志愿者开发机器学习工具以识别罕见云层

某些种类的云层表现“异常”——它们出现的频率比以往更高，且高度更低。为了帮助确定影响这些变化（例如地球（Earth）长期天气模式的转变）的因素，科学家们已邀请全球各地的摄影爱好者提交这些云层的新图像，作为美国国家航天局（NASA）支持的 Space Cloud Watch 项目的一部分。现在，一名志愿者开发了一种新工具，以帮助其他 Space Cloud Watch 志愿者更高效地开展工作。

Certain kinds of clouds are misbehaving – appearing more often and lower in the sky than they used to. To help identify the factors influencing these changes (e.g. shifts in Earth’s long-term weather patterns), scientists have asked people around the world with cameras to submit fresh images of these clouds as a part of the NASA-supported Space Cloud Watch project. Now, one volunteer has developed a new tool to help other Space Cloud Watch volunteers work more efficiently.

这些表现异常的云层被称为“夜光云（noctilucent cloud，NLC）”。这些云层在日落后很久和日出前很久仍能散射太阳光，从而呈现出银色的光芒。然而，尽管有这种光芒，要将夜光云与低空外观相似的云层区分开来仍有难度。这种混淆给项目负责人带来了额外的工作量。

The misbehaving clouds are “noctilucent” or “night-shining” clouds (NLCs). These clouds scatter light from the Sun long after sunset and long before sunrise, giving them a silvery glow. But despite this glow, it can be hard to differentiate NLCs from lower-altitude look-alikes. That confusion has meant extra work for project leaders.

志愿者 Namai Chandra 分享道：“我注意到夜光云图像正由项目负责人进行人工核实。我觉得这项任务非常适合采用‘人在回路（human-in-the-loop）’的机器学习流程，它可以自动处理重复性的筛选工作，同时在处理关键图像时保持人类判断的核心地位。”换句话说，Namai 找到了一种方法，可以帮助观测者核实他们所看到的究竟是否为夜光云。

Volunteer Namai Chandra shared, "I noticed that NLC images were being manually verified by the project leaders. It felt like a task well-suited for a human-in-the-loop machine learning pipeline, one that could handle the repetitive screening automatically, while keeping human judgment central for the images that matter most.” In other words, Namai found a way to help observers verify when they are indeed seeing NLCs and when they’re not.

Namai 联系了 Space Cloud Watch 的科学家 Chihoko Cullens 博士和 Brentha Thurairajah 博士，他们对他的想法感到非常高兴。Namai 很快开发了一个机器学习流程，并利用各种云层图像对其进行了训练，其中包括夜光云以及经常被提交给 Space Cloud Watch 的低空外观相似云层。该流程结合了图像预筛选、云层分类以及基于置信度的审查路由。经过多轮开发、测试和改进，他将自己的夜光云识别工具发布到了项目中。该工具目前正被那些不确定自己是否观测到夜光云的云层贡献者，以及希望标记图像以供审查的项目科学家所使用。

Namai reached out to the Space Cloud Watch scientists Drs. Chihoko Cullens and Brentha Thurairajah, who were delighted with his idea. Namai soon developed a machine learning pipeline, training it on a variety of cloud images, including both the NLCs and the lower altitude look-alikes that are often submitted to Space Cloud Watch. The pipeline combines image pre-screening, cloud classification, and confidence-based review routing. After several rounds of development, testing, and refinement, he released his NLC identification tool to the project. This tool is now being used by cloud contributors who are unsure whether they have observed NLCs, as well as project scientists that want to flag images for review.

拿起相机，今天就加入 Space Cloud Watch 项目吧！如果你曾因为不确定自己看到的是否为夜光云而对参与 Space Cloud Watch 犹豫不决，那么现在多亏了 Namai，你在分享之前就有了一种核实的方法。

Grab a camera and join the Space Cloud Watch project today! If you’ve hesitated to contribute to Space Cloud Watch because you were not certain if what you were seeing was a noctilucent cloud, you now have a way to check before you share – thanks to Namai.