Kottayam, Kerala · 9°32′N 76°36′E · IMD daily data 2005–2019

32.2°C

Dry-bulb temperature · dataset maximum

Estimating daily dew point temperature in a tropical region of India

We compare four machine learning models, Random Forest, XGBoost, Support Vector Regression and Decision Tree Regression, for estimating daily dew point temperature from 15 years of meteorological observations in Kottayam, Kerala.
Paper: Comparison of Different Machine Learning Algorithms for Predicting Dew Point Temperature in the Indian Tropics

5,262days of valid observations (2005–2019)
12 → 5variables, selected by Pearson correlation
124runs: 31 input combinations × 4 models
0.0777 °Clowest RMSE (RF with WBT + DBT + VP)

Error by input combination and model

Each bar is one run. The long axis holds the 31 input combinations, grouped by the number of input variables; the short axis holds the four models. Lower, bluer bars mean lower error. Bar height uses a log scale, because on a linear scale the best several dozen runs would look identical.

WebGL is not available in this browser, so the 3D view cannot be shown.

Drag to rotate · Click a bar for details RMSE low high

RFIncludes VP
Scenario#16
RMSE0.0777
R²0.9985
Rank among 1241

Dew point from the Magnus formula

When a glass is colder than the dew point of the surrounding air, water vapour condenses on its outer surface. Adjust the values below to see it happen. The dew point here is not predicted by a model. It is calculated directly with the Magnus formula, as a physical reference for the machine learning results in the paper.

Dew point Td 23.3°C
Machine learning
Error of the best model in the paper (RF-16)0.0777 °C

Trained on 15 years of data, with WBT, DBT and VP as inputs.

Physics formula
VP extremes from Table 2 through the Magnus formula10.76 / 27.53

The minimum and maximum DPT reported in the paper are 10.7 and 27.5 °C. The formula needs only VP and no training.