TROPOMI CSF

Abstract

This document presents an analysis of methane ( ${\mathrm{CH}}_4$) emission estimates for the Hail Creek open cut mine as reported by Sadavarte, et al., 2021 . It is accompanied by a Python software package available on GitHub GitHub TROPOMI_CSF that reproduces and examines the underlying methodology. The primary focus of this work is the unusually high emission factor reported for Hail Creek relative to previous estimates. Such a large deviation from established methodologies typically reflects either a fundamentally new understanding of the emission processes or, alternatively, potential methodological limitations that warrant careful examination.

Our analysis indicates that the calculations are subject to substantial uncertainty, driven primarily by the automated determination of plume shape and background concentration.

Because detailed plume modelling requires the use of a Chemical Transport Model (CTM) at spatial resolutions beyond those available to the author, a definitive conclusion cannot be reached. Instead, we provide a set of open source Python tools to facilitate further investigation of this problem. We hope that these tools will enable readers to conduct their own analyses and form independent judgements.

Summary

The problem of atmospheric dispersion, assuming known emission rates, has been extensively studied since the 1950s, resulting in a substantial body of scientific and engineering literature (see Zanetti, ed. 2003-2010 ). Introductory treatments are provided by Seinfeld, et al., 2016 . A wide range of chemical transport and dispersion models ( CTM ) has been developed to address this problem, including AERMOD , CALPUFF , CMAQ , GEOS-Chem , HYSPLIT and , WRF-Chem , among others.

A recent development in this field is the availability of satellite borne spectrometer measurements. When combined with numerical model data, these observations enable estimation of ${\mathrm{CH}}_4$ mixing ratios for individual pixels, often resolved vertically through post processing algorithms. Numerical models provide information on the atmospheric state, including temperature profiles, pressure fields, geopotential heights, and water vapour content, while satellite observations supply spatially resolved spectra. This reverses the traditional dispersion problem: instead of predicting concentrations from known emissions, concentrations are observed and emission rates must be inferred. This inverse problem is commonly referred to as the plume inversion problem (see Jacob, et al., 2016 ).

A taxonomy of approaches used to address the plume inversion problem is presented by Jacob, et al., 2016 and Varon, et al., 2018 . The former proposes the use of CTM combined with Bayesian optimisation. In this framework, initial emission rates and concentration fields are assumed, the CTM is run forward, and the simulated concentrations are compared with subsequent satellite observations. This process is iterated, and the set of emission rates that best reproduces the observed concentrations provides the solution to the inversion problem.

The second paper focuses on inversion methods applied to individual plumes. The Cross Sectional Flux method (CSF) used by Sadavarte, et al., 2021 falls within this category, as does the Integrated Mass Enhancement method (IME) approach, which represents another plume based inversion technique.

The Introduction restates the problem and provides background on the available data sources, modelling approaches, and commonly used terminology in chemical transport modelling and meteorology. Readers already familiar with these topics may skip this section without loss of continuity.

The second section, Hail Creek Mine , summarises publicly available information on the Hail Creek mine, including its geographic setting, climatology, coal production, emission estimates, and the implied emission factors used.

The third section, Software , describes the implementation of a software package that replicates the algorithm used by Sadavarte, et al., 2021 . It outlines the individual steps of the algorithm, discusses potential ambiguities, and examines its computational stability. This section forms the core of the report, as it enables readers to reproduce the calculations and generate their own emission estimates.

The fourth section, Plume Patterns , analyses three distinct plume regimes and provides a systematic framework for evaluating the software output. The first regime is characterised by the formation of a Stable Nocturnal Boundary Layer (SNBL) . For this case, we examine TROPOMI orbit 09956 (15 September 2019). The analysis combines AWS observations with HYSPLIT trajectory modelling to extend the meteorological assessment. We show that under SNBL conditions the plume is not in steady state, rendering the CSF method inapplicable. Because CSF estimates for this orbit were used in Sadavarte, et al., 2021 , we compare those results with emission estimates derived using the TM approach.

The second plume pattern is characterised by continuous overnight turbulent mixing, which prevents the formation of a SNBL. We use TROPOMI orbit 11332 (21 December 2020) as a representative example. These conditions are the most favourable for applying the CSF method, while simultaneously rendering the TM and IME approaches inapplicable.

The final plume pattern corresponds to one of only two cases in which the wind exhibited a pronounced westerly component during the 24 hours preceding the satellite overpass. These conditions allow backward HYSPLIT modelling to be used to gain insight into background concentration levels. For this case, we analyse TROPOMI orbit 09445 (10 August 2019).

The fifth section, Results , summarises the findings obtained from reproducing the CSF algorithm. The main outcomes are as follows:

1. A fundamental challenge for plume based inversion methods is the determination of both plume extent and background concentration; these two quantities are inherently interdependent. An objective determination of plume shape would ideally rely on output from a CTM. In the absence of such modelling, plume identification becomes subjective, and automated algorithms may or may not yield physically meaningful results. Background concentrations are therefore typically estimated using statistical approaches, and the choice of background directly influences the inferred plume shape.
2. TROPOMI data have been processed using multiple algorithm versions over time, resulting in several versions of the same observations. Within the dataset analysed here (April 2018 to December 2019), two processed versions are available, both of which exhibit significant striping artefacts that strongly affect the CSF algorithm. In the original study, these striping effects were mitigated through additional filtering. Because such filtering requires subjective judgement regarding the treatment of TROPOMI data, this step was not applied in the present analysis.
3. The CSF algorithm, as implemented in this report, exhibits low numerical stability, with relatively minor configuration changes leading to substantial variations in the estimated emission rates.
4. Manual verification of the underlying methodological assumptions remains essential. In particular, the CSF method should not be applied when the plume is not in steady state, most notably under SNBL conditions. Conversely, the IME and TM approaches are not appropriate when plume coverage is incomplete or when there is non negligible mass flow across the boundaries of the analysis box.
5. The Hail Creek mine is situated within complex terrain, a relatively deep valley containing a deep open cut pit, which poses significant challenges for both CTM and inverse emission estimation.

The report concludes with a description of selected commands provided in the accompanying software package. It is expected that readers will use these tools to run the software themselves and form their own independent assessments.