Streams and rivers can substantially modify organic carbon (OC) inputs from terrestrial landscapes, and much of this processing is the result of microbial respiration. While carbon dioxide (CO2) is ...the major end-product of ecosystem respiration, methane (CH4) is also present in many fluvial environments even though methanogenesis typically requires anoxic conditions that may be scarce in these systems. Given recent recognition of the pervasiveness of this greenhouse gas in streams and rivers, we synthesized existing research and data to identify patterns and drivers of CH4, knowledge gaps, and research opportunities. This included examining the history of lotie CH4 research, creating a database of concentrations and fluxes (MethDB) to generate a global-scale estimate of fluvial CH4 efflux, and developing a conceptual framework and using this framework to consider how human activities may modify fluvial CH4 dynamics. Current understanding of CH4 in streams and rivers has been strongly influenced by goals of understanding OC processing and quantifying the contribution of CH4 to ecosystem C fluxes. Less effort has been directed towards investigating processes that dictate in situ CH4 production and loss. CH4 makes a meager contribution to watershed or landscape C budgets, but streams and rivers are often significant CH4 sources to the atmosphere across these same spatial extents. Most fluvial systems are supersaturated with CH4 and we estimate an annual global emission of 26.8 Tg CH4, equivalent to ∼15-40% of wetland and lake effluxes, respectively. Less clear is the role of CH4 oxidation, methanogenesis, and total anaerobic respiration to whole ecosystem production and respiration. Controls on CH4 generation and persistence can be viewed in terms of proximate controls that influence methanogenesis (organic matter, temperature, alternative electron acceptors, nutrients) and distal geomorphic and hydrologie drivers. Multiple controls combined with its extreme redox status and low solubility result in high spatial and temporal variance of CH4 in fluvial environments, which presents a substantial challenge for understanding its larger-scale dynamics. Further understanding of CH4 production and consumption, anaerobic metabolism, and ecosystem energetics in streams and rivers can be achieved through more directed studies and comparison with knowledge from terrestrial, wetland, and aquatic disciplines.
The rapid growth of data in water resources has created new opportunities to accelerate knowledge discovery with the use of advanced deep learning tools. Hybrid models that integrate theory with ...state‐of‐the art empirical techniques have the potential to improve predictions while remaining true to physical laws. This paper evaluates the Process‐Guided Deep Learning (PGDL) hybrid modeling framework with a use‐case of predicting depth‐specific lake water temperatures. The PGDL model has three primary components: a deep learning model with temporal awareness (long short‐term memory recurrence), theory‐based feedback (model penalties for violating conversation of energy), and model pretraining to initialize the network with synthetic data (water temperature predictions from a process‐based model). In situ water temperatures were used to train the PGDL model, a deep learning (DL) model, and a process‐based (PB) model. Model performance was evaluated in various conditions, including when training data were sparse and when predictions were made outside of the range in the training data set. The PGDL model performance (as measured by root‐mean‐square error (RMSE)) was superior to DL and PB for two detailed study lakes, but only when pretraining data included greater variability than the training period. The PGDL model also performed well when extended to 68 lakes, with a median RMSE of 1.65 °C during the test period (DL: 1.78 °C, PB: 2.03 °C; in a small number of lakes PB or DL models were more accurate). This case‐study demonstrates that integrating scientific knowledge into deep learning tools shows promise for improving predictions of many important environmental variables.
Key Points
Process‐Guided Deep Learning (PGDL) models integrate advanced empirical techniques with process knowledge
We used PGDL to accurately predict lake water temperatures for various conditions
PGDL performance improved significantly when pretraining data included diverse conditions generated by an existing process‐based model
It is standard practice to compare the status of performance indicators between restoration and reference sites to monitor restoration progress and demonstrate restoration success. However, standard ...methods for defining the reference ecosystem, selecting reference sites, and measuring success are surprisingly lacking. Our study develops these methods based on the acceptable range of variation (ARV) within the desirable stable (reference) state as a measure of restoration success. The method (1) constrains application to the contemporary landscape to avoid the problematic historical range of variation concept and idealized restoration targets; (2) acknowledges the theory of alternative stable states and ecosystem dynamics and posits that the reference ecosystem should be clearly defined as a desirable stable (reference) state; and (3) shows that identifying an acceptable thematic (classification) scale and an acceptable management timeframe is essential to defining the desirable stable (reference) state. We present two approaches to calculating an ARV and a simulation method to explore reference site replication sufficiency. We apply the methods to two contrasting Australian restoration case studies and recommend that routine adoption of these methods would make a significant contribution to the science and practice of restoration ecology and to the assessment of restoration success.
Most environmental data come from a minority of well‐monitored sites. An ongoing challenge in the environmental sciences is transferring knowledge from monitored sites to unmonitored sites. Here, we ...demonstrate a novel transfer‐learning framework that accurately predicts depth‐specific temperature in unmonitored lakes (targets) by borrowing models from well‐monitored lakes (sources). This method, meta‐transfer learning (MTL), builds a meta‐learning model to predict transfer performance from candidate source models to targets using lake attributes and candidates' past performance. We constructed source models at 145 well‐monitored lakes using calibrated process‐based (PB) modeling and a recently developed approach called process‐guided deep learning (PGDL). We applied MTL to either PB or PGDL source models (PB‐MTL or PGDL‐MTL, respectively) to predict temperatures in 305 target lakes treated as unmonitored in the Upper Midwestern United States. We show significantly improved performance relative to the uncalibrated PB General Lake Model, where the median root mean squared error (RMSE) for the target lakes is 2.52°C. PB‐MTL yielded a median RMSE of 2.43°C; PGDL‐MTL yielded 2.16°C; and a PGDL‐MTL ensemble of nine sources per target yielded 1.88°C. For sparsely monitored target lakes, PGDL‐MTL often outperformed PGDL models trained on the target lakes themselves. Differences in maximum depth between the source and target were consistently the most important predictors. Our approach readily scales to thousands of lakes in the Midwestern United States, demonstrating that MTL with meaningful predictor variables and high‐quality source models is a promising approach for many kinds of unmonitored systems and environmental variables.
Key Points
Meta‐transfer learning (MTL) learns from models trained on data‐rich systems to inform predictions in systems where no observations exist
We use MTL with process‐based and process‐guided deep learning models to accurately predict lake temperatures in the Midwest United States
The most important predictor of transfer model success is the difference in maximum depth between the data‐rich and unmonitored lake
1. Grazing is one of the most widespread forms of intensive management on Earth and is linked to reductions in soil health. However, little is known about the relative influence of herbivore type, ...herbivore intensity and site productivity on soil health. This lack of knowledge reduces our capacity to manage landscapes where grazing is a major land use. 2. We used structural equation modelling to assess the effects of recent (cattle, sheep, goats, kangaroos and rabbit dung) and historic (cattle, sheep/goat livestock tracks) herbivore activity on soil health at 451 sites across 0.5 M km² of eastern Australia. We assessed the direct and indirect effects of increasing herbivore intensity, using dung and livestock tracks, on 15 morphological, physical and chemical attributes that are indicative of soil health, and we used these attributes to derive three indices representing the capacity of the soil to maintain its structural integrity (stability), cycle nutrients (nutrients) and maintain water flow (infiltration). 3. Grazing had negative effects on the three soil health indices, but these effects varied with productivity. Grazing intensity was associated with strong reductions in the stability and nutrient indices under low productivity, but these effects diminished with increasing productivity. Herbivore effects on individual attributes varied in relation to productivity level and were strongly herbivore specific, with most due to cattle grazing, and to a lesser extent, sheep, goats and rabbits. Few effects due to kangaroos or historic grazing by livestock were observed. 4. Synthesis and applications. Our study shows that livestock and rabbits degrade soil health through grazing, and its effects are strongest under low or moderate productivity; however, kangaroo effects are benign. Our findings support calls for resource management agencies to consider site productivity, as well as herbivore type and intensity, when developing strategies to manage grazing by livestock, and feral and native herbivores.
Scientists have largely neglected the effects of grazing on soil microbial communities despite their importance as drivers of ecosystem functions and services. We hypothesized that changes in soil ...properties resulting from grazing regulate the diversity of soil microbes by releasing/suppressing subordinate microbial taxa via competition. To test this, we examined how intensity of vertebrate herbivores influences the diversity and composition of soil bacteria and fungi at 216 soil samples from 54 sites across four microsites. Increasing grazing intensity reduced soil carbon, suppressing the dominant bacterial phylum Actinobacteria (indirectly promoting bacterial diversity) and increasing the dominant fungal phylum Ascomycetes (indirectly reducing fungal diversity). Our data provide novel evidence that grazing modulates the diversity and composition of soil microbes via increases or reductions in competition by dominant taxa. Our results suggest that grazing can potentially alter soil function by altering microbial community composition, providing a clear link between grazing management, carbon availability and ecosystem functions.
Deep learning (DL) models are increasingly used to make accurate hindcasts of management‐relevant variables, but they are less commonly used in forecasting applications. Data assimilation (DA) can be ...used for forecasts to leverage real‐time observations, where the difference between model predictions and observations today is used to adjust the model to make better predictions tomorrow. In this use case, we developed a process‐guided DL and DA approach to make 7‐day probabilistic forecasts of daily maximum water temperature in the Delaware River Basin in support of water management decisions. Our modeling system produced forecasts of daily maximum water temperature with an average root mean squared error (RMSE) from 1.1 to 1.4°C for 1‐day‐ahead and 1.4 to 1.9°C for 7‐day‐ahead forecasts across all sites. The DA algorithm marginally improved forecast performance when compared with forecasts produced using the process‐guided DL model alone (0%–14% lower RMSE with the DA algorithm). Across all sites and lead times, 65%–82% of observations were within 90% forecast confidence intervals, which allowed managers to anticipate probability of exceedances of ecologically relevant thresholds and aid in decisions about releasing reservoir water downstream. The flexibility of DL models shows promise for forecasting other important environmental variables and aid in decision‐making.
Although spatial and temporal variation in ecological properties has been well‐studied, crucial knowledge gaps remain for studies conducted at macroscales and for ecosystem properties related to ...material and energy. We test four propositions of spatial and temporal variation in ecosystem properties within a macroscale (1000 km's) extent. We fit Bayesian hierarchical models to thousands of observations from over two decades to quantify four components of variation – spatial (local and regional) and temporal (local and coherent); and to model their drivers. We found strong support for three propositions: (1) spatial variation at local and regional scales are large and roughly equal, (2) annual temporal variation is mostly local rather than coherent, and, (3) spatial variation exceeds temporal variation. Our findings imply that predicting ecosystem responses to environmental changes at macroscales requires consideration of the dominant spatial signals at both local and regional scales that may overwhelm temporal signals.
1. Grazing by domestic livestock is sometimes promoted as a management tool to benefit biodiversity. In many situations, however, it can produce negative outcomes. 2. Here, we examine the impacts of ...recent and historic livestock grazing on bird communities in the semi-arid woodlands in eastern Australia, testing the notion that grazing removes the suppressive effect of structurally complex vegetation on miners, thereby reducing the richness and abundance of small birds. 3. We used time- and area-limited searches of 108 sites varying in livestock grazing history and intensity, to explore the direct and indirect effects of grazing, habitat complexity and the abundance of aggressive, large-bodied birds on smaller-bodied birds using two-way analysis of variance and structural equation modelling. 4. Small birds were less abundant and had lower richness in the presence of miners. Our structural equation models indicated that recent grazing had direct suppressive effects on the abundance of miners, and both richness and abundance of all but the largest-bodied bird groups. However, higher levels of historic livestock grazing reinforced the competitive exclusion of the six small-bodied bird groups (insectivores, nectarivores, declining woodland birds, small ground-foraging birds, all small birds and all non-miners) by aggressive miners via reductions in habitat complexity. Moreover, the strength of any suppressive effects on small birds or positive effects on large birds by miners increased with increasing miner abundance. 5. Synthesis and applications. Our results highlight the importance of vegetation structural complexity, not only for providing habitat for woodland birds, but also as barriers to the invasion and competitive dominance of miners. Our findings suggest that management actions aimed at reducing tree and shrub density to promote open woodlands are likely to have significant negative consequences for the conservation of small woodland birds.
Identifying threatened ecosystem types is fundamental to conservation and management decision‐making. When identification relies on expert judgment, decisions are vulnerable to inconsistent outcomes ...and can lack transparency. We elicited judgements of the occurrence of a widespread, critically endangered Australian ecosystem from a diverse pool of 83 experts. We asked 4 questions. First, how many experts are required to reliably conclude that the ecosystem is present? Second, how many experts are required to build a reliable model for predicting ecosystem presence? Third, given expert selection can narrow the range opinions, if enough experts are selected, do selection strategies affect model predictions? Finally, does a diverse selection of experts provide better model predictions? We used power and sample size calculations with a finite population of 200 experts to calculate the number of experts required to reliably assess ecosystem presence in a theoretical scenario. We then used boosted regression trees to model expert elicitation of 122 plots based on real‐world data. For a reliable consensus (90% probability of correctly identifying presence and absence) in a relatively certain scenario (85% probability of occurrence), at least 17 experts were required. More experts were required when occurrence was less certain, and fewer were needed if permissible error rates were relaxed. In comparison, only ∼20 experts were required for a reliable model that could predict for a range of scenarios. Expert selection strategies changed modeled outcomes, often overpredicting presence and underestimating uncertainty. However, smaller but diverse pools of experts produced outcomes similar to a model built from all contributing experts. Combining elicited judgements from a diverse pool of experts in a model‐based decision support tool provided an efficient aggregation of a broad range of expertise. Such models can improve the transparency and consistency of conservation and management decision‐making, especially when ecosystems are defined based on complex criteria.
La importancia de seleccionar expertos para identificar ecosistemas amenazados
Resumen
La identificación de los tipos de ecosistemas amenazados es fundamental para decidir sobre su conservación y gestión. Cuando la identificación se basa en la opinión de expertos, las decisiones son vulnerables a resultados incoherentes y pueden carecer de transparencia. Recabamos la opinión de 83 expertos sobre la presencia de un ecosistema australiano extendido y en peligro crítico. Se plantearon cuatro preguntas: ¿Cuántos expertos son necesarios para concluir con fiabilidad que el ecosistema está presente?; ¿Cuántos expertos son necesarios para construir un modelo fiable de predicción de la presencia del ecosistema?; ya que la selección de expertos puede reducir el rango de opiniones, si se seleccionan suficientes expertos, ¿afectan las estrategias de selección a las predicciones del modelo; y ¿Una selección diversa de expertos proporciona mejores predicciones del modelo? Utilizamos cálculos de potencia y tamaño de muestra con una población finita de 200 expertos para obtener el número de expertos necesarios para evaluar de forma fiable la presencia de ecosistemas en un escenario teórico. Después usamos árboles de regresión reforzada para modelar la consulta de expertos de 122 parcelas basadas en datos del mundo real. Para obtener un consenso fiable (90% de probabilidad de identificar correctamente la presencia y la ausencia) en un escenario relativamente seguro (85% de probabilidad de ocurrencia), se necesitaban al menos 17 expertos. Se necesitaban más expertos cuando la ocurrencia era menos segura, y menos si se relajaban los porcentajes de error permitidos. En comparación, sólo se necesitaron unos 20 expertos para obtener un modelo fiable que pudiera predecir una serie de escenarios. Las estrategias de selección de expertos modificaron los resultados modelados, a menudo con sobre predicción de la presencia y subestimación de la incertidumbre. Sin embargo, los grupos de expertos más pequeños pero diversos produjeron resultados similares a los de un modelo construido a partir de todos los expertos participantes. La combinación de las opiniones obtenidas de un grupo diverso de expertos en una herramienta de apoyo a la toma de decisiones basada en un modelo proporcionó una agregación eficiente de una amplia gama de conocimientos. Estos modelos pueden mejorar la transparencia y coherencia de la toma de decisiones en materia de conservación y gestión, especialmente cuando los ecosistemas se definen en función de criterios complejos.