Precipitation affects many aspects of our everyday life. It is the primary source of freshwater and has significant socioeconomic impacts resulting from natural hazards such as hurricanes, floods, droughts, and landslides. Fundamentally, precipitation is a critical component of the global water and energy cycle that governs the weather, climate, and ecological systems. Accurate and timely knowledge of when, where, and how much it rains or snows is essential for understanding how the Earth system functions and for improving the prediction of weather, climate, freshwater resources, and natural hazard events. The Global Precipitation Measurement (GPM) mission is an international satellite mission specifically designed to set a new standard for the measurement of precipitation from space and to provide a new generation of global rainfall and snowfall observations in all parts of the world every 3 h. The National Aeronautics and Space Administration (NASA) and the Japan Aerospace and Exploration Agency (JAXA) successfully launched the Core Observatory satellite on 28 February 2014 carrying advanced radar and radiometer systems to serve as a precipitation physics observatory. This will serve as a transfer standard for improving the accuracy and consistency of precipitation measurements from a constellation of research and operational satellites provided by a consortium of international partners. GPM will provide key measurements for understanding the global water and energy cycle in a changing climate as well as timely information useful for a range of regional and global societal applications such as numerical weather prediction, natural hazard monitoring, freshwater resource management, and crop forecasting.

针对目前的技术手段下难以直接获得大范围高精度精细化降水空间分布的问题,本文以闽浙赣地区为研究范围,选用GPM IMERG降水产品,综合应用地面实测降水数据以及水汽与植被指数数据,基于地理加权回归（GWR）法构建了基于水汽因子的降尺度模型,同时基于最小二乘（OLS）法构建了基于水汽因子与植被指数的对比模型,将降水产品的分辨率从0.1°提升至1 km,最终获得2015年闽浙赣地区各月精细化降水空间分布,使用验证站点实测数据进行验证。结果表明：① 构建的 3个降尺度模型中,GWR模型与2种OLS模型相比,拟合优度分别提升了102.9%和93.9%,模型降尺度结果整体优于2种OLS模型,且月际差异小,稳定性更高;2种OLS模型中,采用了水汽因子的模型拟合效果有8个月份更优;② 融合多源数据的GWR降尺度模型获得的结果在研究区内是可靠的,与GPM降水产品相比,在提升空间分辨率的同时,平均相对误差与均方根误差月均分别下降了42%和32%,精度明显改善。

高时空分辨率降雨数据的获取对陆地水循环研究至关重要。遥感卫星反演降水产品虽然能有效再现降雨的空间格局,但存在空间分辨率较低的问题。以植被指数NDVI(Normalized Difference Vegetation Index)和海拔高度为自变量,通过构建太行山区GPM降水(Global Precipitation Measurement Mission)的时滞地理加权回归模型,得到了2014—2016年研究区1 km分辨率GPM降水数据。研究结果表明：利用植被指数和海拔高度构建的时滞地理加权回归模型能够有效地对太行山月尺度GPM降雨数据进行尺度下延,在提高GPM数据空间分辨率的同时保留了原始数据的观测精度。考虑NDVI的时滞性提高了地理加权回归模型的降尺度效果,相对于多元线性回归模型和不考虑NDVI时滞效应的地理加权回归模型,时滞地理加权回归模型的降尺度结果与站点实测数据的确定性系数更高,RMSE更低。冬季降雨与第二年春季植被NDVI的关系较为密切,虽然采用第二年春季的NDVI作为解释变量构建降尺度模型能有效地提高冬季降雨的降尺度效果,但基于植被指数和海拔高度构建的时滞地理加权回归模型更加适用于植被生长季GPM降雨数据的降尺度研究。

As a fundamental component in water circulation, the spatio-temporal pattern of precipitation is critical for terrestrial hydrological cycle simulation. The satellite-based precipitation can describe the spatial pattern of precipitation properly, but the relatively low spatial resolution of the product limited its application in terrestrial hydrological cycle simulation. By taking Taihang Mountains as an example, based on the relationship between precipitation, altitude and NDVI (Normalized Difference Vegetation Index), the monthly GPM (Global Precipitation Measurement Mission) data from 2014 to 2016 are disaggregated to 1-km resolution with a GWRDL model (Geographically Weighted Regression Model coupled with Distributed Lagging). The results showed that with the aid of the altitude and NDVI, the GWRDL model could effectively downscale monthly GPM data. The spatial resolution of downscaled GPM data was increased by the GWRDL model, and the accuracy of the original GPM data was retained at the same time. Compared with precipitation downscaled by the Geographically Weighted Regression Model and Multiple Linear Regression Model, the precipitation downscaled by GWRDL model has highest coefficients of determination (R2), lowest root mean square error (RMSE) and lowest mean absolute error (MAE) with the observed data, indicating that using NDVI in other months as explanatory variable is better than that using NDVI in current month, and this practice improved the downscaling algorithm and highlighted the accuracy of downscaled precipitation. Because the relationship between precipitation and NDVI was closer in the next spring than that in winter, using NDVI in the following 2-3 months (NDVI in next spring) as an explanatory variable in GWRDL model can improve precipitation downscaling precision in winter. Although the GWRDL model, which gives a consideration of time lagging of NDVI, has a better performance than GWR model in winter, it is more suitable for precipitation downscaling in vegetation growing stage than in winter. Compared with original GPM data, the coefficients of determination between downscaled GPM data and observed precipitation was averagely increased by 0.02 with GWRDL model in vegetation growing stage (April to October), which is higher than that in winter (0.002). Therefore, we suggest that the GWRDL model should be used in GPM downscaling in vegetation growing seasons.

Satellite precipitation products have been largely improved in the recent years particularly with the launch of the global precipitation measurement (GPM) core satellite. Moreover, the development of techniques for exploiting the information provided by satellite soil moisture to complement/enhance precipitation products have improved the accuracy of accumulated rainfall estimates over land. Such satellite enhanced precipitation products, available with a short latency (< 1 day), represent an important and new source of information for river flow prediction and water resources management, particularly in developing countries in which ground observations are scarcely available and the access to such data is not always ensured. In this study, three recently developed rainfall products obtained from the integration of GPM rainfall and satellite soil moisture products have been used; namely GPM+SM2RAIN, PRISM-SMOS, and PRISM-SMAP. The prediction of observed daily river discharge at 10 basins located in Europe (4), West Africa (3) and South Africa (3) is carried out. For comparison, we have also considered three rainfall products based on: (1) GPM only, i.e., the Early Run version of the Integrated Multi-Satellite Retrievals for GPM (GPM-ER), (2) rain gauges, i.e., the Global Precipitation Climatology Centre, and (3) the latest European Centre for Medium-Range Weather Forecasts reanalysis, ERA5. Three different conceptual and lumped rainfall-runoff models are employed to obtain robust and reliable results over the 3-year data period 2015-2017. Results indicate that, particularly over scarcely gauged areas (West Africa), the integrated products outperform both ground- and reanalysis-based rainfall estimates. For all basins, the GPM+SM2RAIN product is performing the best among the short latency products with mean Kling-Gupta Efficiency (KGE) equal to 0.87, and significantly better than GPM-ER (mean KGE = 0.77). The integrated products are found to reproduce particularly well the high flows. These results highlight the strong need to disseminate such integrated satellite rainfall products for hydrological (and agricultural) applications in poorly gauged areas such as Africa and South America.

We show how to use "complementary priors" to eliminate the explaining-away effects that make inference difficult in densely connected belief nets that have many hidden layers. Using complementary priors, we derive a fast, greedy algorithm that can learn deep, directed belief networks one layer at a time, provided the top two layers form an undirected associative memory. The fast, greedy algorithm is used to initialize a slower learning procedure that fine-tunes the weights using a contrastive version of the wake-sleep algorithm. After fine-tuning, a network with three hidden layers forms a very good generative model of the joint distribution of handwritten digit images and their labels. This generative model gives better digit classification than the best discriminative learning algorithms. The low-dimensional manifolds on which the digits lie are modeled by long ravines in the free-energy landscape of the top-level associative memory, and it is easy to explore these ravines by using the directed connections to display what the associative memory has in mind.

Precipitation nowcasting can predict and alert for any possibility of abrupt weather changes which may cause both human and material risks. Most of the conventional nowcasting methods extrapolate weather radar echoes, but precipitation nowcasting is still a challenge, mainly due to rapid changes in meteorological systems and time required for numerical simulations. Recently video prediction deep learning (VPDL) algorithms have been applied in precipitation nowcasting. In this study, we use the VPDL PredRNN++ and sequences of radar reflectivity images to predict the future sequence of reflectivity images for up to 1-h lead time for São Paulo, Brazil. We also verify the feasibility for the continuous use of the VPDL model, providing the meteorologist with trends and forecasts in precipitation edges regardless of the weather event occurring. The results obtained confirm the potential of the VPDL model as an additional tool to assist nowcasting. Even though meteorological systems that trigger natural disasters vary by location, a general solution can contribute as a tool to assist decision-makers and consequently issue efficient alerts.

The spatial resolution of precipitation predicted by general circulation models is too coarse to meet current research and operational needs. Downscaling is one way to provide finer resolution data at local scales. The single-image super-resolution method in the computer vision field has made great strides lately and has been applied in various fields. In this article, we propose a novel reference-based and gradient-guided deep learning model (RBGGM) to downscale daily precipitation considering the discontinuity of precipitation and ill-posed nature of downscaling. Global Precipitation Measurement Mission (GPM) precipitation data, variables in ERA5 re-analysis data, and topographic data are selected to perform the downscaling, and a residual dense attention block is constructed to extract features of them. By exploring the discontinuous feature of precipitation, we introduce gradient feature to reconstruct precipitation distribution. We also extract the feature of high-resolution monthly precipitation as a reference feature to resolve the ill-posed nature of downscaling. Extensive experimental results on benchmark data sets demonstrate that our proposed model performs better than other baseline methods. Furthermore, we construct a daily precipitation downscaling data set based on GPM precipitation data, ERA5 re-analysis data and topographic data.

<p>This paper analyzed the temporal and spatial responses of vegetation NDVI to the variations of temperature and precipitation in each ten-day period in the whole year, spring, summer and autumn covering 1998 to 2007&nbsp; based on the SPOT VGT-NDVI data and daily temperature and precipitation data from 205 meteorological stations in eastern China. The results indicate that on the whole, the response of vegetation NDVI to the variation of temperature is greater than to that of precipitation in eastern China. Vegetation NDVI maximally responds to the variation of temperature with a lag of about 10 days, and it maximally responds to the variation of precipitation with a lag of about 30 days. The response of vegetation NDVI to temperature and precipitation is the greatest in autumn, and the lag time is longer in summer. Spatially, the maximum response of vegetation NDVI to the variation of temperature is greater in the northern and central parts than in the southern part of eastern China. The maximum response of vegetation NDVI to the variation of precipitation is greater in the northern part than in the central and southern parts of eastern China. There is more lag time of vegetation NDVI to the variation of temperature in the northern and southern parts, while less in the central part. The lag time of vegetation NDVI to the variation of precipitation gradually increases from the northern to the southern part according to the latitude. The response of vegetation NDVI to the variations of temperature and precipitation in eastern China is mainly consistent with other results obtained in eastern and southern China.</p>

Learning to store information over extended time intervals by recurrent backpropagation takes a very long time, mostly because of insufficient, decaying error backflow. We briefly review Hochreiter's (1991) analysis of this problem, then address it by introducing a novel, efficient, gradient-based method called long short-term memory (LSTM). Truncating the gradient where this does not do harm, LSTM can learn to bridge minimal time lags in excess of 1000 discrete-time steps by enforcing constant error flow through constant error carousels within special units. Multiplicative gate units learn to open and close access to the constant error flow. LSTM is local in space and time; its computational complexity per time step and weight is O(1). Our experiments with artificial data involve local, distributed, real-valued, and noisy pattern representations. In comparisons with real-time recurrent learning, back propagation through time, recurrent cascade correlation, Elman nets, and neural sequence chunking, LSTM leads to many more successful runs, and learns much faster. LSTM also solves complex, artificial long-time-lag tasks that have never been solved by previous recurrent network algorithms.

Recurrent neural networks (RNNs) have been widely adopted in research areas concerned with sequential data, such as text, audio, and video. However, RNNs consisting of sigma cells or tanh cells are unable to learn the relevant information of input data when the input gap is large. By introducing gate functions into the cell structure, the long short-term memory (LSTM) could handle the problem of long-term dependencies well. Since its introduction, almost all the exciting results based on RNNs have been achieved by the LSTM. The LSTM has become the focus of deep learning. We review the LSTM cell and its variants to explore the learning capacity of the LSTM cell. Furthermore, the LSTM networks are divided into two broad categories: LSTM-dominated networks and integrated LSTM networks. In addition, their various applications are discussed. Finally, future research directions are presented for LSTM networks.

卫星降水产品在山区的适用性具有较高的不确定性。论文基于天山山区2014&mdash;2015年167个气象站点数据,应用较为广泛的两套卫星降水产品&mdash;&mdash;TRMM与CMORPH,选用均方根误差（RMSE）、相关系数（R）、相对误差（PB）,以及分类统计分析指标（错报率FAR、探测率POD、公正先兆评分ETS、频率偏差BIAS）等,评估了新一代卫星降水产品&mdash;&mdash;Global Precipitation Measurement（GPM）在天山山区的适用性。结果表明：相对于以上两种产品,GPM在山区的精度最高。具体表现在：1）3套产品在降水较多的夏秋季均表现较好,相对于TRMM与CMORPH,GPM与观测数据的相关系数最高（R&ge;0.6）,相对误差最小（PB&asymp;10%）;2）在整个天山山区,GPM相对于其他两套产品表现出较低的误差范围（-55%~55%）;3）GPM在不同的高程带内,均表现出同观测站点较低的误差与较高的相关系数;4）3套降水产品均表现出对探测弱降水事件较高的准确性（POD&asymp;0.58）和较低的错误率（FAR&asymp;0.63）,但综合分析4种指数,GPM表现最佳,能够以较准确的精度和较低的误差估测降水系统。

The comparison of satellite and high-quality, ground-based estimates of precipitation is an important means to assess the confidence in satellite-based algorithms and to provide a benchmark for their continued development and future improvement. To these ends, it is beneficial to identify sources of estimation uncertainty, thereby facilitating a precise understanding of the origins of the problem. This is especially true for new datasets such as the Integrated Multisatellite Retrievals for GPM (IMERG) product, which provides global precipitation gridded at a high resolution using measurements from different sources and techniques. Here, IMERG is evaluated against a dense network of gauges in the mid-Atlantic region of the United States. A novel approach is presented, leveraging ancillary variables in IMERG to attribute the errors to the individual instruments or techniques within the algorithm. As a whole, IMERG exhibits some misses and false alarms for rain detection, while its rain-rate estimates tend to overestimate drizzle and underestimate heavy rain with considerable random error. Tracing the errors to their sources, the most reliable IMERG estimates come from passive microwave satellites, which in turn exhibit a hierarchy of performance. The morphing technique has comparable proficiency with the less skillful satellites, but infrared estimations perform poorly. The approach here demonstrated that, underlying the overall reasonable performance of IMERG, different sources have different reliability, thus enabling both IMERG users and developers to better recognize the uncertainty in the estimate. Future validation efforts are urged to adopt such a categorization to bridge between gridded rainfall and instantaneous satellite estimates.