Some tiles have no predictions

Thank you very much for sharing the notebook (run_chm_model.ipynb). The model performs great in forests and urban areas!

The input image I am using is a USDA NAIP image with a spatial resolution of 0.6 meters. I have observed that a large portion of the predictions are all (0) in forested areas. In the screenshot, you can see that some tiles have no predictions. Most of the scenes occur on the south side of the mountains. Is this because of shadows affecting the model performance? I appreciate any suggestions to avoid prediction failures.