Google DeepMind发布WeatherNext
WeatherNext 3: More accurate, timely, and local weather forecasts
WeatherNext 3在时效性和分辨率上实现了显著跃迁，直接整合原始卫星数据实现每小时更新，这对能源调度和灾害预警有实际价值，值得关注其落地效果。
Predicting weather is one of the oldest, most complicated challenges.
预测天气是最古老、最复杂的挑战之一。
When we can tell how the weather may change, we can plan ahead.
当我们能够预判天气的变化时，就可以提前进行规划。
And with extreme weather, the stakes are getting higher. Traditional weather forecast use mathematical equation based on physics. So basically they taking all the different variables of the atmospheres and trying to simulate it step by step. Now, if you're trying to do it at a high resolution, at a global scale, takes a long period of time to do that and it's also super expensive.
而在极端天气情况下，风险也在不断升高。传统的天气预报使用基于物理学的数学方程。基本上，它们会收集大气的所有不同变量，并尝试一步步进行模拟。现在，如果你试图以高分辨率在全球范围内进行这种模拟，需要花费很长时间，而且成本也极其高昂。
You kind of had to trade off between having a regional high resolution model or a global but lower resolution models. AI takes a different approach. It learns from historical observations of the atmosphere and figures out how to best predict the weather. Now, we can close this gap and we can deliver high resolution all over the globe. Our previous Weather Next models demonstrated that AI can produce global forecasts faster and more accurately than traditional weather forecasting models.
你必须在区域高分辨率模型和全球低分辨率模型之间做出权衡。人工智能采取了不同的方法。它从大气的历史观测数据中学习，找出最佳预测天气的方式。现在，我们可以弥补这一差距，在全球范围内提供高分辨率的预报。我们之前的 Weather Next 模型证明，人工智能比传统天气预报模型能更快、更准确地生成全球预报。
So, Weather Next 3 is the first global operational weather model that produces a new forecast every hour of the day with up to 5 km resolution with hourly time steps and state-of-the-art skill across most variables. When we talk about variables in weather, what we mean is things like temperature, wind, pressure, rain, the things that describe what's going on in the atmosphere.
因此，Weather Next 3 是首个全球业务化天气预报模型，它每天每小时生成一次新预报，空间分辨率高达 5 公里，时间步长为每小时，并且在大多数变量上都达到了最先进的精度水平。当我们谈论天气中的“变量”时，指的是温度、风、气压、降雨等描述大气状况的因素。
This ability to bring rich speed and details together all at once make it better at delivering actionable data where people and industry are operating.
这种将丰富的速度、细节和准确性集于一体的能力，使其能够更好地为人类活动和工业运营提供可操作的数据。
Weather next 3 directly incorporates real world observation such as those coming from satellites or weather stations on the ground. This is different from what AI models have done previously where they were purely based on an analysis. This means that they can only be refreshed so often, typically every 6 hours.
Weather Next 3 直接整合了现实世界的观测数据，例如来自卫星或地面气象站的数据。这与以往的人工智能模型不同，后者完全基于分析结果。这意味着它们只能每隔一段时间更新一次，通常是每 6 小时。
The atmosphere does not move in a 6h hour lips, right? Whether it's precipitation or shift in the wind, changes can happen in minutes.
大气的运动并不是以 6 小时为间隔进行的，对吧？无论是降水还是风向变化，都可能在几分钟内发生。
With weather 3, we just use raw satellite data and that allows us to produce a new forecast every hour. So you've also improved the spatial resolution of the model. What does that mean?
通过 Weather Next 3，我们直接使用原始卫星数据，这使得我们能够每小时生成一次新预报。所以你也提高了模型的空间分辨率。那是什么意思呢？
I need a whiteboard for that.
我需要一块白板来解释这个。
A whiteboard.
一块白板。
Special resolution is the amount of details we want to provide in a specific area. So all of us knows from our phones, right? Like if you have a picture with high resolution that mean that it has many more pixels on the picture itself. Now imagine that you taking the globes and you divided it to different squares. If you're trying to predict weather at square of 25 kilome or 25 kilometer that mean that you're basically doing some kind of averaging to the weather at that scale when you want to predict the weather that people are actually feeling on the ground this is where you need a higher resolution of the weather forecast.
特殊分辨率是指我们希望在特定区域提供的细节程度。我们都从手机中了解过这一点，对吧？比如，如果你有一张高分辨率的图片，那就意味着图片本身包含更多的像素。现在想象一下，你把地球仪划分成不同的方格。如果你试图预测一个25公里见方的区域内的天气，这意味着你基本上是在该尺度上对天气进行某种平均处理。而当你想要预测人们在地面上实际感受到的天气时，这就是你需要更高分辨率天气预报的地方。
This really matters if you for example live near the coast or near mountains where you know from experience that weather can change very quickly on short distances. Weather next three actually provide with three native resolutions in single pass. We have the 25 km resolution for the broader atmosphere changes.
这非常重要，例如，如果你住在沿海或山区附近，根据经验你知道天气在短距离内可能会迅速变化。Weather Next 3 在一次传递中提供三种原生分辨率。我们拥有用于更广泛大气变化的25公里分辨率。
It can now predict all surface variables so wind or pressure at a 9 km resolution
它现在可以预测所有表面变量，如9公里分辨率下的风速或气压。
and then we have up to 5 kilome for temperature and humidity.
然后我们有高达5公里的温度和湿度分辨率。
One of the areas that we decided to focus is renewable energy. We added wind speed and direction at 100 mters, which is useful for managing wind farms because it's at the typical height of a wind turbine. We also added cloud cover and radiation to predict how much sunlight will hit solar panels.
我们决定重点关注的领域之一是可再生能源。我们添加了100米高度的风速和风向，这对于管理风电场非常有用，因为这与风力涡轮机的典型高度一致。我们还添加了云量和辐射数据，以预测有多少阳光会照射到太阳能电池板上。
Weather is one of those rare domains where cutting edge AI can directly improve decisions affected billions of people every single day.
天气是那些前沿人工智能可以直接改善数十亿人每天所受影响决策的罕见领域之一。
So imagine a world where we can give people the forecasts that they need. So think of a farmer making a decision of when to harvest. Think of flood forecasting, making sure that we evacuate people in time if there's a big flood coming or just personal use if you want to know whether you can do an event on the weekend or not. Weather next 3 will be available across all Google surfaces. So that's on search that's uh in Gemini on maps that allows us to reach billions of people with our forecasts.
所以想象一个我们可以为人们提供所需预报的世界。想想农民决定何时收割。想想洪水预报，确保如果发生大洪水能及时疏散人员，或者仅仅是个人用途，比如你想了解周末是否能举办活动。Weather Next 3 将在所有 Google 平台上提供。也就是说，在搜索中、在 Gemini 中、在地图中，这使我们能够通过我们的预报触达数十亿人。
I think improving weather forecasts can have big impacts across society. It's a really important challenge
我认为改进天气预报可以在整个社会产生重大影响。这是一个非常重要的挑战。
because so many decisions depends on getting it right.
因为许多决策都取决于能否做出正确的判断。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力