By: Ole Petter Laksforsmo Vindstad, John-André Henden and Rolf Anker Ims // UiT The Arctic University of Norway. Jane Uhd Jepsen // Norwegian Institute for Nature Research
Using models to make predictions is a cornerstone of the scientific method. The accuracy of predictions can be assessed by comparing them to new observations, and when models predict well, we gain confidence that they are good representations of the study system. Prediction is thus essential to the testing of scientific theory.
Predictions can also have great practical value. This is especially true for predictions about the likelihood of future events, i.e., forecasts. The utility of forecasts is often exemplified with meteorology, where the development of accurate short-term weather forecasts has provided society with an indispensable planning tool.
In ecology, the use of forecasts has focused mainly on the response of ecological systems to changes in climate or habitat over timescales of multiple decades. Here, the observations that we need to assess forecast accuracy will only become available in the far future. Hence, the opportunities for model validation are limited. Moreover, there is often a mismatch between the decadal timescale of long-term ecological forecasts and the functioning of management institutions, where planning and actions are typically concerned with periods of days to a few years. For instance, quotas for many harvested populations are set on an annual basis – following annual reproduction cycles – while the abundance of harmful organisms like toxic algae or parasites may fluctuate on even shorter timescales. In such cases, short-term forecasts can be a great aid to planning and decision-making.
In recognition of these issues, focus on near-term forecasting has increased among ecologists. When we make predictions about the near future, new observations needed to assess the accuracy of forecasts will become available within a short time. Thus, we are also rapidly alerted to poor forecasts that may signal inadequate models. This enables efficient learning, where we iteratively make forecasts, confront them with data, update models and make new forecasts. As near-term forecasts have a timescale that is better aligned with management, they can also make ecological science more management-relevant and contribute to the operationalisation of the monitoring programmes that feed the forecast models with data. Hence, near-term forecasting is a suitable paradigm for an age when ecosystems are undergoing rapid and extensive change, demanding continuous update of knowledge and adaptable and responsive management. Near-term forecasting is now being integrated in several of COAT’s modules and we present two examples below.
Forecasting ptarmigan abundance
In the ptarmigan module of COAT, we have employed an ecosystem-based approach to develop a model of willow ptarmigan population dynamics in Finnmark, where our collaborating panel of stakeholders decided that forecasting ability should be a priority. A conceptual food-web model was built to highlight the biotic interactions thought to affect short‐term ptarmigan population dynamics and longer‐term trends.
We tested the model’s ability to forecast near-term population dynamics, iteratively referring to long-term population monitoring data to corroborate accuracy in forecasting ability.
The food-web model accurately predicted near-term population dynamics and trends, returning forecasts that, on average, aligned well with monitoring data. Hence, in 2021 we forecasted and published, for the first time, the upcoming ptarmigan densities, two months prior to the official population surveys. Again, the model’s predictive ability aligned very well with estimates from the official survey data. While the forecasting ability of the model varies from year to year, we expect it to improve as datasets grow and our understanding of what determines good and poor forecasting ability improves.
This work now provides active stakeholders such as the local management authorities in Finnmark (FeFo) a tool to make better science-based assessments about the prospects for this year’s hunt, prior to the official surveys.
Forecasting moth outbreaks and impacts
In the tundra-forest module of COAT, near-term forecasting models for the occurrence and impact of moth outbreaks in the mountain birch forest of Troms and Finnmark are under development. This is in response to a need, expressed by both the public and forest management authorities, to know where and when outbreaks are likely to develop, how long they will last, and what impact they will have on the forest. As for the ptarmigan case, a collaborative process between researchers and stakeholders (County Governor, Director of Forestry, and FeFo) identified forecasting ability as a high priority.
The development and survival of all moth life cycle stages is strongly linked to climatic variables – especially temperature – and forest resilience is also expected to be linked to climate. Hence, our approach is to build forecasting models based on downscaled climate data that capture both topographical climate gradients and temporal weather patterns. Using Bayesian time series models, the gridded data is wed to long-term timeseries of moth abundance, forest damage and defoliation maps (from satellite images), to establish the quantitative relationships between these response variables and climatic variation in both space and time.
The ambition is to use the models to forecast outbreak intensity and risk of forest mortality 1-3 years into the future, thereby facilitating planning and mitigation measures for forest managers and the public.
The work is funded by the Research Council of Norway (RCN projects NORTHERN FOREST and SUSTAIN) and the Fram Centre Terrestrial flagship (supporting grants to projects NORTHERN FOREST and SUSTAIN).
Dietze MC, Fox A, Beck-Johnson LM, Betancourt JL, Hooten MB, Jarnevich CS, Keitt TH, Kenney MA, Laney CM, Larsen LG, Loescher HW, Lunch CK, Pijanowski BC, Randerson JT, Read EK, Tredennick AT, Vargas R, Weathers KC, White EP (2018) Iterative near-term ecological forecasting: Needs, opportunities, and challenges. Proceedings of the National Academy of Sciences 115(7): 1424-1432
Hamel S, Yoccoz NG, Ims RA (2021) How to engage stakeholders in strategic foresight for ecosystem research. Fram Forum 2021 p 22-27,
Henden JA, Ims RA, Yoccoz NG, Asbjørnsen EJ, Stien A, Mellard JP, Tveraa T, Marolla F, Jepsen JU (2020) End-user involvement to improve predictions and management of populations with complex dynamics and multiple drivers. Ecological Applications 30(6): e02120