Forecasting ecological communities: NEON ground beetles case study

[sokole]

There are many approaches that folks use to model populations and communities. Yet, there seems to be a barrier to entry for modelers to create forecasts, as seen with the ground beetles theme for the EFI RCN NEON forecasting challenge. This meeting presents an opportunity for folks interested in forecasting ecological communities to dig into the challenges, information gaps, and barriers to creating forecasts in this space.

What are general barriers to forecasting ecological communities?

• Communities are too abstract?
• Too many metrics to choose from to forecast (e.g., EBVs, 100s of beta diversity metrics, etc.)
• Lag in data collection - e.g., it often takes year or more to ID samples
• Frequency in data collection
• Length of time series datasets that are available
• more?

Discussion of the above could lead to some specific tasks focusing on the ground beetle challenge for the EFI NEON challenge. What are the barriers to entry? Can we make some working examples with the current challenge design? Should we propose updates to the challenge?

Folks interested in this topic would also be interested in #6

[khumbdr]

I like the idea of taking beetle as model organism to do ecological forecasting. I believe NEON has long term observational data to find out what are the barriers for forecasting communities. I believe good group members with continuous discussion; workflow can lead into publication of research paper.

[GlendaWardle]

Great topic. We could consider some of these ideas with comparison to the methods for [Red List of Ecosystems.](https://iucnrle.org/rle-program. The assessment process includes forecasts in change of extent, for example.

The RLE is a global standard for how we assess the conservation status of ecosystems, applicable at local, national, regional and global levels. By monitoring the status of ecosystems, current degradation as well as positive impacts of conservation measures can be recognized.

The Red List of Ecosystems (RLE) evaluates whether ecosystems have:

reached the final stage of degradation (a state of Collapse), whether they are threatened at Critically Endangered, Endangered or Vulnerable levels, or if they are not currently facing significant risk of collapse (Least Concern).

[vihangagunadasa]

What are general barriers to forecasting ecological communities?

I believe from my experience, one additional barrier that goes unnoticed is the presence of missing values in time series data. While we are tempted to simply fill these with zero or with simple measures of central tendencies, effectively tackling missing values through proper imputations can be quite helpful for novel forecasters who are facing this issue.

[mdietze]

@vihangagunadasa while I'd agree that missing data is a universal challenge, I'd recommend using methods that are robust to missing values over imputation / gap-filling (and in particular over single imputation methods). It's dangerous to treat any sort of gap-filled observation as if it were true data.

I also think it is absolutely essential that anyone working with missing data understand the "missing at random" assumption, as analyses where the missing data is not random (e.g., more likely to be missing winter over summer data) can be significantly biased and traditional measures of uncertainty won't give you any indication that these biases are present.

Same thing occurs with spatial missingness, and in spatial problems you're almost always guaranteed to have massive amounts of missingness.

Ideas like these could be part of a broader "best practices" discussion.

[sokole]

Interesting paper recently came out that might be relevant to this topic https://onlinelibrary.wiley.com/doi/10.1111/geb.13670