A holistic framework to estimate the origins of atmospheric moisture and heat using a Lagrangian model

The commands for the analysis were as described below. For simplicity, we omit the years (set by --ayyyy and --ryyyy for the 10 analysis and run years, respectively) and the analysis months (set by --am), which covered all months between January 1980 and December 2016 for this analysis. Furthermore, commands are only illustrated for one experiment (--expid “ALL-ABL”) and one city (Denver, --maskval 1001). The analysis is split in two parts:

The Lagrangian analysis in the main manuscript was run with the Heat And MoiSture Tracking framEwoRk HAMSTER v1.0.0, as published on https://github.com/h-cel/hamster.
The commands for the analysis were as described below. For simplicity, we omit the years (set by --ayyyy and --ryyyy for the 10 analysis and run years, respectively) and the analysis months (set by --am), which covered all months between January 1980 and December 2016 for this analysis. Furthermore, commands are only illustrated for one experiment (--expid "ALL-ABL") and one city (Denver, --maskval 1001). The analysis is split in two parts: 1. The global detection and quantification of fluxes based on two-step trajectories 15 2. The attribution of source regions for heat and precipitation, and their bias-correction using data from the previous step.
A short description of the commands used for both parts is given below. 20

35
Both hamster steps 0 and 1 were run with a paths.txt file that looks as follows: The output of the first step are 6-hourly h5-files which contain parcel positions and properties for a specific date and the prior 40 time step. The output of the second step is a monthly netCDF file which contains three variables, i.e., P, E and H, gridded onto a regular 1x1° global grid, for all 6-hourly time steps of the month.

50
To extract 16-day trajectories, we run python main.py --steps 0 --ctraj_len 16 --maskval 1001 --pathfile paths.txt To get (biased) estimates of the source regions of precipitation and heat for Denver, we then employ the same settings as for 55 the global "ALL-ABL" analysis, and we evaluate the source contributions over 15 days into the past using the linear discounting and attribution.
python main.py --steps 2--ctraj_len 15 --expid "ALL-ABL" --cheat_dtemp 0 --fheat_drh False --fevap_drh False $fheat_rdq False --cevap_dqv 0 --cpbl_method "max" --cpbl_strict 1 --cpbl_factor 1 --cprec_dqv 0 --cprec_rh 80 --60 mattribution "linear" --maskval 1001 --pathfile paths.txt Finally, to bias-correct these source regions, we use the diagnosed fluxes from the previous step and a references data sethere ERA-Interimand adjust the source-receptor relationships day by day (--bc_time "daily"): python main.py --steps 3 --expid "ALL-ABL" --maskval 1001 --bc_useattp True --bc_aggbwtime False --bc_time "daily" --pathfile paths.txt The output of the first step are 6-hourly h5-files which contain parcel positions and properties for a specific date and the 16days prior to that date. The output of the second step is a monthly netCDF file which contains the source regions of precipitation 70 ("E2P") and the source region of heat ("Had") for the city of Denver. These variables are of dimension time x level x latitude x longitude, where time refers to the 'arrival date', level refers to the backward days (ranging from 0 to -15 with 0 being the 'arrival date') and latitude x longitude being again a global 1x1° grid. By setting --bc_aggbwtime False in the bias-correction step, we retain the 4D dimension for this analysis; the netCDF output of this step contains 6 variables: "E2P" as the biased estimates of daily source regions for Denver, "E2P_Es" as the source-corrected estimates of precipitation origins, "E2P_Ps" 75 as the sink-corrected estimates of precipitation origins, and "E2P_EPs", which is the sourceand sink-corrected estimate of these source regions. For heat advection, the final output file "Had", which is a copy of the biased estimates, and "Had_Hs", which represents the source-corrected source regions of heat.

Analysis performed with hamster.
The settings for all experiments contained in the main manuscript are as listed in Table S1 (listing detection criteria only). To assess the impact of the attribution methodology for the estimation of precipitation source regions, all experiments were repeated with --mattribution random in addition to the linear discounting / attribution (--mattribution linear). Overview of hamster flags used for the detection of P, E and H in this study. * not used due to --cpbl_strict 0. ** not used due to --fevap_drh False. *** not used to due --cheat_drh False. **** not used due to --fheat_rdq False.

110
The specific humidity loss associated with precipitation cannot always be fully attributed to the source locations identified along the trajectory:  fraction. While one could argue that this is an argument against process-based detection criteria and the restriction of source locations within the ABL, we argue that above-ABL moisture increases are likely a result of mixing processes and that the within-ABL locations are (biased) representatives of surface processes. Contrary to Sodemann (2020), we argue that these above-ABL "source locations" do not reflect surface processes, even if the moisture mixed into these parcels originated from surface evaporation in the first placewhich in turn may have taken place prior to the mixing and at a different location.

125
These above-ABL moisture increases are, however, assumed to contribute to the specific humidity of parcels en route and are thus indirectly considered indirectly in the discounting procedure. In addition, expecting ∆ to be biased as it reflects only the net flux ( − ), one may assume that the corresponding source region contributions reflect a reliable detection of source region contributions, and the corresponding weights can be upscaled to 100% of the desired quantity using , = ∑ .

130
It is noted here, however, that bias-correcting for precipitation yields the same effect.