I learned that Log Pearson Type III was the favored method based on its ability to calculate and adjust to a skewed distribution, often common with flood peak data.  I am not a statistician and have little recollection to applying the other methods.

In looking over the raw data, I guess that is why I decided to comment, as there were abnormalities in gage height and flow rate for the GARUDESHWAR site (inflow) that suggests relooking at the stage discharge relationship and explain if it has had major adjustments over time, for example 1997 and 2006, about same stage, yet major differences in flow estimates.  I assume the flow in Bharuch is below the dam, but I was concerned in 2008 why you did not have a stage for the upstream site to go with the flow data for inflow, even though I understand that there was no downstream releases, which downstream peak releases except in 2008 was presented only as a stage value.  Most dams have ways to control water levels and outflows and storage capacity can be modified by lowering lake before storm or retaining water during storms to increase lake levels.  Initial dam storage can also be variable, and you might want to report on this aspect of dam management to modify downstream releases.  It might be useful to identify instances of emergency channel or dam overflow releases, if your concern was also related to downstream flooding from lake releases.  

I would also order the floods from low to high, and plot them on a log or log-log scale to see how the curve looks relative to years or anticipated frequency probability.  I do not recognize the flow units (LACS), but you may have some that are more meaningful to you.  Since you are not comparing multiple sites, I would use cubic meters per second or cubic feet per second instead of area weighted flow values such as cubic meters per second per square kilometer or acre inches per hour as examples.  When in doubt, International Units are best, but if you have local units, then continue to use them for better understanding an reporting.

If your data is normally distributed, perhaps some of the other methods could be considered, and this would best be discussed with a statistician, which I am not.  I have had some concerns that the standard statistical methods may not be completely reliable with evaluating peak flow data from short term data sets common to flow data.  Since we often want to address vary rare events, even 50 years of data may or may not represent these extreme rare events.  So I would consider presenting the confidence limits, and perhaps these alternative values to suggest that these estimates with variability presented and alternative methods are apt to contain the flood frequency presented, but there remains some statistical uncertainty, even though plotting the data has a supports a high r squared correlation value.

Also helpful to me is understanding the storms behind the floods, and with the weather channel and doppler imagery of today, that is becoming more possible.  Are your major floods associated to some specific type of rare events, such as Typhoons, etc.?  The severity of these types events are difficult to predict and depend alot on location relative to the storm.

The below link provided number of practical methods:

https://www.researchgate.net/post/What_kind_of_software_provide_flood_frequency_analysis

You can use Kolmogorov Smirnov, Anderson Darling or Chi Square Test to find the most suitable distribution to fit your flow data!! You can use EasyFit software to carryout these test to find the Goodness of Fit!!

http://www.mathwave.com/easyfit-distribution-fitting.html

Thank you Mr William. That was really helpful. 

Thank u Mr Manikandan 

Thank You Mr Raoof. 

Based on the available 60 years data of flood, Is it possible to built some kind of model which can predict the future floods with certainty? how to build that?

Can we plot future floods with autocorrelation?

Get a copy of the foloowing article, it will help. The pare is old, yet still current.

Manuel A Benson

Uniform Flood Frequency Estimation Methods for Federal Agencies 

Water Resources Research Oct 1968

MIKE FLOOD

MIKE FLOOD is a complete toolbox for flood modelling with flexibility in model couplings. It allows you to simulate virtually any flood problem in rivers, floodplains, urban and coastal areas.

Dear Umer

You have an annual maximum series of flows i.e. one that provides only the flow value per year. Extreme value theory shows that annual maximum series are typically well approximated by a Generalised Extreme Value (GEV) distribution whatever the distribution of the underlying processes as long as the process is stationary with many independent events (see Coles, S. 2001. An introduction to statistical modeling of extreme values. Springer, London). The GEV distribution provides a solid theoretical basis on which to base extrapolations of return periods beyond the timespan of the data.

There are numerous software packages that can be used to fit a GEV distribution to your data (e.g. R see https://www.r-project.org/).

William above makes some very useful points that you should consider. However, I think that it is unlikely that your time series of flow data will be normally distributed.

I hope that this is of help.

Best regards

Darren