As far as I understand there should be two different calibrations in modeling with SWMM. I believe that one is to balance water volumes within the model and another one is to generate a time series that accurately simulate of the real flows in a watershed.

I believe that once the water balance is within acceptable limits, then I should start checking existing data. I also think that the first thing to check are the voulmes of runoff (peaks, baseflow, etc.) to find out if the model is giving a proper representation of the real scenario. Then a temporal analysis should be done to find out if flows (peaks/baseflow) occur at the proper season and finally if the individual flows represent the observed ones.

In between each checked it is probable that we want to check the balance of the water volumes.

My questions are:

1. Is there a method that takes into consideration the temporal variation of the storms? It is not always the situation that you get discharges at the same time during the day when modeling a precipitation which can be due to many factors: time gaps in recording precipitation, distance between rain gages and other factors. I thought about a temporal average of temporal volume of water but would like to know if there are proved methods. I have used Nash Sutcliff but it considers difference between point flows and if the flow is off by a time step then the method gives a bad relation index when in fact the discharge is off by time steps and the model predictions may be acceptable.
2. Is there a procedure for calibrating models that I can follow? or is it just up to the modeler to get the best relation he cosiders?
3. I have tried to divided flows into three cathegories: baseflow, peaks flows and mid flows and set different limts for each one based on observations and sometimes a better correlation can be found and some other it just gets worse. Do you have any ideas or literature that guide through the calibration process of the model vs. real data? I am reluctant to use normal correlation coefficient from statiscs based on the minimum square errors because the time series of rain events are not independent events but they are seasonal and are correlated, therefore I think that moving into the correlation analyses of time series wouls be better.

I was waiting for and hoping someone else would reply before this. Here are a few pointers: There are many publications on this topic. Personally I have had a 25 year interest in SWMM optimization, and recently completed the 3rd edition (300 page) of the booklet "Rules for responsible modeling" which deals with the issues that you raise (and forms the background for the approaches used in PCSWMM2002). Regrettably, it would be hard to summarise the book and program in one email. For instance, some 14 different objective functions, and 20 different evaluation functions are listed (6 are programmed), that should be matched to your study or design objectives, and semi-automatic calibration is achieved using sensitivity-based and uncertainty-based genetic algorithms. The procedure can handle very large SWMM data files.