WeatherWave: A Machine Learning-Integrated Web Application for Localized Weather Forecasting in Nepal

Abstract

Accurate and accessible weather information is
critical for decision-making in regions characterized by complex
terrain and climatic variability. This paper presents Weather
Wave, a machine learning–enhanced web application for localized
weather forecasting across Nepal. The system integrates real
time meteorological data from external weather APIs with a
Random Forest regression model trained on NASA POWER
satellite-derived reanalysis data covering all 77 administrative
districts of Nepal. Experimental evaluation on a held-out test
set of 24,187 samples shows that the proposed Random Forest
model achieves a Mean Absolute Error (MAE) of 0.424◦C,
RMSE of 0.695◦C, and an R2 score of 0.9934. This high
predictive performance is attributed in part to strong short-term
temperature persistence characteristic of Nepal’s continental
climate zones. Comparative and statistical analyses demonstrate
that the model significantly outperforms standard API-based
persistence forecasting across districts. The results highlight the
feasibility of combining lightweight machine learning models
with modern web architectures to deliver accurate, localized,
and accessible weather information for geographically diverse
regions.