This map shows the NEON sites with and their CRPS score relative to the climatology model for the seven day ahead forecast, indicated by color. The size of each bubble shows the percent of models which are more skilled than climatology. Included in each site description is the site name, site type, number of models submitted, and the number of models that are more skilled than climatology. We define a skillful model as having a lower CRPS forecast score than the climatology model.
These plots show the percent of skillful water temperature forecasts over the thirty day forecast horizon. The first plot shows the percent of skillful forecasts over for NEON lake sites and river sites separately.
The second plot shows the percent of skillful forecasts for each NEON lake site.
Forecasts
These plots show the most recently submitted forecast (a single reference_datetime) for which we have at least 10 observations.
Models which did not submit a forecast on the given reference date are not shown.
Leaderboard
Average skill scores of each model across all sites. Scores are shown by reference date and forecast horizon (in days). Scores are averaged across all submissions of the model with a given horizon or a given reference_datetime out of submissions made since the cutoff date, 2023-07-10.
Submission statistics
Between 2023-07-10 and 2023-10-08:
- 55 models have submitted a total of 3778 forecasts to this challenge
Long-term patterns
Here we look at some long term patterns in the median scores across all models. Are some months more predictable than others? Note that logs score penalties for observations that fall outside expected ranges are much higher than CRPS penalties.