ManyWaves: A Visual Analytics Approach for Exploring Multivariate Spatiotemporal Data Based on the Graph Wavelet Transform

Georeferenced time series are central to domains such as urban planning and climate studies, yet insights remain difficult to obtain when patterns span multiple variables and scales. Analysts often seek events—abrupt changes that may co-occur across variables, space, and time. The Graph Wavelet Transform (GWT) is effective for detecting such changes; however, previous visualization tools have primarily applied it to univariate signals. We introduce ManyWaves, a visual analytics system that applies GWT separately to each variable and integrates the resulting event information in a coordinated multivariate environment. Each variable is modelled as a signal on a shared spatiotemporal graph, and its high-frequency wavelet coefficients characterize local spatial and temporal changes. A 1D convolutional autoencoder then learns a low-dimensional representation from these coefficients, which drives linked views: a projection of locations by event similarity, a heatmap summarizing when variables change, and spatial and time-series views for detailed inspection. Through three case studies—U.S. flight disruptions, traffic and crime in São Paulo, and land-use change in the Brazilian Legal Amazon—we show how ManyWaves supports the exploration of multivariate events. Expert feedback suggests that the system's event-oriented views and interactions support sense-making, while also highlighting the need for onboarding to interpret wavelet-based representations.