I get a lot of questions about how I made this video:
There is obviously more to it than just model output. The video incorporates visual elements such as arrows and wind barbs to illustrate the surface forcing and uses camera motion to reveal different aspects of the circulation. In my opinion, the coolest thing about the video is the transition from the cartoon schematic to the 3D view that happens right at the beginning. This makes a strong impression during a presentation when the audience thinks they are looking at a still figure and then realizes it is a 3D movie!
It was not easy to make this movie. I literally spent weeks on it. I have decided to share my process in hopes that it will be useful to other people. I encourage you to leave comments below!
The basic steps were as follows. First I ran the model to output the necessary data. Then I generated images of the temperature fields in Matlab. I them assembled the 3D view and created the supplementary graphics in Apple Motion, a professional motion graphics program. Finally, I compressed the video with the H.264 codec using Apple Compressor. Below I go over these steps in more detail.
The Model Output
In order to generate a high-quality animation, you need high-resolution data to begin with. In particular, you need to output data from your model with a frequency compatible with the video frame rate you intend to use. Most video formats use either 24, 25, or 30 frames per second (fps). I prefer 24 fps. The goal is to output a data snapshot for each frame; the frequency of output will determine how “fast” time runs in the movie. For the movie above, which is 24 fps, I output a snapshot of the temperature field every two days. This means that it takes 7.6 seconds to show one year of simulation, a reasonable time for this particular simulation. I also decided to output the spinup of the model from rest, rather than simply showing an equilibrium state, since this is by definition more dynamic, and therefore more exciting to watch.
The Matlab Stage
The next step is to read the data into Matlab (or python, or whatever program you use prefer to use to analyze model output) and render figures for each frame. Presumably you know how to make contour or pcolor plots of your data already. The important point here is to recognize that a “movie” is nothing more than a sequence of these plots, saved in image format. At this stage of the processes, where we are not concerned about file size or bandwidth limitations, I choose to generate enormous images.
For this movie, I generated 3 filled-contour plots for each timestep: one of the surface (x/y), a side cross section (y/z), and a front cross section (x/z). I caused the axis to fill the figure, and I made all axes elements (ticks, lines, etc.) invisible. This left just the pure colors of the contour plot completely filling the figure. I output each image as a .png file, a lossless format which is much better than .jpg for this purpose. I think the dimensions were 2000 x 4000 pixels for the surface and side views. Huge! I wrote a script to save these figures using sequential numbering, i.e. top_00001.png, side_00001.png, etc. This makes the next step much easier.
Assembling in Apple Motion
This is the step that is unfamiliar to most scientists. Apple Motion is a professional motion graphics program designed for artists / animators / creative people. (Adobe After Effects is a competing product.) These are basically graphic design programs, like Illustrator or Photoshop, with a timeline, allowing the elements of the design to evolve and change with time. You can also move the virtual camera around your scene to show different views. You manipulate everything visually. To give you a sense of how the program works, here is a screenshot:
Motion is able to interpret sequences of images as movies, which you can then arrange and manipulate in 3D space. Each of the cube faces I generated in Matlab appears as a flat, two-dimensional plane in Motion. I arranged these faces into the shape of a cube and voila, a 3D animation.* In fact, it was very tedious figuring out how to do this. Trial and error was my method for learning how to use the program.
Once I had the basic 3D geometry of the channel figured out, I added the extra elements such as spinning heat-flux arrows and wind barbs to represent the surface forcing. I think these add a lot to the animation. Motion has some cool built-in effects that helped with generating these elements. Again, it took a lot of experimentation (and reading of the manual) to figure out how to make things look how I wanted.
The final step was to figure out the timing and camera movement. This basically comes down to subjective, artistic choices about how I wanted the movie to unfold. The hardest part was deciding what I wanted to happen, and what views were the most interesting. Then I just dragged the camera around to make it happen.
I worked the whole project in 1080p HD format, and I output the movie from Motion as an Apple ProRes 12-bit Quicktime video. This produced 2 GB file for a one-minute movie. The image quality was impeccable, but obviously this is unsuitable for online use. Which leads to the final stage…
I host my videos online with Vimeo. I am a huge fan of this site for many reasons. Vimeo provides very clear guidelines about how to optimally prepare your video for uploading. But even if you don’t plan to use Vimeo, these guidelines are a great reference. They key to it all is the magic H.264 codec, which reduces HD video to manageable bit rates while maintaining high quality. I used Apple Compressor to compress my video, and in the process I also down-scaled it to 720p resolution. The final file was only 47 MB. Quite a reduction!
There are alternatives to Compressor for H264 encoding. The most prominent is ffmpeg, an open-source, all-purpose video toolkit. As with many open-source alternatives, it may require a bit more effort to get things working properly.
I hope this post is useful for someone. I welcome any feedback you might have.