IMI Best Practices

Choosing an inversion time period

The IMI can be applied to any period of interest beginning 1 May 2018, when the TROPOMI methane record begins.
Common choices for the length of the inversion period are one year, one season (~3-6 months), one month, or one week.
We recommend choosing time periods of one week or more to ensure there are enough satellite observations for a successful inversion.
The IMI Preview feature can be used to refine the choice of inversion period.

Defining a region of interest

The IMI can be applied to any region of interest, from the global scale down to small focus areas such as cities, oil and gas basins, and agricultural areas.
The region of interest can be specified in several ways:
Setting latitude/longitude bounds for a rectangular domain.
Using a shapefile.
Interactively in the Integral Earth web user interface.
We recommend users select regions of interest larger than about 10,000 km² (100x100 km²) to ensure there are enough satellite observations for a successful inversion.
Larger regions of interest require more computational resources. This can be mitigated by optimally reducing the effective resolution of the inversion via smart state vector clustering.

Configuring the inversion domain

Regional inversions focus on a region of interest within a larger rectilinear inversion domain.
The inversion domain includes both the region of interest and an external buffer region.
The buffer region is broken into a collection of buffer emission elements representing emissions outside the region of interest.
We recommend using ≥ 8 buffer elements to pad the region of interest by ≥ 2°. The default number is 8.

Reducing the dimension of the state vector for large regions of interest

Inversions for large regions of interest at the IMI native 0.25°x0.3125° grid resolution can be computationally expensive.
This can be mitigated by reducing the dimension of the state vector using the state vector clustering options.
Smart state vector clustering combines 0.25°x0.3125° into coarser grid elements where the prior emission estimates are low and/or where TROPOMI provides few observations

Interpreting the IMI Preview

Examine the expected information content for the region and period of interest. This includes the map of expected averaging kernel sensitivities and the expected degrees of freedom for signal (DOFS).
- The averaging kernel sensitivities should be higher where the prior emission estimates are higher and where more observations are available.
- \(DOFS > 0.5\) is a bare minimum to achieve any solid information about emissions.
- \(DOFS < 2\) is marginal for most applications.
If the expected information content is low, consider:
- Increasing the inversion period to incorporate more observations.
- Increasing the prior error estimate.

Choosing the TROPOMI data product for the inversion

The IMI supports inversions with two versions of the TROPOMI methane record:
- The operational TROPOMI retrieval product developed by the SRON Netherlands Institute for Space Research.
- The Blended TROPOMI+GOSAT retrieval product developed by Balasus et al. (2023) to mitigate retrieval artifacts in the operational product.
Choosing a product depends on the application. The operational product is updated every few days. The blended product is updated intermittently and is currently available through 2023.
We recommend using the blended product when available (currently until 2024-01-01) to mitigate retrieval artifacts.

Creating an inversion ensemble

The IMI calculates the posterior error covariance matrix as part of the analytical solution. However,

if the error covariance matrices are not well-conditioned, the posterior error covariance results may be unreliable. * The best practice for calculating uncertainty is to use the ensemble generation feature of the IMI, to create an ensemble of results based on realistic ranges of hyperparameters. The spread of the ensemble members can be used to estimate the uncertainty in the inversion results.