城市热岛作为城市化进程中最为显著的气候效应，与局部背景气候、大气污染、城市的植被物候变化以及人类的健康问题息息相关。随着遥感技术的发展，由热红外传感器获取的地表温度逐渐成为城市热岛研究中的重要数据来源。但由于现有热红外传感器的时空分辨率问题，同时具有高空间分辨率和高时间分辨率的地表温度仍然难以获取，并且热红外影像易受云层干扰，造成大量数据缺失，导致热红外数据的可利用率大幅降低。为了应对以上问题，本文构建了地表温度空间降尺度模型和时间域内插模型，讨论了空间降尺度和时间域内插模型的耦合策略，旨在生产具有高时空分辨率的地表温度数据，并以北京市为例，对时空降尺度数据驱动下多尺度城市热岛效应的演变模式开展定量分析与研究。本文的主要创新点与结论如下：

（1）梳理了地表温度时空降尺度的概念、类型、研究现状及存在的问题。定义了地表温度空间降尺度、地表温度时间域内插以及地表温度时空降尺度的概念，将地表温度的时空信息增强过程分为空间降尺度和时间域内插两部分，并对两个过程中各自存在的问题进行梳理，继而指出时空降尺度地表温度在城市热岛研究中的应用前景。

（2）耦合核驱动降尺度和时空融合降尺度模型，构建了一种顾及误差传递的空间降尺度模型（CKFM）。针对两种经典的耦合模型R-F和F-R，建立两种模型的选择标准，再基于R-F和F-R，构建一个顾及传递误差的基于核驱动降尺度和时空融合降尺度的加权组合模型（CKFM）。相较于单一的核驱动降尺度和时空融合降尺度，CKFM模型精度平均能提高0.1-0.6 K，相较于R-F和F-R模型，CKFM模型不传递误差，而是最小化两个降尺度过程的误差。

（3）构建了一个顾及植被物候特征的温度循环（PATC）模型。模型将植被物候特征对地表温度的影响作为地表温度变化的主要驱动因素之一，而天气状况对地表温度的影响则作为短期内的温度震荡，添加温度循环模型的残差项中，用于地表温度的年内时间域内插。相较于增强的温度循环模型，PATC模型能更好地模拟由植被组分动态变化和局部天气变化引起的温度震荡，在植被覆盖浓密的白天，精度平均能提高1 K。

（4）梳理了四种遥感地表温度降尺度时空耦合策略，揭示了直接温度时空降尺度性能的优越性。对四种时空耦合策略进行性能对比，揭示了在同时顾及实施效率和精度的情况下，直接温度时空降尺度-III的性能要优于其它策略，并且耦合CKFM模型和DTC模型的策略相对于耦合核驱动降尺度和DTC模型的策略，精度平均提高了0.5 K。

（5）基于降尺度时空耦合策略，结合PATC模型、温度日内循环（DTC）模型及CKFM模型生产了北京市逐日和逐小时100 m分辨率的地表温度数据，时空维度上有效提升了城市热岛的检测能力。在逐日和逐小时的局地气候区城市热岛效应研究中，相较于MODIS地表温度，降尺度地表温度在白天和夜间探测的城市热岛强度平均提高了0.3和0.6 K，相较于瞬时Landsat 8地表温度，能动态地刻画各局地气候区的城市热岛变化。

本文提出的遥感地表温度时空降尺度方法，能为地表温度的时空信息增强提供有效的理论指导与可借鉴的实例。基于本文的算法与降尺度实例得到的高时空分辨率地表温度数据集，可为多尺度的局地气候区城市热岛研究提供重要的数据支撑。

Urban heat island (UHI) is the most significant climate effect of urbanization, and is close related to local climate, air pollution, urban vegetation phenology and human health. With the development of remote sensing technology, land surface temperature (LST) derived from the thermal infrared (TIR) sensor becomes important data source of researches on UHI. However, due to the tradeoff between spatial and temporal resolution of TIR sensors, it is still difficult to obtain LSTs with high-spatiotemporal resolution. Meanwhile, TIR sensors are not sensitive to clouds which makes lots of missing data due to cloud coverage, accordingly, results in a low availability of TIR data. In order to cope with these challenges, we propose a spatial downscaling model and a temporal interpolation method for LSTs; then we discuss the concept and method of existing spatial downscaling and temporal interpolation to generate high-spatiotemporal-resolution LSTs; and we quantitatively study the spatiotemporal variations of UHI in Beijing by the downscaled LSTs. The main innovations and conclusions of this paper are as follows:

(1) we summarized the concepts, types, research status and existing problems of spatiotemporal downscaling of LST. We defined the spatiotemporal downscaling of LST, and divided it into two parts, i.e., spatial resolution enhancement and temporal resolution enhancement. Then we summarized the existing problems of these two methods, and pointed out the application prospect of spatiotemporal downscaling of LST in studying UHI.

(2) we proposed a combination of spatial downscaling methods (i.e., kernel-driven method and fusion-based methods) to take the transitive error into account. First, we introduced two classic methods, i.e., “regression-then-fusion” (R-F) and “fusion-then-regression” (F-R) methods, and put forward the selection criterion to make a choice between these two methods. Then, we proposed a combination of kernel-driven and fusion-based methods (CKFM) to minimize errors between two downscaling processes. Compared with single kernel-driven method and fusion-based method, CKFM can improve the accuracy by 0.1-0.6 K. Compared with R-F and F-R method, CKFM can minimize the errors of kernel-driven and fusion-based processes, rather than transmit error from the first process to the second.

(3) we proposed a phenology-based annual temperature cycle (PATC) model for daily LST reconstruction. PATC takes the influence of vegetation fraction variation on LSTs as the main driving factor, and the influence of weather conditions on LSTs as the residual term of the temperature cycle model. Compared with the enhanced annual temperature cycle (ATCE), PATC can well simulate the temperature fluctuation caused by vegetation fraction variations and weather change, with an improved accuracy by 1 K in the daytime over where densely covered by vegetation.

(4) we summarized and compared four spatiotemporal downscaling strategies, and comclued that the direct spatiotemporal downscaling strategy performs better than other strategies when we take the efficiency and accuracy into account. In addition, the strategy which combines CKFM and diurnal temperatue cycle (DTC) model together improves the accuracy by 0.5 K compared with that combines kernel-driven method and DTC model together.

(5) we combined CKFM with PATC and DTC model to generate daily and hourly 100 m-resolution LSTs of Beijing for UHI analysis in local climate zones (LCZs), and found that the downscaled LSTs improves the ability to detect UHIs at spatiotemporal scales. Compared with MODIS LSTs, the downscaled LST improves the detected UHI intensities by 0.3 K and 0.6 K for daytime and nighttime, respectively. Compared with Landsat 8, it can dynamically describe the temporal variations of UHI.

This study put forwards spatiotemporal downscaling methods of LST, and provides effective theoretical guidance and referential case for spatiotemporal downscaling of LSTs. In addition, high-spatiotemporal-resolution LSTs obtained from the method play an important role in supporting analyzing multi-scale UHIs in local scales.