| ConvLSTM Layer | |
| The model leverages ConvLSTM2D, a combination of convolutional layers and LSTM units, to predict cloud cover: | |
| ConvLSTM2D Layer: Captures spatiotemporal features by applying convolutional operations on the input sequence of images. | |
| Batch Normalization: Normalizes the output of each convolutional layer to improve training speed and stability. | |
| Residual Connections: Introduced in the model to allow for deeper layers without vanishing gradients, ensuring that important features are passed forward. | |
| TimeDistributed Layer: Allows the model to apply 2D convolution to each frame in a sequence independently. | |
| Sigmoid Activation: The final output layer uses the sigmoid activation function to output predicted cloud cover values in the range [0, 1]. | |
| Model Evaluation | |
| SSIM (Structural Similarity Index) | |
| SSIM is a metric used to measure the similarity between two images. It evaluates luminance, contrast, and structure. The closer the SSIM score is to 1, the more similar the two images are. | |
| MSE (Mean Squared Error) | |
| MSE calculates the average squared differences between the predicted and actual values, providing a quantitative measure of the prediction error. Lower MSE indicates better predictions. | |