WEATHER
FORECASTING USING
RNN
Presented by:
Name- Anwesha Sur
Roll no.- UG/02/BCA/2023/124
Name- Sayoni Das
Roll no.- UG/02/BCA/2023/121
Name- Diya Sharma
Roll no.- UG/02/BCA/2023/109
Name- Sangita das
Roll no.- UG/02/BCA/2023/115
INTRODUCTION
• Forecasting helps in planning agriculture, travel,
and disaster management
• Traditional models are slow and computationally
heavy
• ML models can learn patterns from past weather data
• RNNs handle sequential data using memory of past
inputs
• LSTM and GRU are advanced RNN types used for
better accuracy
• Can predict temperature, rainfall, humidity, etc.
• Suitable for short-term forecasting
• Faster and cheaper than physics-based simulations
WHAT IS RNN?
•Used in weather, language, and speech predictions
•Same weights are reused for all time steps
•Helps model data over time (like daily temperature)
•Can handle variable-length input (e.g., one week, one month)
•Better than normal neural networks for time-based data
•Can be combined with LSTM or GRU for improved results
RNN ARCHITECTURE
•Input Layer: Takes weather data like temperature, humidity, etc.
•Hidden Layer:
•Has a loop connection to remember previous data
•Stores "memory" of past values (called hidden state)
•Output Layer: Gives the predicted result (like next day's temperature)
•At each time step:
•New input + past hidden state → new hidden state
•New hidden state → output
•Same weights are shared across all time steps (good for sequences)
•Works step-by-step on time-series data (like Day 1 → Day 2 → Day 3...)
DATASET USED ?
•Source: Weather dataset from Kaggle / IMD / NOAA
•Features:
•Temperature
•Humidity
•Wind speed
•Pressure
•Rainfall
•Date/Time
WHY USE RNN FOR WEATHER
FORECASTING?
•Captures sequential dependencies in data.
•Good for time-dependent predictions.
•Can learn patterns like temperature trends, humidity, rainfall, etc.
•Handles historical weather data efficiently.
LSTM IN RNN:
•LSTM = Long Short-Term Memory
•Solves the vanishing gradient problem
•Keeps information for longer periods
•Often used in weather prediction instead of vanilla RNN
LIMITATIONS
• Needs large dataset
• Struggles with long-term accuracy
• Sensitive to data quality
• Requires careful tuning (hyperparameters)
FUTURE SCOPE
• Use of LSTM & GRU
Advanced RNN models like LSTM and GRU give better results by remembering long-
term patterns in weather.
•Combining with CNN
Use CNN (Convolutional Neural Networks) along with RNN to analyze satellite images for
more accurate forecasting.
• Real-time Predictions
Weather data from IoT sensors (temperature, wind, etc.) can be used to give live and
local forecasts.
•Transfer Learning
Use pre-trained remodels that already understand global weather patterns, saving time
and improving sults.
• Regional Weather Models
Train models for specific locations (like Kolkata, Mumbai, etc.) for more accurate local
forecasting.
CONCLUSION
•Weather forecasting is important for daily planning and safety.
•Traditional methods are slow and use physics-based models.
•RNN models give faster and data-driven predictions.
•LSTM/GRU improve accuracy by learning long-term patterns.
•RNNs are useful for short-term, real-time weather forecasting.
THANK YOU