Billions in Monitoring, Forecasts Still Fail

We cannot produce reliable ecological forecasts. Despite billions invested in environmental monitoring (NEON, LTER, satellite remote sensing), the field of ecology lacks the forecast skill needed to anticipate and mitigate widespread ecosystem change during the biodiversity and climate crises. There has been no cross-ecosystem synthesis of near-term ecological forecasts, making it impossible to assess ecological predictability systematically — we do not even know which ecological variables are forecastable, at what timescales, and in which ecosystems. As global change pushes ecosystems beyond historical conditions, empirical models trained on past data become unreliable precisely when they are needed most.

Ecosystem services — water purification, pollination, carbon sequestration, flood buffering, fisheries — underpin trillions of dollars of economic activity and are declining globally. The IPBES Global Assessment found that 1 million species face extinction. Lake managers need algal bloom forecasts to protect drinking water. Fire managers need fuel load predictions. Fisheries managers need population projections. Conservation planners need habitat suitability predictions under future climates. All of these decisions are currently made with inadequate predictive tools, leading to reactive rather than proactive management — by the time a problem is observed, intervention options are limited and expensive.

Historical patterns cannot be relied upon as ecosystems are pushed into novel conditions that have no historical precedent — model transferability to unprecedented environments is unproven. Strict assumptions about system dynamics (stationarity, equilibrium, linearity) are difficult to justify but underlie most ecological models. Biophysical process models for complex phenomena (harmful algal blooms, species invasions, disease outbreaks) demand data accuracy, intricate parameterization, and initial/boundary conditions that are rarely available. Most prediction projects make only limited variables publicly available, preventing cross-ecosystem comparison. Monitoring all Essential Biodiversity Variables requires aggregating data from field observations, remote sensing, satellites, NEON, LTER, and citizen science platforms, but these data sources use different protocols, formats, and spatial/temporal resolutions, making integration a persistent bottleneck.

Near-term iterative ecological forecasting with uncertainty quantification, applied across multiple ecosystem types — treating ecology like weather forecasting, with regular forecast-verification cycles that improve skill over time. Nonparametric data-centric methods (empirical dynamic modeling, attractor reconstruction) that embrace ecosystem complexity without oversimplifying. Operationalized Essential Biodiversity Variables built from integrated multi-source data. Cross-ecosystem synthesis frameworks comparing forecast skill across scales. Using NEON's continental-scale standardized data for model development and validation.

A student team could build a near-term ecological forecast for a specific NEON site (e.g., predicting water temperature, chlorophyll-a, or phenology metrics 1–4 weeks ahead) using publicly available NEON data, implementing a simple forecast-verification cycle and quantifying forecast skill relative to climatological baselines. The Ecological Forecasting Initiative's NEON Forecast Challenge provides an existing framework and community for this work. Alternatively, a team could compare machine learning vs. process-based model performance for a specific ecological prediction task using standardized datasets. Relevant skills: ecology, data science, machine learning, statistics, environmental monitoring.