Thunderstorms Prediction In Coastal and Urban areas

How are Thunderstorms Formed 

and whats total x  ?

totalx – Could be a placeholder or derived variable
u10 – 10-meter U-component of wind (east-west wind)
v10 – 10-meter V-component of wind (north-south wind)
d2m – 2-meter dew point temperature
t2m – 2-meter air temperature
blh – Boundary layer height
cape – Convective available potential energy
cin – Convective inhibition
cp – Convective precipitation
kx – K-index (convective instability index)
slhf – Surface latent heat flux
ssr – Surface solar radiation
sp – Surface pressure
sshf – Surface sensible heat flux
ssrdc – Surface solar radiation downwards (clear sky)
ssrd – Surface solar radiation downwards
tcc – Total cloud cover
tp – Total precipitation
swvl1 – Volumetric soil water layer 1 (top layer soil moisture)
p84.162 – Vertical integral of divergence of moisture flux

add units

why are we doing on march april and may (premonsoon)

acquire data from era5 and whats era 5 

lat long to gps conversion 

connect to wandb 

how does lstm work better  ?

whats reanalysis

grib cdf and hdf

use cds api for live prediction 

visual crossing api for verfication 

and maybe mosdac and meterology reports

what is era5 and ecmwf how it gets data by satellites  ?

order data using cds hourly data with variables (days only when it rained i guess) select variables and months , days with lat and long 

select net cdf or grib 

convert grib to csv 

how to extract featyues and how to forecast ?

using a bash script to compile grib2 

ls

compile eccodes

calc tti using formulae

Introduction:

Study and prediction of thunderstorms in coastal and urban areas , here for the study we have taken three areas vizag as coastal , tirupati as hilly area and vijayawada as plains area . 

data is collected from ecmwf era5 (satellite data) from the areas and many parameters were collected with extract the data next 

we have computed a parameter total x which when crosses its threshold value we can say that thunderstorms will happen in that certain areas 

Case study & History of location 

premonsoon & about thunderstorms and clouds formation for thunderstorm and causalities of thunderstorms and loss & collateral loss & identification and mitigation of loss both human and collateral

Motivation 

we want model to be inferenced even on edge devices and low computational devices and finetunable model

Research 

Data collection and Extraction

data is acquired from ecmwf copernicus era5 reanalysis data for the particular locations and this data has so many environmental variables over the years 2017 - 2022

this data is crucial in order to train a machine learning model which can be able to predict a thunderstorm

Methodology 

first we will get total x parameter by which we can predict thunderstorms if its threshold value crosses 

now we correlate all the variables with targer variable and utilize only variables which are related to the target 

Modelling & Optimization

we can take many models but we decided to utilize basic neural network which will identify underlying patterns and predict if it thunderstorms or not 

lstm we can use this model because it can identify sequences and helpful for predictions 

Results

the early results were quite impressive as the model was basic and simple we were able to acquire predictions with very less loss 

we can acquire live data or use cds api for dataset for finetuning model 

References