Category: Machine learning
Ground Truth
Ground truth is reference information about real conditions on the ground used to train or validate models.
Also known as: reference data, labels
Expanded definition
Ground truth can come from field surveys, farm records, authoritative maps, or high-quality manual labeling.
Ground truth is rarely perfect. Timing mismatches, labeling errors, and differences in definitions can dominate model performance.
When citing a model result, the quality and definition of ground truth is often more important than the model architecture.
Related terms
Training Data
Training data is the labeled examples used to fit a machine learning model.
Validation
Validation tests model performance on data not used for training, to estimate how well it will generalize.
Change Detection
Change detection identifies meaningful differences between dates, such as harvest, flooding, deforestation, or construction.