All models are wrong: some models are useful

[Dr. Sudhanshu Kumar]

All models are wrong: some models are useful
– Noel de Nevers in his book ‘Air Pollution Control Engineering’.

It's true whether you believe in it or not. Ideally, the perfect air pollutant concentration model would allow us to predict the concentrations that would result from any specified set of pollutant emissions, for any specified meteorological conditions, at any location, for any time period, with total confidence in our prediction.

However, the best currently available models are far from this ideal. In fact, all these models are simplifications of a reality, leading to my belief too that all atmospheric models are wrong but some of them are useful for the qualitative study only.

Share your thoughts.

[Dr. Sudhanshu Kumar]

Dear Dr Naik,
Many thanks for your thought and advice; but your opening and closing sentences
are contradicting each other. Anyway, the citation may sounds as a general
statement for ‘model’ however, my post was categorically for the atmospheric
dispersion modelling.

Honestly, I did agree with the line in that book and shared with the members;
now it’s up to an individual to agree with the quote or not – you shouldn't be
surprised, I am not a genius.

I am just a meteorologist, doing the dispersion modelling stuff for quite some
time and have little experience of PBL profiling studies, model development and
the model evaluation works. Unfortunately, I never find a model during this
period which is beyond assumptions and opportunity to utilize ‘best & perfect
meteorological data’.

As a meteorologist, the only thing I understand – dispersion primarily depends
on the atmospheric variables and these variables are extremely dynamic and
sensitive and no way, these atmospheric dispersion models are designed to take
care of these variables as absolute. Therefore, atmospheric models are just
good enough for qualitative assessments.

It was just my thought may be wrong, but I will be quite happy to learn from
your experiences.

Regards,
Sudhanshu  

[Dr. Hemalkumar Naik]

Dear Dr. Sudhanshu, I am really surprised after seeing this post. Perhaps you have interpretated the term model as Forecast or prediction for perfect condition/matter, which is not at all possible in this world. Only god can reply what can be there in future. Well, models are based on standard formula for various calculation giving suitable probability and exact result. Further, all modeling software as you said for Air quality/dispersion, need skill like use of best & perfect meteorological data, climatic data, topography, landuse, emissions, stack height /internal dia at top and exit gas temerature & velocity, building down wash etc. The model works even with use of only two three data like emission, stack details and basic meteorology. When you run model with these basic data, your outcome is absolutely wrong as you said. this is not fault of modeling software, its fault of operator who uses minimum data. Further, Maximum people uses mixing height as given by IMD/CPCB which is years old and common for wast geographical area. If you will noticed in the IMD/CPCB data of Mixing height for many zone, one Mix Height zone will include coastal area as well as Hill area. Now here is the serious and biggest fault in input for modeling. This will give 100% wrong outcome for incremental GLC and predicted air quality. well its a big matter to discuss and its not possible to explian over here by typing few words. Further, the variable like exit gas temeprature & velocity will also affect the quality of outcome. So there are lot of issue to consider while doing modeling.

But, I would like to convey that models are predictive tolld requires numbers of prefect & correct &  acurate input data. If, maximum input data (with hghest quality) used for modeling, one can get more accurate idea/prediction of modeling outcome parameter like GLC/Tide/Sedimentation etc. Dependibility on secondary data will reduce the reliability & quality of the outcome. and So its all about the fault of Operator/modeller.

Try once using all prefect & actual data, use all input data and try to evaluate the outcome quality. You will have satisfaction with the outcome.

[RAGHAVA]

Sorry Sir,

 

I contradict this statement. All models are wrong i.e all technologies are wrong. if that is the case then we should not be having electricity, telephone, mobiles, televisions, aeroplanes, computers, calculator etc. But we are having all of these and we are in a phase that we cannot exist with atlease one of the things cited above. Modeling is also a scientific invention were it is constantly being updated to mimimize the errors as much as possible.

 

Most of the time it becomes a contradiction in air modeling because many of those modelers are running the model without understanding the basic subject knowledge i.e the dynamics of atmospheric physics. how many people are verifying their predictions. most of the time the prediction are carried out for a worst case scenario rather than for an actual existing situation. Most probably this is the reason why the people are thinking that the models are wrong.

[Vivek Mariappan]

i agree to a certain extent. i was able to get a a bit more than 3% error on my CALPUFF model for a refinery setup. it definitely depends on the capability of the model and the input parameters and its quality.

[Vinodh Valluri]

Human intellect has four fallibilities: it is subject to illusion, its tools -the senses -are limited severely in their reach, it has a propensity to cheat others, and of course, it is only human to commit mistakes. No wonder, no model will ever be perfect, or even 'right' enough. The irony is that science proceeds with haughty facades even after it has been laid out clearly how ethical mores can guide better living than scientific formulas. Without the guiding ethics, the models are absolutely of NO use.