随着全球城市化进程的快速推进，大量自然地表被城市不透水面替代，从而导致城市内部能量平衡发生显著变化。快速城市化进程导致地表覆盖的剧烈变化以及人类活动的显著增强使城市热岛效应日益凸显，城市热岛效应是指城市温度高于周围乡村温度的一种典型城市化气候。由于城市热岛效应对生态环境的破坏作用以及对人类健康的显著负面影响，准确监测全球城市热岛的时空特征、厘清全球城市热岛的变化模式以及掌握城市热岛的产生机制是优化城市人居环境以及合理规划城市发展的理论基础。然而，全球尺度城市热岛特征、模式及驱动的系统研究尚未充分开展，影响了全球城市热岛时空变化监测的准确性，阻碍了城市热岛消减工作的有效开展。针对以上问题，本文基于MODIS地表温度数据，以全球城市热岛“时空特征—变化模式—驱动分析”为研究脉络，首先刻画了全球城市热岛的时空变化特征，进而开展了全球城市热岛变化模式研究，以此为基础探讨了驱动因素和热浪事件对全球城市热岛效应的影响。本文的主要工作和成果如下：

（1）构建了全球城市年度地表温度数据，改进了温度基准划定方法，基于热岛强度指标刻画了全球城市热岛变化特征。

开展了温度数据的影像质量控制和时空聚合方法的精度验证和方法比选，构建了全球城市年度地表温度数据，改进了适用于全球尺度研究的热岛强度温度基准划定方法，定量评价了不同温度基准划定方法对全球城市热岛研究的影响，以此为基础刻画了全球城市热岛强度的时空变化特征。研究表明，①均值法（EQC）应用于影像质量控制的精度最高，空间-时间-阈值聚合法（FSAT-T）处理温度数据缺失的鲁棒性最优，基于EQC和FSAT-T构建全球城市年度地表温度数据能够有效提高对影像质量下降和数据缺失等问题的鲁棒性；②基于改进的等面积法（MEA-R）划定温度基准范围克服了其他方法应用于全球城市热岛研究的局限性，不同基准划定方法的应用会导致全球城市热岛强度研究的时空分布特征不可比；③全球城市热岛强度的昼夜均值分别为1.73和1.22 K，昼夜城市热岛纬度变化和气候带差异呈相反趋势，全球61%的城市白天热岛强度高于夜间。

（2）构建了全球城市无缝日尺度地表温度数据，定义了城市热岛时间指标，基于热岛时间指标刻画了全球城市热岛变化特征。

基于ATC-SKT模型构建了全球城市无缝日尺度地表温度数据，开展了基于像元尺度的地表温度精度验证和基于城市尺度的热岛强度精度验证，定义了城市热岛时间指标，包括频率和最大持续时间，基于热岛时间指标定量刻画了全球城市热岛变化特征。研究表明，①精度验证结果证明基于ATC-SKT模型构建的全球城市无缝日尺度地表温度数据准确性高，可应用于城市热岛研究，能够识别城市热岛极值出现的时间；②全球城市热岛频率的昼夜均值分别为214和175天/年，全球20%的城市全年白天每天都发生热岛效应，45%的城市夏季每日都发生热岛效应，白天赤道气候带热岛频率最高而夜间冷温带热岛频率最高，全球城市冷岛频率的昼夜均值分别为41和3天/年，全球33%的城市全年昼夜都未发生城市冷岛效应；③全球城市热岛最大持续时间的昼夜均值分别为147和58天，全球11%的城市白天热岛最大持续时间超过350天，城市冷岛最大持续时间很短，夜间几乎没有城市冷岛效应连续发生；④受连续发生的热岛效应影响，昼夜城市热岛强度分别增加了15%和8%，对冷温带的增强作用最大，该气候带昼夜热岛强度分别增加23%和14%，而受连续发生的冷岛效应影响，昼夜热岛强度分别降低了8%和5%，可从限制热岛效应连续发生的角度开展消减工作。

（3）系统地归纳了城市热岛的模式特征，开展了全球城市热岛年度、季节和昼夜变化模式研究。

结合全球城市的热岛强度和时间指标，基于聚合层次聚类（AHC）开展了全球城市热岛年度变化模式研究，基于季节调整的年变化模型（ATCS）和昼夜差异年变化模型（DNC-ATCS）开展了全球城市热岛季节和昼夜变化模式研究。结果表明，①全球城市热岛年度变化模式主要分为冷岛、不显著、低值热岛、中值热岛和高值热岛五类，白天高值热岛模式的城市比例为34%，主要分布于赤道气候带，而夜间高值热岛模式的城市比例为26%，主要分布于暖温带，全球6%的城市昼夜都属于高值热岛模式，应予以重点关注；②全球城市热岛季节变化模式主要分为暖季、冷季、夏秋、春冬和不显著五类，白天城市热岛的季节变化以暖季和夏秋模式为主，全球66%的城市夜间热岛效应无显著季节变化；③全球城市白天热岛强度高于夜间的天数比例为60%，受白天热岛效应季节变化的显著影响，全球城市热岛昼夜差异的年内变化以暖季和夏秋模式为主，全球35%的城市全年白天城市热岛强度都大于夜间，48%的城市热岛昼夜差异的正负随着时间变化。

（4）顾及不同城市热岛评价指标和变化模式，基于TabNet厘定了驱动因素对城市热岛的影响，进一步探讨了热浪事件对城市热岛的影响。

考虑不同城市热岛评价指标和变化模式的差异，基于可解释的深度学习模型（TabNet）厘定了驱动因素对不同热岛评价指标和变化模式的贡献权重，进一步探讨了热浪事件对城市热岛的影响。结果表明，①白天城市热岛效应受城乡植被差异的显著影响，夜间城市热岛效应主要的驱动因素为城乡反照率差异和夜间灯光强度，驱动因素对不同热岛评价指标及变化模式的贡献程度呈显著差异，因此开展城市热岛消减工作时需要考虑城市热岛的变化模式；②白天全球城市热岛强度、频率和最大持续时间受热浪事件的影响分别减少20%、37%和87%，而夜间城市热岛效应受热浪影响而增强，不同热岛变化模式对热浪事件的响应程度不同，低值和中值热岛模式的热岛频率受到热浪事件的显著影响，白天热岛效应受热浪影响而减弱，因为植被气孔受极端高温影响而闭合导致植被降温作用削弱，而夜间城市热岛效应由于热浪期间地表储热增加而增强。

本文围绕全球城市热岛遥感研究开展，以城市热岛“时空特征—变化模式—驱动分析”为研究脉络，揭示了全球城市热岛效应的时空变化规律，厘清了全球城市的热岛变化模式，探明了全球城市热岛效应的影响机制，对开展全球城市治理有重要借鉴意义。

With the rapid development of urbanization at a global scale, a large number of natural surfaces have been replaced by impervious areas, leading to significant perturbation to the surface energy balance. Massive surface changes and considerable human activities cause increasingly prominent of surface urban heat island (SUHI), a classic urban climate effect with an urban temperature higher than rural temperature. Since the negative effects from SUHI on ecological environment and human health, accurately monitoring spatiotemporal variations, identifying patterns and mastering the driving mechanism of SUHI is a theoretical basis for environmental optimization and urban planning. However, systematic research on the global-scale SUHI variations, patterns and controls remained further investigation, which challenged the accurate monitoring of spatiotemporal variations and hampered effective mitigation work. To address these issues, this paper conducted a systematic research on the global-scale SUHI following the research thread of “variations-pattern -control” based on MODIS land surface temperature (LST). This paper investigated spatiotemporal variations of SUHI across global cities, and conducted pattern classification. On this basis, the impacts from related factors and heat wave explored. The major contents are as follows:

(1) This paper reconstructed annual LST, modified the method for defining temperature reference, and investigated spatiotemporal variations of SUHI intensity (SUHII) across global cities.

We conducted accuracy assessments and method comparison for quality control and spatiotemporal aggregations, and reconstructed the annual LST across global cities. We modified the method for defining temperature reference and evaluated the impacts from different defining methods on SUHII. On this basis, spatiotemporal variations of SUHII were investigated across global cities. There were three major findings. ① The equally quality control (EQC) method was proven with higher accuracy among quality control methods. The first spatial and after temporal aggregation with threshold removal (FSAT-T) method was demonstrated with greater robustness to data missing over other methods. Reconstructing annual LST based on EQC and FSAT-T can improve the robustness of image quality reduction and data missing. ② The modified equal area[1]rural (MEA-R) method could overcome the limitations of the other methods for defining background reference at a global scale. Different reference defining methods led to inaccurate and incomparable results among different studies across global cities. ③ The daytime and nighttime SUHII averages were 1.73 and 1.22 K at a global scale, with opposing tendency of latitudinal and climate-based variations between day and night. Daytime SUHII is higher of 0.41 K than nighttime SUHII, and sixty-one percentage of global cities indicated higher SUHII during the day than at night.

(2) This paper reconstructed seamless daily LST, defined SUHI temporal indicators, and investigated spatiotemporal variations of temporal indicators across global cities.

We reconstructed Seamless daily LST across global cities based on ATC-SKT and conducted pixel-based LST and city-based SUHII accuracy assessments. We proposed temporal indicators including frequency and maximum duration to quantitatively characterize the SUHI temporal variations across global cities. There were four major findings. ① The accuracy assessments validated the accuracy and feasibility of seamless daily LST reconstructed based on ATC-SKT, which can be employed in the studies on SUHI temporal indicators and can improve the ability to identify the extreme period. ② The daytime and nighttime averages of SUHI frequency were 214 and 175 days/year. The cities exhibiting occurrences of SUHI effects for almost every day accounted for 20% and 10% during daytime and nighttime. Forty-five percent of cities exhibited occurrences of SUHI effects for almost every day during the summer daytime. The highest frequency of SUHI indicated in the equatorial climate zone at daytime while in the snow climate zone at nighttime. The daytime and nighttime averages of surface urban cool island (SUCI) frequency were 41 and 3 days/year, and 33% of global cities were found no occurrence of SUCI effects at both day and night throughout the year. ③ The daytime and nighttime averages of SUHI maximum duration were 147 and 58 days. Eleven percent of global cities were found the maximum duration of daytime SUHI over 350 days. The maximum duration of SUCI was found short, and there was almost no continuous occurrence of nighttime SUCI effects. ④ Continuous occurrences of surface heat island effects enhanced SUHII by 15% and 8% at day and night, and it was the most effective for the cold temperate zone with increase 23% and 14% at day and night. Continuous occurrences of surface cool island effects decreased SUHII by 8% and 5% at day and night. Thus, the mitigation work can be conducted from avoiding its consecutive occurrences.

(3) This paper systematically identified SUHI patterns, and conducted annual, seasonal and day-night-contrast SUHI patterns classification.

Based on intensity and temporal indicators of SUHI, we conducted annual pattern classification based on agglomerative hierarchical clustering (AHC) and conducted seasonal and day-night-contrast (DNC) pattern classification based on annual temperature cycle with seasonal adjustment (ATCS) and DNC-ATCS at a global scale. There were three major findings. ① There were five classic annual patterns, namely, SUCI, insignificant, low, medium and high-level SUHI. The high-level SUHI accounted for 34% and 26% at day and night. Cities with high-level SUHI were densely distributed in the equatorial climate zone at day, while they were densely located in the temperate climate zone at night. Six percent of global cities with high-level SUHI at both day and night should be paid close attention. ② There were five major seasonal patterns, namely, warming (WA), cooling (CO), summer-autumn (SA), spring-winter (SW) and insignificant season patterns. WA and SA seasonal patterns dominated during daytime. Sixty-six percent of global cities indicated insignificant seasonal variations at night. ③ Sixty percent of days throughout a year indicated higher SUHII at day than night. Due to significant impacts from daytime seasonal patterns, major DNC annual variations were WA and SA. Thirty-five percent of global cities were found higher SUHI than night throughout the year, while 48% of cities showed found positive and-negative variations of DNC with time.

(4) Considering different evaluating indicators and patterns, this paper investigated driving factors of SUHI based on TabNet and further explored the impacts from heat wave.

 Based on SUHI indicators and patterns, we investigated the attribution of related factor based on the interpretable deep learning model and explored the impacts from heat wave on SUHI across global cities. There were two major findings. ① The major factor for explaining spatiotemporal variations of daytime SUHI was vegetation difference between urban and rural areas, while the major factors of nighttime SUHI were albedo difference and nighttime light. The attribution of related factors indicated notable difference among different SUHI indicators and patterns. Therefore, consideration of different variations patterns can help to improve the efficiency of mitigation work of SUHI. ② At daytime, the intensity, frequency and maximum duration of SUHI decreased by 20%, 37% and 87%, respectively, while the nighttime intensity and frequency of SUHI increased during heat wave. Different patterns indicated various response to heat wave, SUHI frequency of the low and medium-level patterns significant changed during heat wave. Daytime SUHI was reduced during heat wave because of the vegetation closure caused by the extremely high temperature, while nighttime SUHI was enhanced due to increase of surface heat storage when heat wave strikes.

 This paper conducted a systematic research on the global-scale SUHI following the research thread of “variations-pattern -control” to characterize spatiotemporal variations, identify patterns and explore driving mechanism of SUHI across global cities, which can support for global urban governance.