Spatial Prediction of N2O Emissions in Pasture: A Bayesian Model Averaging Analysis

Xiaodong Huang, Peter Grace, Wenbiao Hu, David Rowlings, Kerrie MengersenPLoS 2013 · DOI

Abstract

Nitrous oxide ($N_2O$) is a significant greenhouse gas contributing to global warming, yet its spatial variability often leads to large uncertainties in prediction. Traditional studies frequently ignore spatial dependency when quantifying the relationship between $N_2O$ and environmental factors. This study assesses the impact of three spatial correlation structures—independence, distance-based (EXP), and neighborhood-based (CAR)—on spatial predictions of $N_2O$ emissions. Using replicated, irregular point-referenced data from a pasture in southeast Australia, we calibrated spatial predictions using Bayesian Model Averaging (BMA). Our results demonstrate that incorporating spatial dependency substantially improves prediction accuracy and better quantifies parameter uncertainties. The hybrid model structure obtained via BMA provided the most accurate spatial prediction across the study region.

Introduction: The Challenge of "Patchy" Emissions

Nitrous oxide ($N_2O$) is a potent greenhouse gas, with agricultural soils serving as a primary source. A major hurdle in managing these emissions is their inherent "patchiness"—high spatial variability even within small field plots. Conventional modeling often relies on simple linear regression, which assumes that emission measurements are independent of their location. However, soil properties like moisture, temperature, and nutrient levels are spatially correlated. Ignoring these dependencies can lead to biased estimates and "oversmoothed" maps that fail to identify critical emission "hotspots."

Methodology and Data

This research employed a hierarchical Bayesian framework to compare three distinct spatial structures:

  1. Independent Model: A baseline linear regression assuming independent errors.
  2. Geostatistical (EXP) Model: A distance-based model where spatial correlation decays as a function of physical distance.
  3. Conditional Autoregressive (CAR) Model: A neighborhood-based model. Given the irregular spacing of sampling sites, Thiessen polygons were used to define neighborhood relationships.

To address model uncertainty, the researchers utilized Bayesian Model Averaging (BMA). Instead of selecting a single "best" model, BMA assigns weights to each based on the Deviance Information Criterion (DIC), creating a robust hybrid prediction.

The study was conducted on a 271 $m^2$ subtropical pasture in southeast Australia. Data were collected from 17 randomly placed chambers over a year (October 2007 – September 2008). Seven environmental covariates were tracked, including soil moisture, temperature, $NO_3^-$ concentration, pH, and soil texture.

Results: Comparing Spatial Predictions

The analysis revealed that spatial models significantly outperform independent regression. The CAR model provided the best fit (DIC = 745.75), followed closely by the EXP model (DIC = 746.1), while the independent model performed poorly (DIC = 786.69).

Comparison of spatial prediction maps for N2O emissions using different modeling approaches: Observed data, CAR, EXP, Independent, and the BMA hybrid.

As shown in the comparison above, the independent structure fails to capture the spatial patterns observed in the field. The CAR model was particularly effective at identifying extreme hotspots, whereas the EXP model tended to produce smoother transitions. Through BMA, the CAR model received a posterior probability of 54.3% and the EXP model 45.7%, while the independent model was effectively discounted (0%).

Key Insights

  • Spatial Dependency is Critical: Incorporating spatial correlation reduced the sum of squared residuals and improved the quantification of uncertainty for soil parameters.
  • Environmental Drivers: Soil moisture and temperature were confirmed as primary positive drivers. Interestingly, a negative relationship with $NO_3^-$ suggested that nitrification was the dominant pathway for $N_2O$ production in this specific environment.
  • BMA Calibration: The BMA approach provided a "calibrated" map that more closely mirrored observed field data than any individual model, effectively balancing the strengths of neighborhood and distance-based approaches.

Conclusion and Future Directions

While the study successfully demonstrated the power of BMA in spatial prediction, some limitations remain. The reliance on 17 chambers provides a starting point, but larger-scale sensor networks could offer higher resolution. Additionally, while this work focused on spatial dynamics, $N_2O$ emissions are also highly variable over time. Future research integrating spatio-temporal BMA could further refine our ability to predict and mitigate greenhouse gas emissions in agricultural landscapes.

Summary updated

Abstract

Nitrous oxide (N2O) is one of the greenhouse gases that can contribute to global warming. Spatial variability of N2O can lead to large uncertainties in prediction. However, previous studies have often ignored the spatial dependency to quantify the N2O - environmental factors relationships. Few researches have examined the impacts of various spatial correlation structures (e.g. independence, distance-based and neighbourhood based) on spatial prediction of N2O emissions. This study aimed to assess the impact of three spatial correlation structures on spatial predictions and calibrate the spatial prediction using Bayesian model averaging (BMA) based on replicated, irregular point-referenced data. The data were measured in 17 chambers randomly placed across a 271 m(2) field between October 2007 and September 2008 in the southeast of Australia. We used a Bayesian geostatistical model and a Bayesian spatial conditional autoregressive (CAR) model to investigate and accommodate spatial dependency, and to estimate the effects of environmental variables on N2O emissions across the study site. We compared these with a Bayesian regression model with independent errors. The three approaches resulted in different derived maps of spatial prediction of N2O emissions. We found that incorporating spatial dependency in the model not only substantially improved predictions of N2O emission from soil, but also better quantified uncertainties of soil parameters in the study. The hybrid model structure obtained by BMA improved the accuracy of spatial prediction of N2O emissions across this study region.
BibTeX
@inproceedings{huang2013spatial,
  author = {Xiaodong Huang and Peter Grace and Wenbiao Hu and David Rowlings and Kerrie Mengersen},
  title = {{Spatial Prediction of N2O Emissions in Pasture: A Bayesian Model Averaging Analysis}},
  booktitle = {PLoS},
  year = {2013},
  doi = {10.1371/journal.pone.0065039},
}
Figure