scholarly journals Improving the prediction of an atmospheric chemistry transport model using gradient boosted regression trees

2019 ◽  
Author(s):  
Peter D. Ivatt ◽  
Mathew J. Evans
Keyword(s):  
Atmospheric Chemistry ◽  
Transport Model ◽  
Boosted Regression Trees ◽  
Training Data ◽  
Model Bias ◽  
Chemistry Transport Model ◽  
Pearson's R ◽  
The Impact ◽  
Process Based Models ◽  
Pearson’S R

Abstract. Predictions from process-based models of environmental systems are biased, due to uncertainties in their inputs and parameterisations, reducing their utility. We develop a predictor for the bias in tropospheric ozone (a key pollutant) calculated by an atmospheric chemistry transport model (GEOS-Chem), based on outputs from the model and observations of ozone from both the surface (EPA, EMEP and GAW) and the ozone-sonde networks. We train a gradient-boosted decision tree algorithm (XGBoost) to predict model bias, with model and observational data for 2010–2015, and then test the approach using the years 2016–2017. We show that the bias-corrected model performs significantly better than the uncorrected model. The root mean square error is reduced from from 16.21 ppb to 7.48 ppb, the normalised mean bias is reduced from 0.28 to −0.04, and the Pearson's R is increased from 0.479 to 0.841. Comparisons with observations from the NASA ATom flights (which were not included in the training) also show improvements but to a smaller extent reducing the RMSE from 12.11 ppb to 10.50 ppb, the NMB from 0.08 to 0.06 and increasing the Pearson's R from 0.761 to 0.792. We attribute the smaller improvements to the lack of routine observational constraints of the remote troposphere. We explore the choice of predictor (bias prediction versus direct prediction) and conclude both may have utility. We show that the method is robust to variations in the volume of training data, with approximately a year of data needed to produce useful performance. Data denial experiments (removing observational sites from the algorithm training) shows that information from one location (for example Europe) can reduce the model bias over other locations (for example North America) which might provide insights into the processes controlling the model bias. We conclude that combining machine learning approaches with process based models may provide a useful tool for improving performance of air quality forecasts or to provide enhanced assessments of the impact of pollutants on human and ecosystem health, and may have utility in other environmental applications.

scholarly journals Improving the prediction of an atmospheric chemistry transport model using gradient-boosted regression trees

2020 ◽  
Vol 20 (13) ◽  
pp. 8063-8082 ◽  
Author(s):  
Peter D. Ivatt ◽  
Mathew J. Evans
Keyword(s):  
Root Mean Square Error ◽  
Mean Square Error ◽  
Root Mean Square ◽  
Atmospheric Chemistry ◽  
Transport Model ◽  
Mean Square ◽  
Model Bias ◽  
Chemistry Transport Model ◽  
Pearson's R ◽  
Process Based Models

Abstract. Predictions from process-based models of environmental systems are biased, due to uncertainties in their inputs and parameterizations, reducing their utility. We develop a predictor for the bias in tropospheric ozone (O3, a key pollutant) calculated by an atmospheric chemistry transport model (GEOS-Chem), based on outputs from the model and observations of ozone from both the surface (EPA, EMEP, and GAW) and the ozone-sonde networks. We train a gradient-boosted decision tree algorithm (XGBoost) to predict model bias (model divided by observation), with model and observational data for 2010–2015, and then we test the approach using the years 2016–2017. We show that the bias-corrected model performs considerably better than the uncorrected model. The root-mean-square error is reduced from 16.2 to 7.5 ppb, the normalized mean bias is reduced from 0.28 to −0.04, and Pearson's R is increased from 0.48 to 0.84. Comparisons with observations from the NASA ATom flights (which were not included in the training) also show improvements but to a smaller extent, reducing the root-mean-square error (RMSE) from 12.1 to 10.5 ppb, reducing the normalized mean bias (NMB) from 0.08 to 0.06, and increasing Pearson's R from 0.76 to 0.79. We attribute the smaller improvements to the lack of routine observational constraints for much of the remote troposphere. We show that the method is robust to variations in the volume of training data, with approximately a year of data needed to produce useful performance. Data denial experiments (removing observational sites from the algorithm training) show that information from one location (for example Europe) can reduce the model bias over other locations (for example North America) which might provide insights into the processes controlling the model bias. We explore the choice of predictor (bias prediction versus direct prediction) and conclude both may have utility. We conclude that combining machine learning approaches with process-based models may provide a useful tool for improving these models.

scholarly journals Impact of uncertainties in inorganic chemical rate constants on tropospheric composition and ozone radiative forcing

2017 ◽  
Vol 17 (23) ◽  
pp. 14333-14352 ◽  
Author(s):  
Ben Newsome ◽  
Mat Evans
Keyword(s):  
Rate Constant ◽  
Atmospheric Chemistry ◽  
Rate Constants ◽  
Tropospheric Ozone ◽  
Radiative Forcing ◽  
Transport Model ◽  
Surface Ozone ◽  
Chemistry Transport Model ◽  
The Impact ◽  
Chemical Rate

Abstract. Chemical rate constants determine the composition of the atmosphere and how this composition has changed over time. They are central to our understanding of climate change and air quality degradation. Atmospheric chemistry models, whether online or offline, box, regional or global, use these rate constants. Expert panels evaluate laboratory measurements, making recommendations for the rate constants that should be used. This results in very similar or identical rate constants being used by all models. The inherent uncertainties in these recommendations are, in general, therefore ignored. We explore the impact of these uncertainties on the composition of the troposphere using the GEOS-Chem chemistry transport model. Based on the Jet Propulsion Laboratory (JPL) and International Union of Pure and Applied Chemistry (IUPAC) evaluations we assess the influence of 50 mainly inorganic rate constants and 10 photolysis rates on tropospheric composition through the use of the GEOS-Chem chemistry transport model. We assess the impact on four standard metrics: annual mean tropospheric ozone burden, surface ozone and tropospheric OH concentrations, and tropospheric methane lifetime. Uncertainty in the rate constants for NO2 + OH →M  HNO3 and O3 + NO  →  NO2 + O2 are the two largest sources of uncertainty in these metrics. The absolute magnitude of the change in the metrics is similar if rate constants are increased or decreased by their σ values. We investigate two methods of assessing these uncertainties, addition in quadrature and a Monte Carlo approach, and conclude they give similar outcomes. Combining the uncertainties across the 60 reactions gives overall uncertainties on the annual mean tropospheric ozone burden, surface ozone and tropospheric OH concentrations, and tropospheric methane lifetime of 10, 11, 16 and 16 %, respectively. These are larger than the spread between models in recent model intercomparisons. Remote regions such as the tropics, poles and upper troposphere are most uncertain. This chemical uncertainty is sufficiently large to suggest that rate constant uncertainty should be considered alongside other processes when model results disagree with measurement. Calculations for the pre-industrial simulation allow a tropospheric ozone radiative forcing to be calculated of 0.412 ± 0.062 W m−2. This uncertainty (13 %) is comparable to the inter-model spread in ozone radiative forcing found in previous model–model intercomparison studies where the rate constants used in the models are all identical or very similar. Thus, the uncertainty of tropospheric ozone radiative forcing should expanded to include this additional source of uncertainty. These rate constant uncertainties are significant and suggest that refinement of supposedly well-known chemical rate constants should be considered alongside other improvements to enhance our understanding of atmospheric processes.

scholarly journals The effects of lightning-produced NO<sub>x</sub> and its vertical distribution on atmospheric chemistry: sensitivity simulations with MATCH-MPIC

2005 ◽  
Vol 5 (7) ◽  
pp. 1815-1834 ◽  
Author(s):  
L. J. Labrador ◽  
R. von Kuhlmann ◽  
M. G. Lawrence
Keyword(s):  
Vertical Distribution ◽  
Atmospheric Chemistry ◽  
Large Scale ◽  
Transport Model ◽  
Chemistry Transport Model ◽  
Model Match ◽  
Circulation Patterns ◽  
Range Transport ◽  
Best Fitting ◽  
The Impact

Abstract. The impact of different assumptions concerning the source magnitude as well as the vertical placement of lightning-produced nitrogen oxides is studied using the global chemistry transport model MATCH-MPIC. The responses of NOx, O3, OH, HNO3 and peroxyacetyl-nitrate (PAN) are investigated. A marked sensitivity to both parameters is found. NOx burdens globally can be enhanced by up to 100% depending on the vertical placement and source magnitude strength. In all cases, the largest enhancements occur in the tropical upper troposphere, where lifetimes of most trace gases are longer and where they thus become more susceptible to long-range transport by large-scale circulation patterns. Comparison with observations indicate that 0 and 20 Tg(N)/yr production rates of NOx from lightning are too low and too high, respectively. However, no single intermediate production rate or vertical distribution can be singled out as best fitting the observations, due to the large scatter in the datasets. This underscores the need for further measurement campaigns in key regions, such as the tropical continents.

scholarly journals The effects of lightning-produced NO<sub>x</sub> and its vertical distribution on atmospheric chemistry: sensitivity simulations with MATCH-MPIC

2004 ◽  
Vol 4 (5) ◽  
pp. 6239-6281 ◽  
Author(s):  
L. J. Labrador ◽  
R. von Kuhlmann ◽  
M. G. Lawrence
Keyword(s):  
Vertical Distribution ◽  
Long Range ◽  
Atmospheric Chemistry ◽  
Transport Model ◽  
Chemistry Transport Model ◽  
Model Match ◽  
Circulation Patterns ◽  
Range Transport ◽  
Best Fitting ◽  
The Impact

Abstract. The impact of different assumptions concerning the source magnitude as well as the vertical placement of lightning-produced nitrogen oxides is studied using the global chemistry transport model MATCH-MPIC. The responses of NOx, O3, OH, HNO3 and peroxyacetyl-nitrate (PAN) are investigated. A marked sensitivity to both parameters was found. NOx burdens globally can be enhanced up to 100% depending on the vertical placement and source magnitude strength. In all cases, the largest enhancements occur in the tropical upper troposphere, where lifetimes of most trace gases are longer and where they thus become more susceptible to long-range transport by long-range circulation patterns. Comparison with observations indicate that the 0 and 20 Tg/yr(N) production rates of NOx from lightning are too low and too high a source magnitude, respectively. However, no single intermediate production rate or vertical distribution can be singled out as best fitting the observations due to the large scatter in the datasets. This underscores the need for further measurement campaigns in key regions. The vertical profiles of Pickering et al. (1998) have been implemented in MATCH-MPIC for this study.

scholarly journals Impact of uncertainties in inorganic chemical rate constants on tropospheric composition and ozone radiative forcing

2017 ◽  
Author(s):  
Ben Newsome ◽  
Mat Evans
Keyword(s):  
Rate Constant ◽  
Atmospheric Chemistry ◽  
Rate Constants ◽  
Tropospheric Ozone ◽  
Radiative Forcing ◽  
Transport Model ◽  
Surface Ozone ◽  
Photolysis Rates ◽  
The Impact ◽  
Chemical Rate

Abstract. Chemical rate constants determine the composition of the atmosphere and how this composition has changed over time. They are central to our understanding of climate change and air quality degradation. Atmospheric chemistry models, whether online or offline, box, regional or global use these rate constants. Expert panels synthesise laboratory measurements, making recommendations for the rate constants that should be used. This results in very similar or identical rate constants being used by all models. The inherent uncertainties in these recommendations are, in general, therefore ignored. We explore the impact of these uncertainties on the composition of the troposphere using the GEOS-Chem chemistry transport model. Based on the JPL and IUPAC evaluations we assess 50 mainly inorganic rate constants and 10 photolysis rates, through simulations where we increase the rate of the reactions to the 1σ upper value recommended by the expert panels. We assess the impact on 4 standard metrics: annual mean tropospheric ozone burden, surface ozone and tropospheric OH concentrations, and tropospheric methane lifetime. Uncertainty in the rate constants for NO2 + OH    M →  HNO3, OH + CH4 → CH3O2 + H2O and O3 + NO → NO2 + O2 are the three largest source of uncertainty in these metrics. We investigate two methods of assessing these uncertainties, addition in quadrature and a Monte Carlo approach, and conclude they give similar outcomes. Combining the uncertainties across the 60 reactions, gives overall uncertainties on the annual mean tropospheric ozone burden, surface ozone and tropospheric OH concentrations, and tropospheric methane lifetime of 11, 12, 17 and 17 % respectively. These are larger than the spread between models in recent model inter-comparisons. Remote regions such as the tropics, poles, and upper troposphere are most uncertain. This chemical uncertainty is sufficiently large to suggest that rate constant uncertainty should be considered when model results disagree with measurement. Calculations for the pre-industrial allow a tropospheric ozone radiative forcing to be calculated of 0.412 ± 0.062 Wm−2. This uncertainty (15 %) is comparable to the inter-model spread in ozone radiative forcing found in previous model-model inter-comparison studies where the rate constants used in the models are all identical or very similar. Thus the uncertainty of tropospheric ozone radiative forcing should expanded to include this additional source of uncertainty. These rate constant uncertainties are significant and suggest that refinement of supposedly well known chemical rate constants should be considered alongside other improvements to enhance our understanding of atmospheric processes.

scholarly journals Model – TCCON comparisons of column-averaged methane with a focus on the stratosphere

2016 ◽  
Author(s):  
Andreas Ostler ◽  
Ralf Sussmann ◽  
Prabir K. Patra ◽  
Sander Houweling ◽  
Marko De Bruine ◽  
...  
Keyword(s):  
Atmospheric Chemistry ◽  
Transport Model ◽  
Meteorological Data ◽  
Michelson Interferometer ◽  
Total Carbon ◽  
Data Sets ◽  
Transport Models ◽  
Improved Model ◽  
The Impact ◽  
Chemistry Experiment

Abstract. The distribution of methane (CH4) in the stratosphere can be a major driver of spatial variability in the dry-air column-averaged CH4 mixing ratio (XCH4), which is being measured increasingly for the assessment of CH4 surface emissions. Chemistry-transport models (CTMs) therefore need to simulate the tropospheric and stratospheric fractional columns of XCH4 accurately for estimating surface emissions from XCH4. Simulations from three CTMs are tested against XCH4 observations from the Total Carbon Column Network (TCCON). We analyze how the model-TCCON agreement in XCH4 depends on the model representation of stratospheric CH4 distributions. Model equivalents of TCCON XCH4 are computed with stratospheric CH4 fields from both the model simulations and from satellite-based CH4 distributions from MIPAS (Michelson Interferometer for Passive Atmospheric Sounding) and MIPAS CH4 fields adjusted to ACE-FTS (Atmospheric Chemistry Experiment Fourier Transform Spectrometer) observations. In comparison to simulated model fields we find an improved model-TCCON XCH4 agreement for all models with MIPAS-based stratospheric CH4 fields. For the Atmospheric Chemistry Transport Model (ACTM) the average XCH4 bias is significantly reduced from 38.1 ppb to 13.7 ppb, whereas small improvements are found for the models TM5 (Transport Model, version 5; from 8.7 ppb to 4.3 ppb), and LMDz (Laboratoire de Météorologie Dynamique model with Zooming capability; from 6.8 ppb to 4.3 ppb), respectively. MIPAS stratospheric CH4 fields adjusted to ACE-FTS reduce the average XCH4 bias for ACTM (3.3 ppb), but increase the average XCH4 bias for TM5 (10.8 ppb) and LMDz (20.0 ppb). These findings imply that the range of satellite-based stratospheric CH4 is insufficient to resolve a possible stratospheric contribution to differences in total column CH4 between TCCON and TM5 or LMDz. Applying transport diagnostics to the models indicates that model-to-model differences in the simulation of stratospheric transport, notably the age of stratospheric air, can largely explain the inter-model spread in stratospheric CH4 and, hence, its contribution to XCH4. This implies that there is a need to better understand the impact of individual model transport components (e.g., physical parameterization, meteorological data sets, model horizontal/vertical resolution) on modeled stratospheric CH4.

scholarly journals HETEROFOR 1.0: a spatially explicit model for exploring the response of structurally complex forests to uncertain future conditions. II. Phenology and water cycle

2019 ◽  
Author(s):  
Louis de Wergifosse ◽  
Frédéric André ◽  
Nicolas Beudez ◽  
François de Coligny ◽  
Hugues Goosse ◽  
...  
Keyword(s):  
Climate Change ◽  
Water Balance ◽  
Tree Growth ◽  
Leaf Development ◽  
Forest Growth ◽  
Spatially Explicit ◽  
Model Assessment ◽  
Pearson's R ◽  
The Impact ◽  
Pearson’S R

Abstract. Climate change affects forest growth in numerous and sometimes opposite ways and the resulting trend is often difficult to predict for a given site. Integrating and structuring the knowledge gained from the monitoring and experimental studies into process-based models is an interesting approach to predict the response of forest ecosystems to climate change. While the first generation of such models operates at stand level, we need now individual-based and spatially-explicit approaches in order to account for structurally complex stands whose importance is increasingly recognized in the changing environment context. Among the climate-sensitive drivers of forest growth, phenology and water availability are often cited as crucial elements. They influence, for example, the length of the vegetation period during which photosynthesis takes place and the stomata opening, which determines the photosynthesis rate. In this paper, we describe the phenology and water balance modules integrated in the tree growth model HETEROFOR and evaluate them on six Belgian sites. More precisely, we assess the ability of the model to reproduce key phenological processes (budburst, leaf development, yellowing and fall) as well as water fluxes. Three variants are used to predict budburst (Uniforc, Unichill and Sequential), which differ regarding the inclusion of chilling and/or forcing periods and the calculation of the coldness or heat accumulation. Among the three, the Sequential approach is the least biased (overestimation of 2.46 days) while Uniforc (chilling not considered) best accounts for the interannual variability (Pearson’s R = 0.68). For the leaf development, yellowing and fall, predictions and observation are in accordance. Regarding the water balance module, the predicted throughfall is also in close agreement with the measurements (Pearson’s R = 0.856, bias = −1.3 %) and the soil water dynamics across the year is well-reproduced for all the study sites (Pearson’s R comprised between 0.893 and 0.950, and bias between −1.81 and −9.33 %). The positive results from the model assessment will allow us to use it reliably in projection studies to evaluate the impact of climate change on tree growth and test how diverse forestry practices can adapt forests to these changes.

scholarly journals The influence of biogenic emissions on upper-tropospheric methanol as revealed from space

2007 ◽  
Vol 7 (24) ◽  
pp. 6119-6129 ◽  
Author(s):  
G. Dufour ◽  
S. Szopa ◽  
D. A. Hauglustaine ◽  
C. D. Boone ◽  
C. P. Rinsland ◽  
...  
Keyword(s):  
Northern Hemisphere ◽  
Biomass Burning ◽  
Transport Model ◽  
Tropospheric Chemistry ◽  
Biogenic Emissions ◽  
Chemistry Transport Model ◽  
Mixing Ratios ◽  
Oxygenated Compounds ◽  
Global Space ◽  
The Impact

Abstract. The distribution and budget of oxygenated organic compounds in the atmosphere and their impact on tropospheric chemistry are still poorly constrained. Near-global space-borne measurements of seasonally resolved upper tropospheric profiles of methanol (CH3OH) by the ACE Fourier transform spectrometer provide a unique opportunity to evaluate our understanding of this important oxygenated organic species. ACE-FTS observations from March 2004 to August 2005 period are presented. These observations reveal the pervasive imprint of surface sources on upper tropospheric methanol: mixing ratios observed in the mid and high latitudes of the Northern Hemisphere reflect the seasonal cycle of the biogenic emissions whereas the methanol cycle observed in the southern tropics is highly influenced by biomass burning emissions. The comparison with distributions simulated by the state-of-the-art global chemistry transport model, LMDz-INCA, suggests that: (i) the background methanol (high southern latitudes) is correctly represented by the model considering the measurement uncertainties; (ii) the current emissions from the continental biosphere are underestimated during spring and summer in the Northern Hemisphere leading to an underestimation of modelled upper tropospheric methanol; (iii) the seasonal variation of upper tropospheric methanol is shifted to the fall in the model suggesting either an insufficient destruction of CH3OH (due to too weak chemistry and/or deposition) in fall and winter months or an unfaithful representation of transport; (iv) the impact of tropical biomass burning emissions on upper tropospheric methanol is rather well reproduced by the model. This study illustrates the potential of these first global profile observations of oxygenated compounds in the upper troposphere to improve our understanding of their global distribution, fate and budget.

scholarly journals Modelling atmospheric OH-reactivity in a boreal forest ecosystem

2011 ◽  
Vol 11 (18) ◽  
pp. 9709-9719 ◽  
Author(s):  
D. Mogensen ◽  
S. Smolander ◽  
A. Sogachev ◽  
L. Zhou ◽  
V. Sinha ◽  
...  
Keyword(s):  
Boreal Forest ◽  
Gas Phase ◽  
Atmospheric Chemistry ◽  
Transport Model ◽  
Inorganic Compounds ◽  
Chemistry Transport Model ◽  
One Dimensional ◽  
Inorganic Species ◽  
Individual Contributions ◽  
The Individual

Abstract. We have modelled the total atmospheric OH-reactivity in a boreal forest and investigated the individual contributions from gas phase inorganic species, isoprene, monoterpenes, and methane along with other important VOCs. Daily and seasonal variation in OH-reactivity for the year 2008 was examined as well as the vertical OH-reactivity profile. We have used SOSA; a one dimensional vertical chemistry-transport model (Boy et al., 2011a) together with measurements from Hyytiälä, SMEAR II station, Southern Finland, conducted in August 2008. Model simulations only account for ~30–50% of the total measured OH sink, and in our opinion, the reason for missing OH-reactivity is due to unmeasured unknown BVOCs, and limitations in our knowledge of atmospheric chemistry including uncertainties in rate constants. Furthermore, we found that the OH-reactivity correlates with both organic and inorganic compounds and increases during summer. The summertime canopy level OH-reactivity peaks during night and the vertical OH-reactivity decreases with height.

Sign in / Sign up

Export Citation Format

Share Document