Object-oriented spatial-temporal association rules mining on ocean remote sensing imagery

Abstract

Abstract. Using the long term marine remote sensing imagery, we develop an object-oriented spatial-temporal association rules mining framework to explore the association rules mining among marine environmental elements. Within the framework, two key issues are addressed. They are how to effectively deal with the related lattices and how to reduce the related dimensions? To deal with the first key issues, this paper develops an object-oriented method for abstracting marine sensitive objects from raster pixels and for representing them with a quadruple. To deal with the second key issues, by embedding the mutual information theory, we construct the direct association pattern tree to reduce the related elements at the first step, and then the Apriori algorithm is used to discover the spatio-temporal associated rules. Finally, Pacific Ocean is taken as a research area and multi-marine remote sensing imagery in recent three decades is used as a case study. The results show that the object-oriented spatio-temporal association rules mining can acquire the associated relationships not only among marine environmental elements in same region, also among the different regions. In addition, the information from association rules mining is much more expressive and informative in space and time than traditional spatio-temporal analysis. IntroductionIn recent three decades, advanced earth-observing technologies make it possible to acquire long time series of marine bio-optical parameters and dynamic elements from multi-remote sensing imagery, and the inter-annual, intra-annual variations and spatial distribution of marine environmental elements are studied by means of mathematical statistics and empirical orthogonal functions (EOF), i.e. spatiotemporal variation of sea surface temperature (SST)[1], spatial distribution and annual, seasonal and monthly properties of sea surface chlorophyll-a concentration and marine primitive productivity [2], spatio-temporal characteristics of sea surface precipitation [3]. Moreover, many algorithms are developed to extract these characteristic parameters from remote sensing imagery, such as ENSO indices, western Pacific warm pool, cold tongue of SST, oceanic rain pool, oceanic desert [4]. To the best of our knowledge, the spatio-temporal variation of marine environmental elements is a complicated system. Usually the inter-annual variability of marine environmental elements is driven by ENSO events, and on the contrary, it can intensify or weaken ENSO events [5][6]. So far few studies are carried on investigating the associated relationship among three or more elements, especially at macroscale using remote sensing imagery.Geographical spatio-temporal association rules mining is a novel data-driven methodology in spatio-temporal analysis, derived from geography, information science and computer technology [7][8].

Global Land Surface Dry/Wet Conditions Mining Based on Spatial‐Temporal Association Rules

“…Spatial-temporal data mining methods refer to analytic processes designed to search for consistent patterns and/or systematic relationships between variables from large volumes of spatial-temporal data (Jain & Srivastava, 2013;Xie et al, 2018), which are not used to prove or disprove preexisting hypotheses but rather to identify patterns embedded within spatial-temporal data (Mennis & Liu, 2005;Raheja & Rajan, 2012;Xu et al, 2017). As one of the most important spatial-temporal data mining tasks (Agrawal & Srikant, 1994;Schlüter & Conrad, 2010), spatial-temporal association rule mining is expected to discover the presence of pair conjunctions appearing in a spatial-temporal data set (Qin et al, 2015),which has the advantage to discover nontrivial, implicit, previous unknown, but potentially useful knowledge from large data sets and can be used to mine the spatial-temporal associations of dry/wet conditions in meteorological data (Han et al, 2011;Laube et al, 2008;Pei et al, 2020;Xue et al, 2014).…”