Regardless of what the scene in front of your camera is, its light meter assumes (as it must) that the average brightness of the scene in front of you is medium grey. A nighttime scene is, however, very, very far from medium grey. It is predominantly dark: it contains large areas of extreme darkness and a small amount of bright highlights. Your camera, however, will come up with settings that will reproduce this scene as one that is medium grey on average. Your image becomes too bright: highlights become overexposed and rich blacks and deep shadows become murky dark greys.
This is where exposure compensation comes in handy. Experiment with various values and find out how many stops less exposure you need in order to create a photo which you feel to be a psychologically “valid” representation of darkness.
The original value that the metering system suggested was way too bright, so I started by “telling” the camera to underexpose the scene by 2 stops. (Exposure compensation of -2.) I gradually went down to -5 stops.
And here is a set of crops that shows an enlarged area of the original picture.
Observe what happens to the highlights and what happens to the shadows. From another perspective what you see is an example of the limited dynamic range of digital cameras. Either your highlights are blown out or you lose precious information in the shadows. Looking at these images: when blacks are finally blacks and the lights on the buildings actually contain plenty of colour, the colour reflection on the water is almost lost. But when the reflections on the water are sufficiently bright and saturated highlights become white hole without any information whatsoever. A tough choice.
And finally, another set of images – just see how the metering of the camera got it all wrong and how exposure compensation saves the day!