updata

Sampling time as known information can certainly enhance the reliability of density estimation for space-time trajectories. Established kernel density estimation (KDE) methods utilize the sampling time to reshape the kernel or weights, but it is difficult to control two types of errors at the same time: overestimation or underestimation of the density due to temporal autocorrelation, and assignment of non-zero densities to non-reachable points. For this reason, this paper introduces a divide-and-conquer strategy for movement data with irregular sampling intervals, and proposes a time-KDE method under the dual drive of time geography and temporal autocorrelation. It establishes two mappings between time to kernels and to weights respectively in order to reconcile the above two types of errors by recalibrating the KDE, with the aim of maximizing the temporal information to abate the uncertainty of the density estimation, and thus providing a theoretical basis for unbiased density estimation.

已知的采样时间信息，可有效提升时空轨迹密度估计的可靠性。现有核密度估计（Kernel Density Estimation, KDE）方法会利用采样时间对核函数或权重进行重塑，但难以同时管控两类误差：一是由时间自相关导致的密度高估或低估，二是将非零密度赋予不可达点位。为此，本文针对采样间隔不规则的移动轨迹数据提出分治策略，并基于时间地理学与时间自相关的双重驱动，提出了时间KDE方法。该方法分别构建了时间到核函数、时间到权重的两类映射，通过对核密度估计进行重校准以调和上述两类误差，旨在最大化利用时间信息以降低密度估计的不确定性，从而为无偏密度估计提供理论支撑。