Testing the validation metrics

Consider the model comparisons in order to check we are identify the optimal model for PVNet:
- [x] Create an image showing the difference between the model and the ground truth and add it to the wandb videos #76
- [x] Calculate the metrics on a crop of GB rather than the full image (as we don't care about the accuracy of the forecast for the edges of the images) #79 #78
- [ ] Consider whether we need additional metrics (e.g. PSNR)
- [ ] Add plots showing the daily variation in prediction accuracy