Smeet KatariaRoll No. 16014225042

Tamil Nadu & Puducherry Weather Atlas

A full year of gridded weather — temperature, humidity, rain, wind, cloud and pressure — cleaned, described and explored with the core toolkit of data science. Every panel responds to the filters below.

Live feed · insight engine

Right now, and what stands out

What stands out in your selection
    Geospatial

    Where it happens

    A smooth surface interpolated (bicubic) from 2,100 grid cells spaced 0.12° apart (≈ 13 km); faint contours mark each tenth of the colour scale. Hover shows the exact value of the nearest cell. The scale is shared across the state so districts stay comparable. Click to filter to a district.
    Time series · smoothing

    Through the year

    View monthly table
    Categorical encoding

    How often it rained

    Share of grid-days in each decoded Condition_encoded class, by month. Faded months are outside the season filter.

    View counts table
    Calendar heatmap

    Every day of 2022

    Directional statistics

    Where the wind comes from

    Distribution · outliers

    Shape of the data

    Correlation

    What moves together

    Regression

    Fitting a line

    Aggregation · ranking

    District league table

    Machine learning · validation

    Modelling lab

    Two models for the selected variable, each judged on data it never saw while being fitted.

    Classification · logistic regression

    Can we tell a rainy day?

    A logistic regression learns to separate rainy district-days from dry ones using the other seven weather variables. It trains on 80% of rows and is scored on the 20% it never saw.

    ROC curve
    Confusion matrix
    What-if simulator drag the sliders

    Unsupervised learning · PCA

    Districts that behave alike

    PCA projection

    Each shape is a district; grey arrows show how each climate feature pulls on the two principal components. Click a district to filter.

    Cluster profiles standardised means (z)
    Choosing k

    Top: inertia, the spread inside clusters, falls as k grows; look for the bend. Bottom: silhouette score; higher means better-separated clusters.

    Anomaly detection

    Days that broke the pattern

    Hypothesis testing

    Is the difference real?

    vs
    Descriptive statistics

    The numbers behind it all

    Exact values over all 766,500 cleaned observations (whole state, full year). Outliers use Tukey's fences: below Q1 − 1.5·IQR or above Q3 + 1.5·IQR.

    Data quality · wrangling

    From raw CSV to clean data

    The cleaning steps applied before any chart was drawn, in the order they ran.