SalienTime: User-driven Selection of Salient Time Steps for Large-Scale Geospatial Data Visualization

TL;DR

A multifaceted definition of salient time steps is established via extensive need-finding studies with domain experts to understand their workflows and a novel approach that leverages autoencoders and dynamic programming to facilitate user-driven temporal selections is proposed.

Abstract

The voluminous nature of geospatial temporal data from physical monitors and simulation models poses challenges to efficient data access, often resulting in cumbersome temporal selection experiences in web-based data portals. Thus, selecting a subset of time steps for prioritized visualization and pre-loading is highly desirable. Addressing this issue, this paper establishes a multifaceted definition of salient time steps via extensive need-finding studies with domain experts to understand their workflows. Building on this, we propose a novel approach that leverages autoencoders and dynamic programming to facilitate user-driven temporal selections. Structural features, statistical variations, and distance penalties are incorporated to make more flexible selections. User-specified priorities, spatial regions, and aggregations are used to combine different perspectives. We design and implement a web-based interface to enable efficient and context-aware selection of time steps and evaluate its efficacy and usability through case studies, quantitative evaluations, and expert interviews.

Figures (12)

• Figure 1: The platforms NASAWorldview2023SentinelHubEO2023MetOceanView2023 we used for the observational study.
• Figure 2: The compiled results of our need-finding study. The participants' domains are illustrated in backgrounds with green (Earth Science) or blue (Data Visualization) color.
• Figure 3: The architecture of our network. The encoder operates from top to bottom while the decoder operates from bottom to top. The flow connecting the output of a decoder layer to the next layer's contact operation is omitted for clarity.
• Figure 4: The cost matrix for the first 100 frames from data tmp2m (Air Temperature at 2 Meters), using AVG as aggregation method for $\mathcal{C}_{\mathrm{stat}}$. Each item $(i, j)$ in the matrix denotes the cost value for frames $i$ and $j$.
• Figure 5: The impact of the distance cost parameter $\gamma$ on the selection of salient time steps. Subfigures a, b, c, and d show the results with different $\gamma$s. Each dot represents a frame's latent code $Z_i$, dimension-reduced with t-SNE maatenVisualizingDataUsing2008 and colored in chronological order. Salient time steps are marked in red circles. On the left are the colorized frames 19 to 22, where drastic changes occur between frames 20 to 21, as reflected in the latent space. The selection is performed on 80 frames of hs (significant height of wind and swell waves) data of the East China Sea area with k = 10.