Can ecological forecasts help us in our fight to bend the curve of biodiversity loss?

Forecasting is fundamental in our quest to stem biodiversity loss, providing actionable information, but many challenges persist.

Climate change and biodiversity loss are intricately linked. Biodiversity has declined globally over recent decades due primarily to a few key stressors like land use change, invasive species, pollution, and over exploitation. And of course, climate change has played a key role and is predicted to increasingly do so as world continues to warm¹.

As we’ve argued a number of times (e.g. for rivers² and for Antarctic ecosystems³), forecasts are imperative to help prepare ecosystems for the uncertain future they face. Without robust forecasts we’re effectively flying blind into the future.

I recently attended the Ecological Forecasting Initiative meeting in Helsinki, which was a fantastic few days. I used this as a chance to reflect on the role of forecasting in helping us to bend the curve of biodiversity loss. Here, I touch on a few key highlights from the conference.

Andy Purvis gave a fantastic plenary talk and emphasised that if we can’t make good ecological forecasts, we can’t meet our global biodiversity goals. I’d agree with this sentiment. Forecasts help to identify whether actions are likely to work or not and help to redefine actions iteratively. Without them, progress would likely be much slower.

But there are major challenges to implementing forecasts. Primarily, we face major data constraints.

Data constraints

These constraints differ depending on the type of model one wants to build, which depends on the goal of the model. I like to class predictive models that are designed to look into the future into two different categories: near-term forecasts and long-term projections. Near-term forecasts typically require time-series data, such as species abundances collected weekly or annually. Importantly, these models learn from the transitions between time points (i.e. population abundance at time t + 1 is a result of the population abundance at time t plus the environmental drivers). Because they are so dependent on the historical data they are better at predicting into the near future (e.g. a few time steps forward in time) rather than further out into the future. However, their prediction horizon (the point at which the predictions become uninformative) depends on a number of factors. So, here, continued investment into monitoring efforts that collect biodiversity information through time is critical. Breaking such time-series is a major loss to science and our understanding of biodiversity change.

By contrast, long-term projections are often built from fundamental mechanistic relationships. These models don’t necessarily require time-series data, but instead often require a lot of detailed natural history information about how species respond to particular environmental conditions. This might include demographic aspects like birth rates, mortality rates of different stages of trees from seedlings through to adults and so on. This is important because if the conditions we’re moving into are going to be different from anything species have perceived in the past data, then models based on correlations will fall down over these time scales. If we build models from fundamental mechanisms that capture the full range of variation in the environment, they can look beyond the historical conditions. Here, continued investments into natural history is critical. Natural historians, however, are a dying breed. It’s hard to fund basic natural history because it doesn’t often result in the headline grabbing news items that funders and universities are interested in. But for forecasting, such information is all too regularly a critical gap.

Of course, these model definitions are both gross generalisations. There are any number of approaches one can use, including many hybrid models that leverage both time-series information and mechanistic information. However, the point being is that robust forecasts or projections require quality data. And we continually lack such data. Hybrid models have offered a solution to this challenge by offering ways to make the most of poor data, such as borrowing strength from common or well-known species to help inform models for those species we have less data for. My experience with these approaches so far have had limited success, however.

We’ve come a long way

Nevertheless, the Helsinki meeting showed me how far we’ve come in leveraging and building statistical tools to deal with the tricky nature of the data we work with. A highlight for me was Gabrielle Koerich’s (a PhD student in my lab) talk on modelling moss distributions in Antarctica. (I wasn’t alone in this sentiment — she won the best student presentation.) She has been able to build robust models that make sense despite the fact mosses have been rarely sampled in Antarctica in very unstructured ways, they’re very sparsely distributed, and Antarctica is simply huge. So there is hope. Advanced statistical approaches are beginning to help us deal with the sparse patchy data that dominates what we have available at our fingertips.

I also loved seeing so much discussion on dealing with uncertainty in models. All too often, we overlook uncertainty in our modelling efforts. This is extremely problematic. Being overly confident in model predictions will not help anyone and will lead to decision makers losing faith. Clearly defined uncertainties are fundamentally important. Tom Johnson, who gave a great philosophical talk, is leading the way in this space with a fantastic paper published recently in Nature⁴ on this topic. A highlight for me were the interesting discussions on how to make the most of uncertainty. After all, this is the place where, if we target it explicitly, we can make the most gains in our understanding of biodiversity change.

Connections between stakeholders and scientists is key for actionable forecasts, but high quality long-term field data is necessary

Nigel Yoccoz shared some fantastic examples from his work at the COAT Observatory in northern Norway, a biodiversity observatory now funded by the Norwegian government. The funders here explicitly required forecasts as part of the observatory and key stakeholders were involved from the outset in deciding on and setting priorities. The involvement of stakeholders has been critical in this observatory as their insights helped to direct which parameters to include in models, some of which were fundamental drivers of ptarmigan populations in contrast to the scientists expectation. These near-term forecasts are being used to inform managers about ptarmigan numbers during the hunting season, helping to adjust their efforts with actionable information. This is a great example of a successful stakeholder-scientist collaboration where both parties benefit from a robust two-way dialogue.

Actionable information in informative ways

“Farmers don’t consume data, farmers consume services.” This quote from Masilin Gudoshava really captured my attention. Forecasts offer so much in this space, helping to provide such information to managers to act on. And cyber infrastructure tools are becoming increasingly useful for sharing forecasts in real time with diverse audiences. But as Mike Dietze put it, we need to build a community of practice in forecasting to help avoid continually building boutique solutions, reinventing the wheel and generally muddling through. Communities of practice can help to standardise approaches, easing our workflows and sharing our work in more informative and actionable forms. Without this, we’re not responding to societal needs as well as we could be.

Of course, forecasts aren’t limited to purely applied pursuits. We argued a couple of years ago⁵ that ecological forecasts can play an important role in advancing our basic understanding of how ecosystems work. I’m a big fan of this way of thinking. If we focus on building forecasts as opposed to more traditional ways of doing science, I’d argue we have the chance to gain a lot more for both basic and applied science.

Summary

We’ve certainly come a long way in forecasting in ecology, but we have a long way to go. There are many exciting examples of successful applications of forecasts but the common theme among the conference is that many of us are starved of high quality data. We must continue to advance our biodiversity monitoring globally and the best time to do that was yesterday (or 100 years ago), but it’s never too late to start. Initiatives like GEO BON (Group on Earth Observations Biodiversity Observatory Network) are doing a lot of good in this space. But we must continue to build such initiatives and, in particular, we absolutely must continue to service long-term data collection programmes as these are priceless to science.

I’m excited to be working in this space with my lab and increasingly look forward to connecting iterative models with cyber infrastructure to share actionable information with the public and decision makers.

Thanks again for reading. Don’t be shy in sharing with your networks outside of Substack. It appears most of my subscribers are coming from within the Substack network. There is so much goodness being produced on this platform from a wide diversity of writers.

1. Pereira, H. M., I. S. Martins, I. M. D. Rosa, H. Kim, P. Leadley, A. Popp, D. P. van Vuuren, G. Hurtt, L. Quoss, A. Arneth, D. Baisero, M. Bakkenes, R. Chaplin-Kramer, L. Chini, M. Di Marco, S. Ferrier, S. Fujimori, C. A. Guerra, M. Harfoot, T. D. Harwood, T. Hasegawa, V. Haverd, P. Havlík, S. Hellweg, J. P. Hilbers, S. L. L. Hill, A. Hirata, A. J. Hoskins, F. Humpenöder, J. H. Janse, W. Jetz, J. A. Johnson, A. Krause, D. Leclère, T. Matsui, J. R. Meijer, C. Merow, M. Obersteiner, H. Ohashi, A. De Palma, B. Poulter, A. Purvis, B. Quesada, C. Rondinini, A. M. Schipper, J. Settele, R. Sharp, E. Stehfest, B. B. N. Strassburg, K. Takahashi, M. V. Talluto, W. Thuiller, N. Titeux, P. Visconti, C. Ware, F. Wolf, and R. Alkemade. 2024. Global trends and scenarios for terrestrial biodiversity and ecosystem services from 1900 to 2050. Science 384:458–465. https://doi.org/10.1126/science.adn3441

2. Tonkin, J. D., N. L. Poff, N. R. Bond, A. Horne, David. M. Merritt, L. V. Reynolds, J. D. Olden, A. Ruhi, and D. A. Lytle. 2019. Prepare river ecosystems for an uncertain future. Nature 570:301–303. https://doi.org/10.1038/d41586-019-01877-1

3. Koerich, G., C. I. Fraser, C. K. Lee, F. J. Morgan, and J. D. Tonkin. 2023. Forecasting the future of life in Antarctica. Trends in Ecology & Evolution 38:24–34. https://doi.org/10.1016/j.tree.2022.07.009

4. Johnson, T. F., A. P. Beckerman, D. Z. Childs, T. J. Webb, K. L. Evans, C. A. Griffiths, P. Capdevila, C. F. Clements, M. Besson, R. D. Gregory, G. H. Thomas, E. Delmas, and R. P. Freckleton. 2024. Revealing uncertainty in the status of biodiversity change. Nature 628:788–794. https://doi.org/10.1038/s41586-024-07236-z

5. Lewis, A. S. L., C. R. Rollinson, A. J. Allyn, J. Ashander, S. Brodie, C. B. Brookson, E. Collins, M. C. Dietze, A. S. Gallinat, N. Juvigny-Khenafou, G. Koren, D. J. McGlinn, H. Moustahfid, J. A. Peters, N. R. Record, C. J. Robbins, J. D. Tonkin, and G. M. Wardle. 2023. The power of forecasts to advance ecological theory. Methods in Ecology and Evolution 14:746–756. https://doi.org/10.1111/2041-210X.13955