Melckykaisha/Weather_Forcust_Kenya_DL_Models
Domain:
climateRecord type:
model
Creator:
Mel
Host:
Deep learning system for hyper-local weather forecasting across all 47 Kenya counties. Trains and compares LSTM, GRU & ConvLSTM models on 10 years of historical meteorological data. Interactive 7-day forecast map built with Streamlit & Folium.
# 🌦️ Kenya Local Weather Forecasting System
A deep learning-based weather forecasting system that predicts local weather conditions across all **47 counties in Kenya** using historical meteorological data. Built with LSTM, GRU, and ConvLSTM neural networks and deployed as an interactive Streamlit web application.
---
## 📌 Overview
Kenya's diverse geography — from coastal plains to central highlands and arid semi-arid lands (ASALs) — creates complex microclimates that are poorly served by broad national forecasts. This system addresses that gap by training deep learning models on 10 years of historical weather data to generate **county-level, 7-day forecasts** for temperature, rainfall, humidity, wind speed, and atmospheric pressure.
---
## 🚀 Live Demo
> **Launch App →**
---
## 📸 App Pages
| Page | Description |
|------|-------------|
| 🗺️ **Kenya Map** | Interactive map with all 47 county markers. Click any county for a 7-day forecast popup |
| 📍 **County Forecast** | Select any county to view detailed daily predictions with trend charts |
| 📊 **Model Comparison** | Side-by-side evaluation of LSTM, GRU, and ConvLSTM using MAE, RMSE, and R² |
| 📈 **Predicted vs Actual** | Scatter and time-series plots validating model accuracy against real observations |
---
## 🧠 Models
Three deep learning architectures were trained and compared:
| Model | Description |
|-------|-------------|
| **LSTM** | Long Short-Term Memory — captures long-range temporal dependencies |
| **GRU** | Gated Recurrent Unit — faster, lighter alternative to LSTM |
| **ConvLSTM** | Convolutional LSTM — combines spatial feature extraction with temporal modeling |
**Best performing model: GRU** (lowest RMSE across all variables)
---
## 📊 Predicted Variables
| Variable | Unit |
|----------|------|
| Mean Temperature | °C |
| Rainfall | mm |
| Relative Humidity | % |
| Wind Speed | km/h |
| Atmospheric Pressure | hPa |
- **Input window:** 30 days of historical data
- **Forecast horizon:** …