Search

Agricultural nitrous oxide (N2O) emission accounts for a non-trivial fraction of global greenhouse gas (GHG) budget. To date, estimating N2O fluxes from cropland remains a challenging task because the related microbial processes (e.g., nitrification and denitrification) are controlled by complex interactions among climate, soil, plant and human activities. Existing approaches such as process-based (PB) models have well-known limitations due to insufficient representations of the processes or uncertainties of model parameters, and due to leverage recent advances in machine learning (ML) a new method is needed to unlock the “black box” to overcome its limitations such as low interpretability, out-of-sample failure and massive data demand. In this study, we developed a first-of-its-kind knowledge-guided machine learning model for agroecosystems (KGML-ag) by incorporating biogeophysical and chemical domain knowledge from an advanced PB model, ecosys, and tested it by comparing simulating daily N2O fluxes with real observed data from mesocosm experiments. The gated recurrent unit (GRU) was used as the basis to build the model structure. To optimize the model performance, we have investigated a range of ideas, including (1) using initial values of intermediate variables (IMVs) instead of time series as model input to reduce data demand; (2) building hierarchical structures to explicitly estimate IMVs for further N2O prediction; (3) using multi-task learning to balance the simultaneous training on multiple variables; and (4) pre-training with millions of synthetic data generated from ecosys and fine-tuning with mesocosm observations. Six other pure ML models were developed using the same mesocosm data to serve as the benchmark for the KGML-ag model. Results show that KGML-ag did an excellent job in reproducing the mesocosm N2O fluxes (overall r2=0.81, and RMSE=3.6 mgNm-2d-1 from cross validation). Importantly, KGML-ag always outperforms the PB model and ML models in predicting N2O fluxes, especially for complex temporal dynamics and emission peaks. Besides, KGML-ag goes beyond the pure ML models by providing more interpretable predictions as well as pinpointing desired new knowledge and data to further empower the current KGML-ag. We believe the KGML-ag development in this study will stimulate a new body of research on interpretable ML for biogeochemistry and other related geoscience processes.

The atmospheric carbon dioxide (CO2) mixing ratio and its carbon isotope (δ13C-CO2) composition contain important CO2 sink and source information spanning from ecosystem to global scales. The observation and simulation for both CO2 and δ13C-CO2 can be used to constrain regional emissions and better understand the anthropogenic and natural mechanisms that control δ13C-CO2 variations. Such work remains rare for urban environments, especially megacities. Here, we used near-continuous CO2 and δ13C-CO2 measurements, from September 2013 to August 2015, and inverse modeling to constrain the CO2 budget and investigate the main factors that dominated δ13C-CO2 variations for the Yangtze River delta (YRD) region, one of the largest anthropogenic CO2 hotspots and densely populated regions in China. We used the WRF-STILT model framework with category-specified EDGAR v4.3.2 CO2 inventories to simulate hourly CO2 mixing ratios and δ13C-CO2, evaluated these simulations with observations, and constrained the total anthropogenic CO2 emission. We show that (1) top-down and bottom-up estimates of anthropogenic CO2 emissions agreed well (bias < 6 %) on an annual basis, (2) the WRF-STILT model can generally reproduce the observed diel and seasonal atmospheric δ13C-CO2 variations, and (3) anthropogenic CO2 emissions played a much larger role than ecosystems in controlling the δ13C-CO2 seasonality. When excluding ecosystem respiration and photosynthetic discrimination in the YRD area, δ13C-CO2 seasonality increased from 1.53 ‰ to 1.66 ‰. (4) Atmospheric transport processes in summer amplified the cement CO2 enhancement proportions in the YRD area, which dominated monthly δs (the mixture of δ13C-CO2 from all regional end-members) variations. These findings show that the combination of long-term atmospheric carbon isotope observations and inverse modeling can provide a powerful constraint on the carbon cycle of these complex megacities.

The atmospheric CO2 mixing ratio and its δ13C-CO2 composition contain important CO2 sink and source information spanning from ecosystem to global scales. The observation and simulation for both CO2 and its carbon isotope ratio (δ13C-CO2) can be used to constrain regional emissions and better understand the anthropogenic and natural mechanisms that control δ13C-CO2 variations. Such work remains rare for urban environments, especially megacities. Here, we used near-continuous CO2 and δ13C-CO2 measurements, from September 2013 to August 2015, and inverse modeling to constrain the CO2 budget and investigate the main factors that dominated δ13C-CO2 variations for the Yangtze River Delta (YRD) region, one of the largest anthropogenic CO2 hotspots and densely populated regions in China. We used the WRF-STILT model framework with category-specified EDGAR v432 CO2 inventories to simulate hourly CO2 mixing ratios and δ13C-CO2, evaluated these simulations with observations, and constrained the anthropogenic CO2 emission categories. Our study shows that: (1) Top-down and bottom-up estimates of anthropogenic CO2 emissions agreed well (bias δ13C-CO2 variations; (3) Anthropogenic CO2 emissions played a much larger role than ecosystems in controlling the δ13C-CO2 seasonality. When excluding ecosystem respiration and photosynthetic discrimination in the YRD area, δ13C-CO2 seasonality increased from 1.53 ‰ to 1.66 ‰; (4) Atmospheric transport processes in summer amplified the cement CO2 enhancement proportions in the YRD area, which dominated monthly δs variations. These findings support that the combination of long-term atmospheric carbon isotope observations and inverse modeling can provide a powerful constraint on the carbon cycle of these complex megacities.

Terrestrial ecosystems are simultaneously the largest source and a major sink of volatile organic compounds (VOCs) to the global atmosphere, and these two-way fluxes are an important source of uncertainty in current models. Here, we apply high-resolution mass spectrometry (proton transfer reaction-quadrupole interface time-of-flight; PTR-QiTOF) to measure ecosystem-atmosphere VOC fluxes across the entire detected mass range (m/z 0-335) over a mixed temperate forest and use the results to test how well a state-of-science chemical transport model (GEOS-Chem CTM) is able to represent the observed reactive carbon exchange. We show that ambient humidity fluctuations can give rise to spurious VOC fluxes with PTR-based techniques and present a method to screen for such effects. After doing so, 377 of the 636 detected ions exhibited detectable gross fluxes during the study, implying a large number of species with active ecosystem-atmosphere exchange. We introduce the reactivity flux as a measure of how Earth-atmosphere fluxes influence ambient OH reactivity and show that the upward total VOC (-VOC) carbon and reactivity fluxes are carried by a far smaller number of species than the downward fluxes. The model underpredicts the -VOC carbon and reactivity fluxes by 40-60% on average. However, the observed net fluxes are dominated (90% on a carbon basis, 95% on a reactivity basis) by known VOCs explicitly included in the CTM. As a result, the largest CTM uncertainties in simulating VOC carbon and reactivity exchange for this environment are associated with known rather than unrepresented species. This conclusion pertains to the set of species detectable by PTR-TOF techniques, which likely represents the majority in terms of carbon mass and OH reactivity, but not necessarily in terms of aerosol formation potential. In the case of oxygenated VOCs, the model severely underpredicts the gross fluxes and the net exchange. Here, unrepresented VOCs play a larger role, accounting for ∼30% of the carbon flux and ∼50% of the reactivity flux. The resulting CTM biases, however, are still smaller than those that arise from uncertainties for known and represented compounds.

We present top-down constraints on global monthly N2O emissions for 2011 from a multi-inversion approach and an ensemble of surface observations. The inversions employ the GEOS-Chem adjoint and an array of aggregation strategies to test how well current observations can constrain the spatial distribution of global N2O emissions. The strategies include (1) a standard 4D-Var inversion at native model resolution (4° × 5°), (2) an inversion for six continental and three ocean regions, and (3) a fast 4D-Var inversion based on a novel dimension reduction technique employing randomized singular value decomposition (SVD). The optimized global flux ranges from 15.9 TgNyr-1 (SVD-based inversion) to 17.5-17.7 TgNyr-1 (continental-scale, standard 4D-Var inversions), with the former better capturing the extratropical N2O background measured during the HIAPER Pole-to-Pole Observations (HIPPO) airborne campaigns. We find that the tropics provide a greater contribution to the global N2O flux than is predicted by the prior bottom-up inventories, likely due to underestimated agricultural and oceanic emissions. We infer an overestimate of natural soil emissions in the extratropics and find that predicted emissions are seasonally biased in northern midlatitudes. Here, optimized fluxes exhibit a springtime peak consistent with the timing of spring fertilizer and manure application, soil thawing, and elevated soil moisture. Finally, the inversions reveal a major emission underestimate in the US Corn Belt in the bottom-up inventory used here. We extensively test the impact of initial conditions on the analysis and recommend formally optimizing the initial N2O distribution to avoid biasing the inferred fluxes. We find that the SVD-based approach provides a powerful framework for deriving emission information from N2O observations: by defining the optimal resolution of the solution based on the information content of the inversion, it provides spatial information that is lost when aggregating to political or geographic regions, while also providing more temporal information than a standard 4D-Var inversion.

We present top-down constraints on global monthly N[subscript 2]O emissions for 2011 from a multi-inversion approach and an ensemble of surface observations. The inversions employ the GEOS-Chem adjoint and an array of aggregation strategies to test how well current observations can constrain the spatial distribution of global N[subscript 2]O emissions. The strategies include (1) a standard 4D-Var inversion at native model resolution (4° × 5°), (2) an inversion for six continental and three ocean regions, and (3) a fast 4D-Var inversion based on a novel dimension reduction technique employing randomized singular value decomposition (SVD). The optimized global flux ranges from 15.9 Tg N yr[superscript −1] (SVD-based inversion) to 17.5–17.7 Tg N yr[superscript −1] (continental-scale, standard 4D-Var inversions), with the former better capturing the extratropical N[subscript 2]O background measured during the HIAPER Pole-to-Pole Observations (HIPPO) airborne campaigns. We find that the tropics provide a greater contribution to the global N[superscript 2]O flux than is predicted by the prior bottom-up inventories, likely due to underestimated agricultural and oceanic emissions. We infer an overestimate of natural soil emissions in the extratropics and find that predicted emissions are seasonally biased in northern midlatitudes. Here, optimized fluxes exhibit a springtime peak consistent with the timing of spring fertilizer and manure application, soil thawing, and elevated soil moisture. Finally, the inversions reveal a major emission underestimate in the US Corn Belt in the bottom-up inventory used here. We extensively test the impact of initial conditions on the analysis and recommend formally optimizing the initial N[superscript 2]O distribution to avoid biasing the inferred fluxes. We find that the SVD-based approach provides a powerful framework for deriving emission information from N2O observations: by defining the optimal resolution of the solution based on the information content of the inversion, it provides spatial information that is lost when aggregating to political or geographic regions, while also providing more temporal information than a standard 4D-Var inversion., United States. National Oceanic and Atmospheric Administration (Grant NA13OAR4310086), United States. National Oceanic and Atmospheric Administration (Grant NA13OAR4310081)

An important aspect that determines the ability of crop growth models to predict growth and yield is their ability to predict the rate of water consumption or evapotranspiration (ET) of the crop, especially for rain-fed crops. If, for example, the predicted ET rate is too high, the simulated crop may exhaust its soil water supply before the next rain event, thereby causing growth and yield predictions that are too low. In a prior inter-comparison among maize growth models, ET predictions varied widely, but no observations of actual ET were available for comparison. Therefore, another study has been initiated under the umbrella of AgMIP (Agricultural Model Inter-Comparison and Improvement Project). This time observations of ET using the eddy covariance technique from an 8-year-long experiment conducted at Ames, IA are being used as the standard. Simulation results from 29 models have been completed. In the first “blind” phase for which only weather, soils, and management information were furnished to the modelers, estimates of seasonal ET varied from about 200 to about 700 mm. A detailed statistical analysis of the daily ET data from 2011, a “typical” rainfall year, showed that, as expected, the median of all the models was more accurate across several criteria (correlation, root mean square error, average difference, regression slope) than any particular model. However, some individual models were better than the median for a particular criteria. Predictions improved somewhat in later stages when the modelers were provided additional leaf area and growth information that allowed them to “calibrate” some of the parameters in their models to account for varietal characteristics, etc.