Radar rainfall nowcasts for 1,533 events spread over 12 Dutch catchments

This repository contains the radar rainfall nowcasts for 1,533 rainfall events spread over 12 basins in the Netherlands, from Imhoff et al. (2020). The basins, a combination of freely draining catchments and polder areas, are: Aa, Beemster, Delfland, Dwarsdiep, Grote Waterleiding, Hupsel, Linde, Luntersebeek, Regge, Reusel, Rijnland and Roggelsebeek. More catchment information can be found in figure 1 in Imhoff et al. (2020). The nowcasts were run with the radar composite of the Royal Netherlands Meteorological Institute for the period 2008 – 2018. Subsequently, the nowcasts were clipped to the extent of one of the basins and saved. <br>The events were systematically selected as follows (see also section 2.2.2. in Imhoff et al., 2020): Per catchment and for each season, eight events were selected per event duration (1, 3, 6, and 24 hr). An event was defined as a period with one of the four aforementioned durations in which it does not have to rain continuously. Based on this description of an event duration, the highest rainfall sums per duration, basin, and season were ranked from high to low, and the eight events with the highest rainfall sum for that duration (basin and season) were selected. In the list, an event could not occur within the time span of a previously selected event with a higher rainfall sum. This ensures that there is no overlap in the chosen events, given the duration, basin and season. Hence, summed over all durations (4), seasons (4) and basins (12), this selection procedure leads to 4 × 4 × 12 × 8 = 1,536 events. Of these 1,536 events, 1,533 ran successfully and are thus available in this repository.<br> For every five-minute time step in each of the events (with durations from 1 to 24 hours), a nowcast was made with one of the following nowcasting algorithms from the open-source Rainymotion (v0.1 used) and pySTEPS (v0.2 used) libraries (Ayzel et al., 2019; Pulkkinen et al., 2019): Rainymotion Sparse (referred to as RM-S in the files and corresponding paper), Rainymotion DenseRotation (referred to as RM-DR), pySTEPS deterministic (PS-D) and pySTEPS probabilistic (PS-P) with 20 ensemble members. All nowcasts have a time step of 5 min and a forecast horizon of six hours. Hence, in total, over 940,000 separate 5-min forecasts can be analyzed per algorithm. More information regarding the nowcasting algorithms can be found in Imhoff et al. (2021), Ayzel et al. (2019) and Pulkkinen et al. (2019). <br>The nowcasts are saved as netCDF files in folders with the following structure: catchment name &gt; nowcasting algorithm (RMS, RMDR, PSD and PSP) &gt; the individual nowcasts. The filename indicates the issue time of the nowcast in UTC. The events and information about the rainfall sum during the event are present in a separate folder called events. In this folder, the structure is as follows: season &gt; duration &gt; basin. The csv-files provide information on the date of the event (the end time is given) and the catchment-averaged rainfall sum during the event.

本数据集仓库收录了源自Imhoff等（2020）研究的、覆盖荷兰12个流域的1533次降雨事件的雷达降雨临近预报数据。这些流域由自由排水汇水区与圩田区域共同组成，分别为：Aa、Beemster、Delfland、Dwarsdiep、Grote Waterleiding、Hupsel、Linde、Luntersebeek、Regge、Reusel、Rijnland以及Roggelsebeek。更多流域相关信息可参见Imhoff等（2020）的图1。
本数据集采用荷兰皇家气象研究所的雷达组合反射率产品，在2008至2018年期间生成临近预报。随后将预报结果裁剪至对应单个流域的范围并保存。

降雨事件的筛选流程如下（详见Imhoff等（2020）的2.2.2节）：针对每个汇水区与每个季节，按照降雨事件持续时长（1、3、6、24小时）分别选取8次事件。本次研究将降雨事件定义为具有上述四种持续时长之一的时段，时段内无需持续降雨。基于上述事件时长的定义，按单次时长、单个流域与单个季节的总降雨量从高到低排序，选取对应时长、流域和季节下总降雨量最高的8次事件。在筛选列表中，已选取的高降雨量事件的时间范围内不得再纳入其他事件，以此确保在相同时长、流域与季节条件下，所选事件之间不存在时间重叠。综上，综合4种时长、4个季节与12个流域，本次筛选共得到1536次事件。其中1533次事件成功生成临近预报，因此收录于本数据集仓库中。

针对每场时长1至24小时的降雨事件，按每5分钟的时间步长，采用开源库Rainymotion（使用v0.1版本）与pySTEPS（使用v0.2版本）中的以下临近预报算法生成预报结果（Ayzel等，2019；Pulkkinen等，2019）：Rainymotion Sparse（在文件与相关论文中记为RM-S）、Rainymotion DenseRotation（记为RM-DR）、pySTEPS确定性预报（PS-D）以及带有20个集合成员的pySTEPS概率预报（PS-P）。所有临近预报的时间步长均为5分钟，预报时效为6小时。因此，每种算法总计可生成超过94万条独立的5分钟预报结果，可供分析使用。有关临近预报算法的更多细节可参见Imhoff等（2021）、Ayzel等（2019）以及Pulkkinen等（2019）的研究。

临近预报结果以netCDF格式文件保存，文件夹结构如下：汇水区名称 > 临近预报算法（RMS、RMDR、PSD与PSP） > 单个预报文件。文件名标注了该临近预报的发布时间，采用协调世界时（UTC）。降雨事件及事件期间的总降雨量信息存储在名为`events`的独立文件夹中，该文件夹内的结构为：季节 > 时长 > 流域。其中的CSV文件提供了事件日期（标注事件结束时间）以及汇水区平均总降雨量的相关信息。