True.  However, I will be updating a 3d image on the basis on the data
in the bitmap and editing the bitmap, converting the results and
updating the 3d image will be much slower than getting the data directly
from the bitmap in the correct format.

More significantly, as the edits to the buffered image are being used to
alter the heightmap, if the image uses 16bit int accuracy, it will apply
that limitation to the float buffer.  This is the problem I'm trying to
avoid in the first place.  If it weren't for this problem I could just
use a 16bit greyscale image for the data and live with the resolution
restraints.

Errors corrected from last try; test suite added.   ;)

public static void main(String[] args)
{
 final int w = 320;
 final int h = 240;

 SampleModel sm = new ComponentSampleModel(
  DataBuffer.TYPE_FLOAT, w, h, 1, w, new int[] {0});
 DataBuffer db = new DataBufferFloat(w * h);
 WritableRaster wr = Raster.createWritableRaster(sm, db, null);
 ColorSpace cs = ColorSpace.getInstance(ColorSpace.CS_GRAY);
 ColorModel cm = new ComponentColorModel(
  cs, false, true, Transparency.OPAQUE, DataBuffer.TYPE_FLOAT);
 final BufferedImage bi = new BufferedImage(cm, wr, true, null);

 Graphics2D g2 = bi.createGraphics();
 g2.setRenderingHint(
  RenderingHints.KEY_ANTIALIASING, RenderingHints.VALUE_ANTIALIAS_ON);
 g2.setColor(Color.BLACK);
 g2.fillRect(0, 0, w, h);
 g2.setColor(Color.WHITE);
 g2.drawLine(0, 0, w, h);
 g2.drawLine(w, 0, 0, h);
 g2.drawOval(w/4, h/4, w/2, h/2);
 g2.dispose();

 Frame f = new Frame("Float Graphics");
 f.addWindowListener(new WindowAdapter() {
  public void windowClosing(WindowEvent we) {
   System.exit(0);
  }
 });
 f.setLayout(new BorderLayout());
 Canvas c = new Canvas() {
  public void paint(Graphics g) {
   g.drawImage(bi, 0, 0, null);
  }
 };
 c.setSize(w, h);
 f.add(c, BorderLayout.CENTER);
 f.pack();
 f.show();
}