Dear Muhammad Abrar Faiz ,

Thank you very much. From viewpoint of practical use of the model, your opinion is good. In many cases, we simply use all available data to calibrate the model after remaining some for verification. In this case, the question becomes if the result is good enough or satisfactory. We need to make such a judgement.

Three accurate readings

Hello,

It seems reasonable to calibrate your model with 6 years. I can use different strategy for the calibration and validation. To guide you, here you will find an interesting study from Klemes.

http://www.tandfonline.com/doi/pdf/10.1080/02626668609491024

Best regards,

Thomas

Dear Thomas ,

Thank you very much. Affected by WMO or Klemes or not, we tried to collect 6 years to calibrate our model up to now. Sometime, it is not easy to have a data set longer than 6 years especially in developing countries. We need to be very careful while using the results.

For a continuous hydroloigc model the length of the data series is less significant than whether the forcings are sufficient to drive the model states to multiple realizations of upper and lower bounds on key model parameters (e.g. upper bound on upper zone free water storage). If not, than the "true" value of that parameter is not observable within the calibration period of record - regardless of its length in years. Sorooshian & Gupta's work explored this problem in very thoughtful detail.

Dear Stuart,

Thank you for your good point. A data set covering wider climatic condition will make all parts of the model experience more, and make the model parameters more "observable".

Is Sorooshian & Gupta's work published in following paper.

@article {WRCR:WRCR3286, author = {Sorooshian, Soroosh and Gupta, Vijai Kumar}, title = {Automatic calibration of conceptual rainfall-runoff models: The question of parameter observability and uniqueness}, journal = {Water Resources Research}, volume = {19}, number = {1}, issn = {1944-7973}, url = {http://dx.doi.org/10.1029/WR019i001p00260}, doi = {10.1029/WR019i001p00260}, pages = {260--268}, keywords = {Hydrology}, year = {1983}, }

Dear friend:

If you need to calibrate a continuous model to estimate long runoff periods for applying, for example to water resources management, my suggestion is that you need a period to calibrate that includes at least one or two complete cycles of inter-annual wet and dry periods. For example, in Spain, the wet periods and the dry periods remains aprox. 6-8 years and the calibration period is around 20-30 years. So that, the calibrated model will simulate well most of the situation that you will find in the basin and your system will be exposed to several climate condition.

Dear Alvaro,

Thank you very much. A period of 20-30 years is not a bad idea if we always have enough data. Considering a period of 30 years is used to calculate climatological averages in many countries, it will be a good choice to make the model to experience or to be exposed to different climate condition though I don't know the scientific base for 30-year period.

Regards,

Minjiao

Dear Minjiao,

The point mentioned by Stuart is a good one, i.e. include as much hydrological variability as possible in the calibration. Obviously, the hydrological variability to be included in the calibration will depend on the purpose of modelling (and the related and relevant hydrological conditions), and objective functions should be selected accordingly.

An interesting article about the impact of the number of data points on the performance of hydrological models can be found through Article Impact of limited streamflow data on the efficiency and the ...

Regards,

Martijn

Dear Minjiao: the explanation for considering 20, 30 or XX years will depend to the inter annual climate cycles existing in your basin. If you calibrate the model under several configurations, the validation will be successful and when you are simulating, the model will have the ability of represent properly the reality.

interesting discussion, how would you consider land use change if you go for longer periods?

Surely longer data records eventually produce better future management for land use.

Dear all, Thank you for all your participation. This discussion will soon reach 100 reads. I would like to make a interim summary here.

It looks quite commonly acceptable that the calibration needs to include hydrological variability, in other words to make the model experienced or observable as possible. As Martijn mentioned, the degree of this inclusion depends on the purpose of modelling .

 Alvaro discussed this point from viewpoint of forcing data and its variability.  Sorooshian & Gupta's work introduced by  Stuart showed us a good point how parameter values can be observed in calibration. It is a viewpoint from model side. The conclusions (Alvaro:20-30 years; Thomas and I: >6 year; Stuart: several years) seem to be quite different. An explanation is needed.

Nowaday, most modern hydrological models have many sub-models which mimic many hydrological processes occuring at different time scales. Looking at different process could result in different record length to assure the model accuracy which also could differ from researcher to researcher. When we focus on total performance of a model, attention will be paid to the processes having longer time scales, e.g. baseflow or evapotranspiration, a long record will be expected.

It looks we still do not have a simple yet scientific answer. I really expect more comments to deepen the discussion about this topic.

Regards,

Minjiao

Dear Fredrick Mashingia ,

About land use change, I am not quite sure if you are discuss changed basins or changing basins.

For changed basins, I think we need to treat the basin before and after land use change as two different basins at least for sub-models related to land use. For changing basins, I have no idea. A changing basin will require changes in model structures or model parameters, and will make the calibration difficult.

Regards,

Minjiao

Dear Minjiao,

When assessing the performance of the model (in the calibration and/ or validation), we might not necessarily focus on hydrological processes with a large temporal scale. This will depend on the modelling objective and the importance of a particular process for the model performance (relevant for the modelling objective). When assessing the 'total model performance' in a large, groundwater dominated catchment, probably larger temporal scales are important (and longer time series for calibration) than in a small, flashy catchment. So I am afraid there isn't a single answer to your interesting question.

Regards,

Martijn

interesting discussion and comments, how would you consider land use change if you go for longer periods?

Hello,

A new interesting study concerning this question is :

Zheng, F., Maier, H. R., Wu, W., Dandy, G. C., Gupta, H. V., & Zhang, T. (2018). On lack of robustness in hydrological model development due to absence of guidelines for selecting calibration and evaluation data: Demonstration for data-driven models. Water Resources Research, 54. https://doi.org/10.1002/2017WR021470

Best regards,

Thomas

Dear Fredrick Mashingia ,

Cc: Omar Al-Khashman

It looks I did not correctly understand your questions.

> how would you consider land use change if you go for longer periods?

>

When the period becomes longer, we will face changes including land use change.

Personally, I think the calibration process has an implicit assumption: the hydrological processes, e.g. runoff generation, evapotranspiration are stationary. For changing basins, we need to evaluate the effects of the changes, and make a trade off. Also we need to keep in mind the calibration result including some uncertainties caused by these changes.

Regards,

Minjiao

Dear Thomas,

Thank you very much for your information about latest research.

Regards,

Minjiao

Dear,

it is difficult to recommend an optimal length of the calibration period since different subsets of the streamflow series do not provide enough information to enable parameter identification. There is an interesting paper on this topic:

Juston, J., Seibert, J., Johansson, P.-O., 2009. Temporal sampling strategies and uncertainty in calibrating a conceptual hydrological model for a small boreal catchment. Hydrol. Process. 3109, 3093–3109. doi:10.1002/hyp

Consequently, parameter identifiability varies in time:

Wagener, T., McIntyre, N., Lees, M.J., Wheater, H.S., Gupta, H. V., 2003. Towards reduced uncertainty in conceptual rainfall-runoff modelling: dynamic identifiability analysis. Hydrol. Process. 17, 455–476. doi:10.1002/hyp.1135

Therefore, length of the calibration period might be viewed with respect to the parameter identifiability.

Concerning changing catchments, some studies present time/varying parameterisations, such as e.g.

Fenicia, F., Savenije, H.H.G., Avdeeva, Y., 2009. Anomaly in the rainfall-runoff behaviour of the Meuse catchment. Climate, land-use, or land-use management? Hydrol. Earth Syst. Sci. 13, 1727–1737. doi:10.5194/hess-13-1727-2009

Kind regards,

Andrijana