Fig. 3

The performance of multi-trait selection methods. A The learning workflow of TOP algorithm. It iteratively learned the trait weights, the whole process firstly learned the weights by towards trait prediction accuracy (assessed by correlation between each round of simulated weight and prediction accuracy), then attempted to learning the trait balance status. The process converged until the identification rate goes stable. B The performance of multi-trait selection methods. Three methods are tested, including (i) independent culling levels (red line); (ii) three scenarios of weights on economic value for index selection called Index1, Index2, and Index3 (blue dash line, blue bold dash line, and blue line); and (iii) TOP (orange line)