Progress on a comprehensive earth system model evaluation framework
Bibliographic record
Abstract
Thorough evaluation of model fidelity is critical to have faith in a model’s capability to simulate surprise; i.e. in the model’s capability to accurately simulate hydrologic behavior under changing conditions. We define three elements that should support such thorough model evaluation. First, test cases with known solution should be used to isolate the implementation of specific processes in a model and to make sure that the model code can reproduce these known solutions. Second, model simulations of individual processes should be compared to observations of these processes to determine the accuracy and appropriateness of the equations used to represent these processes in the model. Third, benchmarks need to be defined that both provide a lower limit to the accuracy we require the model to have and an upper limit to system’s predictability based on the information contained in the input and output observations. We present progress on all three themes covering (1) the development of test cases with known solutions called “laugh tests”; (2) progress in setting up a continental-scale model for process-based model evaluation; (3) critical notes about the still common use of aggregated efficiency criteria for model evaluation; (4) a summary of existing work on the need of lower and upper benchmarks which will inform further benchmarking work. A guiding principle in our work is to make our code available as open-source, so that the community can reproduce and use our work if desired. As such, we explicitly invite the community to share their own thoughts about these topics with us.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.167 | 0.156 |
| Meta-epidemiology (narrow) | 0.003 | 0.001 |
| Meta-epidemiology (broad) | 0.004 | 0.004 |
| Bibliometrics | 0.010 | 0.009 |
| Science and technology studies | 0.002 | 0.004 |
| Scholarly communication | 0.018 | 0.019 |
| Open science | 0.013 | 0.010 |
| Research integrity | 0.004 | 0.007 |
| Insufficient payload (model declined to judge) | 0.008 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".