Knowledge-guided machine learning can improve carbon cycle quantification in agroecosystems.

Accurate and cost-effective quantification of the carbon cycle for agroecosystems at decision-relevant scales is critical to mitigating climate change and ensuring sustainable food production. However, conventional process-based or data-driven modeling approaches alone have large prediction uncertainties due to the complex biogeochemical processes to model and the lack of observations to constrain many key state and flux variables. Here we propose a Knowledge-Guided Machine Learning (KGML) framework that addresses the above challenges by integrating knowledge embedded in a process-based model, high-resolution remote sensing observations, and machine learning (ML) techniques. Using the U.S. Corn Belt as a testbed, we demonstrate that KGML can outperform conventional process-based and black-box ML models in quantifying carbon cycle dynamics. Our high-resolution approach quantitatively reveals 86% more spatial detail of soil organic carbon changes than conventional coarse-resolution approaches. Moreover, we outline a protocol for improving KGML via various paths, which can be generalized to develop hybrid models to better predict complex earth system dynamics.

Ensemble machine learning approach for examining critical process parameters and scale-up opportunities of microbial electrochemical systems for hydrogen peroxide production.

Hydrogen peroxide (H2O2) production in microbial electrochemical systems (MESs) is an attractive option for enabling a circular economy in the water/wastewater sector. Here, a machine learning algorithm was developed, using a meta-learning approach, to predict the H2O2 production rates in MES based on the seven input variables, including various design and operating parameters. The developed models were trained and cross-validated using the experimental data collected from 25 published reports. The final ensemble meta-learner model (combining 60 models) demonstrated a high prediction accuracy with very high R2 (0.983) and low root-mean-square error (RMSE) (0.647 kg H2O2 m-3 d-1) values. The model identified the carbon felt anode, GDE cathode, and cathode-to-anode volume ratio as the top three most important input features. Further scale-up analysis for small-scale wastewater treatment plants indicated that proper design and operating conditions could increase the H2O2 production rate to as high as 9 kg m-3 d-1.

Accuracy of tropical peat and non-peat fire forecasts enhanced by simulating hydrology.

Soil moisture deficits and water table dynamics are major biophysical controls on peat and non-peat fires in Indonesia. Development of modern fire forecasting models in Indonesia is hampered by the lack of scalable hydrologic datasets or scalable hydrology models that can inform the fire forecasting models on soil hydrologic behaviour. Existing fire forecasting models in Indonesia use weather data-derived fire probability indices, which often do not adequately proxy the sub-surface hydrologic dynamics. Here we demonstrate that soil moisture and water table dynamics can be simulated successfully across tropical peatlands and non-peatland areas by using a process-based eco-hydrology model (ecosys) and publicly available data for weather, soil, and management. Inclusion of these modelled water table depth and soil moisture contents significantly improves the accuracy of a neural network model in predicting active fires at two-weekly time scale. This constitutes an important step towards devising an operational fire early warning system for Indonesia.

Exploration of machine learning algorithms for predicting the changes in abundance of antibiotic resistance genes in anaerobic digestion.

The land application of digestate from anaerobic digestion (AD) is considered a significant route for transmitting antibiotic resistance genes (ARGs) and mobile genetic elements (MGEs) to ecosystems. To date, efforts towards understanding complex non-linear interactions between AD operating parameters with ARG/MGE abundances rely on experimental investigations due to a lack of mechanistic models. Herein, three different machine learning (ML) algorithms, Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Artificial Neural Network (ANN), were compared for their predictive capacities in simulating ARG/MGE abundance changes during AD. The models were trained and cross-validated using experimental data collected from 33 published literature. The comparison of model performance using coefficients of determination (R2) and root mean squared errors (RMSE) indicated that ANN was more reliable than RF and XGBoost. The mode of operation (batch/semi-continuous), co-digestion of food waste and sewage sludge, and residence time were identified as the three most critical features in predicting ARG/MGE abundance changes. Moreover, the trained ANN model could simulate non-linear interactions between operational parameters and ARG/MGE abundance changes that could be interpreted intuitively based on existing knowledge. Overall, this study demonstrates that machine learning can enable a reliable predictive model that can provide a holistic optimization tool for mitigating the ARG/MGE transmission potential of AD.

A critical review of process parameters influencing the fate of antibiotic resistance genes in the anaerobic digestion of organic waste.

The overuse and inappropriate disposal of antibiotics raised severe public health risks worldwide. Specifically, the incomplete antibiotics metabolism in human and animal bodies contributes to the significant release of antibiotics into the natural ecosystems and the proliferation of antibiotic-resistant bacteria carrying antibiotic-resistant genes. Moreover, the organic feedstocks used for anaerobic digestion are often highly-rich in residual antibiotics and antibiotic-resistant genes. Hence, understanding their fate during anaerobic digestion has become a significant research focus recently. Previous studies demonstrated that various process parameters could considerably influence the propagation of the antibiotic-resistant genes during anaerobic digestion and their transmission via land application of digestate. This review article scrutinizes the influences of process parameters on antibiotic-resistant genes propagation in anaerobic digestion and the inherent fundamentals behind their effects. Based on the literature review, critical research gaps and challenges are summarized to guide the prospects for future studies.

How could simulated dewatering of slurry mitigate nitrous oxide emissions from fall and spring injections? - A modelling study in a Chernozem soil in Western Canada.

Process-based ecosystem models, such as ecosys, can be useful tools to gain insights and accurately project nitrous oxide (N2O) inventories in national, regional and global scales, and to explore potential emission reduction strategies. Our objectives are to investigate how the ecosys model simulate the effects of fall and spring slurry injections on N2O production and if de-watering slurry could become a potential N2O mitigation strategy for both fall and spring injections. The ecosys model was used to simulate hourly N2O fluxes from 2014 to 2017 in a cropping system with and without slurry (fall and spring additions) in comparison with field measurements in Alberta, Canada. Furthermore, we performed simulations of de-watered fall and spring slurry applications in the same scenarios. Our results showed ecosys adequately simulated soil temperatures and moisture contents at 10 and 20 cm depths [correlation coefficients (r) ≥ 0.929 for temperatures; r ≥ 0.529 for moistures]. The divergences of modelled and measured soil water contents during spring thaws could be attributed to uncertainties in model inputs for soil hydrological parameters as well as uncertainties in field measurements. The model captured reasonably well the dynamics of N2O fluxes from soils receiving fall and spring slurry (r = 0.356). However, the concurrent discrepancies of N2O fluxes between modelled and measured values during the wetter spring thaw of 2017 might be a result of an unsatisfactory simulation of snowmelt infiltration and runoff. Compared to whole slurry, simulated de-watered slurry resulted in considerable reductions in cumulative N2O emissions by 16-36 and 23-29% for fall and spring slurry injections, respectively. The model results indicate that de-watering slurry would potentially be an efficient emission mitigation strategy; however, there is still a paucity of studies addressing the feasibility of dewatering as a practice and further research can focus on this knowledge gap.

Non-stationary response of rain-fed spring wheat yield to future climate change in northern latitudes.

The non-stationary response of crop growth to changes in hydro-climatic variables makes yield projection uncertain and the design and implementation of adaptation strategies debatable. This study simulated the time-varying behavior of the underlying cause-and-effect mechanisms affecting spring wheat yield (SWY) under various climate change and nitrogen (N) application scenarios in the Red Deer River basin in agricultural lands of the western Canadian Prairies. A calibrated and validated Soil and Water Assessment Tool and Analysis of Variance decomposition methods were utilized to assess the contribution of crop growth parameters, Global Climate Models, Representative Concentration Pathways, and downscaling techniques to the total SWY variance for the 2040-2064 period. The results showed that the cause-and-effect mechanisms, driving crop yield, shifted from water stress (W-stress) dominated (27 days of W-stress days) during the historical period to nitrogen stress (N-stress) dominated (27 to 35 N-stress days) in the future period. It was shown that while higher precipitation, warmer weather, and early snowmelts, along with elevated CO2 may favor SWY in cold regions in the future (up to 50% more yields in some sub-basins), the yield potentials may be limited by N-stress (only up to 0.7% yield increase in some sub-basins). The N-stress might be partially related to the N deficiency in the soil, which can be compensated by N fertilizer application. However, inadequate N uptake due to limited evapotranspiration under elevated atmospheric CO2 might pose restrictions to SWY potentials even in the least N deficient regions. This study uncovers important information on the understanding of spatiotemporal variability of hydrogeochemical processes driving crop yields and the non-stationary response of yields to changing climate. The results also underscore spatiotemporal variability of N-stress due to N deficiency in the soil or N uptake by crops, both of which may restrain SWY by changes in atmospheric CO2 concentrations in the future.

Modelling spatio-temporal patterns of soil carbon and greenhouse gas emissions in grazing lands: Current status and prospects.

The sustainability of grazing lands lies in the nexus of human consumption behavior, livestock productivity, and environmental footprint. Due to fast growing global food demands, many grazing lands have suffered from overgrazing, leading to soil degradation, air and water pollution, and biodiversity losses. Multidisciplinary efforts are required to understand how these lands can be better assessed and managed to attain predictable outcomes of optimal benefit to society. This paper synthesizes our understanding based on previous work done on modelling the influences of grazing of soil carbon (SC) and greenhouse gas emissions to identify current knowledge gaps and research priorities. We revisit three widely-used process-based models: DeNitrification DeComposition (DNDC), DayCent, and the Pasture Simulation model (PaSim) and two watershed models: The Soil & Water Assessment Tool (SWAT) and Variable Infiltration Capacity Model (VIC), which are widely used to simulate C, nutrient and water cycles. We review their structures and ability as process-based models in representing key feedbacks among grazing management, SOM decomposition and hydrological processes in grazing lands. Then we review some significant advances in the use of models combining biogeochemical and hydrological processes. Finally, we examine challenges of incorporating spatial heterogeneity and temporal variability into modelling C and nutrient cycling in grazing lands and discuss their weakness and strengths. We also highlight key research direction for improving the knowledge base and code structure in modelling C and nutrient cycling in grazing lands, which are essential to conserve grazing lands and maintain their ecosystem goods and services.