[{"data":1,"prerenderedAt":220},["ShallowReactive",2],{"content-query-y3lWDSWV0O":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"date":10,"cover":11,"type":12,"body":13,"_type":214,"_id":215,"_source":216,"_file":217,"_stem":218,"_extension":219},"/news/en/2874","en",false,"","New Member of MindSpore Scientific Computing Suites: MindSpore Earth 0.1","At the MindSpore forum of HUAWEI CONNECT 2023 with the theme of "Accelerate Intelligence", the MindSpore community officially released a new MindSpore scientific computing suite, MindSpore Earth 0.1.","2023-10-08","https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2023/11/15/2a7e19ddb0724bdcadd44e9ad75629f4.png","news",{"type":14,"children":15,"toc":211},"root",[16,24,30,35,40,45,53,58,67,72,80,85,93,98,105,110,118,123,131,136,144,157,164,169,174,181,186,191,198],{"type":17,"tag":18,"props":19,"children":21},"element","h1",{"id":20},"new-member-of-mindspore-scientific-computing-suites-mindspore-earth-01",[22],{"type":23,"value":8},"text",{"type":17,"tag":25,"props":26,"children":27},"p",{},[28],{"type":23,"value":29},"On the afternoon of September 21, 2023, at the MindSpore forum of HUAWEI CONNECT 2023 with the theme of \"Accelerate Intelligence\", the MindSpore open-source community officially released a new MindSpore scientific computing suite, MindSpore Earth 0.1.",{"type":17,"tag":25,"props":31,"children":32},{},[33],{"type":23,"value":34},"This suite integrates the AI weather forecast SOTA model at multiple spatiotemporal scales, provides tools such as data preprocessing and forecast visualization, and integrates ERA5 reanalysis, radar echo, and high-resolution DEM datasets. It is committed to efficiently enabling AI+ meteorology and ocean forecast convergence research.",{"type":17,"tag":25,"props":36,"children":37},{},[38],{"type":23,"value":39},"Weather forecast is closely related to people's work and life, and is one of the most widely concerned application scenarios in the AI4Sci domain. MindSpore, as an all-scenario AI convergence framework, has the ability to natively support foundation models and AI4Sci, two leading innovative capabilities.",{"type":17,"tag":25,"props":41,"children":42},{},[43],{"type":23,"value":44},"Figure 1 shows the MindSpore Earth architecture planning. Multiple SOTA models, including GraphCast, ViT-KNO, FourCastNet, and DGMR, are used in scenarios such as short-term precipitation, medium-range weather forecast, and super-resolution. The model coverage is industry-leading, the forecast accuracy is higher than that of the traditional numerical mode, and the forecast speed is more than 1000 times higher than that of the traditional numerical mode.",{"type":17,"tag":25,"props":46,"children":47},{},[48],{"type":17,"tag":49,"props":50,"children":52},"img",{"alt":7,"src":51},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2023/11/15/37e04b002823442ea3f06b7706752b06.png",[],{"type":17,"tag":25,"props":54,"children":55},{},[56],{"type":23,"value":57},"Figure 1 MindSpore Earth suite architecture planning",{"type":17,"tag":25,"props":59,"children":60},{},[61],{"type":17,"tag":62,"props":63,"children":64},"strong",{},[65],{"type":23,"value":66},"1. Mid-term Weather Forecast",{"type":17,"tag":25,"props":68,"children":69},{},[70],{"type":23,"value":71},"The medium-range global weather forecast refers to the forecast of weather for about three to ten days in the future globally. Such forecasts are usually based on numerical mode simulations of atmospheric conditions, such as changes in the temperature, humidity, pressure, wind speed and direction, and precipitation. MindSpore Earth provides multiple medium-range forecast AI models:",{"type":17,"tag":25,"props":73,"children":74},{},[75],{"type":17,"tag":62,"props":76,"children":77},{},[78],{"type":23,"value":79},"FourCastNet",{"type":17,"tag":25,"props":81,"children":82},{},[83],{"type":23,"value":84},"MindSpore Earth provides the FourCastNet model, which uses the adaptive Fourier neural operators (AFNOs). This neural network architecture is an improvement on the Vision Transformer (ViT) model. It constructs mixed operations into continuous global convolutions and effectively implements them through fast Fourier transform (FFT) in the Fourier domain, reducing the spatial mixing complexity to O(Nlog N). This allows for flexible and scalable modeling of dependencies across spatial and channel dimensions. This model is the first AI weather forecast model that is comparable with the high-resolution integrated forecasting system (IFS) model of the European Centre for Medium-Range Weather Forecasts (ECMWF) in terms of forecast accuracy.",{"type":17,"tag":25,"props":86,"children":87},{},[88],{"type":17,"tag":62,"props":89,"children":90},{},[91],{"type":23,"value":92},"ViT-KNO",{"type":17,"tag":25,"props":94,"children":95},{},[96],{"type":23,"value":97},"MindSpore Earth provides a lightweight grid-independent Koopman neural operator model designed based on the Koopman global linearization theory and the idea of neural operators. Figure 2 shows the model architecture. This model is jointly launched by Huawei Advanced Computing and Storage Laboratory and Tsinghua University. It can capture complex nonlinear behavior while maintaining model lightweight attribute and computational effectiveness. Compared with FourCastNet, ViT-KNO has more efficient training and better forecast accuracy.",{"type":17,"tag":25,"props":99,"children":100},{},[101],{"type":17,"tag":49,"props":102,"children":104},{"alt":7,"src":103},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2023/11/15/44d40dafd2dd4e739bc55399df8c1161.png",[],{"type":17,"tag":25,"props":106,"children":107},{},[108],{"type":23,"value":109},"Figure 2 ViT-KNO model architecture",{"type":17,"tag":25,"props":111,"children":112},{},[113],{"type":17,"tag":62,"props":114,"children":115},{},[116],{"type":23,"value":117},"GraphCast",{"type":17,"tag":25,"props":119,"children":120},{},[121],{"type":23,"value":122},"GraphCast is a model from Google's DeepMind. It uses graph neural networks (GNNs) to automatically generate forecast results through an \"encode-process-decode\" architecture in an autoregressive manner. The encoder maps the latitude-longitude girds of meteorological elements at historical moments to a multi-scale internal grid representation; the processor performs multi-round message passing on the multi-grid representation; the decoder maps the multi-grid representation back to the latitude-longitude grids while outputting the forecast result. MindSpore Earth has open-sourced the regular icosahedron grid generation module, achieving automated multi-scale grid construction. In addition, to address the accuracy decay in multi-step forecast, MindSpore Earth has implemented multi-step iterative training to reduce the accumulation of model errors.",{"type":17,"tag":25,"props":124,"children":125},{},[126],{"type":17,"tag":62,"props":127,"children":128},{},[129],{"type":23,"value":130},"2. Short-term Precipitation Forecast",{"type":17,"tag":25,"props":132,"children":133},{},[134],{"type":23,"value":135},"MindSpore Earth offers the DGMR precipitation model. The main body of the model is a generator that works with the losses of time and space discriminators and additional regularization items for adversarial training. The model learns contextual representations from the first four frames of radar sequences as inputs to a sampler. The sampler is a recursive network composed of convolutional gated recurrent units (GRUs). It takes the contextual representations and latent vectors sampled from a Gaussian distribution as inputs, and forecast the next 18 radar fields. Based on MindSpore Earth and Ascend, you can perform efficient training and inference on precipitation intensity and spatial distribution.",{"type":17,"tag":25,"props":137,"children":138},{},[139],{"type":17,"tag":62,"props":140,"children":141},{},[142],{"type":23,"value":143},"3. Digital Elevation Model Super-resolution",{"type":17,"tag":25,"props":145,"children":146},{},[147,149,155],{"type":23,"value":148},"MindSpore, Ascend AI4Sci Lab, and Professor Huang Xiaomeng's team from the Tsinghua University jointly launched a super-resolution 3-arc-second (90-meter) global digital elevation model (DEM) and related data product (Figure 3). The related article has been published in ",{"type":17,"tag":150,"props":151,"children":152},"em",{},[153],{"type":23,"value":154},"Science Bulletin",{"type":23,"value":156},". This model outperforms widely adopted super-resolution models in terms of RMSE, clarity, and detail. This achievement is the first global DEM dataset with a resolution of less than 100 meters. It meets the needs of ocean sounding data in different fields and at different levels and provides important support for research on the relationship between the global land-sea gravity field and topography with different terrain complexity, exploration of the equilibrium mechanism of different land-sea tectonic units, and research on the influence of land-sea topography on ocean tidal current movement.",{"type":17,"tag":25,"props":158,"children":159},{},[160],{"type":17,"tag":49,"props":161,"children":163},{"alt":7,"src":162},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2023/11/15/34071ea05ead47c6857943f6721365fa.png",[],{"type":17,"tag":25,"props":165,"children":166},{},[167],{"type":23,"value":168},"Figure 3 High-resolution global land-sea DEM dataset",{"type":17,"tag":25,"props":170,"children":171},{},[172],{"type":23,"value":173},"In addition, MindSpore Earth also provides a forecast visualization module, such as wind field visualization (Figure 4). It has built-in ERA5 reanalysis dataset, radar echo dataset, and high-resolution DEM data, and supports model training and evaluation for short-term forecast, medium-range forecast, and more. In the future, MindSpore Earth will continue to provide cutting-edge, efficient AI-based meteorology and ocen models and tools, including Pangu-Weather model inference, long-term climate forecast, and scale reduction, enabling AI+ integrated research on meteorology and ocean.",{"type":17,"tag":25,"props":175,"children":176},{},[177],{"type":17,"tag":49,"props":178,"children":180},{"alt":7,"src":179},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2023/11/15/8a8f8bec5c844a3f9075977eba1d40c0.png",[],{"type":17,"tag":25,"props":182,"children":183},{},[184],{"type":23,"value":185},"Figure 4 Wind speed visualization effect",{"type":17,"tag":25,"props":187,"children":188},{},[189],{"type":23,"value":190},"For more details, please join the MindSpore Flow & Earth SIG.",{"type":17,"tag":25,"props":192,"children":193},{},[194],{"type":17,"tag":49,"props":195,"children":197},{"alt":7,"src":196},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2023/11/15/e72f686d23e84f0b9b64ef58802d15fd.png",[],{"type":17,"tag":25,"props":199,"children":200},{},[201,203],{"type":23,"value":202},"Address of the MindSpore Earth code repository: ",{"type":17,"tag":204,"props":205,"children":209},"a",{"href":206,"rel":207},"https://gitee.com/mindspore/mindscience/tree/master/MindEarth",[208],"nofollow",[210],{"type":23,"value":206},{"title":7,"searchDepth":212,"depth":212,"links":213},4,[],"markdown","content:news:en:2874.md","content","news/en/2874.md","news/en/2874","md",1776506046464]