🔥 Major Improvement (Nov 2025):
Added a lightning-based weight map into the loss function, resulting in significantly improved detection of localized convective showers.

Dataset:
Training and validation dataset is at https://osf.io/ehwmv/files/osfstorage
Train DL-Precipitation/Lightning Forecasting model (check paths for dataset):
python train.py --epoch ${epoch}
Demo inference:
https://colab.research.google.com/github/hunter3789/Deep-Learning-QPF/blob/main/sample_inference.ipynb
Model architecture (updated!!!):

Input file description:
[channels, height, width] = [80, 720, 576]
Resolution: 2km x 2km
Variables are as follows (altitude (topo) is not included in individual files) :
| Number | Variable | Level (hPa) |
|---|---|---|
| 1 | U-component wind (U) | 1000 |
| 2 | U-component wind (U) | 950 |
| 3 | U-component wind (U) | 925 |
| 4 | U-component wind (U) | 900 |
| 5 | U-component wind (U) | 850 |
| 6 | U-component wind (U) | 800 |
| 7 | U-component wind (U) | 700 |
| 8 | U-component wind (U) | 600 |
| 9 | U-component wind (U) | 500 |
| 10 | U-component wind (U) | 400 |
| 11 | U-component wind (U) | 300 |
| 12 | U-component wind (U) | 250 |
| 13 | U-component wind (U) | 200 |
| 14 | U-component wind (U) | 150 |
| 15 | U-component wind (U) | 100 |
| 16 | V-component wind (V) | 1000 |
| 17 | V-component wind (V) | 950 |
| 18 | V-component wind (V) | 925 |
| 19 | V-component wind (V) | 900 |
| 20 | V-component wind (V) | 850 |
| 21 | V-component wind (V) | 800 |
| 22 | V-component wind (V) | 700 |
| 23 | V-component wind (V) | 600 |
| 24 | V-component wind (V) | 500 |
| 25 | V-component wind (V) | 400 |
| 26 | V-component wind (V) | 300 |
| 27 | V-component wind (V) | 250 |
| 28 | V-component wind (V) | 200 |
| 29 | V-component wind (V) | 150 |
| 30 | V-component wind (V) | 100 |
| 31 | Relative humidity (R) | 1000 |
| 32 | Relative humidity (R) | 950 |
| 33 | Relative humidity (R) | 925 |
| 34 | Relative humidity (R) | 900 |
| 35 | Relative humidity (R) | 850 |
| 36 | Relative humidity (R) | 800 |
| 37 | Relative humidity (R) | 700 |
| 38 | Relative humidity (R) | 600 |
| 39 | Relative humidity (R) | 500 |
| 40 | Relative humidity (R) | 400 |
| 41 | Relative humidity (R) | 300 |
| 42 | Relative humidity (R) | 250 |
| 43 | Relative humidity (R) | 200 |
| 44 | Relative humidity (R) | 150 |
| 45 | Relative humidity (R) | 100 |
| 46 | Temperature (T) | 1000 |
| 47 | Temperature (T) | 950 |
| 48 | Temperature (T) | 925 |
| 49 | Temperature (T) | 900 |
| 50 | Temperature (T) | 850 |
| 51 | Temperature (T) | 800 |
| 52 | Temperature (T) | 700 |
| 53 | Temperature (T) | 600 |
| 54 | Temperature (T) | 500 |
| 55 | Temperature (T) | 400 |
| 56 | Temperature (T) | 300 |
| 57 | Temperature (T) | 250 |
| 58 | Temperature (T) | 200 |
| 59 | Temperature (T) | 150 |
| 60 | Temperature (T) | 100 |
| 61 | Geopotential Height (GH) | 1000 |
| 62 | Geopotential Height (GH) | 950 |
| 63 | Geopotential Height (GH) | 925 |
| 64 | Geopotential Height (GH) | 900 |
| 65 | Geopotential Height (GH) | 850 |
| 66 | Geopotential Height (GH) | 800 |
| 67 | Geopotential Height (GH) | 700 |
| 68 | Geopotential Height (GH) | 600 |
| 69 | Geopotential Height (GH) | 500 |
| 70 | Geopotential Height (GH) | 400 |
| 71 | Geopotential Height (GH) | 300 |
| 72 | Geopotential Height (GH) | 250 |
| 73 | Geopotential Height (GH) | 200 |
| 74 | Geopotential Height (GH) | 150 |
| 75 | Geopotential Height (GH) | 100 |
| 76 | 2m Temperature (2T) | Surface |
| 77 | 2m Dew-point temperature (2D) | Surface |
| 78 | Mean Sea Level pressure (MSL) | Surface |
| 79 | Surface net Solar RaDiation (SSRD) | Surface |
| 80 | Altitude | Surface |
