Quantifying and modelling decay in forecast proficiency indicates the limits of transferability in land-cover classification

1. The ability to provide reliable projections for the current and future distribution patterns of land-covers is fundamental if we wish to protect and manage our diminishing natural resources. Two inter-related revolutions made map productions feasible at unprecedented resolutions- the availability of high-resolution remotely-sensed data and the development of machine-learning algorithms. However, the ground-truth data needed for training models is in most cases spatially and temporally clustered. Therefore, map production requires extrapolation of models from one place to another and the uncertainty cost of such extrapolation is rarely explored. In other words, we focus mainly on projections, and less on quantifying how reliable they are. 2. Following the concept of ‘forecast horizon’, we suggest that the predictability of land-cover classification models should be methodologically explored with quantitative tools as a continuum against distances measured along multiple dimensions. Fo...1. 若要保护并管理日益枯竭的自然资源，对土地覆盖（land cover）当前与未来分布格局提供可靠预测的能力乃是核心前提。两项相互关联的变革使得以空前分辨率开展地图制图成为可能——高分辨率遥感数据的可获取性，以及机器学习算法的发展。然而，训练模型所需的实地真值（ground-truth）数据，在多数情况下于空间和时间维度上均呈现聚集特性。因此，地图制图工作需要将模型从一个区域外推至其他区域，但这类外推所带来的不确定性成本却鲜有探讨。换言之，当前研究多聚焦于预测结果本身，却较少对其可靠性进行量化评估。 2. 基于“预测视界（forecast horizon）”这一概念，我们提出应通过定量工具，以多维度距离为连续统一体，从方法论层面探究土地覆盖分类模型的可预测性。Fo...