Data-driven modeling on the global annual soil nitrous oxide emissions: Spatial pattern and attributes

Abstract

Previous assessments generated divergent estimates of global terrestrial soil nitrous oxide (N2O) emission and its spatial distributions, which did not match the observed data well. The objectives of this study were to generate a global map of terrestrial soil N2O emissions based on field observations (n = 5549) and quantify the contribution of different variables for predicting the global variation of N2O emissions. We provided spatially explicit maps of annual soil N2O emission rates across forest, grassland and cropland using the random forest approach. The global mean soil N2O emission rate in our data-driven model was 0.059 ± 0.006 g N m−2 year−1, which was lower than the estimates from previous model ensembles. Soil N2O emissions were higher in the northern than southern hemisphere. The average annual soil N2O emission rate of cropland (0.094 ± 0.009 g N m−2 year−1) was higher than that of forest (0.039 ± 0.004 g N m−2 year−1) and grassland (0.045 ± 0.007 g N m−2 year−1). In addition, we found that soil nitrogen substrates dominated the changes in soil N2O emissions and the relative importance of nitrate, ammonium, and fertilizer in predicting soil N2O emissions was greater than that of mean annual temperature and precipitation. Our data-driven model results implied that previous process-based model may overestimate the global soil N2O emission rates due to limited validation data and incomplete assumptions on related-mechanisms. This study highlights the importance of global field observations in N2O emission estimation, which can provide an independent dataset to constrain previous process-based models for better prediction.