MODIS OWL flood inundation mapping using hydrodynamic modelling. Natural Hazards, 78, 803–820.Find this resource:

Tofani, V., Segoni, S., Agostini, A., Catani, F., & Casagli, N. (2013). Technical note: Use of remote sensing for landslide studies in Europe. Natural Hazards and Earth System Science, 13, 299–309.Find this resource:

Union of Concerned Scientists. (2016). Retrieved from

Van Westen, C. (2000). The modelling of landslide hazards using GIS. Surveys in Geophysics, 21, 241–255.Find this resource:

Varga, T. A., & Asner, G. P. (2008). Hyperspectral and lidar remote sensing of fire fuels in Hawaii volcanoes national park. Ecological Applications, 18(3), 613–623.Find this resource:

Venkatachalam, G., Nagesha, M., & Dodagoudar, G. (2002). Landslide modelling using remote sensing and GIS. Geoscience and Remote Sensing Symposium, 4, 2045–2047.Find this resource:

Vu, T. T., Nguyen, P. K., Chua, L. H., & Law, A. W. (2015). Two dimensional hydrodynamic modelling of flood inundation for a part of the Mekong river with TELEMAC-2D. British Journal of Environment and Climate Change, 5, 162–175.Find this resource:

Wu, X., Kumar, V., Quinlan, J. R., Ghosh, J., Yang, Q., & Motoda, H., et al. (2007). Top 10 algorithms in data mining. Knowledge and Information Systems, 14(1), 1–37.Find this resource:

Xie, M., Esaki, T., Zhou, G., & Mitani, Y. (2003). Geographic Information Systems-based three-dimensional critical slope stability analysis and landslide hazard assessment. Journal of Geotechnical and Geoenvironmental Engineering, 129, 1109–1118.Find this resource:

Zhang, P., Lu, J., Feng, L., Chen, X., Zhang, L., & Xiao, X., et al. (2015). Hydrodynamic and inundation modeling of China’s largest freshwater lake aided by remote sensing data. Remote Sensing, 7, 4858–4879.Find this resource: