Wednesday, July 31, 2013

(Almost) a perfect Answer

I read this in a LinkedIn discussion, and I found it the perfect complement of my previous post "Essential for Hydrologists"



I have a question specified in the following and hope some of you can guide me a

There is a lake which is the source of water for an industrial project. The project is known for being water intensive, approximately 20 tons fresh water is needed for production of 1 ton final product. According to satellite images, the lake area has been shrinking for approximately 62% since the project started operation about 5 years ago. Local farmers also complained that they had to dig water as deep as 100 meters in order to get water, indicating a severe drop of ground water level. We would like to know quantitatively how water extraction has caused local water scarcity problem. I wonder whether we need a hydrological model to do such assessment; if a hydrological model is not available for that specific lake, is that possible use time series data to run a regression model where lake area is seen as a function of water extraction, precipitation, temperature, etc.

I wonder how analysis like is done typically, if any of you is aware of any exisiting studies, please do let me know. Many thanks in advance.

Best wishes


Dear Sissi,

Five year data is not adequate for detailed analysis. The reason of decline is due to inadequate rainfall in these five years or increase of draft for different purposes during the five year.  Recently I have done a similar study in one of the lakes in India, though not for the same purpose. I would like to suggest the following points.

1. Study the rainfall / inflow pattern for at least for 50 years

2. Use the Soil and Water Assessment Tool (SWAT)and estimate the water balance taking the rainfall and temperature and other data for at least 15 year daily data.

3. If you have actual inflow data, compare the results with the actual inflow data.

4. If you do not have actual inflow data, asses the inflow by SCS-CN Method (Soil Conservation services (SCS), HEC_HMS  and compare the data with SWAT.

5. Use HEC-RAS and HEC-GEORAS for flood modelling if required to delineate the areas of flood in the lake catchment.


With best wishes,


My comment:

The answer is certainly correct. HEC* and SWAT constitute very well engineered standard models used in operational (and operative) hydrology. SCS-CN method, was already commented. Certainly a term of reference. However, if you look at the physics under the hood, you can recognise the existence of many of those "old-style" parameterisations and simplifications that I tried to avoid in favour of more physically sound choices in GEOtop, more modernity, both in the way to implement the model solutions and the way to do simplifications, in JGrass-NewAGE and  my lectures in hydrology. 
Therefore the answer is perfect if you have to do (possibly is also perfect if you, at the end are an engineer) but I feel uncomfortable with it. Can't we do it better ?  


  1. Dear Sir,

    Many thanks for this blog and the comments. I found it interesting and useful.

    Please, let me introduce you to another hydrological problem. I also have an issue on lake modeling. In this case there are several small terminal lakes. The climate is semi-arid and strong continental with long lasting harsh winters. There is a raising problem of lakes shallowing. The goal is to evaluate the water balance of the lakes and the entire area and to create a hydrological model (for the moment it will be something conceptual and simplified I guess, because of the lack of data and measurements). The only time series that I have for the moment are mean monthly values of precipitation and air temperature (for 30 years) and some general information. In fact the water balance of the lakes is strongly influenced by spring snowmelting. However there is no data on any inflow estimation.

    How I should proceed in this case? What software/numerical tools should I use? Hope to get some advice. Thank you in advance.

    Best regards,

  2. Dear Mira,

    with so a few data is very difficult. I believe that more than those data you should be able to get a Digital Elevation Model of the area from which working out the catchment of the lake, for instance using uDig (see other posts about). For the operation you will need to single out all the catchment flowing into the lake, since we do not have (yet) a Watershed model that select and treat the water bodies like lakes properly.

    If the area is not too small you can probably access also to some satellite data that can give a further rough information about precipitation and other meteorological variables.

    So at this point you will have mean monthly precipitations, and temperature (I did not understand if you had it daily or monthly, and the DEM with all the catchments delineated.

    The zero approximation will be to estimate the water budget for each catchment. You have the total water inputs (the rainfall). The first step would be to estimate the snow part of precipitation. You can do it by using the temperatures (and a lapse rate suitable for your region). The second step would be to try to estimate the potential evapotranspiration. With the help of the DEM and a solar radiation model, you can use the Priestley Taylor approximation (but you have to cut the radiation with literature for accounting of transmissivities - see for instance our paper on GMD also referred in this blog). Then you can think, that the liquid precipitation minus the potential evapotranspiration (PET) is a first estimate of water inflow into the lakes. This is obviously a defective approximation, since you would like to use the actual evapotranspiration (AET) instead. To reduce PET to AET you should use some guesses. For what regards snow, you can use a degree-day approach, even if estimating the coefficients will be another great guess. For snow I prefer an approach based again on radiation (see our JGrass-NewAGE model) but this, anyway face the fact that you have to estimate some radiation reduction (not different from the one of estimating ET). These my 2 cents.

    For a rough, large scale estimate, you could refer to global data as derived from satellite, as proposed by Eric Wood and coworkers.

    Hope this helps,


  3. A more meditated comment:

    - Use weather generators for generating meteo-forcing data with the statistics you can get from your data.

    - Then use a model that work at daily scale (e.g. SWAT) or a model that works at hourly scale (e.g. like JGrass-NewAGE)

    - Perform many simulations with different weather generations (i.e. Montecarlo simulations series) and assess in this way the uncertainty in your forecast.