I suggest download the 1km resolution datasets then you can easily apply one of the data resampling techniques such as nearest neighbourhood algorithm to get a 3km resolution datasets using Envi or Erdas.

Using GIS systems like ArcGIS you can calculate the flow direction and the flow accumulation based on your DEM.

My apologies if I do not understand.  You prefer 3 km resolution over 1 km.  Usually for flow accumulation, we use the most detail we can muster.  In the earliest days in the 1990s, we had a 30m2 pixel, then we moved to 10 m2 and now for coastal delineation we moved to 5 m2 and perhaps less with LiDAR data.  Some of this was the capability of the early computers to handle large areas as I believe you are suggesting.  Yes, larger basin size hydrologic units will take more storage to calculate DEMs, convert to raster base and conduct the flow routing.  My expertise is hydrology and not so much GIS, but you might want to look over our paper in researchgate where we applied LiDAR to low gradient, marine terrace terrain.  There is the paper and a powerpoint with some issues on a few slides in converting it to a pdf document.  The normal ArcMap hydrologic tools use an 8 direction system for flow direction, but there is also another one that can use any direction that some suggest it has helped them.  But we have not found that.  The biggest problem in flow accumulation model is that if it hits flat areas such as floodplains, terraces, or possibly flats when you fill depressions in the landscape, the flow path does not know where to go.  Also with the detail we use, if there is a log, beaverdam, old road crossing that pushed the flow out of the channel, it gets on the floodplain and does not know how to get off until forced back.  So we have been identifying many streams and burning them in, removing culverts or other barriers and it is time consuming.  Don't expect this to be just pressing a button and getting a perfect flow accumulation result, and on a large river basin, that would be a substantial project if you already undertaken.  Having hilly and mountainous terrain helps some, but my question would be if your resolution is 3 km, how do you expect to even pick up the small streams?  The ArcMap tools have explanations and I have seen descriptions on how to use them from the internet.  I am fortunate in getting to work, with some excellent GIS professionals, and I know there are many others out there.

I think I figured out question.  You want to be able to download the finished work at the 3 km level of detail such as an individual tile that is 3 km on a side.  Ok, I messed up and cannot help you.  You need to consult with your area GIS specialists and they would know if information is available.  I would suggest if you find such a product, you do some quality control checking because in our experience, this is alot of work and is unreliable for some types of work if you expect the streams to be properly georeferenced, rather than just a line down a valley of between two slopes.  The Indian Creek Hydrologic Analysis (researchgate about 1998) has the streams ordered for the watershed of about 100 square miles and you will see even without LiDAR how dense streams can be on the landscape (about 10+ miles of stream per square mile) in the mountains and piedmont of SC.  So within your 3 km tile, assuming mountain and piedmont (rolling hill) terrain, you may have 35 miles (55 km) of streams if similar.  LiDAR tends to pick up another couple of stream orders over the old topographic maps which were analyzed by cartographers generally many years ago.

You can download SRTM or ASTER DEM data and for your area of interest and then use terrain processing tool to generate the flow direction and others. ARCGIS can help in this

https://lta.cr.usgs.gov/SRTM

https://asterweb.jpl.nasa.gov/gdem.asp

U can try this Website:

http://earthexplorer.usgs.gov/

u can download from LandSat, ASTER DEM and SRTM DEM......

If you have/ download dem of 1km resolution, you can use a GIS software package to resample to 3km. Then you can generate your own flow direction(fdr) from that. If you have ARC GIS the resample and flow direction module are built in. 

Don't resample fdr, 'cause that doesn't give accurate results.

I usually download DEM from DIVA-GIS ( http://www.diva-gis.org/gdata ). It has the resolution of 30 Sec or 0.00833 degree SRTM data. This allows you to download DEM data by country. You can use the mask tool in Arc-GIS ko crop out the area of your interest.

For Flow direction and Flow accumulation, you can use the in build tool box in Arc-GIS.

Goto: Spatial Analyst >>Hydrology>>Flow duration / Flow accumulation

But if you really want DEM data in 3km you can convert the 30 Sec DEM to 3Km by using Resample tool box.

Goto: Data Management>>Raster>>Resample

Hope this was useful.

Thank you all for your nice answer, but let me explain you more detail about my problem. Actually, I have downloaded DEM, Flow Direction and Flow Accumulation from USGS HydroSHEDS and SRTM with 30 sec (~1km). But my target study area is Mekong river, too large area to apply with 30 sec resolution. That's why I need to change to 0.025 degree or 3km resolution. When I use ArcGIS tool to resample resolution, whole river is disconnected and break into many outlets. I have tried this with original 30 sec from USGS Hydrosheds, SRTM already. the results are the same. I think that I should download Original data with 0.025deg or 3km, but I can't find any data available with this resolution. Can you suggest me other solution?

The delineation of the river is messed up because it was originally created  on 1 km resolution or even finer, to avoid this problem you should try:

(1) resample the DEM to 3 km resolution and repeat the delineation of the river using flow direction and flow accumulation rasters created from the resampled 3 km DEM.

(2) if the # (1) didn't work and still the river is not properly outlined try to extract the original river delineation (i.e. 1km resolution) and impose the extracted outline on the resampled DEM using conditional analysis., but this requires a little bit familiarity with ArcGIS. 

Alan Ghafoor,

I tried your #(1) already, but it still doesn't work. Do you have any document regarding resampling DEM using conditional analysis?

Thank you,

http://global.jaxa.jp/press/2015/05/20150518_daichi.html