Bias Correction Issues in Limited Area Models: Strategy for HIRLAM