Meta-modeling methods for estimating ammonia volatilization from nitrogen fertilizer and manure applications

Ramanantenasoa, Maharavo Marie Julie; Génermont, Sophie; Gilliot, Jean-Marc; Bedos, Carole; Makowski, David

doi:10.1016/j.jenvman.2019.01.066

Cited by 27 publications

(19 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…delayed planting in 2019 in Iowa due to an exceptionally wet spring). Our results concur with previous research (Ramanantenasoa et al 2019), and highlight the potential of coupling simulation models and ML toward developing dynamic decision support tools for agricultural management.…”

Section: Discussionsupporting

confidence: 91%

“…In view of increasing data availability in agriculture and the maturity of analytics from descriptive to prescriptive (National Academy of Sciences, E A M 2019), more ML applications are currently taking place. For example, Ramanantenasoa et al (2019) evaluated the performance of various ML based meta-models to emulate the complex process-based models in predicting ammonia emissions produced by agricultural activities and demonstrated the superiority of random forests compared to LASSO regression. Lawes et al (2019) used ML and APSIM modeling to predict optimum N rates for wheat, Puntel et al (2019) and Qin et al (2018) used ML and experimental data to predict optimum N rates to maize, while others are exploring coupling ML and simulation models to develop faster and more flexible tools for impact regional assessments (Fienen et al 2015) and simulation model parameterization (Gladish et al 2019).…”

Section: Discussionmentioning

confidence: 99%

“…By replacing a more detailed simulation model, a meta-model can provide faster execution, reduced storage needs, and added flexibility to extrapolate across temporal and spatial scales than a more detailed simulation model (Simpson et al 2001). Metamodeling has been widely implemented for extrapolating hydrological (Fienen et al 2015, Nolan et al 2018 and biogeochemical (Britz and Leip 2009, Villa-Vialaneix et al 2012, Ramanantenasoa et al 2019 simulations across regional scales, as well as to streamline sensitivity analyzes for parameterization of crop simulation models (Stanfill et al 2015, Pianosi et al 2016, Gladish et al 2019. To our knowledge, metamodeling approaches for decision support and forecasting applications in crop production have not been previously evaluated.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Maize yield and nitrate loss prediction with machine learning algorithms

Shahhosseini

Martinez-Feria

et al. 2019

Environ. Res. Lett.

153

View full text Add to dashboard Cite

Pre-growing season prediction of crop production outcomes such as grain yields and nitrogen (N) losses can provide insights to farmers and agronomists to make decisions. Simulation crop models can assist in scenario planning, but their use is limited because of data requirements and long runtimes. Thus, there is a need for more computationally expedient approaches to scale up predictions. We evaluated the potential of four machine learning (ML) algorithms (LASSO Regression, Ridge Regression, random forests, Extreme Gradient Boosting, and their ensembles) as meta-models for a cropping systems simulator (APSIM) to inform future decision support tool development. We asked:(1) How well do ML meta-models predict maize yield and N losses using pre-season information? (2) How many data are needed to train ML algorithms to achieve acceptable predictions? (3) Which input data variables are most important for accurate prediction? And (4) do ensembles of ML meta-models improve prediction? The simulated dataset included more than three million data including genotype, environment and management scenarios. XGBoost was the most accurate ML model in predicting yields with a relative mean square error (RRMSE) of 13.5%, and Random forests most accurately predicted N loss at planting time, with a RRMSE of 54%. ML meta-models reasonably reproduced simulated maize yields using the information available at planting, but not N loss. They also differed in their sensitivities to the size of the training dataset. Across all ML models, yield prediction error decreased by 10%-40% as the training dataset increased from 0.5 to 1.8 million data points, whereas N loss prediction error showed no consistent pattern. ML models also differed in their sensitivities to input variables (weather, soil properties, management, initial conditions), thus depending on the data availability researchers may use a different ML model. Modest prediction improvements resulted from ML ensembles. These results can help accelerate progress in coupling simulation models and ML toward developing dynamic decision support tools for pre-season management.

show abstract

Section: Discussionsupporting

confidence: 91%

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Maize yield and nitrate loss prediction with machine learning algorithms

Shahhosseini

Martinez-Feria

et al. 2019

Environ. Res. Lett.

153

View full text Add to dashboard Cite

show abstract

“…Unlike process‐based models, the Random Forest algorithm learning is facilitated through an iterative process of recursive data partitioning and constructing hundreds of decision trees to partition the observations into distinct groups characterized by different properties of the predictor variables. Meta‐modeling approach using Random Forest and process‐based models has been used in predicting soil N biogeochemistry (Ramanantenasoa, Génermont, Gilliot, Bedos, & Makowski, 2019; Shahhosseini, Martinez‐Feria, Hu, & Archontoulis, 2019); however, its use in predicting highly variable temporal soil N 2 O fluxes is limited. Saha, Basso, and Robertson (in review) developed a Random Forest model based on automated flux chamber data from corn in the upper Midwest that predicted 51% variability in daily N 2 O fluxes from an unknown site.…”

Section: Current Mechanistic Understanding About the Timing And Magnimentioning

confidence: 99%

21st‐century biogeochemical modeling: Challenges for Century‐based models and where do we go from here?

et al. 2020

View full text Add to dashboard Cite

21st‐century modeling of greenhouse gas (GHG) emissions from bioenergy crops is necessary to quantify the extent to which bioenergy production can mitigate climate change. For over 30 years, the Century‐based biogeochemical models have provided the preeminent framework for belowground carbon and nitrogen cycling in ecosystem and earth system models. While monthly Century and the daily time‐step version of Century (DayCent) have advanced our ability to predict the sustainability of bioenergy crop production, new advances in feedstock generation, and our empirical understanding of sources and sinks of GHGs in soils call for a re‐visitation of DayCent's core model structures. Here, we evaluate current challenges with modeling soil carbon dynamics, trace gas fluxes, and drought and age‐related impacts on bioenergy crop productivity. We propose coupling a microbial process‐based soil organic carbon and nitrogen model with DayCent to improve soil carbon dynamics. We describe recent improvements to DayCent for simulating unique plant structural and physiological attributes of perennial bioenergy grasses. Finally, we propose a method for using machine learning to identify key parameters for simulating N2O emissions. Our efforts are focused on meeting the needs for modeling bioenergy crops; however, many updates reviewed and suggested to DayCent will be broadly applicable to other systems.

show abstract

“…This paper develops a statistical model (NH 3 _STAT) to predict NH 3 emissions using the physicochemical properties of agricultural soils from different regions and analyzes the spatial distribution of NH 3 emissions from agricultural soils. Our model is different from other models following a metamodeling approach, such as Ramanantenasoa et al () who utilized emission simulations from a process‐based model to generate its NH 3 emissions. Rather, our model uses measured NH 3 emissions from field experiment as the inputs.…”

Section: Introductionmentioning

confidence: 99%

Characterization of the Global Sources of Atmospheric Ammonia from Agricultural Soils

Aneja

Schlesinger

et al. 2020

JGR Atmospheres

View full text Add to dashboard Cite

Global ammonia (NH 3 ) emissions to the atmosphere are projected to increase in the coming years with the increased use of synthetic nitrogen fertilizers and cultivation of nitrogen-fixing crops. A statistical model (NH 3 _STAT) is developed for characterizing atmospheric NH 3 emissions from agricultural soils and compared to the performance of other global and regional NH 3 models (e.g., Emission Database for Global Atmospheric Research, Magnitude and Seasonality of Agricultural Emissions, MIX, and U.S. Environmental Protection Agency). The statistical model was developed from a multiple linear regression between NH 3 emission and the physicochemical variables. The model was evaluated for 2012 NH 3 emissions. The results indicate that, in comparison to other data sets, the model provides a lower global NH 3 estimate by 58%, (NH 3 _STAT: 13.9 Tg N yr −1 ; Emission Database for Global Atmospheric Research: 33.0 Tg N yr −1 ). We also performed a region-based analysis (United States, India, and China) using the NH 3 _STAT model. For the United States, our model produces an estimate that is a~1.4 times higher in comparison to the Environmental Protection Agency. Meanwhile, the NH 3 _STAT estimate for India shows NH 3 emissions between 0.8 and 1.4 times lower when compared to other data sets. A lower estimate is also seen for China, where the model estimates NH 3 emissions 0.4 to 5 times lower than other data sets. The difference in the global estimates is attributed to the lower estimates in major agricultural countries like China and India. The statistical model captures the spatial distribution of global NH 3 emissions by utilizing a simplified approach compared to other readily available data sets. Moreover, the NH 3 _STAT model provides an opportunity to predict future NH 3 emissions in a changing world. − ), and nitrous oxide (N 2 O); and organic nitrogen such as urea, amino acids, and proteins. Many of these are biologically active, chemically reactive, and radiatively active in the Earth's atmosphere and biosphere, in contrast to nonreactive nitrogen gas (N 2 ).

show abstract

Meta-modeling methods for estimating ammonia volatilization from nitrogen fertilizer and manure applications

Cited by 27 publications

References 35 publications

Maize yield and nitrate loss prediction with machine learning algorithms

Maize yield and nitrate loss prediction with machine learning algorithms

21st‐century biogeochemical modeling: Challenges for Century‐based models and where do we go from here?

Characterization of the Global Sources of Atmospheric Ammonia from Agricultural Soils

Contact Info

Product

Resources

About