在总结多年来研究GIS智能计算的理论与实践基础上,提出地理信息科学发展的新方向:空间智能.空间智能强调发现与应用空间模式,以增强GIS处理复杂数据和解决复杂问题的能力.空间智能主要的技术体系由空间分析、空间优化和空间模拟三大模块构成,其技术基础包括空间统计与索引、智能代理、高级启发式,以及数学规划等系列智能技术.由于空间智能融合了机器学习、统计分析和人工智能等多个学科理论,面向解决实际工程需求中大量存在的复杂时空问题,因此理论上具有广阔的发展空间,实践上也有重大的应用需求.随着空间智能体系的完善和技术的进一步成熟,它将在实际应用中具有巨大的价值.

Notwithstanding the astounding growth achieved by Geographic Information Systems (GIS) in recentdecades,some criticalbottlenecks still continue to pose challenges to the advancement of this technology.The inability to handle large and diverse datasets and the lack of functionalities to solve complex spatial problems such as combinatorial location problems is essential one among these hurdles.Recent advances in Computational Intelligence (CI) and Operations Research have, however, opened up new avenues to overcome these obstacles to the development of GIS cience.Judicious integration of the aforementioned techniques into GIS cience could possibly lead to an innovative discipline,namely spatial intelligence. This paper presents our prelmi inary investigation on this subject, including its framework,major acquisitionmethods, and sample applications. The basic concept of spatial intelligence derived from psychology refers to themental process associated with the brain’s attempts to interpret certain types of information received. This\ninformation basically includes anykind ofmental input, such asvisualpictures, maps, and plans. Based on this concept,we propose that the introduction of spatial intelligencewithin the domain of GIS cience is an ability to discover and apply spatial patterns, which is usually elicited through analysis/mining, optmi ization, and smi ulation. Two characteristics of spatial intelligence are highlighted here. One is the ability of spatial cognition, and the other is the self learning capability. Spatialcognition refers to the processofrecognizing, encoding, saving, expressing, decomposing, constructing and generalizing spatialobjects, which can be obtained from spatialobservation, spatial perception, spatial indexing, and spatial deductive inference. Self learning includes enforcing learning, adaptive learning, and knowledge acquiring abilities to actively dig up knowledge from observation data. Promoting spatial intelligence is a logical requirement forhigher level analysis and application ofGIScience. Through active learning and searching process in complex spatiotemporal data,spatial intelligence discovers unknown spatial patterns, trends, and regularities.From the technical perspective, we emphasize the use ofmeta-heuristic and other intelligent algorithms to address complex geospatial problems. A three-tiered structure is proposed for the spatial intelligence framework. At the bottom of the framework, spatial statistics,programming, and intelligence computation are used to provide the foundation of spatial analysis, smi ulation, and optmi ization. The middle level consists of spatial intelligence, self learning, and spatial cognition foranalyzing, smi ulating, interpreting, and decisionmaking forgeospatialprocesses and phenomena. At the top of the framework, GIScience laws and regularities are used to mine unknown patterns. To realize the goal of spatial intelligence, a solid research that integrates key topics with the concepts and modeling approaches derived from Information Science andOperationsResearch to advanceGIS theories have been developed. The core supportingmethods and techniques pertinent to the proposed framework include spatial analysis, smi ulation, and optmi ization. The development of spatial analyticalmodels to represent spatial and temporal features and their relationships forms a vital aspectof this research. Spatial optmi ization is employed tomaxmi ize orminmi ize a planning objective, given the lmi ited area, finite resources, and spatial relationships for a location-specific problem,once spatiotemporal patterns have been discovered. Smi ulation is an mi portant tool to evaluate and mi provemodels and spatial patterns.Some successfulapplicationsofspatialanalysis, optmi ization, and smi ulation are also reported in thispaper. Logistic regression models,e.g.,binomia,l multinomia,l and Nested Logit,are applied and examined to predict various spatiotemporal changes, including rural to commercia,l rural to recreationa,l and rural to other land uses.A range of heuristic algorithms, such as Geneti Algorithms (GA) and Ant Colony Systems (Ant), to troubleshoot complex routing and location problems,andmulti-objective optmi ization forspatialdecisionmaking are studied.A case study of integrating agent-basedmodelingwith analyticalmodelsby drawing uponmicroscopic traffic smi ulations for emulating real-tmi e traffic conditions is also conducted.