[{"data":1,"prerenderedAt":221},["ShallowReactive",2],{"content-query-Isno3nz5FG":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"date":10,"cover":11,"type":12,"body":13,"_type":215,"_id":216,"_source":217,"_file":218,"_stem":219,"_extension":220},"/news/zh/2776","zh",false,"","昇思科学计算套件喜迎新成员：MindSpore Earth 0.1地球科学套件全新发布","2023年9月21日下午，在以“加速行业智能化”为主题的华为全联接大会2023(Huawei Connect2023)昇思MindSpore专题论坛上，昇思MindSpore开源社区发布了MindSpore Earth 0.1地球科学套件。","2023-10-08","https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2023/10/12/79b456a307ed4f07b978606a21268c65.png","news",{"type":14,"children":15,"toc":212},"root",[16,24,29,34,39,44,53,58,70,75,83,88,96,101,108,113,121,126,137,142,153,158,165,170,175,182,187,192,199],{"type":17,"tag":18,"props":19,"children":21},"element","h1",{"id":20},"昇思科学计算套件喜迎新成员mindspore-earth-01地球科学套件全新发布",[22],{"type":23,"value":8},"text",{"type":17,"tag":25,"props":26,"children":27},"p",{},[28],{"type":23,"value":9},{"type":17,"tag":25,"props":30,"children":31},{},[32],{"type":23,"value":33},"该套件集成了多时空尺度下的AI气象预报SOTA模型，提供了数据前处理、预报可视化等工具，并集成了ERA5再分析、雷达回波、高分辨率DEM数据集，致力于高效使能AI+气象和海洋预报的融合研究。",{"type":17,"tag":25,"props":35,"children":36},{},[37],{"type":23,"value":38},"气象预报与人们的工作生活息息相关，也是科学智能(AI4Science)领域受到最广泛关注的应用场景之一。昇思MindSpore作为全场景AI融合框架，具有原生支持大模型与AI4Science两大引领创新的能力。",{"type":17,"tag":25,"props":40,"children":41},{},[42],{"type":23,"value":43},"MindSpore Earth架构规划如图1所示，涵盖气象预报短临降水、中期预报、超分辨率等多个场景的业界SOTA模型，包括GraphCast、ViT-KNO、FourCastNet、DGMR等，模型覆盖度业界领先，预报精度超越传统数值模式，预报速度较传统数值模式提升千倍以上。",{"type":17,"tag":25,"props":45,"children":46},{},[47],{"type":17,"tag":48,"props":49,"children":52},"img",{"alt":50,"src":51},"image.png","https://fileserver.developer.huaweicloud.com/FileServer/getFile/cmtybbs/e64/154/b38/90a1d5d431e64154b387b3660e356ff5.20231011021625.51232818337217176445796363416249:50541011062820:2400:0651856F0D723B6D61F8FD175F2648AED666E26C0B7E1D817FEFA6AE394DBA1D.png",[],{"type":17,"tag":25,"props":54,"children":55},{},[56],{"type":23,"value":57},"图1 MindSpore Earth套件架构规划",{"type":17,"tag":25,"props":59,"children":60},{},[61],{"type":17,"tag":62,"props":63,"children":64},"strong",{},[65],{"type":17,"tag":62,"props":66,"children":67},{},[68],{"type":23,"value":69},"1、 中期天气预报",{"type":17,"tag":25,"props":71,"children":72},{},[73],{"type":23,"value":74},"中期全球天气预报是指在全球范围内预测未来约3至10天的天气。这类预测通常基于数值模式模拟大气条件，如温度、湿度、气压、风速和方向以及降水的变化。MindSpore Earth提供了多个AI中期预报模型：",{"type":17,"tag":25,"props":76,"children":77},{},[78],{"type":17,"tag":62,"props":79,"children":80},{},[81],{"type":23,"value":82},"FourCastNet",{"type":17,"tag":25,"props":84,"children":85},{},[86],{"type":23,"value":87},"MindSpore Earth提供FourCastNet 模型，该模型采用自适应傅里叶神经算子 AFNO，这种神经网络架构是对 Vision Transformer(ViT) 模型地改进，它将混合操作步骤构建成连续的全局卷积，在傅里叶域中通过 FFT 有效实现，将空间混合复杂度降低到 O(Nlog N)，这允许灵活且可扩展地建模跨空间和通道维度的依赖关系。该模型为第一个预报精度能与欧洲中期天气预报中心（ECMWF）的高分辨率综合预测系统（IFS）模型比较的AI预报模型。",{"type":17,"tag":25,"props":89,"children":90},{},[91],{"type":17,"tag":62,"props":92,"children":93},{},[94],{"type":23,"value":95},"ViT-KNO",{"type":17,"tag":25,"props":97,"children":98},{},[99],{"type":23,"value":100},"MindSpore Earth提供了一个基于Koopman的全局线性化理论并结合神经算子的思想设计的一个轻量化的、网格无关的Koopman Neural Operator模型，模型架构如图2所示。该模型由华为先进计算与存储实验室与清华大学合作推出。该模型能够捕获复杂的非线性行为，同时保持模型轻量级和计算有效性。与FourCastNet相比，ViT-KNO具有更高效的训练性能与更优的预测精度。",{"type":17,"tag":25,"props":102,"children":103},{},[104],{"type":17,"tag":48,"props":105,"children":107},{"alt":50,"src":106},"https://fileserver.developer.huaweicloud.com/FileServer/getFile/cmtybbs/e64/154/b38/90a1d5d431e64154b387b3660e356ff5.20231011021647.36712380495109354325962714339951:50541011062820:2400:50A0624E4E29016FB3DB7D99EBEFF6FC33B7FA19FB316108A99412ACCA313ABD.png",[],{"type":17,"tag":25,"props":109,"children":110},{},[111],{"type":23,"value":112},"图2 ViT-KNO模型架构",{"type":17,"tag":25,"props":114,"children":115},{},[116],{"type":17,"tag":62,"props":117,"children":118},{},[119],{"type":23,"value":120},"GraphCast",{"type":17,"tag":25,"props":122,"children":123},{},[124],{"type":23,"value":125},"GraphCast来自谷歌的DeepMind，该模型使用GNN在“编码-处理-解码”架构中自动回归地生成预测结果。编码器将历史时刻的气象要素的纬度-经度输入网格映射到多尺度内部网格表示；处理器在多网格表示上执行多轮消息传递；解码器将多网格表示映射回纬度-经度网格，同时输出预测结果。MindSpore Earth开源了正二十面体网格生成模块，实现多尺度网格自动化构建。另外，针对多步预测精度衰减，MindSpore Earth实现了rollout多步迭代训练，降低模型误差累积。",{"type":17,"tag":25,"props":127,"children":128},{},[129],{"type":17,"tag":62,"props":130,"children":131},{},[132],{"type":17,"tag":62,"props":133,"children":134},{},[135],{"type":23,"value":136},"2、短临降水预报",{"type":17,"tag":25,"props":138,"children":139},{},[140],{"type":23,"value":141},"MindSpore Earth提供了DGMR降水模型，该模型的主体是一个生成器，配合时间和空间判别器损失以及额外的正则化项进行对抗训练。模型从前四帧雷达序列学习上下文表示，用作采样器的输入，采样器是一个由卷积门控循环单元（GRU）构成的递归网络，它将上下文表示和从高斯分布中取样的潜向量作输入，对未来18个雷达场进行预测。基于MindSpore Earth+昇腾可以进行对降水强度与空间分布进行高效训练与推理。",{"type":17,"tag":25,"props":143,"children":144},{},[145],{"type":17,"tag":62,"props":146,"children":147},{},[148],{"type":17,"tag":62,"props":149,"children":150},{},[151],{"type":23,"value":152},"3、数字高程模型超分",{"type":17,"tag":25,"props":154,"children":155},{},[156],{"type":23,"value":157},"昇思MindSpore团队、华为AI4Sci LAB与清华大学黄小猛团队联合推出适用于全球区域的DEM超分模型，同时发布了全球3弧秒（90 m）海陆DEM数据产品（图3），该成果已发表于《科学通报》（Science Bulletin）上。该模型在RMSE指标、清晰度、细节等方面均优于目前广泛采用的超分模型。该成果是首个分辨率在百米以内的全球海陆DEM数据集，可以满足不同领域和不同层次对海洋测深数据的需求，为不同地形复杂度下全球海陆重力场与地形的关系、探索不同海陆构造单元的均衡机制以及海陆地形对海洋潮流运动的影响等方面的研究提供重要支撑。",{"type":17,"tag":25,"props":159,"children":160},{},[161],{"type":17,"tag":48,"props":162,"children":164},{"alt":50,"src":163},"https://fileserver.developer.huaweicloud.com/FileServer/getFile/cmtybbs/e64/154/b38/90a1d5d431e64154b387b3660e356ff5.20231011021709.28982670450446197019188937800900:50541011062820:2400:063DF019F7D193271F78D2615335348CA37BBBFC692A92582CC59D71CBFF5A83.png",[],{"type":17,"tag":25,"props":166,"children":167},{},[168],{"type":23,"value":169},"图3 高分辨率全球海陆DEM数据集",{"type":17,"tag":25,"props":171,"children":172},{},[173],{"type":23,"value":174},"另外，MindSpore Earth还提供了预报可视化模块，如风场可视化（图4）；内置ERA5再分析数据集、雷达回波数据集、高分辨率DEM数据，支持短临预报、中期预报等模型训练与评估。未来MindSpore Earth将继续提供前沿、高效的AI气象海洋模型与工具，包括盘古气象大模型推理功能、长期气候预测、降尺度等，使能AI+气象海洋的融合研究。",{"type":17,"tag":25,"props":176,"children":177},{},[178],{"type":17,"tag":48,"props":179,"children":181},{"alt":50,"src":180},"https://fileserver.developer.huaweicloud.com/FileServer/getFile/cmtybbs/e64/154/b38/90a1d5d431e64154b387b3660e356ff5.20231011021726.81241391039642006704417094056924:50541011062820:2400:63F655F0688704BF1A6D10FDC8C4087E08648B9B0630640DB0AE6812994DD783.png",[],{"type":17,"tag":25,"props":183,"children":184},{},[185],{"type":23,"value":186},"图4 风速可视化效果",{"type":17,"tag":25,"props":188,"children":189},{},[190],{"type":23,"value":191},"如需了解更多详情，欢迎加入MindSpore Flow & Earth联合SIG群。",{"type":17,"tag":25,"props":193,"children":194},{},[195],{"type":17,"tag":48,"props":196,"children":198},{"alt":50,"src":197},"https://fileserver.developer.huaweicloud.com/FileServer/getFile/cmtybbs/e64/154/b38/90a1d5d431e64154b387b3660e356ff5.20231011021741.33086512859212140114000675876333:50541011062820:2400:95702360D52384755D3732C1F4B790D05A10B18DFFA15F04F57174D95C12A24F.png",[],{"type":17,"tag":25,"props":200,"children":201},{},[202,204],{"type":23,"value":203},"MindSpore Earth代码仓地址：",{"type":17,"tag":205,"props":206,"children":210},"a",{"href":207,"rel":208},"https://gitee.com/mindspore/mindscience/tree/master/MindEarth",[209],"nofollow",[211],{"type":23,"value":207},{"title":7,"searchDepth":213,"depth":213,"links":214},4,[],"markdown","content:news:zh:2776.md","content","news/zh/2776.md","news/zh/2776","md",1776506072742]