Since the inception of the Intergovernmental Panel on Climate Change in 1988, there has been growing scientific consensus that humans have modified the Earth's environment and that human decisions have the potential to mitigate or exacerbate the effects of future global change (Intergovernmental Panel on Climate Change, 2022). In the face of widespread environmental change, society needs information to make sound decisions. Ecological forecasts provide such information about how ecosystems and their services may respond to different environmental conditions before they happen and considering alternative management scenarios. Following the publication of the first IPCC report in 1990 that called for a better understanding of the future effects of climatic change, the number of ecological publications focused on forecasting increased exponentially (Figure 1). source The notion of forecasting ecosystem change in ecology is not new (e.g. Clark et al., 2001; Hodgson, 1932). However, as the number of ecological papers focused on forecasting grew in the context of climatic change, ecologists from across sub-disciplines became increasingly concerned with various aspects of forecasting. Paleoecologists acknowledged the limits of our ability to forecast into non-analog climatic conditions when using forecast horizons suggested by the IPCC (e.g. anticipating conditions 60–80 years into the future; Williams & Jackson, 2007). Applied and theoretical ecologists questioned the utility of long-term forecasts for advancing decision-making and ecological theory, respectively (Houlahan et al., 2017; Mouquet et al., 2015). At the same time as these concerns, ecologists recognized the transformative potential of near-term, iterative forecasting for making ecological forecasts more relevant to decision makers and spurring innovations in ecological theory if key technical and conceptual advances could be made (Dietze et al., 2018). In 2018, the Ecological Forecasting Initiative (EFI; https://ecoforecast.org/) launched with an emphasis on creating near-term, iterative forecasts with the hopes of using outputs to inform theory and to better align forecasts with management timeframes. EFI is a grassroots effort funded as a research coordination network through the US National Science Foundation aimed at building and supporting an interdisciplinary community around near-term (daily to decadal), iterative ecological forecasts akin to weather forecasts. One way EFI built a community of practice around ecological forecasting was through the National Ecological Observatory Network (NEON) Forecasting Challenge, an open opportunity for groups to submit forecasts using NEON data streams ranging from carbon and water cycling to tick dynamics (Thomas et al., 2022). There are now very active chapters of EFI in the United States, Canada, Europe, and Oceania. In organizing this special feature, our goals were two-fold. First, we strove to bring together contributions from a group of authors representing racial, sectoral (e.g. academic, government), geographic and career stage diversity to provide a broad overview of the state of the art in methods related to forecasting given the recent interest in ecological forecasting. Second, we sought to identify emerging opportunities in ecological forecasting. In modelling the future of complex ecological systems, ubiquitous challenges emerge that the papers in this special feature capture. Examples include modelling non-linearities in response variables, modelling unobserved variables, and quantifying uncertainty. Many of the papers in this special feature deal with these ubiquitous challenges of forecasting, but address them with a range of modelling approaches from machine learning (Clark & Wells, 2023; Lapeyrolerie & Boettiger, 2022) to process-based models (Cameron et al., 2022). To illustrate this, we consider how some of the papers in this special feature approach estimates of unobserved variables or uncertainty with myriad approaches. Ecological communities are complex, as they are composed of many interacting species that respond to the environment in different ways. Given this complexity, it is seldom possible to acquire data on all relevant variables needed for an ecological forecast. Clark and Wells (2023) forecast abundances of single and multiple species beyond the range of observed data by modelling unobserved dynamic components in an extension of generalized additive models (GAMs). GAMs are a popular choice for modelling time series because of their ability to describe smooth, nonlinear relationships between predictor and response variables. However, GAMs perform poorly in forecasts because when the smooth functions extend beyond the observed data, they may produce unrealistic forecasts (e.g. linear extrapolations) or fail to capture temporal dependence (Zurell et al., 2012). To overcome these shortcomings of GAMs for forecasting, Clark and Wells (2023) present dynamic generalized additive models (DGAMs) that jointly estimate the GAM linear predictor and unobserved dynamic components to model an evolving time series for the cases of both univariate and multivariate response variables. Examples with simulations and empirical data on ticks from the EFI NEON forecasting challenge illustrate the utility of DGAMs for forecasting discrete ecological time series, and an r package mvgam with worked examples makes the use of this approach accessible. An alternative approach to compensating for unobserved variables with process noise is empirical dynamic modelling (EDM) that stems from dynamical systems theory and employs time delays as surrogates for missing data in a non-parametric framework. Although this approach was first used in ecology by (Schaffer, 1984) and followed by a handful of studies in the 1990 s (e.g. Sugihara & May, 1990), the vast data requirements of the EDMs at the time prevented their widespread uptake by ecologists. Munch et al. (2023) revisit the utility of EDMs for forecasting given that ecology as a discipline entered a new phase of ‘big data’ over the last several decades with remotely sensed data streams, the maturation of long-term ecological datasets (e.g. international Long Term Ecological Research Network [iLTER]), the curation of individual datasets into community data resources (e.g. Global Biodiversity Information Facility [GBIF]) and through international coordination of environmental and ecological data (e.g. Global Earth Observation System of Systems [GEOSS]; Farley et al., 2018; Hampton et al., 2013; Lautenbacher, 2006). Advances in EDMs such as novel algorithms that account for unequal sampling intervals and merging shorter time series along with recent extensions of EDMs to multivariate time series make this approach a relevant alternative to parametric models for forecasting. When supporting decision making, it is integral for ecological forecasts to characterize uncertainty, so that people making choices may weigh the consequences of their actions appropriately. Calibration of process-based models of complex ecological systems often requires model-data fusion of unbalanced quantities of different data types, such as information collected by field technicians versus sensors. In hopes of reducing model uncertainty, a common approach to address unbalanced data is to weight data sets (Medvigy et al., 2009; Richardson et al., 2010). Cameron et al. (2022) employs a series of simulated data sets with a very simple ecosystem model to explore how unbalanced data versus systematic errors in models and data influence the ability to recover the model parameters and the ‘true’ dynamics of latent variables. These data experiments show that biases in the model or data, not unbalanced data, impede recovery of the correct parameters. Cameron et al. (2022) also presents a diagnostic tool for identifying when systematic errors are problematic and show that adding terms representing model structural errors in the calibration improves predictions with a quantification of uncertainty. Taking an entirely different modelling approach to Cameron et al. (2022), Lapeyrolerie and Boettiger (2022) consider uncertainty estimation with deep learning methods in the context of forecasting abrupt changes in ecosystem dynamics (i.e. critical transitions). Deep neural network models have become a popular method in the last decade for making point forecasts of ecological time series because they preclude making errant prior assumptions about the data by automatically learning temporal dependencies (Makridakis et al., 2018). A major shortcoming of many deep learning models for forecasting, however, is that it is difficult to estimate model uncertainty from them. Lapeyrolerie and Boettiger (2022) explore the performance of several deep learning algorithms that can account for model uncertainty to see how well they are able to forecast different simulated examples of critical transitions in comparison to forecasts produced by Markov chain Monte Carlo (MCMC) and autoregressive integrated moving average (ARIMA) models given the ‘true’ transition dynamics. The deep learning methods capable of estimating uncertainty perform comparably to the MCMC and ARIMA approaches, further distinguishing deep learning as a powerful tool for forecasting critical transitions. The diversity of modelling approaches that can be applied to ecological forecasting presents challenges and opportunities in and of itself with regards to generating relevant forecasts for different audiences. Slingsby et al. (2023) emphasizes the value of integrating across different forecasts focused within a single spatial region motivated by a case study for a global biodiversity hotspot, the Cape Floristic Region of South Africa. They propose a state-space framework linking ecological forecasts from different biological, spatial, and temporal scales within the Cape. Although developing such an integrated forecasting framework presents data and modelling challenges, the creation of such a shared ecoinformatics workflow for multiple forecasts provides a practical approach to addressing some of the logistical, technological, and funding issues South Africa experiences as part of the Global South. Ecoinformatics pipelines that generate analysis-ready data across scales promote iterative, near-term forecasting, whose outputs can be provided to decision makers in a timely fashion. Furthermore, if these ecoinformatics pipelines are open and reproducible, they may be adopted by other regions of the world to make forecasting more inclusive and provide opportunities to test the transferability of forecasts to novel regions. Lewis et al. (2022) also highlights the need for forecast comparison and synthesis, in addition to forecast development, to catalyse refinement and formulation of existing and novel ecological theory, respectively. They present a vision where iterative forecasting in a model-data loop (Dietze et al., 2018) promotes theory development and is fully incorporated into every ecologist's toolbox. A roadmap for fulfilling this vision incorporates open-access code, reproducible workflows, increased data science education, and lower barriers to cyberinfrastructure needed for running forecasts. Furthermore, community-wide investment in standards for forecast outputs and metadata will help to enable synthesis across various approaches to forecasting (Thomas et al., 2022). The collection of papers in this special feature highlights the variety of approaches taken to ecological forecasting and the value of synthesizing across iterative forecasts to provide decision support and spur innovations in ecological theory. A recent paper on priorities in synthesis research authored by over one hundred scientists across career stages, institutions, backgrounds and geographies also identified synthesis of forecasts as a critical focus for future synthesis studies in ecology (Halpern et al., 2023). As interest in ecological forecasting continues to increase, efforts to aid in the synthesis of diverse forecasts (e.g. a forecast archive repository, community standards for forecast outputs and metadata) will be integral to successfully transforming decision support and ecological theory with the help of forecasts. Sydne Record acknowledges support by the USDA National Institute of Food and Agriculture, [Hatch/McIntire-Stennis/Animal Health] Project Number ME0-7003869 through the Maine Agricultural and Forest Experiment Station. Carl Boettiger acknowledges support from the National Science Foundation under grant no. DBI-1942280. The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer-review/10.1111/2041-210X.14070.