Many current models have lots of detail. If we tried to display all the detail in the design database of a Boeing 777, or a detailed procedurally generated landscape, we could hardly expect interactive display rates. More recently, people have automatically created models based on scanned range data, and these data sets can be very detailed.
It would be handy if we could display less detailed versions, at least when the object is far away or moves very fast. Also, when reading a geometric model from the web it would be nice if the browser displayed a low resolution version while reading rest of the model.
Wavelet analysis provides a way to automatically create versions at different resolutions from detailed original models.