Using Statistical Learning Algorithms in Regional Landslide Susceptibility Zonation with Limited Landslide Field Data
Yi-ting Wang, Harry Seijmonsbergen, Willem Bouten, Qing-tao Chen, 2015 in J. Mt. Sci. (2015) 12(2): 268-288
Regional Landslide Susceptibility Zonation (LSZ) is always challenged by the available amount of field data, especially in southwestern China where large mountainous areas and limited field information coincide. Statistical learning algorithms are believed to be superior to traditional statistical algorithms for their data adaptability. The aim of the paper is to evaluate how statistical learning algorithms perform on regional LSZ with limited field data. The focus is on three statistical learning algorithms, Logistic Regression (LR), Artificial Neural Networks (ANN) and Support Vector Machine (SVM). Hanzhong city, a landslide prone area in southwestern China is taken as a study case. Nine environmental factors are selected as inputs. The accuracies of the resulting LSZ maps are evaluated through landslide density analysis (LDA), receiver operating characteristic (ROC) curves and Kappa index statistics. The dependence of the algorithm on the size of field samples is examined by varying the sizes of the training set. The SVM has proven to be the most accurate and the most stable algorithm at small training set sizes and on all known landslide sizes. The accuracy of SVM shows a steadily increasing trend and reaches a high level at a small size of the training set, while accuracies of LR and ANN algorithms show distinct fluctuations. The geomorphological interpretations confirm the strength of SVM on all landslide sizes. Our results show that the strengths of SVM in generalization capability and model robustness make it an appropriate and efficient tool for regional LSZ with limited landslide field samples.
Figure 1: Overview workflow for comparing model performance in regional LSZ (landslide susceptibility zonation) assessment.
Click here: DOI: 10.1007/s11629-014-3134-x for the full article.
Segmentation optimization and stratified object-based analysis for semi-automated geomorphological mapping
Niels Anders, Harry Seijmonsbergen, Willem Bouten, 2011 in Remote Sensing of Environment 115, 2976-2985
Semi-automated geomorphological mapping techniques are gradually replacing classical techniques due to increasing availability of high-quality digital topographic data. In order to efficiently analyze such large amounts of data, there is a need for optimizing the processing of automated mapping techniques. In this context, we present a novel approach to semi-automatically map alpine geomorphology using stratified object-based image analysis. We used a 1 m Digital Terrain Model (DTM) derived from laser altimetry data from a mountainous catchment from which we calculated various Land-Surface Parameters (LSPs). The LSPs ‘slope angle’ and ‘topographic openness’ have been combined into a single composite layer for selecting reference material and delineating training samples. We developed a novel method to semi-automatically assess segmentation results by comparing 2D frequency distribution matrices of training samples and image objects. The segmentation accuracy assessment allowed us to automate optimization of the scale parameter and LSPs used for segmentation. We concluded that different geomorphological feature types have different sets of optimal segmentation parameters. The feature-dependent parameters were used in a new approach of stratified feature extraction for classifying karst, glacial, fluvial and denudational landforms. In this way, we have used stratified object-based image analysis to semi-automatically extract contrasting geomorphological features from high-resolution digital terrain data. A further step would be to also automate the optimization of classification rules. We would then be able to create a library of feature characteristics that could be transferred and applied to other mountain regions and further automate geomorphological mapping strategies.
Click here for the full article.
A potential geoconservation map of the Las Lagunas area, northern Peru, using GIS and remote sensing techniques
Harry Seijmonsbergen, Jan Sevink, Eric Cammeraat and Jorge Recharte, 2010 in Environmental Geoconservation 37 (2):107-115.
The Andean páramo ecosystems host geodiversity of global importance, but also have important societal functions, including agricultural production and delivery of water to people and industry. Páramo geo-ecosystems are highly susceptible to environmental degradation because of their alpine relief, extreme climate and fragile soils. In contrast to other parts of the world, geodiversity assessment studies in the Andes are scarce. We adapted and applied a geodiversity assessment method using automated techniques in a Geographical Information System and remote sensing reconnaissance to produce a potential geoconservation map of this area. The Las Lagunas area has a rich archive for climate proxy data and landscape reconstruction, and plays a key role in the functioning of regional geo-ecosystems. Undisturbed proxies for climate change are contained in four Late-Glacial recessional complexes of its former local ice-cap and in pollen records present in the post-glacial peat cover. These new findings contribute to a refined chronostratigraphy of the Late-Glacial warming period in the Andes of Northern Peru. Geo-ecosystem functions (e.g. water and carbon storage) depend on the environmental vulnerability and disturbance of the landforms and deposits. Using a weighting and ranking method in GIS, potential geoconservation areas that combine a high scientific value and environmental vulnerability with a low disturbance and low frequency of occurrence were mapped.
Click here for the full article.
Structure and contents of a new geomorphological GIS database linked to a geomorphological map - with an example from Liden, central Sweden
Marcus Gustavsson, Harry Seijmonsbergen and Else Kolstrup, 2008 in: Geomorphology 95 (3-4): 335-349
This paper presents the structure and contents of a standardised geomorphological GIS database that stores comprehensive scientific geomorphological data and constitutes the basis for processing and extracting spatial thematic data. The geodatabase contains spatial information on morphography/morphometry, hydrography, lithology, genesis, processes and age. A unique characteristic of the GIS geodatabase is that it is constructed in parallel with a new comprehensive geomorphological mapping system designed with GIS applications in mind. This close coupling enables easy digitalisation of the information from the geomorphological map into the GIS database for use in both scientific and practical applications. The selected platform, in which the geomorphological vector, raster and tabular data are stored, is the ESRI Personal geodatabase. Additional data such as an image of the original geomorphological map, DEMs or aerial orthographic images are also included in the database. The structure of the geomorphological database presented in this paper is exemplified for a study site around Liden, central Sweden.
Click here for the full article.
Application of GIS and logistic regression to fossil pollen data in modeling present and past spatial distribution of the Colombian savanna
S.G.A. Flantua, J.H. van Boxel, H. Hooghiemstra and J.W.N. van Smaalen, 2007 in: Climate Dynamics 29 (7-8): 697-712
Climate changes affect the abundance, geographic extent, and floral composition of vegetation, which are reflected in the pollen rain. Sediment cores taken from lakes and peat bogs can be analyzed for their pollen content. The fossil pollen records provide information on the temporal changes in climate and palaeo-environments. Although the complexity of the variables influencing vegetation distribution requires a multi-dimensional approach, only a few research projects have used GIS to analyse pollen data. This paper presents a new approach to palynological data analysis by combining GIS and spatial modeling. Eastern Colombia was chosen as a study area owing to the migration of the forest savanna boundary since the last glacial maximum, and the availability of pollen records. Logistic regression has been used to identify the climatic variables that determine the distribution of savanna and forest in eastern Colombia. These variables were used to create a predictive land-cover model, which was subsequently implemented into a GIS to perform spatial analysis on the results. The palynological data from the study area were incorporated into the GIS. Reconstructed maps of past vegetation distribution by interpolation showed a new approach of regional multi-site data synthesis related to climatic parameters. The logistic regression model resulted in a map with 85.7% predictive accuracy, which is considered useful for the reconstruction of future and past land-cover distributions. The suitability of palynological GIS application depends on the number of pollen sites, the distribution of the pollen sites over the area of interest, and the degree of overlap of the age ranges of the pollen records.
The figure shows the study area and outcomes of predicted land cover distribution by logistic model: a South-America indicating the location of the study area and the actual land cover distribution, green is forest and yellow is savanna; b location of the Colombian savanna biome between the Andes and the Guyana Shield (03-07_N, 68-71_W); c probability map of savanna occurrence based on random data sampling; d Probability map of savanna occurrence based on regular sampling. Yellow lines indicate the 0.5 threshold, red lines delineate the 0.6 cut-point; e-h Differences between observed land-cover distribution and model prediction; e based on random sampling at 0.5 thresholds; f random sampling at 0.6 thresholds; g regular sampling at 0.5 threshold; h regular sampling at 0.6 threshold. The letters correspond to the letters of Table 1: (a) indicates correctly predicted savanna [Dark yellow]; and (b) represents where the model falsely predicted savanna [Bright green]; (c) indicates correctly predicted forest [Dark green]; while (d) shows where the model failed to predict savanna [Red].
Click here for the full article.
Expert-driven semi-automated geomorphological mapping for a mountainous area using a laser DTM
S. van Asselen and A.C. Seijmonsbergen, 2006 in: Geomorphology 78 (3-4): 309-320
In this paper a semi-automated method is presented to recognize and spatially delineate geomorphological units in mountainous forested ecosystems, using statistical information extracted from a 1-m resolution laser digital elevation dataset. The method was applied to a mountainous area in Austria. First, slope angle and elevation characteristics were determined for each key geomorphological unit occurring in the study area. Second, a map of slope classes, derived from the laser DTM was used in an expert-driven multilevel object-oriented approach. The resulting classes represent units corresponding to landforms and processes commonly recognized in mountain areas: Fluvial terrace, Alluvial Fan, Slope with mass movement, Talus slope, Rock cliff, Glacial landform, Shallow incised channel and Deep incised channel. The classification result was compared with a validation dataset of geomorphological units derived from an analogue geomorphological map. For the above mentioned classes the percentages of correctly classified grid cells are 69%, 79%, 50%, 64%, 32%, 61%, 23% and 70%, respectively. The lower values of 32% and 23% are mainly related to inaccurate mapping of rock cliffs and shallow incised channels in the analogue geomorphological map. The accuracy increased to 76% and 54% respectively if a buffer is applied to these specific units. It is concluded that high-resolution topographical data derived from laser DTMs are useful for the extraction of geomorphological units in mountain areas.
Click here for the full article.
Geomorphological mapping and geophysical profiling for the evaluation of natural hazards in an alpine catchment
A.C. Seijmonsbergen and L.W.S. de Graaff, 2006 in: Natural Hazards and Earth Science Systems 6 (2): 185-193
Liechtenstein has faced increasing records of natural hazards during the last decades: landslides, debris flows, snow avalanches and floods repeatedly endanger the local infrastructure. Geomorphological field mapping and geo-electrical profiling was used for the assessment of hazards near Malbun, which is potentially endangered by landslides and debris flows. The geomorphology has developed on the tectonic contacts of four different nappe slices. The bedrock consists of anhydrite and gypsum, dolomite, shale, marl, and limestone. We evaluate the spatial distribution and occurrence of debris flow and landslide hazards using an expert driven geomorphological inventory, supported by analyses in a geographical information system. In a geo-database a symbol-based 1:3,000 scale geomorphological map is combined with geophysical information and interpolated displacement rates of benchmark measurements. Conversion into digital geomorphological units enables the visualization of thematic maps which show the spatial distribution of the main geomorphological environment, the Quaternary material distribution and of geomorphological processes, which are stored in attribute tables. Two geo-electrical profiles show that the thickness of a potentially endangering landslide is approximately 10-20 m and that subterranean karst influences the topography of the failure plane. The surface displacement measurements show, compared with the geomorphological and geophysical data, that the central landslide is disintegrating into minor slides and therefore not a risk to the village of Malbun. The GIS-supported hazard evaluation does indicate that debris flows are a serious risk if no countermeasures are taken. Furthermore, karst processes may accelerate slope movement, but may locally diminish the impact of debris flows; collapse dolines serve as sediment traps for debris flows. This research illustrates how terrain expert knowledge can be integrated in a GIS for the evaluation of natural hazards on a detailed scale.
Click here for the full article (pdf)