地理科学中的人工智能
参考资料：
- artificial intelligence in geography, S.Openshaw C.Openshow
- handbook of geospatial artificial intelligence, Song Gao, Yingjie Hu, Wenwen Li
地理现象的空间性
- 区域，现象：空间表达
- 关联，联系：空间依赖性
- 变异，差异：空间异质性
- 地表，概括：空间格局
- 表达：MAUP，Ecological Fallacies——场模型/对象模型/网络模型——空间狙击型数据（intensive，空间聚合时不可累加，满足空间稳态假设，连续，采样规则，属性单调，常见于趋势分析、表面建模）/延展性数据（extensive，空间聚合时可累加，稀疏，采样不规则，属性丰富，常用于回归分析及预测），Distinguishing extensive and intensive properties for meaningful geocomputation and mapping [2019] Scheider S., Huisjes M.D
- 建模：空间机理，spatial principles
- 推测：时空切片下的数据不完备性：稀疏、缺失的空间分布；难以观测的空间交互；度量的错位（尺度、采样、地理语境等）
朱递, 地理现象空间分布的人工神经网络建模与分析方法研究, 北京大学博学位论文, 2020
空间显式的地理人工智能
- Geo-spatial data: location(x, y) + value(atttribute)
- Aspatial analysis: non-spatial analysis, look at what instead of where is what, locational invariance
- Spatial analysis: everything changes when location changes
Spatial Explicit Model（SEM）:最早出现在生态学领域：
- Spatially explicit population models: current forms and future uses [1995]
- Theory, data, methods: developing spatially explicit economic models of land use change
- Spatially Explicit Modeling in Ecology: A Review [2017]
AI与地理科学的渊源：
- 联结主义时期：概念雏形，ARTIFICIAL INTELLIGENCE IN GEOGRAPHY: CONJECTURES ON THESHAPE OF THINGS TO COME，
- 计算神经网络：方法雏形，Computational neural networks: a new paradigm for spatial analysis
GeoAI 会议及研讨会:
- ACM sigspatial GeoAl workshops and research articles (GeoAl’17,18, 19, …)
- AAG GeoAl and Deep Learning symposiums (2018, 2019, 2020, …, 2023)
The integration of geography and AI has given rise to the new and exciting interdisciplinary field of geospatial artificial intelligence. (GeoAI)
GeoAI can be regarded as a study subject to develop intelligent computer programs to mimic the processes of human perception, spatiaL reasoning, and discovery about geographical phenomena and dynamics; to advance our knowledge; and to solve problems in human environmental systems and their interactions. With a focus on spatial contexts and roots in geography or geographic information science(GIScience).
GeoAI is an emergent spatial analytical framework for data-intensive GIScience which leverages recent breakthroughs in machine learning and advanced computing to achieve scalable processing and intelligent analysis of geospatial big data.
地理领域的相关定义论文：
- Issues in spatially explicit modeling [2001] Goodchild M.
- GeoAI: spatially explicit artificial intelligence techniques for geographic knowledge discovery and beyond [2019] Krzysztof Janowic, Song Gao, Yingjie Hu & Budhendra Bhaduri et al.
- Agent-based models of land-use and land-cover change. James D.A Millington, John Wainwright
几条准则(2001, Goodchild M.)：
- The invariance test: spatially explicit models are not invariant underrelocation
- The representation test: spatially explicit models include spatial representations in their implementations
- The formulation test:spatially explicit models include spatial concepts in their formulations
- The outcome test: spatial structures of inputs and outcomes are different
人工智能与地理空间的内在联系
空间表达的先验+机理建模的策略+时空预测的场景：
location encoding != geocoding
A review of location encoding for GeoAI: methods and applications [2022] Gengchen Mai, Krzysztof Janowicz, Ling Cai & Ni Lao et al.: A NN-based encoding process which represents a point/location into a high dimensionalvector/embedding such that this embedding can preserve different spatial information (e.g. distance, direction) and, at the same time, be learning-friendly for downstream machine learning (ML) models such as neural nets and support vector machines (SVM). Distance Preservation: nearby locations to have similar embeddings. Direction Awareness: locations that point into similar directions have similar embeddings.
A foundation representation framework for different spatial data types? (points,raster, olylines,polygons, and graphs (networks))
- 卷积和池化：二维卷积一般用于局部连接的神经元（空间依赖先验）；池化合并局部特征，以保证平移不变性（Non-spatial先验）
- 稀疏连接和感受野：空间显式特征是否会网络深度增加而更难学到?如何强调长程的空间效应 (距离衰减) ?
地理空间机理的建模：
- 从表达上，如探究在不同语境下的空间连接关系下的GNN：Understanding place characteristics in geographic contexts through graph convolutional neural networks [2020] Zhu Di
- 从网络结构上，如改造loss function：Region2Vec: Community Detection on Spatial Networks Using Graph Embedding with Node Attributes and Spatial Interactions [2022] Yunlei Liang, Jiawei Zhu, Song Gao et al.
- 异质性：localized/more complex parameters v.s. interpretable/simpler parameters，空间异质性的显式考虑，在transformer中引入position encoding:Spatial-Temporal Transformer Networks for Traffic Flow Forecasting [2020] Mingxing Xu, Wenrui Dai, Hongkai Xiong et al.
空间推测：
spatial represenation –> (collect) –> data samples –> (derive) –> spatial relationships –> (predict/interpolate) –> spatial distribution of target variable
分类：插值/尺度变换/流数据补全/预测/知识推理
e.g.
- Yao, X., Gao, Y., Zhu, D., Manley, E, Wang, J, & Liu, Y. (2020). Spatial origin-destination flow imputation using graphconvolutional networks. IEEE Transactions on intelligent Transportation Svstems, 22(12), 7474-7484.
- Zhu, D, Liu, Y., Yao, X.& Fischer, M.M.(2021) Spatial regression graph convolutional neural networks: A deep learning paradigm for spatial multivariate distributions. GeoInformatica, 1-32.
地理人工智能的城市应用
- 自然环境与地球系统：Deep learning and process understanding for data-driven Earth system science，Deep learning in environmental remote sensing: Achievements and challenges，Skilful precipitation nowcasting using deep generative models of radar
- 城市视觉：Urban visual intelligence: Uncovering hidden city profiles with street view images
- 其他：Machine learning for spatial analyses in urban areas: a scoping review, A review of spatially-explicit GeoAI applications in Urban Geography
GeoAI问题与挑战