Brief

Researchers have developed a new method to evaluate Earth embedding models by assessing their complementarity, which measures the performance gain achieved by fusing multiple embeddings. This approach contrasts with traditional methods that evaluate models in isolation. The study found that fused embeddings outperformed single models in four out of six tested downstream tasks, indicating that isolated evaluations often underestimate the full potential of these models. Complementarity was observed to be dependent on the specific task and geographic location, and for one task, it was influenced by the spatial scale of land cover classes. AI

Researchers have developed a new method for mapping tomato cropping systems in California using Google DeepMind's AlphaEarth geospatial embeddings and a deep learning U-Net model. This approach eliminates the need for manual feature engineering, which was common in previous remote-sensing workflows. The model achieved high accuracy, with over 99% in pixel accuracy, precision, recall, and F1 score on an independent test set. Uncertainty maps generated by the model were highest near field edges, indicating reliable predictions within field interiors. AI