1) As far as I know, MODIS reflectance data at 8-day temporal resolution are freely available from NASA. 2) For the higher temporal resolution you need, that is 3-day, you may take advantage that the daily MODIS reflectance data (distributed at Level 2) are already "corrected" from the effects of cirrus clouds, water vapor, aerosols and atmospheric gases, so your task is not difficult. First, you should merge the daily data within each 3-day window frame. Second, you should compare the data acquired in each pixel, and choose the maximum, the median, the mean or the minimum value. It depends on what you are intending to do with these data. Anyway, you should include this information (maximum / median / mean / minimum) in the metadata attached to the generated image, and take it into account at the interpretation stage (and further, if downwards there are other processing phases). If you cannot decide from the beginning what value to choose, you may generate four composite images, and then analyze which is responding best to your problem.

kind interesting, yes i need more high temporal resolution..i'm work in small water in Bali Islands,because located in tropical region cause the intensity of cloud covered is very high...so by merge 3-daily data may be reduce the NaN data..

we know that MODIS daily Level 2 product is not cover he same area everyday, how we solve the problem when we merge the 3-daily data

One thing that you have to considered is that M*D09GA products (Daily Reflectance) are only available for the first 7 MODIS spectral bands. As you are applying MODIS images for an aquatic system I would suggest you to use daily images not a composed image, since aquatic systems have a very quick response to changes in the environment.

However if you need other spectral bands from MODIS, I think you can use SEADAS to process the data.