student projects related to tropomi methane data · student projects related to tropomi methane...

1
Student Projects Related to TROPOMI Methane Data Dr. Dorit Hammerling , Lewis Blake , William Daniels , Aidan Dykstal , Dr. Sean Crowell Colorado School of Mines , University of Oklahoma Introduction Five student teams investigated TROPOMI data between May and February as part of a semester long statistical practicum. Here, we present a selection of interesting results. Anomaly Detection Applied unsupervised machine learning methods such as local outlier factor and isolation forest to identify anomalous atmospheric CH concentrations. Employed the non-parametric bootstrap to estimate CH distributions and subsequently detect anomalies. Figure : Average CH Mixing Ratio over the Mainland USA (Winter /). Regions Studied are Boxed in Black and Anomalies are Circled in Magenta. Figure : Bootstrapped Pareto Distribution for CH with Anomaly Threshold. The Denver-Julesburg (DJ) Basin Region Created estimates for data scarcity on various timescales. Identied areas containing consistently high CH levels. Figure : Locations of Interest within the DJ Basin. Figure : Monthly Time Series of TROPOMI CH Data. Five Locations in the DJ Basin are Selected to Show the Range of Typical CH Concentrations. The Haynesville Shale Region Created a seasonal model and short-term forecasting tool for atmospheric CH concentrations with ARIMA & DFT. Explored a land-water mask to lter out low quality data. Identied areas with regularly elevated CH concentrations. Figure : Estimated Proportion of Days with Above-Average CH Levels. The Southern Wetlands Region Created a linear model for atmospheric CH concentration as a function of other TROPOMI data products and time. Model explains .% of variability in CH . Figure : Estimated CH Levels versus Time Over the Southern Wetlands. The Bakken Formation Examined relationship between atmospheric CH concentration and natural glas aring. Model reveals slight positive association between aring temperature and CH level. Figure : Flaring Temperatures Over the Bakken Formation in Degrees Kelvin. Acknowledgements We would like to thank the Spring practicum students for their hard work. If you are interested in more technical details, contact Dr. Hammerling at [email protected].

Upload: others

Post on 14-Aug-2020

6 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Student Projects Related to TROPOMI Methane Data · Student Projects Related to TROPOMI Methane Data Dr. Dorit Hammerling1, Lewis Blake1, William Daniels1, Aidan Dykstal1, Dr. Sean

Student Projects Related to TROPOMI Methane DataDr. Dorit Hammerling1, Lewis Blake1, William Daniels1, Aidan Dykstal1, Dr. Sean Crowell2

Colorado School of Mines1, University of Oklahoma2

Introduction• Five student teams investigated TROPOMI data between May 2018and February 2020 as part of a semester long statistical practicum.•Here, we present a selection of interesting results.

Anomaly Detection

•Applied unsupervised machine learning methods such as localoutlier factor and isolation forest to identify anomalousatmospheric CH4 concentrations.• Employed the non-parametric bootstrap to estimate CH4

distributions and subsequently detect anomalies.

Figure 1: Average CH4 Mixing Ratio over the Mainland USA (Winter 2018/2019).Regions Studied are Boxed in Black and Anomalies are Circled in Magenta.

Figure 2: Bootstrapped Pareto Distribution for CH4 with Anomaly Threshold.

The Denver-Julesburg (DJ) Basin Region

• Created estimatesfor data scarcity onvarious timescales.• Identified areascontainingconsistently highCH4 levels.

Figure 3: Locations of Interest within the DJ Basin.

Figure 4: Monthly Time Series of TROPOMI CH4 Data. Five Locations in the DJ Basinare Selected to Show the Range of Typical CH4 Concentrations.

The Haynesville Shale Region

• Created a seasonal model and short-term forecasting tool foratmospheric CH4 concentrations with ARIMA & DFT.• Explored a land-water mask to filter out low quality data.• Identified areas with regularly elevated CH4 concentrations.

Figure 5: Estimated Proportion of Days with Above-Average CH4 Levels.

The Southern Wetlands Region

• Created a linear model for atmospheric CH4 concentration as afunction of other TROPOMI data products and time.•Model explains 65.7% of variability in CH4.

Figure 6: Estimated CH4 Levels versus Time Over the Southern Wetlands.

The Bakken Formation• Examined relationship between atmospheric CH4 concentrationand natural glas flaring.•Model reveals slight positive association between flaringtemperature and CH4 level.

Figure 7: Flaring Temperatures Over the Bakken Formation in Degrees Kelvin.

AcknowledgementsWe would like to thank the Spring 2020 practicum students for their hard work.If you are interested in more technical details, contact Dr. Hammerling at [email protected].