b'Environmental geophysics Environmental geophysicsMy (apparently primitive) approach tousing the simultaneous inversion visualising the change is to invert eachapproach to invert time-lapse transient snapshot dataset separately, and thenelectromagnetic (TEM) data collected to manually highlight areas in eachover a geothermal project in Iceland. inversion generation that appear to haveThey show the results by plotting changed. I have considered calculatingeach inversion separately, and then conductivity residuals on the data setscalculating resistivity ratios from each of but, with noise and natural variability,the inversion snapshots. Their approach this has seemed to me to be unlikely toappears to be quite robust, as, for work. So maybe I need to be inverting forexample, the number of data points change rather than for best image. used between inversions was different (and at slightly different locations). In most of the papers that I have read inThe ratio plots for the simultaneous Mike Hatchpreparing this column (see referencesinversions are big improvements over the Associate Editor below), this is the simplest version of theones prepared using the independent for Environmental geophysicsindependent inversion approach, sinceinversion approach.michael.hatch@adelaide.edu.au each time step is inverted independently of every other. It is possible (and moreI dont usually dabble in the seismic space, sophisticated) to run a normal inversionbut in the process of researching this Inversion for difference on the first data set and then invertarticle I found an interesting book chapter for either conductivity differencesby Lin etal. (2022) on time lapse inversion Welcome readers to this issuesor ratios. Again though, each step isapplied to seismic waveform inversion. In column on geophysics applied to theinverted independently and any noisethis work the authors, once again, firstly environment. For this issue, I was inspiredor other external factor affecting a givenshow how the method works on synthetic by a conversation that I had recently withdata set will affect the final result. data, and then provide a case study based Tim Munday, from CSIRO. He told meon reservoir monitoring in Texas. So far as about some interesting developments inSo far, we have more or less coveredI can tell, they used the cascade approach the data inversion space. He was referringwhat I have at least contemplated for(initial model, and then inverting for specifically to inversion for difference. my big, multiple generation data sets.differences) but replace the normal data According to the literature, the bettermisfit term in the regularisation terms so Most of us take raw data sets, i.e.,approach is to run time constrainedthat their inversion minimises not only some form of voltage data recordedinversions. In these approaches, the costtrue data change between two snapshots, from various ground sensors, and usefunction used to regularise the inversionbut also include a term that minimises various pieces of computer code thatinclude a weighting term that is usedthe simulated/modelled data between invert these data into something thatto minimise the difference betweenthe two snapshots. They call this double we hope reflects the real distribution indata snapshots (see e.g. Equations 6differencing. Ultimately, the two inversions the ground of conductivities, or seismicand 7 in Dimech etal. (2022)). Dimechare differenced to plot an inversion velocities, or whatever information ouretal. (2022) and others then tell us thatresidual. They show the results of a number ground sensors are designed to collect.there are two different approaches toof inversion variations and, to no ones We then use other information (drill datausing this new regularisation scheme.surprise, the inversion version with the or geological info, etc.) and our imagedThe slightly simpler (and apparentlymost sophisticated processing produces inversions to fill in the dots betweenless effective) approach is cascadethe best result - pretty convincingly.sparse data to help us interpret theinversion. The initial data set (the bigger picture. baseline data) is run using normalAs an aside, the Dimech etal. (2022) inversion controls. Each subsequent datapaper discusses how a measured However, especially in the environmental/ snapshot is inverted using the previousresistivity may vary significantly engineering space, we may be collectinginversion results as starting models,depending on the temperature of the data that is intended to quantify aand the constraint equation describeddata medium (i.e. the ground in most localised change in the property thatin Equation 7 (Dimech etal. 2022). Thecases). They state, based on other studies, we are measuring. For instance, overmore sophisticated methodology isthat resistivity decreases by a factor of the last few years I have been involvedthe simultaneous approach where allclose to 2% for an increase of 1o C in the in a project collecting electromagneticdata sets are inverted simultaneouslymedium. They argue that this should induction (EM) data over floodplains onusing the regularisation terms describedbe accounted for when comparing data the River Murray in South Australia. Theabove. This way, all of the data sets arecollected at different times of the year aim of this project is to quantify changesconstrained exactly the same way and(even for diurnal variation). In my work I in salinity levels on these highly salinisedthe results have to really fit together. have always assumed that temperature floodplains when they are either naturallyvariation didnt affect my ground data or artificially flooded. Like my RiverlandAn interesting paper to look at in thismuch, especially below about 1 m depth, surveys, more and more surveys are runspace, at least for me as it focusesand really only affected deeper surveys to collect a baseline data set and thenon ground EM data and inversion,like MT, where the crust and mantle subsequent data snapshots are collected,is Xiao et al. (2022). In this paper thetemperatures are (significantly) different hoping to see change as environmentalauthors first show results on a syntheticat different locations. Im going to have to conditions change. data set, and then on a real data set,look into this further.31 PREVIEW AUGUST 2023'