陆地水循环要素是影响区域陆地水资源时空变化的重要组成部分。在气候变化和人类活动的双重背景下，研究长江流域陆地水循环要素时空变化对于长江流域陆地水资源管理、实现社会经济可持续发展具有重要的意义。降水是陆地水循环过程中的重要要素，决定陆地水资源的时空分布及变化。特别地，在气候变化和人类活动影响下，长江流域极端降水呈现出增多增强的趋势，受到广泛的关注。水体通常泛指陆表水体，包括河流、湖泊、水库坑塘和水田等，同样是陆地水循环过程的重要组成部分。长江流域水体分布众多，是陆地水资源的最主要来源。水储量包括冰/雪水、植被冠层水、地表水、土壤水和地下水，表征区域陆地水资源丰富程度。降水、水体和水储量的密切关联深刻地刻画了区域陆地水资源的时空变化轨迹。

目前，围绕长江流域降水、水体和水储量的时空变化及三者之间的关联已有研究，然而仍存在一定的不足。一方面，现有研究较少利用逐小时降水数据、大范围长时序水体遥感数据和长时序重构水储量数据，研究三种陆地水循环要素的时空变化特征。另一方面，综合分析“降水-水体-水储量”三种陆地水循环要素的关联性研究较少。因此，本文研究分别利用站点观测逐日和逐小时降水、美国地球资源卫星系列（Landsat）遥感数据和地球重力场恢复和气候实验（Gravity Recovery and Climate Experiment，GRACE）卫星遥感数据，研究长江流域降水多尺度时空变化特征、水体长时序变化格局和水储量长时序时空变化趋势，并探讨长江流域降水、水体和水储量三要素的关联性。本文的主要创新及结论如下：

（1）分析了长江流域短时和长时极端降水事件频次、强度和昼夜分布时空变化特征，揭示了逐小时降水和逐日降水在长江流域极端降水强度和频次两方面的趋势差异性。主要结论有：

1）1980–2018年暖季（4–9月），长江流域极端降水事件以短时极端降水事件（降水持续时长在1–6小时）为主（占比58.39%）。2）短时极端降水事件和长时极端降水事件（降水持续时长大于6小时）的频次、最大强度和平均强度的年际趋势具有差异性。短时极端降水事件频次和最大强度呈显著增加趋势（p-value < 0.05)；长时极端降水事件的最大强度和平均强度呈显著增加趋势。3）逐小时尺度上的极端降水强度和频次相比逐日尺度上更能观察到增加趋势，凸显了精细化降水数据的应用价值。

（2）提出了一种耦合稳定分位数样本点和水体分区识别规则的水体遥感提取方法，实现了大范围、长时序Landsat遥感数据的水体快速提取，分析了长江流域常年水体和季节水体长时序时空变化特征。主要结论有：

1）基于长时序Landsat影像，采用分位数合成法，得到不同分位数合成影像，采集“空间位置固定、光谱信息变化”的稳定分位数样本点，减少样本受年际变化和季节波动的影响；结合多个遥感植被指数和遥感水体指数，构建水体分区识别规则，进行水体遥感提取。耦合稳定分位数样本和水体分区识别规则的水体遥感提取的验证水体总体精度为95.03%，卡帕系数为0.895。2）1984–2018年，长江流域常年水体（积水频率>75%）和季节水体（25%<积水频率≤75%）面积分别为29076.70 km²和21526.24 km²。2010–2018年，长江流域常年水体相比1984–1999年增加2731.13 km²，主要由季节水体转化；季节水体减少1562.06 km²，主要转化为非水体。3）水库修建是长江流域水体面积变化的重要因素之一。2010–2018年，长江流域主要大、中型水库常年水体和季节水体总面积相比1984–1999年增加281.07 km²，占整个长江上、中、下游区常年水体和季节水体总面积增加的51.40%。

（3）提出了一种最优特征变量和机器学习组合的水储量数据重构方法，建立了长江流域1985–2018年长时序水储量数据集，分析了长江流域长时序水储量的时空变化特征。主要结论有：

1）考虑特征变量数据的空间差异性，采用滑动窗口相关法逐像元构建最优土壤湿度、降水、温度和植被特征变量组合；逐像元集成多元线性回归、随机森林和人工神经网络中的最优机器学习方法，重构GRACE水储量数据。GRACE水储量重构精度评估表明：像元尺度上，GRACE重构结果的相关系数和归一化均方根误差为0.729和18.762%；流域尺度上，GRACE重构结果的相关系数和归一化均方根误差为0.927和17.292%。2）1985–2002.03时期，长江流域水储量1–12月平均年际趋势强度为0.011 cm/year，增加趋势主要发生在长江流域东南部和西南部，减少区域主要发生在长江流域东北部。2002.04–2018时期，长江流域水储量1–12月平均年际趋势强度为0.246 cm/year，且增加趋势主要发生长江流域中部和长江源区。3）长江流域水储量的经验正交函数分解时空模态主要反映季节波动特征和趋势特征，其中季节波动特征为主，方差解释率在77.44%左右，趋势特征方差解释率仅在6.5%左右。季节波动特征表明长江流域东南部和西南部水储量季节波动相位相反。

（4）从时间关联、空间关联、要素响应关联和洪水事件关联四个层面，定量探讨了长江流域降水、水体和水储量的关联性。主要结论有：

1）时间关联方面，长江流域水储量变化滞后降水1个月，和水体不存在滞后。2）空间关联方面，考虑流域间降水汇流空间关系，降水和水储量的空间偏相关系数在0.237–0.556之间，水储量相对降水存在1–3月的滞后。3）要素响应关联方面，分位数回归法表明随着水储量分位数的增加，降水对水储量的影响增加。构建的定量线性回归模型可以较好地模拟降水、水体和水储量三要素在年、月尺度上的定量响应关系。年尺度上“水体-降水”回归模型和“水储量-水体+降水”回归模型的R²分别位于0.190–0.745之间和0.284–0.848之间，月尺度上“水储量-降水”回归模型的R²位于0.752–0.919之间。4）洪水事件关联方面，以丹江口水库2017年7–10月和鄱阳湖2016年6–7月两个洪水事件为例，通过对洪水事件过程中的降水演变、水体淹没和水储量动态变化分析，表明极端降水导致洪水事件发生，引起水体扩张和水储量增加，体现了洪水事件过程中降水、水体以及水储量变化的一致性。另外，洪水事件过程反映了在极端降水情况下，水储量相对降水的变化滞后期可能缩短。

Spatial-temporal changes of the terrestrial hydrological cycle components control the spatial distribution and temporal variation of water resources. Under the background of climate changes and human activities, studying the spatial-temporal changes of the terrestrial hydrological cycle components is of great significance to the water resources management and sustainable development in the Yangtze River Basin. Precipitation is an important component in terrestrial hydrological cycle, because it determines spatial -temporal changes of terrestrial water resources. In particular, under the influence of climate change and human activities, extreme precipitation in the Yangtze River Basin has shown an increasing trend and has received widespread attention. Water body generally refer to land surface water bodies, including rivers, lakes, reservoirs, ponds, and paddy fields. Water body is also an important part of terrestrial hydrological cycle. There are many water bodies in the Yangtze River Basin, providing the most important source of terrestrial water resources. Water storage includes ice/snow water, vegetation canopy water, surface water, soil water and groundwater, which characterize the abundance of regional terrestrial water resources.

At present, there have been some studies on the spatial-temporal changes of precipitation, water bodies and water storage in the Yangtze River Basin and the correlation among the three, but there were still some shortcomings. On one hand, these existing studies rarely use hourly precipitation data, large-scale long time series water body remote sensing data, and long time series reconstructed water storage data to their spatial-temporal changes, respectively. On the other hand, few studies analyzed the correlation among precipitation, water bodies and water storage. Therefore, this study used station precipitation data, Landsat images and GRACE time series to analyze the spatial-temporal changes of precipitation, water body and water storage. Moreover, this research also investigated the correlation among precipitation, water body, and water storage. The main innovations conclusions of this dissertation are as follows:

(1) This study compared the spatial-temporal changes of short-duration extreme precipitation events and long-duration extreme precipitation events in terms of frequency, intensity and diurnal distribution based on hourly precipitation data, and investigated the differences in extreme precipitation trends between hourly precipitation data and daily precipitation data. The main conclusions are :

1) During the warm season (April-September) of 1980–2018, the extreme precipitation events in the Yangtze River Basin were dominated by the short-duration extreme precipitation events （duration of 1–6 hours). The percentage of short-duration extreme precipitation events is 58.39%. 2) The short-duration extreme precipitation events and long-duration extreme precipitation events (duration >6 hours) showed a different trend in frequency, maximum intensity and mean intensity. The short-duration extreme precipitation events showed a significant increasing trend (p-value < 0.05) in frequency and maximum intensity. The long-duration extreme precipitation events (duration >6 hours) showed a significant increasing trend (p-value < 0.05) in maximum intensity and mean intensity. 3) The increasing trend of extreme precipitation intensity and frequency on an hourly scale was more obvious than that on a daily scale, revealing the application value of fine hourly precipitation data.

 (2) The study developed a water extracting approach by combining stable quantile samples-and zonal water extraction rule, realized the rapid water bodies extraction based on long time series and large scale Landsat image in the Yangtze River Basin, revealed the characteristics of year-long water bodies and seasonal water bodies in the Yangtze River Basin. The main conclusions are :

1) Multiple quantile images were composited to collect the stable quantile samples which have a fixed spatial location and quantile–varing spectral information to reduce the sample bias due to the interannual changes and seasonal variations based on long time series Landsat images; zonal water extraction rule was developed to identify water body by combining the multiple vegetation indexes and water indexes. The accuracy assessment showed that overall accuracy was 95.03% and kappa index was 0.895 for these test water body samples. 2) From 1984–2018, the area of year-long water bodies (water inundation frequency>75%) and seasonal water bodies (25%< water inundation frequency≤75%) in the Yangtze River Basin were 29076.70 km2 and 21526.24 km2, respectively. From 1984–1999 period to 2010–2018 period, the area of year-long water bodies increased by 2731.13 km², and most of the increases were from the conversion of seasonal water bodies. At the same time, seasonal water bodies decreased by 1562.06 km², and most of the decreases converted to non-water bodies. 3) Reservoir construction is one of the important factors in the change of water body area in the Yangtze River Basin. From 1984–1999 period to 2010–2018 period, The total area of year-long water bodies and seasonal water bodies of major large-size and medium-size reservoirs in the Yangtze River Basin increased by 281.07 km², accounting for 51.40% of the increase in the area of year-long water bodies and seasonal water bodies in the upper, middle and lower reaches of the Yangtze River.

(3) The study developed a water storage reconstruction model by combined sliding window correlation method and three machine learning models (multiple linear regression, random forest model and artificial neural network), produced the 1985–2018 GRACE dataset, revealed the long time series spatial-temporal changes of water storage in the Yangtze River Basin. The main conclusions are :

1) Sliding window correlation method was used to select the optimized soil moisture data, precipitation data, temperature data and vegetation data; the optimized machine learning model for individual cell was determined from multiple linear regression, random forest and artificial neural network. The correlation coefficient and the normalized root mean square error of the GRACE reconstruction results on the cell scale were 0.729 and 18.762%; the correlation coefficient and the normalized root mean square error of the GRACE reconstruction results on the basin scale were 0.927 and 17.292%. 2) From 1985 to 2002.03, the average annual trend of water storage in the Yangtze River Basin from January to December was 0.011 cm/year. The increasing trend mainly occurred in the southeast and southwest of the Yangtze River Basin, and the decreasing trend mainly occurred in the northeast of the Yangtze River Basin. From 2002.04 to 2018, the average annual trend of water storage in the Yangtze River Basin from January to December was 0.246 cm/year. The increasing trend mainly occurred in the central area of the Yangtze River Basin and the source of Yangtze River Basin.3) The empirical orthogonal function decomposition of water storage in the Yangtze River Basin mainly revealed the seasonal variations and interannual trends. The variance explanation is about 77.44% and 6.5% for seasonal variations and interannual trends, respectively. The water storage in the southeast of the Yangtze River Basin had the opposite seasonal variations to that in the southwest of the Yangtze River Basin.

(4) The study analyzed the correlation among precipitation, water body and water storage in the Yangtze River Basin from the four aspects: temporal correlation, spatial correlation, component response correlation and flood correlation", and fitted the linear regression model of “precipitation-water body-water storage”.

1) In terms of temporal correlation, there was a lag of 1 month between water storage and precipitation changes. 2) In terms of spatial correlation, the study set up the spatial relationship among sub-basins and used partial correlation coefficient to study the spatial correlation. The spatial partial correlation coefficients between precipitation and water storage vary from 0.237 and 0.556, with a time lag of 1-3 months. 3) In terms of component response correlation, firstly, the quantile regression results indicated the precipitation showed more and more influences on water storage with an increasing quantile value. Moreover, the study established three linear regression models. R² of the linear regression model of “water body-precipitation” at the annual scale varied from 0.190 to 0.745; R² of the linear regression model of “water storage-water body+precipitation” at the annual scale varied from 0.284 to 0.848; R² of the linear regression model of “water storage-precipitation” at the monthly scale varied from 0.752 to 0.919. 4)In terms of flood correlation, taking the two flood events in Danjiangkou Reservoir and Poyang Lake as example, the study analyzed the changes of precipitation, water body and water storage, and the results indicated extreme precipitation led to flood events, further resulted in more water body and more water storage, which reflects the change consistency of "precipitation-water body-water storage" during the flood event.