Integrating a Three-Level GIS Framework and a Graph Model to Track, Represent, and Analyze the Dynamic Activities of Tidal Flats

Abstract

Tidal flats (non-vegetated area) are soft-sediment habitats that are alternately submerged and exposed to the air by changeable tidal levels. The tidal flat dynamics research mainly utilizes the cell-level comparisons between the consecutive snapshots, but the in-depth study requires more detailed information of the dynamic activities. To better track, represent, and analyze tidal flats’ dynamic activities, this study proposes an integrated approach of a three-level Geographic Information Science (GIS) framework and a graph model. In the three-level GIS framework, the adjacent cells are assembled as the objects, and the objects on different time steps are linked as lifecycles by tracking the predecessor–successor relationships. Furthermore, eleven events are defined to describe the dynamic activities throughout the lifecycles. The graph model provides a better way to represent the lifecycles, and graph operators are utilized to facilitate the event analysis. The integrated approach is applied to tidal flats’ dynamic activities in the southwest tip of Florida Peninsula from 1984 to 2018. The results suggest that the integrated approach provides an effective way to track, represent, and analyze the dynamic activities of tidal flats, and it offers a novel perspective to examine other dynamic geographic phenomena with large spatiotemporal scales.

The Spatiotemporal Characteristics and Dynamic Changes of Tidal Flats in Florida from 1984 to 2020

“…It substantially contributed to the tidal flat expansions of the entire study area during those eight years, and the details will be discussed in the next subsection (Section 3.1.2). Additionally, in Zone B, we noticed a peak in 1992 (1017.32 km 2 ), which is consistent with our previous study [25] and reconfirms that Hurricane Andrew (category 5) has brought in considerable sediments along the southwestern coast of Florida [38]. On the other hand, we observed two significant area decreases in Zone A, which are from 2004 (366.75 km 2 ) to 2005 (201.45 km 2 ) and from 2006 (326.02 km 2 ) to 2007 (172.80 km 2 ).…”

“…Recall that every pixel in the classified map has 30 m spatial resolution, a primary consideration is to find the area of tidal flats by year and explore the temporal change patterns by zone. On the other hand, the tidal flat occurrence map of each zone can be synthesized from all 37 years' tidal flat distributions, in which the raster values vary from 0 (without occurrence) to 37 (always occurrence) [25]. To quantify the spatial features on the synthesized map, we also calculated the area of tidal flats weighted by the frequency during the 37 years, which are summarized by longitude and latitude, and displayed as line charts along map edges.…”

“…A higher-level assessment for tidal flat dynamics is based on the objects [25]. As shown in Figure 3, one pixel may have up to eight neighboring pixels, and consequently the contiguous tidal flat pixels can be assembled as a whole entity, which refers to a tidal flat object [25]. Two geometry attributes of the derived objects can be used to examine the temporal characteristics of tidal flat dynamics, including:…”

“…Specifically, it regards the pixels as geo-atoms, and the geo-atoms can further aggregate as geo-fields. Accordingly, the object level analyses, which assemble the contiguous pixels as single patches, have been applied to multiple domains [23][24][25] and verified as an effective way to examine the spatiotemporal characteristics of dynamic geographic phenomena. Specifically speaking, the states of an individual object are characterized by the shapes at specific moments, which can be quantified by a variety of geometry attributes such as area, perimeter, roundness, and so on.…”

Remote Sensing Image Classification with a Graph-Based Pre-Trained Neighborhood Spatial Relationship

“…The graph-based data structure enables more analysis regarding spatial relationships between landscape entities such as subgraph mining and kernel analysis. Recently, Xu et al [ 38 ] used graphs to explore the morphological changes between adjacent tidal flat objects. Aside from directly using graph-based spatial relationships in the process of classification, graph convolutional neural networks (GCN) have been implemented in remote sensing image classification.…”