Data assimilation is a critical component for short-range weather forecasting; a number of algorithms have been developed and applied for assimilation of different kind of observations. However, an important but less explored question is the (optimal) amount of observation for maximum improvement in forecasts through data assimilation. Because the meteorological fields at different spatial and temporal resolutions are not necessarily mutually independent, indefinite increase in resolution of observations may be ineffective; thus data optimality in this sense can be defined as the maximum resolution of observation beyond which no appreciable improvement occurs due to assimilation of data. Based on forecasts of seven events over a complex terrain (urban location, Delhi) with different combinations of observations, we show that improvement in forecast skill does not saturate even with assimilation of observations a few kilometers (<10 km) apart. The improvement due to assimilation of data from each of the profilers is appreciable; however, the impact was generally the highest for assimilation of data from all the four profilers. Applicable strategies for observation system design over high-impact areas are discussed.