| Geographical system is a kind of multi-scale processes with the profound span of space and time. MSTC (Multi-scale spatial and temporal characteristics) are the intrinsic characteristics of geographical phenomena as well as the core concern of geographical analysis. By the coupling effect between multi sphere layers, the earth surface system is manifested as the "class order" features between processes at different space scales and time scales. Existing multi-scale spatial and temporal feature extraction and structural analysis methods for the complex geographical phenomena and geographical processes, it is difficult to analysis GSTD (Geographic Spatial-Temporal Data) with time non-stationary, space nonnormaltiy and time and space integration. Thus, there is a question how to explore methods about MSTC extraction and structures analysis to support non-stationary, nonnormaltiy and time and space unity GSTD, to achieve analysis and diagnose GSTD and evolution pattern of geographical objects and phenomenon from different temporal and spatial scales perspectives, which is one of the important way to explore work mechanism and to improve controlling and forecasting ability of geographical system.Based on the modern mathematical analysis method including time series, spatial statistics, signal processing, tensor analysis and et al., this treatise divides different types GSTD into time series, space panel and space-time cube, and then studies the MSTC extraction and structural analysis methods which are applied to extract geographic MSTC from the time domain, space domain and spatial-temporal domain. The study can support the non-stationary GSTD and non-normal GSTD. In this thesis, automatic identification and diagnosis method about characteristic scales of spatial data is built, which by data driven. Then, multi-scale feature detection methods of GSTD based on spatial-temporal unity framework is established. On this basis, the GSTD multi-scale feature analysis and structural analysis system is designed and is validated utilizing sea-level changes data. The key points and achievements of this treatise are as follows:(1) MSTC extraction and analysis mehods for geographical time series data (GTSD) are designed, wich allow to the steadily trend extraction and periodic/ quasi-periodic automatic identification under high noise environment. And research method about multi-scales coupling relationships between different locations GTSD or variable types GTSD also is built. For univariate time series, there proposed a trend extraction method by statistical smoothing EMD and customizable cycle fluctuations and adaptive extraction technology by statistics about power spectral density of SSA components. For bivariate time series, this paper constructed features analysis and sequences synchronization methods fused BEMD and DTW from time domain and frequency domain integration perspective. Multi-variant time sequence data interactions model is developed by MSSA, which can support the analysis of multi-scale spatial transfer processes among different sequences.Then the experiment and verification are done for sea level changes sequences.(2) For the application requirements of unstructured spatial data with non-stationary, non-normal features, features scales automatic recognition and diagnosis methods are established. Spatial-EMD decomposition method based on regression correction is proposed, which is distribution driven. That can support adaptive multi-scale decomposition for the non-zero mean data and non-stationary data. On this basis, spatial filter group algorithm with kernel smoothing is designed, for the skew-distributed data. Then, the adaptive spatial multi-scale decomposition method based on parallel kernel smoothing is realized. Finally, the accuracy, applicability and robustness of the proposed methods are evaluated in a case study of Chinese population and GDP data of 2003. The results reveal the spatial coupling relationship between the population and economic factors under different level urban system.(3) For GSTD, MSTC analysis methods under space-time unified framework are established. The thesis built the unified multi-dimensional tensor expression model about GSTD. Based on tensor decomposition, the features analysis, dimension perspective and processes reconstruction of STD multi-scale structures and evolution processes are done. Then, to track the spatial and temporal evolution of non-linear and non-stability structural signal, features analysis and exploration methods are developed by multi-signal wavelet decomposition under the idea of time sequence high-dimensional expansion. Taking the global satellite altimetry data as validating case, the evolution structure and track features of different ENSO events are extracted. Also, this paper tracked the multi-scale spatial and temporal evolution processes about strong El Nino event between 1997 and 1998.(4) GSTD multi-scale features extraction and structures analysis system is designed. Firstly, spatial-temporal data tensor expression and analysis streaming templates are constructed. Secondly, the GDAL/ORG is used to integrate different STD, and time objects expression and integration are done, which is meeting POSIX criterion. Thirdly, the R, Matlab and other existing statistical functions can be integrated organically, relying on the DCOM server technology. Then calculation engine is built to extract multi-scale GSTD features and analysis the structures. On this basis, the data structure and data streaming of multi-scale spatial-temporal data decomposition is also modeled. Finally, plug-in algorithm library and integrated framework of GSTD multi-scale analysis is developed.This thesis has extended the method system about extract and analysis on the multi-scale features of GSTD. The proposed methods are systematism, actualization and suitable for different GSTD. Which also can support well extract on MSTC extraction, procedure description, time-space transfer mode and interactions analysis of non-stationary, non-normal and spatial-temporal unified GSTD. All of that has further promotive effect and reference function on the studies about that multi-scale spatial and temporal distribution patterns, evolution processing and multi-factors interactions of geographical objects and phenomena. |
PDF Full Text Request