Flood susceptibility mapping (FSM) is crucial for preparedness and risk mitigation of floods, and machine learning (ML)-based FSM methods have been well-studied. However, due to computational constraints, geographical variations, and differential data availability, FSM is often confined to smaller regions, much less than 1,000 km2. To address the computational constraints of FSM generation in data-rich regions, we propose a data-parallel approach using PySpark with two levels of spatial partitioning, where they address the computational load and geographical variations, respectively. In our case study of the state of Kerala, India, we propose grid lattice partitioning and the Ecologically Sensitive Zone (ESZ) map for the two levels of spatial partitioning, respectively. The latter is used to train random forest classifier models. We investigate the influence of contiguity in training grids on FSM outcomes within different ESZs. Our quantitative and qualitative comparative analysis of observations demonstrate the importance of contiguous training grids, especially in capturing finer details in FSM. Our implementation demonstrates the efficiency of our approach in the study area of Kerala spanning 38,863 km2.

Choropleth maps are widely used geovisualizations due to their simplicity, especially for applications involving political, climate, and other geospatial data for contiguous regions. There is a need for automated data extraction from such maps to aid the human-in-the-loop in handling cognitive overload from large-scale visualization generation and visual impairments. There are gaps in generalizing such a system for choropleth maps with different types of color legends. We propose the choropleth map analytics (CMA) system to address these gaps using a six-step workflow involving deep learning (DL) architectures and tools. We propose a novel method for color-to-data mapping for different color legend types. We finally demonstrate the usability of CMA for a set of choropleth images in climate research for a text summarization application. Our work is a step toward reverse engineering choropleth visualizations. Our code and curated datasets are at:https://github.com/GVCL/Choropleth-CMA

The state-of-the-art visualization tools for multidimensional gaze or eye-tracking data focus on a few dimensions leading to incomplete analysis, e.g., fixations as time series data, scanpath trajectory, etc. We propose EyeExplore, an interactive visualization tool to explore eye-tracking data for a single static 2D stimulus (i.e., a selected image). It has multiple views, namely, single-user, user comparative, and cohort-summary views. We propose the use of ensemble clustering and visualization of co-association matrices for cohort analysis. We propose the use of semantics-aware areas of interest (AOI) through user interactivity leading to AOI transition matrix visualization. Our preliminary results show that EyeExplore provides a more complete data exploration.

Population survey data is important for understanding the status of well-being along any dimensions, i.e., social, economic, political, health, etc. This data generates spatial point patterns which can be explored and analyzed using visualization. Given the spatial aspect of the data, there is a requirement of using cartographic maps, which are mostly limited to visualizing a single variable in most cases. Here, it is also important that the choice of visualizations also enable scale-aware analysis when zooming in and out of the maps, since the data is from the smaller political units and can be aggregated to larger political units. Thus, we explore the different visual compositions which use mathematical operators and the composite layouts for visualizing multiple outcome variables in survey data. The mathematical operators allow the use of univariate and bivariate data modeling and representation, and composite layouts of interest are juxtaposition and superimposed views. We demonstrate the inferences from visualizations using a case study on malnutrition in children under five in India. Our work shows that a visual composition of binary relationships represented in a visualization and a juxtaposed layout of such pairwise variables is effective in making inferences from the multivariate spatial point patterns in population data.

Multiple spatial scales have been used extensively for feature extraction from light detection and ranging (LiDAR) point clouds. These features have been used for semantic classification, segmentation, and other data analysis methods. There is a gap in the adaptive methodology for the effective use of multiple scales here. This stems from determining the best strategy to aggregate the information or features gathered from different scales. The widely used multiscale method is feature extraction at an optimal scale, which is in itself an adaptive method. However, the success of identifying the optimal scale depends on the set of scales used in its determination, as it must include the scale where the global minimum of eigenentropy occurs. An alternative method is to average features across multiple scales, which works in specific scenarios. In order to improve the flexibility of using different methods in the same workflow, we propose an adaptive method for the selection of multiscale feature extraction for semantic classification of LiDAR point clouds, with a focus on airborne laser scans. Our decision-making process for finding the best multiscale method exploits spatial locality of the features. We show how such a control strategy can be implemented in an Apache Spark–Cassandra distributed system for processing large-scale point clouds using voxelization for preserving spatial locality, and binomial logistic regression for selecting voxels to implement a specific multiscale method at. Our results show significant improvement in classification accuracy in the Dayton Annotated Laser Earth Scan (DALES) data, implemented using Spark MLlib in our distributed system.

For semantic classification of LiDAR point clouds, the features derived from the local geometric descriptors are routinely used as features in (supervised) learning algorithms. In this study, our goal is to determine if the aleatoric uncertainty in the input to a supervised semantic classifier influences the outcomes. We consider two sources of such uncertainty - one from the computation of multiscale local geometric descriptors and the other, from class ambiguities at object boundaries. We perform ensembles of experiments to measure the significance of these uncertainties in the semantic classification of airborne LiDAR point clouds, when using random forest classifier. Our case study shows that the presence of aleatoric uncertainty improves the classification outcomes.

Structural or geometric classification of three-dimensional (3D) point clouds of urban regions from airborne LiDAR enables feature (object-based) classification, and 3D reconstruction. Here, we consider positive semidefinite symmetric second-order tensors as local geometric descriptors (LGDs), which gives structural classification. We compute LGDs using local neighborhood, and their eigenvalue-based features are conventionally used for object-based classification and 3D reconstruction. We combine derivative based 2D gradient energy tensor, and anisotropically diffused 3D voting tensor, using a multi-scale approach, to compute a LGD. We represent LGDs as second-order tensors, and compare the relevant eigenvalue-based features and saliency maps obtained from them. We visually compare the outcomes of our LGD with conventionally used covariance matrix.

We propose a novel unsupervised machine learning approach for effective semantic labeling by combining two different multi-class classifications, structural and contextual classification, of points in airborne LiDAR point cloud of urban environment. Structural classification labels a point in the cloud as a point-, line-, or surface-type feature. An additional outcome of this classification is the geometry-preserving downsampling of the point cloud. The contextual classification, on the other hand, labels the points in four classes, namely, buildings, vegetation, natural ground, and asphalt ground, by using data derived from the raw input, which includes the structural classification. Preserving these two classifications in the labeling of the points gives a geometry-aware contextual semantic labeling. We propose: (a) an augmented semantic classification which preserves both structural and contextual classification, (b) an interactive hierarchical clustering method for contextual classification, and (c) an interactive visual analytic framework to aid both the structural and contextual classifications.