气候变化引起的地表升温在一定程度上加速了全球水循环，可能导致地表蒸发加剧，间接导致大气中水汽含量的增加。降雨特征如雨量、历时、降雨发生的频率和强度都可能随之发生改变，尤其是极端降雨事件。这可能导致降雨潜在的侵蚀能力增强，土壤侵蚀风险增加。降雨特征的刻画往往需要使用高分辨率的降雨资料，一些基于过程的水文和土壤侵蚀模型也需要高于日尺度时间分辨率的降雨数据作为输入参数，然而目前长时间序列的高分辨率降雨观测数据存在严重缺失。天气发生器是一种统计模型，可基于观测天气数据计算的输入参数，随机生成任意时间长度、保留观测数据主要统计特征的模拟天气数据。它也是常用的统计降尺度方法，可对GCMs(Global Climate Models, GCMs)输出的较粗时空分辨率的未来气候数据进行降尺度，如从月降至日尺度、从日降至日以下尺度。

本研究基于收集到的中国大陆地区2405个气象台站的小时降雨资料，在分析全球变暖背景下我国黄河流域降雨特征及潜在降雨侵蚀力变化趋势的基础上，探索对国际著名的天气发生器CLIGEN模型输入参数的空间降尺度方法，并对CLIGEN模型中模拟较差的降雨过程指标的模拟方法进行改进，发展适用我国的日降雨生成逐时降雨的随机模拟模型。本研究主要结论如下：

(1) 在过去46年间，我国黄河流域降雨的平均次降雨侵蚀力、平均降雨动能和最大30分钟雨强（I₃₀）在p = 0.1水平上呈显著升高趋势，与基准期(1981-2010年)的平均值相比，全流域增长的相对增速为4.76%/10a、2.88%/10a和2.69%/10a；有超过70%的站点，三个变量均为增加趋势，有10%的站点的变化趋势在p = 0.1水平显著。同时，极端降雨 (定义为次降雨侵蚀力超过全部降雨侵蚀力数据序列中第90百分位的降雨事件)的总降雨侵蚀力也呈现出明显的增加趋势，其相对增长速率为5.8% /10a (p < 0.1)，这主要是由于极端侵蚀性降雨发生得更加频繁，雨强呈增加趋势 (p < 0.1)。日数据及日降雨侵蚀力估算模型也被用于估算降雨侵蚀力，通过对小时与日数据计算的结果的对比发现，日降雨估算极端侵蚀力的变化程度与用小时数据的估算结果相近，但是趋势在统计上不显著，即小时降雨数据在揭示极端降雨侵蚀力变化趋势方面比日数据更为可靠。降雨的动能与峰值雨强强度的增加、极端降雨发生频次的增加，可能使黄河流域的水土流失风险增高，未来应该继续加强黄河流域尤其是黄河中游地区的水土保持措施的制定与落实。

(2) 基于收集到的我国2405个地面气象台站的长时间序列日气温、日降雨和小时降雨数据，以及130个站的太阳辐射数据，建立了我国CLIGEN气温、降雨、辐射的单站参数库。再使用普通克里金法(Oridinary Kriging)，和加入协变量(经度、纬度、海拔和平均年降水量)的泛克里金法(Universal Kriging)，探索CLIGEN的13组输入参数由单站空间降尺度至10km×10km网格的方法。通过对插值结果的充分评估，结果表明，UK插值表现普遍优于OK插值。UK插值与观测日均温的均方根误差小于1.63℃。除偏度系数的Nash-Sutcliffe效率系数为0.78外，降雨和太阳辐射相关参数的Nash-Sutcliffe效率系数均 ≥ 0.87。此外，使用UK插值参数和观测参数分别输入CLIGEN进行模拟，得到的天气时间序列数据显示出一致的统计特征与频率分布特征。两个模拟序列之间均温的平均绝对误差小于0.51℃，太阳辐射、降水量、降雨历时和最大30分钟雨强的平均相对误差（MARE）小于5%。基于本研究建立的中国大陆地区CLIGEN网格参数库的空间精度为10km，可满足用户在中国大陆地区利用WEPP模型进行水文和侵蚀预报，同时，也可为想利用CLIGEN模型进行月至日尺度及日至日以下尺度的降雨、气温数据进行时空降尺度的用户提供便利，数据下载链接：http://clicia.bnu.edu.cn/data/cligen.html。

(3) 本研究基于中国2405个站点的长序列小时降雨数据，发展日降雨降尺度至小时的时间降尺度方法，以改进CLIGEN模型中降雨过程模拟精度较差的缺陷。首先基于倒高斯法来生成日最大小时降水量/雨强(H_max)，并用shuffle法建立模拟H_max与日降水量之间的相关性；其次，采用指数法生成降雨历时，并采用shuffle法建立模拟降雨历时与日降雨量之间的相关性；最后，在日雨量、降雨历时、日最大小时雨量(峰值雨强)已知的情况下，采用CLIGEN中的双指数法，将日降雨量分解至连续的小时降雨数据。通过与观测数据的充分比较，发现倒高斯法和指数法能够模拟观测数据中H_max和降雨历时的基本统计特征（平均值、标准差、极值）以及频率分布特征。与CLIGEN模型相比，观测数据和本研究发展的方法模拟得到的平均H_max和降雨历时之间的MARE分别为5.6%和8.23%，而观测数据和CLIGEN模拟所得的H_max和降雨历时之间的MARE分别为48.93%和61.94%。可以看出本研究开发的随机模拟模型大大提高了日以下尺度降雨峰值雨强和降雨历时的模拟精度。

The increase in atmospheric temperatures caused by climate change induce a more intense global water cycle, lead to greater evaporation rates, and increase the water vapor content in the atmosphere. Rainfall is expected to change under warming with respect to its amount, duration, frequency, and intensity, especially for extreme heavy events, which may enhance the likelihood of soil erosion. The characterization of heavy storms depends on high-resolution rainfall data, and some process-based hydrological or soil erosion models also require rainfall inputs in higher resolution than daily scale. However, long-term observing precipitation data in high resolution is seriously insufficient currently. When observed data are not sufficient in spatial and temporal coverages, simulations of the data may be required. Weather generators also plays an important role in making climate science accessible to impact studies and other applications. In this study, we collect hourly precipitation to detect the influence of global warming on storm characteristics and rainfall erosivity and explore the downscaling method of rainfall from daily scale to sub-daily scale. Main conclusions are as follows,

(1) The average rainfall erosivity, rainfall energy, and maximum 30-min intensity (I₃₀) increased significantly in the whole basin (p < 0.1) and the middle reach (p < 0.05) over the 46 years. The rates of the three variables averaged over the whole basin increased by 4.76, 2.88, and 2.69% decade^-1, respectively, compared to the 1981-2010 average values. More than 70% of the stations showed increasing trends for the three variables, and trends in 10% of the stations were statistically significant (p < 0.1). In addition, the total extreme erosivity, defined as storms that exceeded the 90^th percentile in erosivity, showed significant upward trends at a relative rate of 5.8% decade^-1 (p < 0.1). Extreme erosivity storms identified using hourly data occurred more frequently and with more intensity during the study period (p < 0.1). Trends for extreme erosivity were also estimated with daily precipitation, but these changing rates were not statistically significant. This may indicate that analysis of hourly precipitation is more reliable in revealing trends in extreme events. Results achieved in this study demonstrate that the increasing rainfall energy and I₃₀, and the more frequent and intensive extreme storms, are driving an increased regional rainfall erosivity in the Yellow River basin. This suggests that soil and water conservation strategies and vegetation projects conducted within the basin should be continued and enhanced in the future, especially in the middle reach.  

(2) Long-term daily temperature, daily and hourly precipitation data from 2405 stations and daily solar radiation from 130 stations distributed across mainland China were collected to develop the most critical set of site-specific parameter values for CLIGEN. Ordinary Kriging (OK) and Universal Kriging (UK) with auxiliary covariables, i.e. longitude, latitude, elevation, and the mean annual rainfall were used to interpolate parameter values into a 10 km × 10 km grid and the interpolation accuracy was evaluated based on the leave-one-out cross-validation. Results showed that UK generally outperformed OK. The root mean square error between UK-interpolated and observed temperature related parameters was < 1.63℃ (2.94℉). The Nash-Sutcliffe efficiency coefficient for precipitation and solar radiation related parameters was ≥ 0.87, apart from that for the skewness coefficient, which was 0.78. In addition, CLIGEN-simulated daily weather sequences using UK-interpolated and observed parameters showed consistent statistics and frequency distributions. The mean absolute discrepancy between the two sequences for temperature was < 0.51℃, and the mean absolute relative discrepancy for solar radiation, precipitation amount, duration and maximum 30-min intensity was < 5% in terms of the mean and standard deviation. These CLIGEN parameter values at 10 km resolution would meet the minimum data requirements for WEPP application throughout mainland China. The dataset is available at http://clicia.bnu.edu.cn/data/cligen.html.

(3) Based on long-term hourly precipitation data from 2405 stations of China, we explored a downscaling method to downscale daily rainfall to rainfall process. Firstly, an inverse-gauss method was developed to generate the maximum-hourly rainfall (H_max), and then a shuffle method was used to re-establish the correlativity between H_max and daily rainfall. Secondly, an exponential method was adopted to generate the rainfall duration for rainy days, and the shuffle method was also used to re-establish the correlativity between simulated duration and daily rainfall. Then we adopted the double-exponential method used in CLIGEN to decompose daily rainfall to hourly rainfall process. The simulation quality of H_max and duration was thoroughly evaluated by comparison with observed hourly data, and results showed that the inverse-gauss and exponential method can perfectly preserve the basic characteristics (mean, standard deviation, extreme value) and frequency distribution of observed H_max and duration. Compared with CLIGEN, the method developed in this study were greatly improved the simulation quality of peak intensity and rainfall duration. The MARE (mean absolute relative error) between observed and simulated mean H_max and duration are 5.6% and 8.23%, respectively, while MARE between observed and CLIGEN simulated are 48.93% and 61.94%.