Machine Learning Approach to Simulate Soil CO2 Fluxes under Cropping Systems

With the growing number of datasets to describe greenhouse gas (GHG) emissions, there is an opportunity to develop novel predictive models that require neither the expense nor time required to make direct field measurements. This study evaluates the potential for machine learning (ML) approaches to predict soil GHG emissions without the biogeochemical expertise that is required to use many current models for simulating soil GHGs. There are ample data from field measurements now publicly available to test new modeling approaches. The objective of this paper was to develop and evaluate machine learning (ML) models using field data (soil temperature, soil moisture, soil classification, crop type, fertilization type, and air temperature) available in the Greenhouse gas Reduction through Agricultural Carbon Enhancement network (GRACEnet) database to simulate soil CO2 fluxes with different fertilization methods. Four machine learning algorithms—K nearest neighbor regression (KNN), support vector regression (SVR), random forest (RF) regression, and gradient boosted (GB) regression—were used to develop the models. The GB regression model outperformed all the other models on the training dataset with R2 = 0.88, MAE = 2177.89 g C ha−1 day−1, and RMSE 4405.43 g C ha−1 day−1. However, the RF and GB regression models both performed optimally on the unseen test dataset with R2 = 0.82. Machine learning tools were useful for developing predictors based on soil classification, soil temperature and air temperature when a large database like GRACEnet is available, but these were not highly predictive variables in correlation analysis. This study demonstrates the suitability of using tree-based ML algorithms for predictive modeling of CO2 fluxes, but no biogeochemical processes can be described with such models.

Estimated regional CO₂ flux and uncertainty based on an ensemble of atmospheric CO₂ inversions

Global and regional sources and sinks of carbon across the earth’s surface have been studied extensively using atmospheric carbon dioxide (CO2) observations and chemistry-transport model (ACTM) simulations (top-down/inversion method). However, the uncertainties in the regional flux (+ve: source to the atmosphere; −ve: sink on land/ocean) distributions remain unconstrained mainly due to the lack of sufficient high-quality measurements covering the globe in all seasons and the uncertainties in model simulations. Here, we use a suite of 16 inversion cases, derived from a single transport model (MIROC4-ACTM) but different sets of a priori (bottom-up) terrestrial biosphere and oceanic fluxes, as well as prior flux and observational data uncertainties (50 sites) to estimate CO2 fluxes for 84 regions over the period 2000–2020. The ensemble inversions provide a mean flux field that is consistent with the global CO2 growth rate, land and ocean sink partitioning of −2.9 ± 0.3 (±1σ uncertainty on mean) and −1.6 ± 0.2 PgC yr−1, respectively, for the period 2011–2020 (without riverine export correction), offsetting about 22–33 % and 16–18 % of global fossil-fuel CO2 emissions. Aggregated fluxes for 15 land regions compare reasonably well with the best estimations for (approx. 2000–2009) given by the REgional Carbon Cycle Assessment and Processes (RECCAP), and all regions appeared as a carbon sink over 2011–2020. Interannual variability and seasonal cycle in CO2 fluxes are more consistently derived for different prior fluxes when a greater degree of freedom is given to the inversion system (greater prior flux uncertainty). We have evaluated the inversion fluxes using independent aircraft and surface measurements not used in the inversions, which raises our confidence in the ensemble mean flux rather than an individual inversion. Differences between 5-year mean fluxes show promises and capability to track flux changes under ongoing and future CO2 emission mitigation policies.

Nongrowing Season CO2 Emissions Determine the Distinct Carbon Budgets of Two Alpine Wetlands on the Northeastern Qinghai—Tibet Plateau

Alpine wetlands sequester large amounts of soil carbon, so it is vital to gain a full understanding of their land-atmospheric CO2 exchanges and how they contribute to regional carbon neutrality; such an understanding is currently lacking for the Qinghai—Tibet Plateau (QTP), which is undergoing unprecedented climate warming. We analyzed two-year (2018–2019) continuous CO2 flux data, measured by eddy covariance techniques, to quantify the carbon budgets of two alpine wetlands (Luanhaizi peatland (LHZ) and Xiaobohu swamp (XBH)) on the northeastern QTP. At an 8-day scale, boosted regression tree model-based analysis showed that variations in growing season CO2 fluxes were predominantly determined by atmospheric water vapor, having a relative contribution of more than 65%. Variations in nongrowing season CO2 fluxes were mainly controlled by site (categorical variable) and topsoil temperature (Ts), with cumulative relative contributions of 81.8%. At a monthly scale, structural equation models revealed that net ecosystem CO2 exchange (NEE) at both sites was regulated more by gross primary productivity (GPP), than by ecosystem respiration (RES), which were both in turn directly controlled by atmospheric water vapor. The general linear model showed that variations in nongrowing season CO2 fluxes were significantly (p < 0.001) driven by the main effect of site and Ts. Annually, LHZ acted as a net carbon source, and NEE, GPP, and RES were 41.5 ± 17.8, 631.5 ± 19.4, and 673.0 ± 37.2 g C/(m2 year), respectively. XBH behaved as a net carbon sink, and NEE, GPP, and RES were –40.9 ± 7.5, 595.1 ± 15.4, and 554.2 ± 7.9 g C/(m2 year), respectively. These distinctly different carbon budgets were primarily caused by the nongrowing season RES being approximately twice as large at LHZ (p < 0.001), rather than by other equivalent growing season CO2 fluxes (p > 0.10). Overall, variations in growing season CO2 fluxes were mainly controlled by atmospheric water vapor, while those of the nongrowing season were jointly determined by site attributes and soil temperatures. Our results highlight the different carbon functions of alpine peatland and alpine swampland, and show that nongrowing season CO2 emissions should be taken into full consideration when upscaling regional carbon budgets. Current and predicted marked winter warming will directly stimulate increased CO2 emissions from alpine wetlands, which will positively feedback to climate change.

O sentido dos bankster para o carbono

A negação da ligação entre poluição e mudanças climáticas faz parte de um antigo debate entre cientistas e política. Mas para entender as raízes da questão não é possível transcender de alguns lugares e personagens pertencentes à esfera do patrocínio e da finança global. Isso levou a um esforço para medir e localizar exatamente as emissões de gases de efeito estufa. As medições visavam estudar os fluxos de CO2 entre terra e atmosfera nas diversas proporções e comparar os resultados entre as diferentes áreas, avaliar as influências com fatores climáticos e meteorológicos, avaliar a influência do manejo florestal e dos incêndios florestais. The denial of the link between pollution and climate change is part of an old debate among scientists and politics. But to understand the roots of the issue it is not possible to transcend from a few places and characters belonging to the sphere of global patronage and finance. This led to an effort to measure and precisely track greenhouse gas emissions. The measurements aimed to study the CO2 fluxes between land and atmosphere in various proportions and compare the results between different areas, evaluate the influences with climatic and meteorological factors, evaluate the influence of forest management and forest fires.

New application of low cost sensors for continuous CO₂ flux measurements

Soil CO2 emissions are one of the largest contributions to the global carbon cycle, and a full understanding of processes generating them and how climate change may modify them is needed and still uncertain. Thus, a dense spatial and temporal network of CO2 flux measurements from soil could help reduce uncertainty in the global carbon budgets. In the present study, low cost Air Enquirer kits, including CO2 and environmental parameters sensors, have been designed, built and applied for the first time to design, develop and test a new Steady-State-Through-Flow (SS-TF) chamber for simultaneous measurements of CO2 fluxes in soil and CO2 concentrations in air. Sensor's responses were previously corrected for temperature, relative humidity, illumination and pressure conditions in order to reduce the uncertainty of measured CO2 values and of the following calculated CO2 fluxes. CO2 soil fluxes measured by the proposed SS-TF and by a standard closed Non-Steady-State-Non-Through-Flow (NSS-NTF) chamber were shortly compared. The use of a multi-parametric fitting reduced the total uncertainty of CO2 concentration measurements by 62 % compared with one where only a simple CO2 calibration was applied, and by a 90 % when compared to uncertainty declared by the manufacturer. The new SS-TF system allows continuous measurement of CO2 fluxes and CO2 ambient air with low cost (~1.2 k€), low energy demand (< 5 W) and low maintenance (twice per year due to sensor calibration requirements).

Response of vegetation and carbon fluxes to brown lemming herbivory in Northern Alaska

The Arctic is warming at double the average global rate, affecting the carbon cycle of tundra ecosystems. Most research on carbon fluxes from Arctic tundra ecosystems has focused on abiotic environmental controls (e.g. temperature, rainfall, or radiation). However, Arctic tundra vegetation, and therefore the carbon balance of these ecosystems, can be substantially impacted by herbivory. In this study we tested how vegetation consumption by brown lemmings (Lemmus trimucronatus) can impact carbon exchange of a wet-sedge tundra ecosystem near Utqiaġvik, Alaska during the summer, and the recovery of vegetation during a following summer. We placed brown lemmings in individual enclosure plots and tested the impact of lemmings’ herbivory on carbon dioxide (CO2) and methane (CH4) fluxes and the normalized difference vegetation index (NDVI) immediately after lemming removal and during the following growing season. During the first summer of the experiment, lemmings’ herbivory reduced plant biomass (as shown by the decrease in the NDVI) and decreased CO2 uptake, while not significantly impacting CH4 emissions. Methane emissions were likely not significantly affected due to CH4 being produced deeper in the soil and escaping from the stem bases of the vascular plants. The summer following the lemming treatments, NDVI and CO2 fluxes returned to magnitudes similar to those observed before the start of the experiment, suggesting recovery of the vegetation, and a transitory nature of the impact of lemming herbivory. Overall, lemming herbivory has short-term but substantial effects on carbon sequestration by vegetation and might contribute to the considerable interannual variability in CO2 fluxes from tundra ecosystems.

Global anthropogenic CO₂ emissions and uncertainties as a prior for Earth system modelling and data assimilation

The growth in anthropogenic carbon dioxide (CO2) emissions acts as a major climate change driver, which has widespread implications across society, influencing the scientific, political, and public sectors. For an increased understanding of the CO2 emission sources, patterns, and trends, a link between the emission inventories and observed CO2 concentrations is best established via Earth system modelling and data assimilation. Bringing together the different pieces of the puzzle of a very different nature (measurements, reported statistics, and models), it is of utmost importance to know their level of confidence and boundaries well. Inversions disaggregate the variation in observed atmospheric CO2 concentration to variability in CO2 emissions by constraining the regional distribution of CO2 fluxes, derived either bottom-up from statistics or top-down from observations. The level of confidence and boundaries for each of these CO2 fluxes is as important as their intensity, though often not available for bottom-up anthropogenic CO2 emissions. This study provides a postprocessing tool CHE_UNC_APP for anthropogenic CO2 emissions to help assess and manage the uncertainty in the different emitting sectors. The postprocessor is available under https://doi.org/10.5281/zenodo.5196190 (Choulga et al., 2021). Recommendations are given for regrouping the sectoral emissions, taking into account their uncertainty instead of their statistical origin; for addressing local hot spots; for the treatment of sectors with small budget but uncertainties larger than 100 %; and for the assumptions around the classification of countries based on the quality of their statistical infrastructure. This tool has been applied to the EDGARv4.3.2_FT2015 dataset, resulting in seven input grid maps with upper- and lower-half ranges of uncertainty for the European Centre for Medium-Range Weather Forecasts Integrated Forecasting System. The dataset is documented and available under https://doi.org/10.5281/zenodo.3967439 (Choulga et al., 2020). While the uncertainty in most emission groups remains relatively small (5 %–20 %), the largest contribution (usually over 40 %) to the total uncertainty is determined by the OTHER group (of fuel exploitation and transformation but also agricultural soils and solvents) at the global scale. The uncertainties have been compared for selected countries to those reported in the inventories submitted to the United Nations Framework Convention on Climate Change and to those assessed for the European emission grid maps of the Netherlands Organisation for Applied Scientific Research. Several sensitivity experiments are performed to check (1) the country dependence (by analysing the impact of assuming either a well- or less well-developed statistical infrastructure), (2) the fuel type dependence (by adding explicit information for each fuel type used per activity from the Intergovernmental Panel on Climate Change), and (3) the spatial source distribution dependence (by aggregating all emission sources and comparing the effect against an even redistribution over the country). The first experiment shows that the SETTLEMENTS group (of energy for buildings) uncertainty changes the most when development level is changed. The second experiment shows that fuel-specific information reduces uncertainty in emissions only when a country uses several different fuels in the same amount; when a country mainly uses the most globally typical fuel for an activity, uncertainty values computed with and without detailed fuel information are the same. The third experiment highlights the importance of spatial mapping.