This dataset, along with Maier et al. (2026), contains output from large-eddy simulations of shallow cumulus clouds, each driven by one of three different radiative transfer solvers. The simulations were performed with the PALM model on a grid with 100 m horizontal and 50 m vertical resolution. After an 11-hour spin-up period, they cover almost one full diurnal cycle, including both daytime and nighttime conditions, with output provided at a temporal resolution of 60 s. The three radiative transfer schemes applied are a conventional 1D delta-Eddington approximation, the fully 3D TenStream solver, and the dynamic TenStream solver, which accelerates 3D radiative transfer calculations by using incomplete solves. Together, the two datasets provide one five-member ensemble for each radiative transfer solver, consisting of one main run and four statistically independent control runs. Apart from the small perturbations used to create the ensemble members, the simulations differ only in the radiative transfer method applied. The combined datasets therefore enable an analysis of how cloud development differs depending on whether 1D or 3D radiative transfer is used, and an assessment of how well the dynamic TenStream solver reproduces key 3D radiative effects.

README

---

This dataset, along with Maier et al. (2026), contains output from large-eddy simulations of shallow cumulus clouds, each driven by one of three different radiative transfer solvers. The simulations were performed with the PALM model on a grid with 100 m horizontal and 50 m vertical resolution. After an 11-hour spin-up period, they cover almost one full diurnal cycle, including both daytime and nighttime conditions, with output provided at a temporal resolution of 60 s. The three radiative transfer schemes applied are a conventional 1D δ-Eddington approximation, the fully 3D TenStream solver, and the dynamic TenStream solver, which accelerates 3D radiative transfer calculations by using incomplete solves. Together, the two datasets provide one five-member ensemble for each radiative transfer solver, consisting of one main run and four statistically independent control runs. Apart from the small perturbations used to create the ensemble members, the simulations differ only in the radiative transfer method applied. The combined datasets therefore enable an analysis of how cloud development differs depending on whether 1D or 3D radiative transfer is used, and an assessment of how well the dynamic TenStream solver reproduces key 3D radiative effects.

This dataset represents the second part of the complete set of simulations. It contains the ensemble members with suffixes _control_2, _control_3, and _control_4. The corresponding _main and _control ensemble members are provided in the companion dataset Maier et al. (2026). Together, both datasets contain the complete five-member ensemble for each radiative transfer solver.

General dataset structure

---

Each simulation is stored in its own folder, with

• INPUT containing the full set of input files.
• OUTPUT containing all the model output files.
• RESTART containing the restart files associated with the simulation.
  
  

Initial simulations

---

The folders initial_main, initial_control, initial_control_2, initial_control_3, and initial_control_4 contain the data from the five statistically independent but otherwise identical initial simulations. initial_control_2, initial_control_3, and initial_control_4 are contained in this dataset, while initial_main and initial_control are provided in the companion dataset Maier et al. (2026). All five were driven by a classical 1D δ-Eddington solver. They were started on 14 June 2023 at 22:00 UTC and ran for 29 hours until 16 June 2023 at 03:00 UTC.

Because these simulations were restarted every 30 minutes, each OUTPUT folder contains 58 output files of each file type. In addition, the corresponding RESTART folders contain 58 restart files, allowing users to continue the simulations from any of the available restart timestamps, for example with modified runtime parameters.

Restart simulations

---

After an 11-hour spin-up period, starting on 15 June 2023 at 09:00 UTC, three different restart simulations were launched from each of the five initial runs. These simulations differ only in the radiative transfer solver applied. Across both dataset parts, they are stored in the following folders:

• delta_eddington_main
• delta_eddington_control
• delta_eddington_control_2
• delta_eddington_control_3
• delta_eddington_control_4
• dynamic_tenstream_main
• dynamic_tenstream_control
• dynamic_tenstream_control_2
• dynamic_tenstream_control_3
• dynamic_tenstream_control_4
• tenstream_main
• tenstream_control
• tenstream_control_2
• tenstream_control_3
• tenstream_control_4

In the present dataset, only the folders ending in _control_2, _control_3, and _control_4 are included. The corresponding _main and _control folders are provided in the companion dataset Maier et al. (2026).

The specific restart file used to initialize each of the simulations can be found in the RESTART directory of the corresponding simulation. In contrast to the initial simulations, these restart runs were conducted without intermediate restarts. Hence, the OUTPUT folders contain only one file of each type. The detailed solver configurations for each of these runs can be found in the ..._ts_options files in the INPUT directory of each simulation.

Reproducing the simulations

---

To rerun a simulation, copy the folder

<Simulation Name>/INPUT

into

<PALMPATH>/build/JOBS

of your PALM installation.

For the restart runs, also copy

<Simulation Name>/RESTART

into

<PALMPATH>/build/tmp

From the build directory of your PALM installation, the initial runs can then be started, for example, using:

bin/palmrun -k -v -r initial_control_2 -c default -a "d3# restart" -X 64 -T 64 -q cluster -b -m 2000 -t 170000

This launches initial_control_2 as a batch job on 64 cores with 2 GB of memory per core and a wallclock limit of 47.2 hours.

Restart runs require a slightly different argument string and can be started, for example, using:

bin/palmrun -k -v -r tenstream_control_2 -c default -a "d3r restart" -X 64 -T 64 -q cluster -b -m 2000 -t 600000

For further details on palmrun options and input files, please refer to the official PALM documentation.

Output files

---

Each simulation produces four main types of output files:

• *_3d — full 3D model fields
• *_pr — vertical profiles
• *_ts — domain-averaged time series
• *_xy — 2D horizontal fields (e.g., surface temperature, liquid water path)

Below, some general remarks on these files as well as lists of notable variables are given.

General remarks

Output quantities are provided either at grid-box centers or at grid-box interfaces. To this end, two different types of spatial coordinates are used:

• Cell-center coordinates such as x, y, zu, or zu_3d represent cell-centered quantities. Using these coordinates, values are assigned to the centers of the corresponding grid boxes.
• Cell-interface coordinates such as xu, yv, zw, or zw_3d, on the other hand, represent values located at grid-box boundaries. For example, the zonal wind u is stored on x-interfaces and therefore uses the staggered coordinate xu. Likewise, radiative fluxes such as downwelling longwave radiation rad_lw_in are stored on vertical interfaces and use zw_3d.

In addition, especially the *_pr files often contain multiple variable-specific vertical coordinates (e.g., zq, zql, zhyp, zrho). In this dataset, however, these are typically identical in values to the main vertical coordinates (zu or zw).

Notable variables

With these conventions in mind, the following tables now list some of the variables that can be used to analyze the dataset, together with their dimensions and units. Note that in all file types, the global attribute origin_time can be used to convert time (seconds) into absolute timestamps.

Notable variables in the *_3d files

Variable	Dimensions	Units	Description
time	30 (initial runs) or 1080 (restart runs)	seconds	time coordinate
zu_3d	82	meters	cell-center vertical coordinate for 3D fields
zw_3d	82	meters	cell-interface vertical coordinate for 3D fields
x	256	meters	cell-center x-coordinate
xu	256	meters	cell-interface x-coordinate (used by u)
y	256	meters	cell-center y-coordinate
yv	256	meters	cell-interface y-coordinate (used by v)
zs_3d	8	meters	soil vertical coordinate (for soil fields)
---	---	---	---
u	(time, zu_3d, y, xu)	m/s	zonal wind component
v	(time, zu_3d, yv, x)	m/s	meridional wind component
w	(time, zw_3d, y, x)	m/s	vertical wind component
q	(time, zu_3d, y, x)	kg/kg	total water mixing ratio
ql	(time, zu_3d, y, x)	kg/kg	liquid water mixing ratio
qc	(time, zu_3d, y, x)	kg/kg	cloud water mixing ratio
qv	(time, zu_3d, y, x)	kg/kg	water vapor mixing ratio
ta	(time, zu_3d, y, x)	°C	air temperature
theta	(time, zu_3d, y, x)	K	potential temperature
m_soil	(time, zs_3d, y, x)	m3/m3	volumetric soil moisture
t_soil	(time, zs_3d, y, x)	K	soil temperature
rad_sw_in	(time, zw_3d, y, x)	W/m2	incoming shortwave radiative flux
rad_sw_out	(time, zw_3d, y, x)	W/m2	outgoing shortwave radiative flux
rad_lw_in	(time, zw_3d, y, x)	W/m2	incoming longwave radiative flux
rad_lw_out	(time, zw_3d, y, x)	W/m2	outgoing longwave radiative flux
rad_sw_hr	(time, zu_3d, y, x)	K/h	shortwave heating rate
rad_lw_hr	(time, zu_3d, y, x)	K/h	longwave heating rate

Notable variables in the *_pr files

Variable	Dimensions	Units	Description
time	30 (initial runs) or 1080 (restart runs)	seconds	time coordinate
zu	82	meters	layer-centered vertical coordinate for u
zv	82	meters	layer-centered vertical coordinate for v
zw	82	meters	level (interface) vertical coordinate for w
zq	82	meters	layer-centered vertical coordinate for q
zql	82	meters	layer-centered vertical coordinate for ql
zqc	82	meters	layer-centered vertical coordinate for qc
zqv	82	meters	layer-centered vertical coordinate for qv
zrh	82	meters	layer-centered vertical coordinate for rh
zrho	82	meters	layer-centered vertical coordinate for rho
zhyp	82	meters	layer-centered vertical coordinate for hyp
---	---	---	---
u	(time, zu)	m/s	zonal wind component
v	(time, zv)	m/s	meridional wind component
w	(time, zw)	m/s	vertical wind component
q	(time, zq)	kg/kg	total water mixing ratio
ql	(time, zql)	kg/kg	liquid water mixing ratio
qc	(time, zqc)	kg/kg	cloud water mixing ratio
qv	(time, zqv)	kg/kg	water vapor mixing ratio
rh	(time, zrh)	%	relative humidity
rho	(time, zrho)	kg/m3	air density
hyp	(time, zhyp)	hPa	hydrostatic pressure

Notable variables in the *_xy files

Variable	Dimensions	Units	Description
time	30 (initial runs) or 1080 (restart runs)	seconds	time coordinate
x	256	meters	cell-center x-coordinate
y	256	meters	cell-center y-coordinate
zu1_xy	1	meters	single-layer “surface plane” vertical coordinate
---	---	---	---
lwp*_xy	(time, zu1_xy, y, x)	kg/m2	liquid water path
shf*_xy	(time, zu1_xy, y, x)	W/m2	surface sensible heat flux
qsws*_xy	(time, zu1_xy, y, x)	W/m2	surface latent heat flux
ta_2m*_xy	(time, zu1_xy, y, x)	°C	2-m air temperature

A complete description of all of the output parameters in the files is provided in the official PALM documentation: https://palm.muk.uni-hannover.de/trac/wiki/doc/app/runtime_parameters#data_output

Quick look at the data

---

For a quick visual inspection of the output files, you may use ncview, a visual browser for netCDF files. For example, you can execute

ncview tenstream_control_2/OUTPUT/tenstream_control_2_3d.022.nc

This provides an easy way to browse through the 3D fields and check the content of the dataset without having to load it into a post-processing environment.

Example plot

---

To give a first glimpse at the data, the following plot shows the temporal evolution of the liquid water path (LWP) in the delta_eddington_control_3 (left), dynamic_tenstream_control_3 (middle), and tenstream_control_3 (right) simulations for five selected time steps between 09:01 and 21:00 UTC. Note that, to enhance the contrast of the plots, the maximum LWP value for the colorbar was set to 50 g m⁻² instead of the global maximum value of 338 g m⁻².