This paper presents a novel approach for the feature extraction of hyperspectral image cubes. In this paper, hyperspectral image cubes are treated as volumetric data sets. Features that are most helpful in separating different targets are effectively extracted from the hyperspectral image cubes using a newly developed high-order texture analysis method. The traditional texture measure of the gray-level cooccurrence matrix is extended to a 3-D tensor field to explore the complicated volumetric data more effectively and to extract discriminant features for better classification. As the kernel size is one of the most important parameters in statistics-based texture analysis, a semivariance analysis and a spectral separability measure are used to determine the most appropriate kernel size in the spatial and spectral domains, respectively, for computing 3-D gray-level cooccurrence. In addition, a few statistical indexes are also extended to third-order forms in order to calculate quantitative texture properties of the generated cooccurrence tensor field. An airborne hyperspectral data set and an EO-1 Hyperion image are used to test the performance of the developed algorithms. Experimental results indicate that the developed 3-D texture analysis outperforms conventional second-order texture descriptors and the support vector machine-based classifier in supervised classifications of both hyperspectral data sets.

This study developed a systematic approach with machine learning (ML) to apply the satellite remote sensing images, geographic information system (GIS) datasets, and spatial analysis for multi-temporal and event-based landslide susceptibility assessments at a regional scale. Random forests (RF) algorithm, one of the ML-based methods, was selected to construct the landslide susceptibility models. Different ratios of landslide and non-landslide samples were considered in the experiments. This study also employed a cost-sensitive analysis to adjust the decision boundary of the developed RF models with unbalanced sample ratios to improve the prediction results. Two strategies were investigated for model verification, namely space- and time-robustness. The space-robustness verification was designed for separating samples into training and examining data based on a single event or the same dataset. The time-robustness verification was designed for predicting subsequent landslide events by constructing a landslide susceptibility model based on a specific event or period. A total of 14 GIS-based landslide-related factors were used and derived from the spatial analyses. The developed landslide susceptibility models were tested in a watershed region in northern Taiwan with a landslide inventory of changes detected through multi-temporal satellite images and verified through field investigation. To further examine the developed models, the landslide susceptibility distributions of true occurrence samples and the generated landslide susceptibility maps were compared. The experiments demonstrated that the proposed method can provide more reasonable results, and the accuracies were found to be higher than 93% and 75% in most cases for space- and time-robustness verifications, respectively. In addition, the mapping results revealed that the multi-temporal models did not seem to be affected by the sample ratios included in the analyses.

This study explores two modeling issues that may cause uncertainty in landslide susceptibility assessments when different sampling strategies are employed. The first issue is that extracted attributes within a landslide inventory polygon can vary if the sample is obtained from different locations with diverse topographic conditions. The second issue is the mixing problem of landslide inventory that the detection of landslide areas from remotely-sensed data generally includes source and run-out features unless the run-out portion can be removed manually with auxiliary data. To this end, different statistical sampling strategies and the run-out influence on random forests (RF)-based landslide susceptibility modeling are explored for Typhoon Morakot in 2009 in southern Taiwan. To address the construction of models with an extremely high false alarm error or missing error, this study integrated cost-sensitive analysis with RF to adjust the decision boundary to achieve improvements. Experimental results indicate that, compared with a logistic regression model, RF with the hybrid sample strategy generally performs better, achieving over 80% and 0.7 for the overall accuracy and kappa coefficient, respectively, and higher accuracies can be obtained when the run-out is treated as an independent class or combined with a non-landslide class. Cost-sensitive analysis significantly improved the prediction accuracy from 5% to 10%. Therefore, run-out should be separated from the landslide source and labeled as an individual class when preparing a landslide inventory.

The universal soil loss equation (USLE) is a widely used empirical model for estimating soil loss. Among the USLE model factors, the cover management factor (C-factor) is a critical factor that substantially impacts the estimation result. Assigning C-factor values according to a land-use/land-cover (LULC) map from field surveys is a typical traditional approach. However, this approach may have limitations caused by the difficulty and cost in conducting field surveys and updating the LULC map regularly, thus significantly affecting the feasibility of multi-temporal analysis of soil erosion. To address this issue, this study uses data mining to build a random forest (RF) model between eight geospatial factors and the C-factor for the Shihmen Reservoir watershed in northern Taiwan for multi-temporal estimation of soil loss. The eight geospatial factors were collected or derived from remotely sensed images taken in 2004, a digital elevation model, and related digital maps. Due to the memory size limitation of the R software, only 4% of the total data points (population dataset) in each C-factor class were selected as the sample dataset (input dataset) for analysis using the stratified random sampling method. Seventy percent of the input dataset was used to train the RF model, and the other 30% was used to test the model. The results show that the RF model could capture the trend of vegetation recovery and soil loss reduction after the destructive event of Typhoon Aere in 2004 for multi-temporal analysis. Although the RF model was biased by the majority class’s large sample size (C = 0.01 class), the estimated soil erosion rate was close to the measurement obtained by the erosion pins installed in the watershed (90.6 t/ha-year). After the model’s completion, we furthered our aim to address the input dataset’s imbalanced data problem to improve the model’s classification performance. An ad-hoc down-sampling of the majority class technique was used to reduce the majority class’s sampling rate to 2%, 1%, and 0.5% while keeping the other minority classes at a 4% sample rate. The results show an improvement of the Kappa coefficient from 0.574 to 0.732, the AUC from 0.780 to 0.891, and the true positive rate of all minority classes combined from 0.43 to 0.70. However, the overall accuracy decreases from 0.952 to 0.846, and the true positive rate of the majority class declines from 0.99 to 0.94. The best average C-factor was achieved when the sampling rate of the majority class was 1%. On the other hand, the best soil erosion estimate was obtained when the sampling rate was 2%.

Treat full-waveform LiDAR point clouds as volumetric data sets 3D texture measures were used for LiDAR feature extraction 3D texture features improve...