Network-Based Precipitation Forecasting (40 Nodes)
A Comprehensive Workflow for Thesis Research
Thesis Workflow Overview
1. Data Prep & Network Construction
- Sort 40 Nodes (Lat $\downarrow$, Long $\uparrow$)
- Spearman $\rho$ (1979-2014)
- Adjacency Matrix ($\rho > 0.8$)
- Split Nodes (80% Train, 20% Test)
→
2. Data Stationarization
- Convert Non-Stationary Data
- Remove Trends & Seasonality
- Prepare for LSTM Input
📈➡️📉
→
3. LSTM Model Training
- Train on 80% Nodes
- Utilize 1979-2014 Data
- Learn Temporal Dependencies
🧠⚙️
→
4. Precipitation Forecasting
- Forecast 20% Test Nodes
- Period: 2015-2023
- Input: Test Node Data + Linked Training Node Data
🔮📊
5. Results Visualization & Evaluation
Time Series Overlay
Actual vs. Forecasted (2015-2023)
Predicted vs. Actual
Scatter Plot for Accuracy
RMSE Scores
Error Metric per Test Node
Network-Based Precipitation Forecasting (40 Nodes)
A Comprehensive Workflow for Thesis Research
Thesis Workflow Overview
1. Data Prep & Network Construction
- Sort 40 Nodes (Lat $\downarrow$, Long $\uparrow$)
- Spearman $\rho$ (1979-2014)
- Adjacency Matrix ($\rho > 0.8$)
- Split Nodes (80% Train, 20% Test)
→
2. Data Stationarization
- Convert Non-Stationary Data
- Remove Trends & Seasonality
- Prepare for LSTM Input
📈➡️📉
→
3. LSTM Model Training
- Train on 80% Nodes
- Utilize 1979-2014 Data
- Learn Temporal Dependencies
🧠⚙️
→
4. Precipitation Forecasting
- Forecast 20% Test Nodes
- Period: 2015-2023
- Input: Test Node Data + Linked Training Node Data
🔮📊
5. Results Visualization & Evaluation
Time Series Overlay
Actual vs. Forecasted (2015-2023)
Predicted vs. Actual
Scatter Plot for Accuracy
RMSE Scores
Error Metric per Test Node