A Strategy for the Approximation of Large Scattered Data Sets