High-resolution prediction and uncertainty estimation of land value and conservation costs

对私有土地权利公平市场价值的准确估计是许多公共政策选择的关键因素，包括财产税和保护规划。然而，这项评估任务具有挑战性，因为土地价值受到许多环境和社会因素的影响，它们的重要性因地理而异，而且现有的物业销售培训数据往往稀缺、聚集，很少能代表地貌。这个项目调查了不同的统计技术的性能，以估计土地价值、保护成本和相关的不确定因素，跨越大而不同的地理区域。这项工作产生的数据旨在支持政府机构、保护组织和学术研究团体确定当前和未来物种栖息地、碳储存、减少洪水风险、农田保护和公平进入开放空间的保护优先事项。该项目还有助于博士后奖学金的教育和培训。由于经济危机和气候变化导致社会对土地福利的需求发生变化，需要对土地的私人利益和公共利益之间的权衡有一个准确的预测性了解，以确定有效和公平的政策解决方案。该项目使用一个新的地理空间数据库，其中包括毗邻的美国的大约1.5亿处房产和400万套销售，比较了传统统计估计器和现代空间机器学习方法在预测空间和时间上的土地价值和公共购置成本方面的性能，以及相关的不确定性。它还考察了用于估计部分土地权利和保护地役权的价值的新机器学习技术的性能，这是美国迅速扩大的一种保护工具。通过这样做，该项目有助于在大型、固定的研究领域对土地价值和保护成本的预测建模和验证方面的可推广的最佳实践。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Accurate estimates of the fair market value of private land rights are key inputs for many public policy choices, including property taxation and conservation planning. However, this estimation task is challenging, because land value is affected by many environmental and social factors, their importance varies geographically, and available training data of property sales tends to be scarce, clustered, and rarely representative of the landscape. This project investigates the performance of different statistical techniques to estimate land value, conservation cost, and associated uncertainties across large and heterogeneous geographical areas. The data produced in this effort are designed to support government agencies, conservation organizations, and academic research groups in the identification of conservation priorities for current and future species habitat, carbon storage, flood risk reduction, farmland protection, and equitable access to open space. The project also contributes to the education and training of a postdoctoral scholar.With societal demand for land-based benefits shifting as a result of economic crises and climate change, an accurate predictive understanding of the trade-offs between private and public benefits from land is needed for the identification of effective and equitable policy solutions. Using a novel geospatial database of approximately 150 million properties and 4 million sales for the conterminous United States, this project compares the performance of traditional statistical estimators and modern spatial machine learning methods in predicting land values and public acquisition costs across space and time, alongside associated uncertainties. It also examines the performance of new machine learning techniques for the estimation of the value of partial land rights and conservation easements, a rapidly expanding protection instrument in the United States. In doing so, the project contributes to generalizable best practice in predictive modeling and validation of land value and conservation costs over large, nonstationary study domains.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.