Hartshorne, Schaefer, and the Question GeoAI Cannot Escape Hartshorne、Schaefer 与 GeoAI 无法逃避的问题

The Hartshorne-Schaefer debate was not a small quarrel over technique. It was a dispute over what geography was supposed to be. Hartshorne, writing from the regional tradition, treated geography as the disciplined study of areal differentiation: how physical, social, historical, and cultural phenomena come together in particular places. Schaefer, writing against what he called "exceptionalism," argued that geography should seek explanatory regularities in spatial arrangement: location, distance, distribution, interaction, and pattern.

The shape of the debate was therefore not "old description" versus "new numbers." It was synthesis versus generalization, place versus law, idiographic judgment versus explanatory science. Hartshorne worried that geography would lose its object if it forgot the complexity of regions. Schaefer worried that geography would lose its scientific force if every place became too unique to explain. Both worries were real.

What made the dispute historically consequential was that Schaefer's side became a program. At the University of Washington, William Garrison's students, later remembered as the space cadets, turned spatial science into courses, statistics, transport models, urban systems, early computing, and dissertations. A philosophical challenge became a training machine. Geography learned to value explanation, comparison, modeling, and prediction in a new way.

This change did not simply erase Hartshorne. It left geography with a productive wound. Quantitative geography, GIS, spatial analysis, and later spatial data science all inherited Schaefer's confidence that spatial pattern can be modeled. But every serious geographic application still runs into Hartshorne's warning: places are not only coordinates, units, or variables. They are historically assembled relations.

This is the point from which to read the AI era, especially GeoAI. Remote sensing foundation models, urban prediction systems, disaster maps, digital twins, mobility models, and spatial knowledge graphs all extend the Schaeferian promise: spatial patterns can be learned, transferred, simulated, and optimized. AI does not merely add a tool to geography. It changes what can be observed, compared, classified, forecast, and acted upon.

But the AI era also revives Hartshorne's question with new force. If a model recognizes a roof but misses tenure, detects a road but misses who can move along it, or maps vulnerability while inheriting the categories that made vulnerability hard to see, then it has not understood place. It has only translated place into features. This is the central danger of GeoAI: context may appear inside the model while geographic judgment disappears from the workflow.

The lesson is not to reject AI, or to return nostalgically to regional description. It is to make geography's old tension operational. GeoAI should explain patterns, but also test where transfer fails. It should use foundation models, but document where they cannot travel. It should produce maps, but keep open the authority to revise them through local knowledge, field evidence, data sovereignty, and community validation.

The Hartshorne-Schaefer debate changed geography because it changed how geographers were trained to think. AI will matter in the same way only if it changes geographic training critically: not just more models, but better questions about scale, place, uncertainty, power, and the limits of generalization.

Hartshorne-Schaefer 之争，并不是一场关于技术细节的小争论。它争的是： 地理学到底应当成为什么。Hartshorne 站在区域传统中，把地理学理解为对 区域差异的有纪律研究：自然、社会、历史与文化现象如何在特定地方组合起来。 Schaefer 则批判所谓"例外主义"，要求地理学寻找空间安排中的解释性规律： 区位、距离、分布、相互作用与格局。

所以，这场论战的格局并不是"旧描述"对"新数字"。它更像是综合与泛化、 地方与法则、个体判断与解释性科学之间的张力。Hartshorne 担心，如果忘记区域 的复杂组合，地理学会失去自己的对象；Schaefer 担心，如果每个地方都独特到 无法解释，地理学会失去科学力量。两种担心都是真的。

它之所以改变了学科，是因为 Schaefer 这一侧后来变成了研究计划。在华盛顿大学， William Garrison 的学生，也就是后来被称为 space cadets 的那群人， 把空间科学变成课程、统计、交通模型、城市系统、早期计算和博士论文。 一个哲学挑战，变成了一套训练机器。地理学从此以新的方式重视解释、比较、 建模与预测。

但这并不意味着 Hartshorne 被简单抹掉了。它给地理学留下了一道有生产力的伤口。 计量地理、GIS、空间分析，以及后来的空间数据科学，都继承了 Schaefer 对空间模式 可建模的信心；但任何严肃的地理应用，仍然会撞上 Hartshorne 的提醒： 地方不只是坐标、单元或变量。地方是历史中组合出来的关系。

从这里看 AI 时代，尤其是 GeoAI，就会更清楚。遥感基础模型、城市预测系统、 灾害地图、数字孪生、流动模型与空间知识图谱，都在延伸 Schaefer 式的承诺： 空间模式可以被学习、迁移、模拟和优化。AI 不只是给地理学增加一个工具。 它正在改变什么可以被观察、比较、分类、预测和行动。

但 AI 时代也以新的强度带回了 Hartshorne 的问题。如果模型认出屋顶，却错过产权； 检测出道路，却错过谁能沿着道路移动；绘制脆弱性，却继承了那些让脆弱性难以 被看见的分类，那么它并没有理解地方。它只是把地方翻译成了特征。这是 GeoAI 的核心危险：语境似乎进入了模型，而地理判断却从工作流程中消失。

教训不是拒绝 AI，也不是怀旧地回到区域描述。教训是把地理学的这道旧张力变成 可操作的原则。GeoAI 应当解释模式，也要测试迁移在哪里失败；可以使用基础模型， 但必须说明它们在哪里不能旅行；可以生产地图，但要保留通过地方知识、田野证据、 数据主权和社区验证来修正地图的权力。

Hartshorne-Schaefer 之争改变地理学，是因为它改变了地理学家被训练去思考的方式。 AI 也只有在这个意义上才真正重要：不是更多模型，而是更好的地理问题，关于尺度、 地方、不确定性、权力，以及泛化的边界。