What do I do if I don't know the answer?

Take a good guess. Use the large image to get some more context, and the help and field guide sections. We will have each image classified several times, so if one image is quite tricky, this will be picked up when we look at the results. As a rule, if you are not convinced that a scene is definately clear, then classify it as cloudy.

Why don't you show me what the proper cloud mask has done for each scene?

We don't want to bias your classification in any way. For this dataset to be used as a proper validation set, we need it to be completely independent.

How can I tell the difference between snow and cloud?

It can be hard to tell the difference between snow and cloud- they are both white-blue in our satellite images! Sometimes you can see the shapes of mountain valleys making it possible to guess that what you are looking at is in fact snow. You also know whereabouts the image is from. For example, if the image is from a country known to have snowy mountains or has a cold climate, that can help in your decision. Finally, using the bigger image to add context will help too. There are further details in the Field Guide.

How good is the automated cloud mask that you use?

We know that the computer cloud mask can sometimes miss cloud, but it also has a tendency to classify too much clear-sky as cloud. Overall, it is between 80-85% accurate, which means out of every 100 pixels, it will get 15-20 pixel classifications wrong. This is still very good for a cloud mask! But we know that in certain regions it is likely to be worse than this (e.g. around coasts, for low warm clouds, over certain land type) and in others it will be much better.

Is this really the best way to check these images for cloud? Isn't there a more technical way of doing it?

It is one of the best ways we have in checking for cloud. Here are some of the other methods that can also be used to evaluate cloud mask performance:

• There is an instrument on another satellite that uses a laser to make measurements of cloud. When our satellite happens to be looking at the same patch of cloud as this laser, we can compare the cloud mask with what the laser is saying about the sky- is it clear or cloudy? But these overlaps only happen in certain places, and not very often.

• We also check the surface temperature measurements by using a load of drifting buoys in the oceans. We compare what the satellite measures with what the buoy thermometer measures. As the presence of cloud affects the surface temperature data, we know that if we get a really bad mismatch in the temperatures, it is very likely to be because some cloud has been missed.

• A small set of images can be visually checked by a cloud expert, marking pixel by pixel which one is cloudy or not. But this very time consuming and a validation exercise like this is not performed regularly, but the dataset it produces is very valuable. What we are asking you to do is similar to this. At CloudCatcher we think everyone is a cloud expert and so by asking many people to inspect the images, we can build up a really good validation set.

How do you select the images that you show us?

We show you images that were taken from March, June and September 2022, right up to January 2023, in order to get a seasonal variation. We then randomly select a number of points across these images. For the first validation set, we are concentrating on land or coastal scenes only.