You can also use SAR images. They are widely used for detecting rivers and flooded areas, and they are available in all weather and illumination conditions.

See for instance: http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=4378540

NDVI is not good enough. You may use

Water Mask = (infra-red) / (green + .0001) * 100

Index values can range from 0 to 255, but river values typically range between 0 to 50

You can also use SAR images. They are widely used for detecting rivers and flooded areas, and they are available in all weather and illumination conditions.

See for instance: http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=4378540

A comparison is made of the initial image-based mapping accuracy of water bodies with values of 14 spectral indices:

http://link.springer.com/article/10.1134/S1875372813030128

I used NDWI for the detection of rivers and it worked quite good.

But you should be careful using the indices, depending on in what kind of area you want to detect rivers! For example, if there are moist areas like swamps etc or think of dry and wet season and so on.

Good Luck!

the area is very densed with forest and the area experiences tropical climate so wet in summer.I also want to know that as we will be working long series of long past temporal data.So may i automatic computational techniques.As stated by Pavel Ukrainski link showes Automated interpretation of water bodies but I could not acces it because its not freely available...suggest some which are freely availble describing automated interpratation of water bodies

I would suggest try NDWI.

If you don't want to go with the models, Then you can try by selecting the single water band from your data set, classify it using un-classify feature of ArcMap into 10 to 20 classes depending upon the data set you are using, and define your range in which water lies. Export your results using spatial analysis.

or Simply

go for eCognition developer, do Object Based Analysis defining Rule set. It will give you good results by correlating other classes.

I have used booth the methods and definitely OBIA gives more accurate results otherwise classifying features using ArcGIS will give you near to real results.

Hi Mustak

In my opinion, depending upon the area, you can simply use unsupervised or supervised classification to determine the extent of the river. Specially unsupervised classification (In erdas) with say 30-40 classes is good enough. Afterwards you can calibrate and merge the classes depending on your experience and using ground control points from that area. It will give you the error matrix and accuracy of the overall classification as well. NDWI will give you only the areas which are covered with water (surface) and in that case if you take satellite image for the lean season the extent of river will be under represented with this model. Because the active river includes the sand part as well. Therefore the choice actually depends upon your necessity.

If the banks are densely populated and the river goes through dense infrastructure built up then NDMI and MNDMI are also some of the options that you have.

Hope this helps you clear out.

Best wishes

Nabajit

I totally agree with Prof. Gloaguen's comment. Moreover, in my opinion, even if the river cover is to be generated using a combination of bands, how it has been proposed properly before, a validation on independent DEM data of the same spatial resolution would be necessary.

Rivers are easily extracted from DEM using Hydrological function in ArcGIS 9.3.

Usng DEM is a good approach. but you should be careful when it is flat.

Then what is the way out when the DEM is flat?

Mr. Akinluyi,

For flat / quasi flat areas, SAR or/and optical images (leading to various indices, e.g., NDWI) should be used, as some of contributors have suggested above.

We are offering an excellent forum for this type of discussions. Why not attending ForestSat 2014? Deadline for abstracts on April 30th, visit http://forestsat2014.com

use ndwi 

its not that much possible, due to depth, Concentration and salinity level in waters may vary their index level...

kindly use supervised classification with band variation

For rivers NDWI will work more better.