湖泊冻融物候是湖泊的重要特征之一，也是气候变化的重要指示器。湖泊冻融物候的台站观测主要依靠目视，耗时耗力，而且存在一定的视野局限性，尤其是大型湖泊，而且对于地广人稀、环境恶劣地区的湖泊观测也较困难。卫星遥感技术的快速发展，有效弥补了传统观测方法的不足，有望实现大范围湖泊冻融物候变化监测，为湖泊冻融物候变化研究提供有力的数据支持。

已有研究中基于卫星遥感湖泊冻融物候方法，由于使用不同数据源及判定阈值导致研究结果存在显著差异。为解决这一问题，本文利用MODIS（中分辨率成像光谱仪）每日四次温度数据，发展了湖泊冻融物候信息自动化提取算法，并应用于青藏高原和中国北方9个大型季节性冰封湖泊，构建了2002-2016年湖泊冻融物候参数集，包括开始冻结日、完全冻结日、开始消融日、完全消融日、完全冰封期和湖冰存在期。在此基础上，对比分析了不同气候区的湖泊冻融物候特征及其年际变化，探讨了气候变化背景下湖泊冻结和消融的过程变化以及其驱动机制，量化了气象要素以及湖泊自身特性（湖泊形态、地理位置和理化因子）对湖泊冻融过程的影响。主要结果如下：

1)     利用实测湖温数据对MODIS温度产品进行了评估，采用上廓线法去除了MODIS湖表温度产品中残留云（云检测产品错误地把云定为晴空）的影响，降低了MODIS湖表温度与实际测量值之间的偏差。借鉴植被物候学中的Logistic生长曲线模型，构建了湖泊冻融过程的变化曲线，实现了对不同类型湖泊冻融物候参数的自动化提取，获取了2002-2016年间青藏高原和中国北方9个大型季节性冰封湖泊冻融物候参数。验证结果表明，该方法与已有结果具有较好的一致性，且在冻结初期可以有效的消除薄冰像元的影响，同时可以避免因数据缺失而对关键时间节点判定造成的影响，使提取湖泊冻融物候参数更为稳健。敏感性分析表明，该方法有效降低了原有卫星遥感方法对阈值或经验系数的依赖性。

2)     揭示了2002-2016年间青藏高原和中国北方9个大型季节性冰封湖泊的冻融物候变化特征。除博斯腾湖和玛旁雍错外，所选湖泊开始冻结日推迟发生，青藏高原地区的湖泊开始冻结日的推迟速率均高于东北部地区；青藏高原北部的湖泊开始冻结日推迟速率高于高原中南部的湖泊，其中青海湖最显著，推迟速率为0.63 day/yr。完全冻结日在东北地区的湖泊呈推迟趋势，而在青藏高原地区呈提前趋势，其中纳木错最为显著，提前的趋势为 0.82 day/yr。东北地区湖泊的消融过程变化显著，开始消融和完全消融日均提前；除色林错和扎日南木错外，青藏高原的湖泊开始消融日提前，其中鄂陵湖的提前趋势最显著，提前速率为2.49 day/yr。

3)     分析了湖泊自身因素与关键气象参数对湖泊冻融过程的影响。发现湖泊自身因素对湖泊冻结过程的影响大于其对消融过程的影响。温度是影响湖泊冻融变化的最直接因子，其中湖温对湖泊的冻结过程影响显著，而气温对消融过程的影响显著。相比于浅湖，深湖的湖温变化对冻结过程的影响更显著。2002-2016年间，青藏高原地区气温上升的速率高于东北地区，且青藏高原北部的湖温上升速率高于中南部，导致青藏高原北部湖泊的开始冻结日推迟；就消融过程来说，东北地区春季升温速率高于青藏高原地区，造成东北地区湖泊消融过程提前的趋势明显于其他地区的湖泊。并以此研究为基础，基于湖泊自身因素和关键气象参数构建了湖泊冻结和消融物候参数的经验预测模型。

本文发展了湖泊冻融物候信息自动化提取算法，构建了湖泊冻融物候数据集，为湖泊冻融的时空特征变化分析提供数据支持，同时也为湖冰演化过程、动力学研究和湖冰模型的建立以及未来湖冰的研究等提供了重要的数据基础，为理解全球气候变暖背景下冰冻圈的变化作出贡献。

Lake ice phenology is one of the important components for lakes in the northern hemisphere, and one of the important indicators of climate change. The observation of lake ice phenology at the ground station mainly relies on eyesight from observers, which has limitations on large lakes. Also, the station-based observation is time-consuming and labor-intensive, and even more difficult to monitor lakes in harsh environments. The rapid development of satellite remote sensing technology effectively makes up for the shortage of traditional observation methods, and also can monitor the freeze-thaw changes in a large scale, providing powerful data support for lake ice phenology researches.

The previous satellite remote sensing-based studies on lake ice phenology were found significant differences in the results when using different data sources and lake-specific thresholds. In this study, we selected nine large seasonal frozen lakes in different climatic regions of Qinghai-Tibet Plateau and northern China. A new automated algorithm to characterize lake ice phenology matrix was developed using Moderate Resolution Imaging Spectroradiometer (MODIS) daily temperature products from 2002 to 2016. The complete set of lake ice phenology was constructed, including the freeze-up start and end date, break-up start and end date, the freeze ice period, and the complete ice cover period. We further analyzed the spatial variation characteristics and interannual variation of lake ice phenology in different climate zones, and evaluate how regional climatic and geographical conditions as well as local lake characteristics attribute to the freeze-up and break-up process of ice phenology. The main findings are summarized as follows:

1)     The MODIS temperature products were evaluated by high-frequency in situ water temperature data. Biased or wrong values caused by residual clouds (incorrectly designated as clear sky by the cloud detection products) were detected and modified by the upper envelope curve method. A new automated extraction algorithm of lakes ice phenology is developed based on the reconstructed MODIS daily lake temperature series. The time series of the freeze-thaw process was created, and firstly fitting by the Logistic growth curve model to characterize the lake ice phenology. With good validation by existing results, our results showed a better performance in the initial phase of the freeze-up process. This new satellite-based algorithm was developed to avoid using any lake-specific or experienced thresholds and could also overcome the misdetection of event dates caused by missing data due to atmospheric conditions, which made the result robust and applicable to other regions.

2)     The spatio-temporal characteristics of ice phenology were analyzed from 2002 to 2016 among nine lakes. For freeze-up patterns, a delayed freeze-up start date (FUS) was found, except Lake Namco and Lake Mapang. The trend of delayed FUS in Tibetan Plateau was larger than in other regions, espercially in the northern Tibetan Plateau. Lake Qinghai showed the most obvious delayed trend of FUS (0.63 day/yr). An earlier freeze-up end date (FUE) was found in lakes in Tibetan Plateau, but a later trend in northeastern China. For break-up patterns, we found the break-up process of the lakes in northeastern China showed the obvious earlier break-up start and end date (BUS and BUE). An earlier break-up start date (BUS) was found in lakes in Tibetan Plateau, and the earliest trend was in Lake Ngoring (-0.82 day/yr).

3)     The dependency of ice phenology variability on climatic conditions and local lake characteristics was statistically analyzed for nine lakes. The freeze-up pattern tends to be more determined by lake characteristics including lake morphology and geographical location than break-up patterns. The temperature is the key parameter for lake ice process. LSWT showed to have a strong influence on freeze-up dates and air temperature showed to be highly correlated to break-up dates. What’s more, the freeze-up dates were affected more significantly by LSWT than air temperature in deeper lakes, although they both showed significant positive correlations with freeze-up dates. Influences from climatic factors were complex and largely depend on the situation of the lake ice cover, but the spring air temperature was found to be the most important factor for the break-up process, which shows by the significant negative correlation of break-up dates. The empirical prediction models of lake freeze-up and break-up process were constructed based on the lake-specific characteristics and key meteorological variables.

The automated extraction algorithm for ice phenology developed in this study, and the lake ice phenology created based on this algorithm provided data support for the analysis of spatial and temporal characteristics changes of ice freeze-thaw changes, and also provided an important data basis for the study of lake ice evolution process, dynamisc and simulation of lake ice phenology and ice thickness as well as future lake ice research, contributing to the understanding of the change of the cryosphere under the background of global warming.