Geographical system, composed of natural and human elements, distributed in the earth's surface space, is provided with the functions of generating, maintaining or transforming material forms and energy forms, or driving the flow of material and energy. It is characterized by natural and human elements and their interactions on spatial distribution, structure, pattern, evolution and function factors. Geographical system is an objective reality system which generally consisted in earth with multi-levels. The geographic system was represented by the geographic information system by a information world with digital form. Then make the plans to transform geographical system by investigating and analyzing the received information. Therefore, two geographic systems, namely real geographic system and information geographic system, are co-existed for a long time and evolved independently. In recent years, with the integration of real-time observation and Internet of Things Technology (IoT), the real-time representation from the real geographic system represented into the information geographic system is coming true. Therefore, two systems evolve no longer independently, and the information geographic system is becoming a mirror of the real geographic system. Once the real geographic system changes, the information geographic system changes simultaneously. At present, the development with an unprecedented speed of human science and technology, especially the cloud services, big data and artificial intelligence, makes the information geographic system no longer be satisfied with being a mirror of the real geographic system. The artificial intelligence and unmanned automation control technology are adopted to achieve the integration of the real world and information world. The real geographic system can be changed through the operation of information geographic system. In the future, the boundary between the real geographic system and the information geographic system will become blurred and the two systems will be fully integrated with no distinguish at last. This article takes the intelligent geographic system to indicate this system, which integrates two worlds, and uses an unattended intelligent system designed for the South Island Reef as an example to show the integration and interoperation of the real geographic system and the information geographic system. The system realizes unmanned and intelligent monitoring, protection, response and operation for the island reef geographic environment and artificial facilities.

The space-air-ground-sea stereo and integrated geospatial sensor web have gradually formed along with the development of remote sensing data network, sensor web, internet of things (IoT), and artificial intelligence. The sensing resources of the geospatial sensor web are of multiple sources, heterogeneity, and dispersion. These characteristics result in the grand technical challenges of sharing and managing heterogeneous resources, real-time access of geospatial information with multiple protocols, spatiotemporally seamless and autonomous sensing of geospatial information, and accurate prediction of key parameters, especially when facing the personalized, instant, and smart application needs of multi-level users. It is hard for static geographical information service to meet the demands of geo-events for integrated monitoring, early warning and decision support, and focusing application. Therefore, there are urgent needs to develop fusion service technologies of the geospatial sensor web, as well as real-time dynamic geographical information service platforms. To solve these problems, this paper proposed online access, integrated management, space-ground fusion, spatiotemporal prediction, and focusing service models and methods. With the online access, dynamic management methods for sensing spectrum resources, and transparent access methods based on heterogeneous sensor protocols pool were proposed; a cyber-physical spatiotemporal information service environment was established, which realized the efficient access of spatiotemporal information with heterogeneous protocols. With the integrated management, sharable and interoperable information models including sensor observation process information description model, observation data description model, observation event description model, and dynamic observation capability index model were proposed, which tackled the coupling problem of sensor web and GIS, and realized large-scale integrated management and sharing of space-air-ground-sea platforms and sensors for the integrated monitoring of fairway, hydrology, soil, meteorology, and ocean. With space-ground fusion, a point-surface-collaboration and seamless reconstruction model, and evaluation-collaboration-reconstruction, cross-scale, seamless and continuous sensing methods were proposed, which improved the sensing quality by 14 times with respect to using satellite only and meanwhile keeping sensing frequency the same as the station networks, providing new ways for continuous monitoring of resources, environments, and disasters. For spatiotemporal predictions, ensemble models of multiple machine learning models, ensemble models of statistical models and dynamic models, and a spatiotemporal deep learning model were proposed, which realized high-resolution and high-precision predictions of meteorological parameters at regional scales. For the focusing service, a geo-control method based on instant sensing feedback, a time-continuous maximal covering location model, and an automatic aggregation sensing method were proposed, which improved the spatiotemporal coverage of sensing by 18%, and realized active and on-demand sensing of spatiotemporal information. Based on the sensor web observation information models and using the architecture of satellite-ground-collaboration spatiotemporal information sensing as a service, a geospatial sensor web spatiotemporal information sensing and service system named GeoSensor was developed, which has the functions of sensing, access, cognition, and control. The GeoSensor has been successfully applied to the sensing management and service of spatiotemporal information in the Yangtze River, the ocean and the smart city. In the future, the theory of smart sensing and cognition of people, water, and city will be further developed, the technology of crowd-sourced sensing, spatial intelligence, and cognition service of space-air-ground-sea-people will be developed, and large-scale applications in the Yangtze River Economic Zone will be conducted as well.

Existing research on human behavior based on the PIR (Passive InfraRed) sensor data is limited by the spatial-temporal distribution of motion, clustering, and so on. The reconstruction of behavior trajectory and analysis of semantic features are relatively few, so it is urgent to develop new modeling and behavior analysis methods for the PIR data. This paper attempts to reconstruct the spatial and temporal trajectories by using the PIR (Passive InfraRed) sensor monitoring data. PIR sensors have the characteristics of low price and privacy protection. However, because only Boolean logical response sequence can be obtained by PIR sensors, it is difficult to accurately obtain movement trajectories. Its application has been relatively limited, and it is difficult to conduct movement behavior feature analysis. Traditional PIR sensor network analysis methods are mostly based on the signal extraction idea, which cannot integrate geometric features and semantic information at the same time. By introducing the geometric algebra theory, the sensor scene network can be constructed to realize the path expression and calculation of dynamic network in the geometric algebra space. This paper analyzed the characteristics of human movement features and semantic features, established semantic units, and realized the transformation from spatial data to semantic features. The spatial and topological characteristics of individual and crowd movements were analyzed. We proposed a generation and transformation-based methods of algebraic structures in the geometric algebra system, which provides a new idea and mathematical basis for solving non-deterministic problems such as the PIR sensor network data based analysis, and can provide reference for the construction of internet of things GIS.

Geographical flow can be defined as the movements of geographical objects between different locations, which are usually displayed as the movement of matter, information, energy and funds, e.g. the jobs-housing flow in a city, communications between different mobile phone holders and the fund transferred between different business entities. Due to the existence of the various flows, the link strength between different locations may not depend on distance only, say one may strongly related to a store faraway through express delivery rather than a store nearby. The traditional knowledge of distance-decay law may be changed. As a result, research on the geographical flow may help to understand geographical patterns and their mechanism from a new point of view. Two conceptual models are introduced for the expression of geographical flows in this paper. In the first model, a flow is abstracted as a coordinate quaternion composed of the origin point and the destination point (called the orthonormal flow model). Thus, the flow space can be defined as a 4-D space which is formed by the Cartesian product of two 2-D spaces. In the second model, a flow is composed of the origin point coordinates, the flow length and the flow angle (called the polar coordinate model). Based on the expression models, four distances are defined, specifically, maximum distance, additive distance, average distance and weighted distance. In addition, this paper defines some other flow measurements, including flow direction, the volume of a flow's r -neighborhood and the flow density. According to the combination of different statistical features (i.e. heterogeneity, homogeneity and randomness) between variables in the polar coordinate model, the spatial patterns of geographical flows are divided into six single patterns including random, clustering, convergent and divergent, community, parallel (angle-clustered) and equilong (length-clustered). The methods for identifying different flow patterns are also analyzed and summarized. Besides the single patterns, the combination of different single patterns will generate mixed patterns, and if more than one type of flows coexists, multi-flow patterns can be produced. Regarding research directions of geographical flow in the future, three aspects should be given more attentions: the basic statistical theory of flow, the mining method of flow pattern and its application in practical problems.

Multi-modal spatio-temporal analysis is aimed at discovering valuable knowledge about the spatio-temporal distributions, associations and revolutions underlying the multi-modal geographic big data. It is a core task of the pan-spatial information system, and is expected to facilitate the study of relationship between human and space. With emerging opportunities and challenges in an era of geographic big data, we systematically summarized four main methods for spatial-temporal analysis based on previous study, including spatio-temporal cluster analysis, spatio-temporal outlier detection, spatio-temporal association mining and spatio-temporal prediction. We discussed the challenges when applying the four methods in multi-scale modeling, multi-view fusion, multi-characteristic cognition, and multi-characteristic expression for spatial-temporal analysis. First, two types of scales (including data scale and analysis scale) are of great importance in the spatio-temporal clustering task. Given the data scale, the best analysis scale for detecting spatio-temporal clusters can be determined using a permutation test method by evaluating the significance of clusters. Second, in the spatio-temporal outlier detection method, the cross-outliers in the context of two types of points are known as the abnormal associations between different types of points and the validity of cross-outliers is assessed through significance tests under the null hypothesis of complete spatial randomness. Third, in the spatio-temporal association mining method, the multi-modal distribution characteristics of each feature quantitatively described in the observed dataset are employed to construct the null hypothesis that the spatio-temporal distributions of different features are independent of each other, and then the evaluation of spatio-temporal associations is modeled as a significance test problem under the null hypothesis of independence. Finally, in the spatio-temporal prediction model, the effects of multiple characteristics of spatio-temporal data (e.g., spatio-temporal auto-correlation and heterogeneity) on the prediction results are fully considered using a space-time support vector regression model. These methods can reveal the geographic knowledge in a more comprehensive, objective, and accurate way, and play a key role in supporting the smart city applications, such as meteorological and environmental monitoring, public safety management, and urban facility planning. For example, the spatio-temporal clustering method can be used to identify the meteorological division, the spatio-temporal outliers can contribute to the detection of the abnormal distribution of urban facilities, the spatio-temporal association mining method can help discover and understand the relationship among different types of crimes, and the spatio-temporal prediction method can be employed to predict the concentration of air pollutants.

In the era of big data, high-resolution Earth observation technologies have been able to provide the most authentic, quantitative, comprehensive-coverage, and fast-updating data about the geographic phenomena and processes on the Earth's surface. Such data provide precise spatiotemporal benchmarks of information aggregation and computation of data mining for new developments of geospatial cognitive research. Geo-parcels are abstract expressions for mapping geographical entities from image-space to geographic-space. Geo-parcels are the smallest units of pattern mining with the construction of geographic scenes and loading various geospatial information. In this paper, a synergistic calculation mechanism with the machine learning methods of visual simulation and symbol inference were analyzed based on the basic unit of geo-parcels. From the dimensions of space, time, and attribute, we constructed an intelligent computation model based on geo-parcels by integrating three sub-models: zoning-stratified perception, spatiotemporal synergistically inversion, and multi-granular decision-making. Furthermore, this paper explored the pattern mining methods of geo-parcels for their distribution, growth, and function via two case studies: the agricultural planting structure mapping in Xifeng County, Guizhou province and the planning decision in Jiangzhou District of Guangxi Zhuang Autonomous Region.

As the representative technology of Artificial Intelligence, deep learning has been the most exciting breakthrough technologies in big data analysis and other domains researches due to its novel data-driven feature representations learning, instead of handcrafting features based on domain-specific knowledge in traditional modeling. Driven by these technological developments. Artificial Intelligence plays a key role in the researches and applications of next-generation geographical information system software technology. Nevertheless, most researches about AI GIS are still in the stage of immature and preliminary exploration. As a method and technology for the novel architecture of GIS fundamental software, AI GIS is widely used in many earth science applications including remote sensing data analysis, water resources research, spatial epidemiology and environmental health. All these technologies are significantly improving capabilities of data processing of traditional GIS, and being able to extract geospatial information and characteristics from unstructured datasets such as street view or remote sensing imagery, texts. These applications are showing great value and developing potential of AI GIS. However, the existing research on the system of software technology of AI GIS is not comprehensive enough. A variety of AI GIS algorithms or models and their scenario-specific applications are commonly considered to be the most important topic. Few researchers have addressed the issues or theory of Artificial Intelligence GIS technologies system and software architecture. This paper presents and analyzes several levels of Geo-intelligence and discuss its relationships to AI GIS technology system , reviewed the research status in AI and GIS technologies from the domestic and abroad perspectives. Then, the system of software technology of AI GIS is proposed according to the relationships between Artificial Intelligence and GIS. This paper define the architecture of AI GIS into three parts including Geospatial Artificial Intelligence(GeoAI), AI for GIS, and GIS for AI. And concepts and examples for each parts of Artificial Intelligence GIS are also analyzed for illustration. Furthermore, in order to deeply explain and investigate the AI GIS software technologies architecture, this paper provide the example of the design and implementation of SuperMap AI GIS software architectures and production. Finally, this paper discusses the problems that need to be solved in the future development of GIS. The tentative study of AI GIS in this paper may provide a theory for establishing the fundamental GIS software technology architecture of Geo-intelligence, which would helps to promote the deep integration and development of AI and GIS technology, and make suggestions for further research about Geo-intelligence.

Traffic signs are essential elements in High Definition (HD) maps and hence very important for vehicles in autonomous driving. Real-time and accurate detection of traffic signs from street level images is of great significance for the development of autonomous driving. Conventional algorithms detect traffic signs based on image color and shape features, and can only work for specific kinds of traffic signs. Algorithms based on image feature and machine learning classifier need artificial designed features, and the detection speed is slow. To date, many approaches using deep learning methods have been developed based on anchor boxes, which introduce extra hyper parameters in network design. When switching to a different detection task, anchor boxes need to be redesigned. Anchor-based methods also generate massive redundant anchor boxes during model training, which easily cause imbalance between positive and negative samples. Inspired by the idea of anchor-free and YOLO, this paper proposed a real-time traffic sign detection network called AF-TSD, which regresses object boundary directly. AF-TSD adopts an effective convolution module named deformable convolution to enhance the feature expression ability of convolutional neural networks. This module adds 2D offsets to the regular grid sampling locations in the standard convolution. It also modulates input feature amplitudes from different spatial locations/bins. Both the offsets and amplitudes are learned from the preceding feature maps, via additional convolutional layers. In addition, AF-TSD introduces attention mechanism. It is inserted after fusion of the feature pyramid, and adaptively recalibrates channel-wise feature responses by explicitly modeling the interdependencies between channels. This module first squeezes global spatial information into a channel descriptor. Then the excitation operator maps the input-specific descriptor to a set of channel weights. The attention mechanism in this paper is lightweight and imposes only a slight increase in model complexity and computational burden. To test the superiority of AF-TSD, extensive comparative experiments were carried out. We first evaluated the influence of different modules on detection precision. The experimental results show that the deformable convolution and attention mechanism can help extract features of traffic signs. Then, AF-TSD was compared with mainstream detection networks, including Faster R-CNN, RetinaNet, and YOLOv3. Our proposed AF-TSD traffic sign detection network achieved 96.80% of mAP on GTSDB traffic sign detection dataset, which was superior to mainstream detection algorithms. The average detection speed was 32ms per images, which can meet the requirements of real-time detection.

The spatial distribution characteristics and activity patterns of urban populations play essential roles in studies of spatial isolation, optimizing urban resource allocation, and so on. Because of the sensitivity of population activity data and socioeconomic data, previous studies focus mostly on the macro level. They have difficulties in dividing the socioeconomic status and quantitatively analyzing human mobility regulation. In recent years, geospatial big data, such as the mobile app data, provide us with a rare opportunity to analyze the human activity of urban internal problems. In this study, we constructed a fine-grained activity portrait of mobile phone users based on the mobile phone signaling data of Shenzhen residents, and coupled the high-resolution Shenzhen house price distribution data to achieve accurate division of people by their economic levels. Then, we extracted six activity indicators, which include the number of active locations, activity entropy, moment of inertia, travel time, travel distance, and travel speed, to quantify the spatial distribution and analyze the activity patterns of people at different economic levels. The results reveal the correlation between mobility and socioeconomic status. The distribution of people's activities at different economic levels in Shenzhen was related to the economic development of each administrative region. The results also demonstrated that three activity indicators (moment of inertia, travel distance, travel speed) were positively related to the economic level. Residents across different socioeconomic classes exhibited different travel patterns. Likely because the rich people live in the southwest of Shenzhen, but their work locations have more self-selectivity. This leads to the distribution of home and work locations in different administrative districts and the home-work distance of high-economic people are larger than others. For the other three activity indicators (number of active locations, activity entropy, travel time) that reflect the similar pattern of activity between different socioeconomic status, we found that people were mainly concentrated in living and working locations on weekdays. These locations share activities on weekdays for people at different socioeconomic levels. The socioeconomic status does not affect the number of daily activities nor the scheduling of activities. This study provides necessary data and policy guidance for government and urban planners.

Spatial accessibility is a basic decision-making question constrained by spatial distribution situations and infrastructure conditions. The isochrone model can help address this question by applying temporal geography principles to measure by time-costs. The spatial accessibility aiming at POI facilities can be conducted from two perspectives in general: (1) on the supply side, the supplier's service accessibility is analyzed starting from specific POIs; (2) on the demand side, the customer’s service convenience is analyzed starting from any location to get the nearest service. However, most studies focus on the service area of specific POIs from the perspective of supply. There is no enough attention on isochrones from the perspective of demands. To help close this gap, this study conducted the isochrone analysis of POI accessibility from the perspective of service demand, to explore the time cost starting from any place to the nearest service facility. The geometric measure applies the network Voronoi diagram model considering the distance difference from that in Euclidean Voronoi diagram. We first theoretically explained the idea of the isochrone model in terms of service demands, and then detailed the steps to apply the isochrone model. To construct the model, the POI data were firstly matched with road networks. Then a network Voronoi diagram model was developed using hospital POIs and road networks to acquire massive useful sampling points along road networks and to calculate their network distance to their nearest POIs. The isochrone model was finally constructed using the sampling points and their distance properties which reflect space-time accessibility of any place in the city in the demand of medical care to get the most convenient service. The study used the isochrone model to reveal the potential relationships between the space-time accessibility of POIs and road networks. The influence of road levels was also examined in comparison with a weighted network Voronoi diagram model. It was found that the morphology of the isochrones had close association with the distribution of POIs and the connectivity of road networks.

China is becoming one of the most air-polluted countries and is experiencing severe PM_2.5 pollution. To acquire spatially continuous PM_2.5 estimates, numerous statistical methods have been developed through the integration of ground-level measurements and satellite-based observations. The estimation of PM_2.5 concentrations in China is characterized by significant spatial nonstationarity and complex nonlinearity due to the complicated terrain variability and wide geographical scope. Mapping the PM_2.5 distributions across China with high accuracy and reasonable details is still challenging. Superior satellite-based PM_2.5 estimation models need to be developed. Taking advantage of a newly proposed Geographically Neural Network Weighted Regression (GNNWR) model that simultaneously accounts for spatial nonstationarity and complex nonlinearity, we developed a satellite-based GNNWR model to obtain spatially continuous PM_2.5 estimates in China. To comprehensively assess the predictive power of the GNNWR model, the widely used Ordinary Linear Regression (OLR) and Geographically Weighted Regression (GWR) models were also carried out for performance comparison. Experimental results demonstrated that the GNNWR model performed considerably better than the OLR and GWR models in terms of multiple statistical indicators, including coefficient of determination (R ²), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). Most notably, the fitting accuracy of GNNWR was slightly better than GWR, but its prediction ability was much superior to GWR since the predictive R ²of GWR was significantly improved from 0.683 to 0.831 and the RMSE value was considerably reduced from 9.359 to 6.837. Moreover, the mapped PM_2.5 distributions derived from the GNNWR model presented more reasonable and finer details at a higher accuracy than the other models. Although the spatial trends estimated by GWR and GNNWR models were quite consistent, the estimates of the GNNWR model were more accurate and reasonable since its values were much closer to the ground monitoring observations than those of the GWR model, especially for areas with high PM_2.5 concentrations, such as Hebei Province and southern Shaanxi Province. In addition, thanks to the excellent learning ability of the neural network, the spatial variations in GNNWR estimates were more sophisticated and displayed a richer hierarchical structure of local changes than that of GWR estimates, which better described the varying details of the PM_2.5 across China. In summary, the GNNWR model is a reliable method to effectively estimate PM_2.5 concentrations and can also be used to model various air pollution parameters.

With the development of information and communication technology, mobile phones have become an indispensable part of human daily life. Human activities have gradually extended from real space to cyberspace. The online behavior of cyberspace in the era of mobile Internet is inseparable from the travel behavior of real space. The current individual travel behavior predictive modeling is less concerned with the relationship between online behavior and travel behavior. A mobile phone user stay behavior prediction model based on the characteristics of online app usage behavior is proposed. Firstly, the time-space constraint is used to define the mobile phone user's stay behavior. Then, from multi-source data, the paper extracts the individual travel behavior's space-time preference, the app usage characteristics such as the APP combination, Internet traffic, Internet access times and other Internet behavior characteristics and weather information, etc. Feature engineering is done in a time and space crossing way, and the mobile phone user stay behavior prediction model with high interpretability from feature to model is constructed. We found the following from the experimental results: (1) The prediction accuracy of the model is 80.31%. After the integration of online behavior characteristics, weather and other external factors, the prediction accuracy is improved by 12.08%, compared with the model only using individual travel characteristics. (2) The prediction accuracy of the model is higher than that of Support Vector Machine (SVM) and Gradient Boosting Decision Tree (GBDT). And it is 1.23% lower than that of Random Forest (RF), but the model in this paper runs faster than RF, and the model solving process is easy to understand and interpretable. Besides, the first-order Markov model has a small amount of calculations and a fast running speed, but the accuracy is low. In general, the model in this paper has higher accuracy and fast running speed, which is more suitable for mobile phone user stay behavior prediction. (3) There is a big difference in the prediction accuracy of different users' stay behaviors prediction. The prediction accuracy of most users is concentrated between 70% and 90%. The highest prediction accuracy is 98.2%, and the worst prediction accuracy is 34.5%. (4) Among the app usage characteristics, the APP combination, Internet traffic and Internet access times contribute more to the prediction of mobile user stay behavior. The use of navigation, news and office apps has a particularly significant impact on the prediction results. In addition, comparing to the historical travel behavior characteristics, travel distance and activity radius in the current period have a stronger impact on the prediction of mobile phone user stay behavior.