@ Satej, there are two types of weather forecasting models. They are Global models or mesoscale/regional models. Among all the models available so far I think AccuWeather model is the most accurate one.

J. C. Tarafdar Thanks. How can I get the process behind the AccuWeather model? I need to use such a model to generate weather advisory.

You may contact AccuWeather.com

"Numerical Weather Prediction" encompasses a huge range of models (i.e., computer codes) that start with the governing equations of motion (essentially Newton's 2nd law, 1st law of thermodynamics, an equation of state or gas law, conservation of mass or "continuity equation", and an equation to track the transport and phase changes of H2O). Those differential equations (which have no analytic solutions) are converted into simpler, but only approximate, algebraic equations, which do at least have "numerical" solutions. That means they can be integrated forward in time ( to make forecasts, or predictions), by using some kind of finite time-stepping method for a finite number of grid-points (or a finite number of wave amplitudes), at a finite number of vertical levels. All the very many different NWP models are mainly differentiated by the various different "discretizations" they use in time and space, and the different parameterizations they use for the sub-gridscale processes that have a significant net effect on the weather despite not being resolved explicitly by the model grid.

Doing this in practice isn't as easy as it sounds. There are very many ways to do it wrong, and only very few ways to do it right. See e.g., https://maths.ucd.ie/~plynch/Publications/ENIAC-BAMS-08.pdf for a description of how the very first "numerical weather predictions" were made, using just one single governing equation at one vertical level on a very coarse grid.

Nowadays, there is a widely-accepted open-source "standard" NWP system called WRF (for "Weather Research and Forecasting"), which as these models go is relatively easy to obtain, build, configure, and run. Check it out from the NCAR web-site at https://www.mmm.ucar.edu/weather-research-and-forecasting-model .

For your purposes, it sounds as though a sensible approach would be to configure WRF at relatively high-resolution over your "local" or "regional" domain of interest, and provide the needed initial and boundary conditions using output from a publicly-available global model such as GFS (see https://www.ncdc.noaa.gov/data-access/model-data/model-datasets/global-forcast-system-gfs ).

When you say "apart from NWP", that suggests some kind of statistical model, or maybe making forecasts based on the closest "historical analogues". However, neither of those methods are as accurate or as reliable as NWP. In other words, it is better overall to rely on the laws of physics (as expressed in NWP models) than on the laws of statistics when making weather forecasts.

Enda William O'Brien Thanks for your elucidation. We are generating a web-based application for disseminating weather advisories, dynamic crop advisories and climate risk-based advisories to help farmers make smart decisions for sustainable agriculture. For this, weather forecasting is very important and it has to be at Three different levels viz. district, village and farm. The details of which are given in the following document.

https://www.cse.iitb.ac.in/~pocra/datafromgom/PIPPoCRA.pdf

Just going on the executive summary of the document you provided, it seems that weather forecasting is not a primary or direct mission of your project, though of course it's easy to see how better weather forecasts could advance your project goals.

If you want to provide any "value-added" service beyond what is available from the Indian meteorology service, probably the best approach is to provide high-resolution "downscaling" - again, using WRF or other similar model - with initial and boundary conditions provided by the Indian regional forecast model. That will provide you with a bit more local detail (esp. in mountainous or hilly regions), without changing the large-scale "synoptic" forecast very much. As far as I know, "downscaling" usually refers to high-resolution embedded climate simulations, but the same idea has been used for weather forecasting almost as long as there has been NWP. Climate downscaling projects have been conducted for many different parts of the world - it has probably been done for India, or parts of India, as well. Worth a google search, at least!

Enda William O'Brien Thanks. Yes, weather forecasting is not a primary objective, but Climate Resilient Agriculture practices development is!

In addition to NWP, there is the Artificial Neural Networks (ANN) which simulates the real human neural systems. However, the ANN techniques are not solid as the case in NWP.