| Traditional Geostatistics is one of the most popular methods to study the spatial variation of the geographical attribute. Despite its widely use in many fields like pedology, environmental sciences and ecology, strong limitations of this method still exist and make it inappropriate to solve some kind of problems. Among those limitations, Gaussian or second-order stationary assumption, unable to make full use of the data with uncertainty and linear estimator are most criticized. Scholars try to propose various kinds of methods to overcome the above limitations. A so-called Bayesian Maximum Entropy(BME) approach was especially developed since 1990 s. Many advantages of this method over traditional Geostatistics can be found like: no assumption like Gaussian or second-order stationary are made, uncertainty information are easily incorporated and it yields the best unbiased but non-linear estimator, along with an entire but non-Gaussian probability density function which can effectively describing the target variable. Nevertheless, being quite a young method, there are still many key issues, practically and theoretically, waiting to be considered and solved. Three of them are listed below:Since high quality soft data are vital element of the success of BME, can we put forward more kinds of simple but effective ways of generating the soft data, pulling down the wall of disciplines and techniques, and to fulfill our understanding of this world based on the giant amount of data and knowledge?How to make full use of the accurate and informative prediction/estimation results obtained by BME to solve the scientific problems and serve the public in a better way?Can we reduce the modeling complexity and improve the prediction efficiency of the BME method, so as to offer timely and accurate solution to the hot or urgent issues of social life?According to that, this thesis will focus on the field of soil science, environmental science and the public health, within which the BME has been most frequently used so far, and aim to explore answers of the above three questions concerning BME in great depth. Some new achievements have been made like:(1) Three approaches of generating the soft data, which are based on environmental discrete relationship(ER), multiple linear regression model(MLR) and geographically weighted regression model, respectively, are put forward and be successfully applied to the spatial prediction of soil total nitrogen content in Shayang County, Hubei Province(China).Utilizing the soft data in rational way would improve the prediction results of BME, yet few kinds of approaches have been created to generate the soft data in soil science. It will be very favorable for reducing the soil sampling cost and promoting the development of soil mapping theory if we make full use of the BME by integrating multi-source data to construct high quality soft data in a simple way. In this study, soft data were obtained by transforming the low-cost DEM based terrain factors and remote-sensing image based indices, as well as the historical soil texture map and soil type map with the help of ER, MLR and GWR, those three kinds of soft data were further integrated into BME, marked as BME_ER、BME_MLR and BME_GWR, respectively. Results indicate that, compared to the ER, MLR, GWR, ordinary kriging(OK) and regression kriging(RK), the proposed BME_ER 、 BME_MLR and BME_GWR have relatively high Pearson correlation coefficient(r) and low mean error(ME), mean absolute error(MAE) and root mean square error(RMSE), along with less uncertainty. Being quite straightforward and easy to implement, the three approaches of generating soft data in this study can make full use of the multisource information and the advantage of BME to achieve the balance of spatial prediction accuracy and efficiency for soil properties.(2) Put forward the so-called spatial predictability criterion(SPC) based on the combination of BME and GWR, and apply it to the ecological risk assessment of the urban soil heavy metals in Wuhan City, Hubei Province.It may be better to produce clear and straightforward results for the policy-making organizations, rather than simply getting the fine but original results from BME, especially in fields like environmental evaluation. In this study, environmental factors which are related to soil heavy metal contents were transformed to soft data, and were introduced into BME to get the spatial distribution maps of SPC. Results indicate that, vehicular emission, industrial activities and household wastes may be the three main sources for trace metals(Pb, Zn, Cu, Cr and Cd) accumulation in Wuhan. Weighted road density(WRD), weighted industry distance(WID) and population density were computed and used to quantify the impact of traffic, industry and human daily activities on the heavy metals. WRD, WID and PD and the potential ecological risk index(RI) were then used to compute the SPC values. According to the spatial distribution maps of SPC, three high risk zones and two low risk zones regarding the soil heavy metals pollution were detected in Wuhan. High risk zones include Gutian(located in Qiaokou District), Southeastern new cities(mainly consist of Baoxie and Liufang) and Southern new cities(mainly consist of Qingling and Jinkou). Low risk zones include areas of Jiangan District along the Yangtze River and the areas between Donghu Lake and Nanhu Lake. High risk zones suffer from high ecological risk due to the recent industrial development planning, high-grade roads network and population migration. Effective measures(such as increasing greening rate) should be taken as earlier as possible to prevent the probable accumulation of heavy metals in these zones, despite content of heavy metals are not very high. On the other hand, the two low risk zones suffer from relatively lower ecological risk due to the traffic speed, the existing environmental protection measures should be strengthen in those zones, and efforts should be made to reduce content of heavy metals. The above results are very informative and could offer effective and powerful suggestion to the policy-making organization of Wuhan City. In a word, the ecological risk assessment method based on SPC is instructive and provides a new way for strengthening the ability of BME to support the policy makers more efficiently.(3) A so-called spatiotemporal dimension reduction model was put forward within the framework of the BME method, the spatiotemporal random field and the spatial random field were connected via the velocity vector, the model was applied to the analysis of disease spatiotemporal dynamics.BME can produce better results because of the auxiliary information besides the sampling data, but would suffer from heavier modeling and computation burden due to the same reasons. It would have considerable modeling benefits theoretically and computationally if modeling complexity is reduced and the prediction efficiency is improved. A spatiotemporal dimension reduction model is proposed in this work. Numerical simulation results suggest that this model could throw insight on our understanding of disease by offering information about direction, speed in an accurate and stable way. The case study find that Plague shows spatially homogeneous variation, and incidence was generally higher along the coastal areas than in the inland areas, gradually decreasing from the year 1945 to 1952, the most serious attack occurred in Zhangzhou County and Quanzhou County of Fujian Province in the year of 1945. Ships traveling between coastal cities have accelerated Plague spread, and the spread speed was about 12.74 km/a according to the velocity vector obtained. In summary, the proposed method can effectively reduce the modeling complexity, it can also offer faster and more powerful information to the policy makers. What’s more, the spatiotemporal dimension reduction model extends the BME framework and leading a possible way for more accurate and efficiency development of the method. |
PDF Full Text Request