[Your short answer is No! it is not true]

It depends on the natural time scales of the phenomenon you are simulating. A general useful prelude is this, your question is NOT a new topic and you can find it in any "experiment design" or "hydrology time series" classic book.

Your calibration period most be as long as possible (of course excluding the data you want to use for validation). Any of those two date sets most be long enough to cover at least one full cycle of phenomenon for you, for example if you model an estuary you should have at least 14 days (one tidal cycle) for validation. Or if you model a reservoir temperature you must have at least 24 hour for validation (the more the better, but you do not have always that luxury, as measurements are expensive and hard to have)

Hope it helps,

Kaveh

there is no thumb rules in truncation of data set. it is based on the model's generalization capability

Dear Naveed Ahmed ,

There is no rule to use the same duration for calibration and validation.

[Your short answer is No! it is not true]

It depends on the natural time scales of the phenomenon you are simulating. A general useful prelude is this, your question is NOT a new topic and you can find it in any "experiment design" or "hydrology time series" classic book.

Your calibration period most be as long as possible (of course excluding the data you want to use for validation). Any of those two date sets most be long enough to cover at least one full cycle of phenomenon for you, for example if you model an estuary you should have at least 14 days (one tidal cycle) for validation. Or if you model a reservoir temperature you must have at least 24 hour for validation (the more the better, but you do not have always that luxury, as measurements are expensive and hard to have)

Hope it helps,

Kaveh

Dear Naveed Ahmed

I think, you have to chose calibration and validation in such a way that the number of wet and dry day would be equal in Calibration period and validation period so that you will find if the model response well in validation period. So this does not mean that it is the same length(same number of years) for calibration and validation

Like most of the other answers here, there is no direct equivalency here. While it's possible validation and Calibration lengths are the same, it's much more likely that calibration time will be measured in multiples for a single validation.

Validation is generally determined to be whether a process will perform the intended task in it's current state. By contrast, Calibration length is generally set to an optimal point the unit will function before failure, balancing in tolerance readings against the drift of the unit from optimal.

So while Validation expects a unit to remain well, calibration expects the unit to slowly drift toward out of tolerance readings. If no component changes, and no item drifts from nominal, ever, there might not be a need to ever validate a second time. If significant components are replaced and/or units are calibrated/adjusted, a second validation becomes necessary whether it's a week later or six decades.

The calibration and validation period should be determined in various way depending on hydrological model applications, time-step simulation, experiences of the modelers, but there is no rule to have equal period of these processes. To my experiences, I used to be recommended to have at least 40% span of total time series period for validation process (for case of my previous study of event based HEC-HMS Modelling).

The period duration of calibration and validation may not be the same

No, it is not necessary to have the same duration for calibration and validation. In my case, I used 60% span of total time series period for calibration and 40% for validation purpose. So there is no any such rule to use the same duration for calibration and validation.

Hi Naveed,

There is no rule that they must have equal lengths.

For example, if I have a data spanning 1970 to 2010, I would divide it into three parts.

My calibration would be 1975 to 1990, 5 years as warm up, then, my validation will be divided into two independent periods to assess the robustness of your calibration parameters. I would choose 1991 to 2000 as validation period one, 2001 to 2010 as validation period two.

If the model is actually a good one, I should still have relatively good performances for the two validation periods. However, if you do not have long span of data, you may use a validation period.

Having two validation periods is a test of model robustness.

Thanks all of you for your valuable comments. @Oluwafemi, your comment has a valid justification.

It is not true

The calibration period must be larger than the vslidation period.

The calibration period approximately 70%of the total period of data

And 30% of data specifies for the verification or validation

Hydrological models are developed for global (common) conditions. Model validation means modifying the model to suit the specific features of a particular catchment. It is done by tuning (adjusting) the model parameters. Normally it is done in two steps. First you may select a period of known input and output. In this step you will match the simulated output with the actual one by changing model parameters. This is called calibration.

Then you must select another period (in depended from the earlier one) with known input and output. You run the model with new input and calibrated model parameters. If your simulated results match with the actual (observed) output, then you know your calibrated parameters are correct. Otherwise you may have to readjust the parameters, may select another verification period and repeat the procedure until you get reasonably good results. This step is called model verification. After completing these two steps your model is validated for your catchment and suitable for future use.

Calibration and verification periods are not necessarily be similar but enough to cover different climatic conditions.

However, if there is a turning point as you mentioned (due to change in climate or catchment) you may skip the period before turning and divide the remaining period in a suitable way (not necessary to be similar) for calibration and verification.

Calibration and validation period need not be same. A simple example being Jack-Knife validation. Each time one year data is kept outside for validation while others are used for calibration. Next time the excluded data is included for calibration while another data is kept outside and this process is continued till all data undergoes the process of inclusion and exclusion. The parameters that ensure minimum error are accepted as the model parameters. So in case is of short length this calibration and validation process is being followed.

It is not necessary to have the same duration for calibration and validation period. However, it is important the calibration period is long enough to capture different meteorological and hydrological conditions (e.g. wet and dry years) in order to get the right model response in all other periods.

See: Article Hydrological modelling of karst catchment using lumped conce...

No, For calibration 2/3 of data

for validation 1/3 of data

Nope, 2/3 for calibration and 1/3 for validation

No, they must not be the same. In my work i collected data for a period of 15 days. I used data for the first 8 days in calibration and data for the remaining 7 days in validation.

Is it not true

The calibration period used in order to build or develop the model, while the validation period used to evaluate the model, so the calibration period should be different from the validation period.

for a hydrology purpose the validation period could be the same or less than calibration period. this is what I have learned from multiple articles. and the reason should be that in calibration period the model starts to understand data and its features, then it uses those understandings to cope with different time period which is called validation time. So if the period is not the same or less than previous period, the simulation goes wrong.

Thanks all of you for valuable comments and explainations.

Not necessary.

Sorry, Frankly speaking, I am slightly embarrassed with such question and answers in hydrological process modeling, in case of any other machine, fine.

Its not necessary

Benali Benzater

Yes, the calibration and validation data period must be the same for the hydrological modeling of a complex watershed, but the watershed should be different, because the hydrological data are not stationary over several periods.

I'm not especially familiar with the literature on watershed hydrology, but "calibration" is normally part of maintaining instrumentation. "Validation" is part of theory-making. So, it appears you are interested in comparison of your computer model (theory) with empirical measurements in the field.

Whatever measuring instrumentation you use should always be maintained on a calibration schedule that depends on the technology involved. Validation schedules should depend on the time scales of the dynamic system you're modeling. Thus, they would be entirely independent. If, however, you're talking about decades-long time scales, most instrument technology that requires regular recalibration at all needs recalibration more often than that!