Name of the Competition: Doppler Weather Radar Echo Data Extrapolation
二、赛题简介 短时极端降水易造成洪涝、滑坡、泥石流等自然灾害，故需对降 水范围、强度进行准确、及时的预警和预报。通常，气象行业部门采 用基于天气雷达回波数据外推的方法以进行短临降水预报。 本赛道中，参赛队伍需建立一个准确的天气雷达回波外推模型， 输入现有的、按时序排列的多普勒天气雷达回波序列数据，外推未来 时刻的、按时序排列的雷达回波序列数据。
Introduction to the Competition: Extreme precipitation is easy to cause natural disasters such as floods, landslides, and debris flows. Therefore, accurate and timely early warning and forecasting of the range and intensity of precipitation is required. Typically, meteorological industry departments use methods based on extrapolation of weather radar echo data for short-term precipitation forecasts.
In this track, the participating teams need to establish an accurate weather radar echo extrapolation model, which receives existing, time-ordered Doppler weather radar echo sequence data as input and extrapolates radar echo sequence data at future moments.
三、功能要求 如图 1 所示，参赛队伍所设计的外推模型以每个雷达回波图序列 的前五张图作为输入，输出其之后的十张回波图。各参赛队伍的目标 是使得模型输出的回波图尽可能和与之对应的真实回波图相似。 原始的回波图为 0-255 的灰度图像，为方便说明，图中的雷达回 波图已使用颜色进行填充。
Functional Requirements: As shown in Figure 1, the extrapolation models designed by the participating teams takes the first five images of each radar echo map sequence as input, and outputs the following ten echo maps. The goal of each participating team is to make the echo maps extrapolated by models as similar as possible to the ground truth.
The original echo maps are grayscale images where the values of pixel in them range from 0 to 255. For the convenience of illustration, the radar echo image in the figure has been filled with color.
Figure 1: Schematic diagram of radar echo extrapolation function requirements
操作系统：Windows 或 Linux
运行环境：PyTorch 或 TensorFlow
Programming language: Python recommended
Operating System: Windows or Linux
Running environment: PyTorch or TensorFlow
五、赛题数据（网址） 本赛题提供某站点天气雷达回波图序列数据集，分为测试集和训 练集，两部分数据格式、大小完全相同，均由若干天气雷达回波序列 构成。详细的数据集参数、说明已在下载链接的“Readme.md”文件 中说明。
Competition Dataset (Download Link): The competition provides a weather radar echo map sequence dataset of a single radar station, which includes a training set and a test set. The data structure and size of the two parts are exactly the same, and both have several weather radar echo sequences. Detailed dataset information and descriptions have been described in the "Readme.md" file in the download link. The training set has been uploaded and the test set will be uploaded on June 15, 2022. The download link of competition dataset is: https://github.com/ihznuy32/FREM-Dataset.
4、测试集全部测试结果的 PNG 图像文件压缩包，参赛队伍需建 立预命名为 Predict.zip 的压缩包，其中含有与测试集序列数相同的文 件夹存储该序列的预测结果 PNG 图片，命名为“Seq_+序列编号”， 如第 20 个序列为“Seq_20”。每一个文件夹中，预测结果 PNG 图片 按其对应时间进行排列，依次命名为“p1.png”、“p2.png”…“p10.png”。
Requirements for submission
1. Model design report, written and submitted using the prescribed English template, the download link of template is: https://www.tocenet.org/?attachment_id=17608&download=1.
2. Model training code, test code and model weights file. The version number of the relevant dependent software package needs to be given to ensure reproducibility.
3. Model demonstration video (≤3 minutes)
4. The compressed file that contains all extrapolated results of test set. All participating teams need to create a compressed file named “Predict.zip”, which contains the same number of sequences as the test set to store the PNG images of the prediction results of the sequence, named as "Seq_" + "sequence number", for example, the 20th sequence is "Seq_20". In each folder, the PNG images of the prediction results are arranged according to their corresponding time, and are named "p1.png", "p2.png"..."p10.png" in turn.
七、评分标准 比赛最终得分由初赛预测结果得分、报告及演示视频得分、答辩 现场表现得分三部分构成，上述三部分得分占比分别为 40%、30%、 30%。预测结果详细评分标准，见数据集下载链接中的“Readme.md” 文件。
The final score of the competition is composed of three parts: the score of the preliminary prediction result, the score of the report and presentation video, and the score of the live performance of the defense, which are 40%, 30% and 30% respectively. The detailed scoring criteria for the prediction results can be found in the "Readme.md" file in the download link.
2022 年 3 月