the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
A neural network-based method for generating synthetic 1.6 μm near-infrared satellite images
Abstract. This study presents an extension to the method for fast satellite image synthesis (MFASIS) to allow simulating reflectances for the 1.6 μm near-infrared channel based on a computationally efficient neural network with improved accuracy. Such a fast forward operator enables using 1.6 μm channels from different satellite instruments in applications like model evaluation or operational data assimilation. It thus paves the way for the exploitation of additional information at this frequency, e.g. on cloud phase and particle sizes, which is complementary to the visible and thermal infrared range.
To achieve similar accuracy for 1.6 μm NIR as for the visible channels 0.4–0.8 μm, it is important to represent vertical gradients of effective cloud particle radii, as well as mixed-phase clouds and molecular absorption. A comprehensive dataset sampled from IFS forecasts is used to develop the method. A new approach for describing the complex vertical cloud structure with a two layer model of water, ice and mixed-phase clouds optimized to obtain small reflectance errors is described and the relative importance of the different input parameters describing the idealized profiles is analyzed. Additionally, a different parameterization of the effective water and ice particle radii was used for testing. Further evaluation uses a month of ICON-D2 hindcasts with effective radii directly determined by the two-moment microphysics scheme of the model. The fast neural network approach itself does not add any significant additional error compared to the profile simplifications. In all cases, the mean absolute reflectance error achieved is about 0.01 or smaller, which is an order of magnitude smaller than typical differences between reflectance observations and corresponding model values.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(2778 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(2778 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-353', Anonymous Referee #1, 20 Jun 2023
Review of the paper “A neural network-based method for generating synthetic 1.6 μm near-infrared satellite images” by Florian Baur et al., MS No.: egusphere-2023-353
This paper focuses on the development of a neural network to estimate the reflectance emerging from the atmosphere at 1.6 micron from SEVIRI on Meteosat Second Generation. This approach should also be suitable for other near-infrared channels on instruments such as AHI, ABI or FCI. This is a nicely written paper discusses the neural network performance when trained on reflectances calculated using DOM with both IFS and ICON-D input profiles.
I believe that this paper would be suitable for publication subject to minor revision addressing the specific comments detailed below.
Specific comments
L37: here you should explain what you mean by 3D effects, e.g. reflection on complex topography etc. A reference could also be useful.
L50: here do you mean little information on discriminating the cloud phase?
L66: It would be really interesting if you could show here an example of the water and ice cloud jacobians for the 1.6 micron channel, possibly by comparing to those for the visible (and/or thermal IR) channels.
L84: How do you assess that the statement “are not very important” is true? Please add a reference to a paper where this is discussed and/or add a few explanations.
L97: Do you mean interpolate the reflectance for the specific value of the albedo at a given location? If so you should expand your text here as you discuss this interpolation only later in the paper.
L133: How many hidden layers, and did you test the effects of having more or fewer layers? how did you initialize the weights?
L136: What distribution did you use for the random numbers? If uniform, which intervals?
L138: How many epochs were used in the training?
Fig 2 caption: Please specify units of effective particle radius
L234 (“exceeds a threshold value of 1”): Is this ok also for low optical depth clouds? From Fig 3 there are quite a few profiles with log(tau_w) ~ 0.1. Did you test having a threshold dependent on cloud optical depth categories? And did you test the radiative effects of the use of different thresholds?
L235 (“where z_sfc is the height…”): do you mean here the height of the highest model level below the bottom cloud layer?
L244 (“exceeds a threshold value…”): I guess this at least partially answers my previous question. But I don't understand if this is done also for cth or only for ctp. And did you test other values (i.e. tau_t/5 etc.)
L255: It is not clear how these parameterizations are consistent with minimum and maximum values of IWP accepted by RTTOV
L262: If I understand correctly, please replace “full vertical profiles” with “full idealised vertical profiles
Fig 12 caption (“8 hidden layers with 25 nodes”): These numbers are inconsistent with those in the figure. Please correct the caption or the figure. Also, please check the total number of weights and biases is as stated.
Citation: https://doi.org/10.5194/egusphere-2023-353-RC1 -
AC1: 'Reply on RC1', Florian Baur, 31 Jul 2023
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-353/egusphere-2023-353-AC1-supplement.pdf
-
AC1: 'Reply on RC1', Florian Baur, 31 Jul 2023
-
RC2: 'Comment on egusphere-2023-353', Hartwig Deneke, 23 Jun 2023
My comments can be found in the supplemental PDF document.
-
AC2: 'Reply on RC2', Florian Baur, 31 Jul 2023
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-353/egusphere-2023-353-AC2-supplement.pdf
-
AC2: 'Reply on RC2', Florian Baur, 31 Jul 2023
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-353', Anonymous Referee #1, 20 Jun 2023
Review of the paper “A neural network-based method for generating synthetic 1.6 μm near-infrared satellite images” by Florian Baur et al., MS No.: egusphere-2023-353
This paper focuses on the development of a neural network to estimate the reflectance emerging from the atmosphere at 1.6 micron from SEVIRI on Meteosat Second Generation. This approach should also be suitable for other near-infrared channels on instruments such as AHI, ABI or FCI. This is a nicely written paper discusses the neural network performance when trained on reflectances calculated using DOM with both IFS and ICON-D input profiles.
I believe that this paper would be suitable for publication subject to minor revision addressing the specific comments detailed below.
Specific comments
L37: here you should explain what you mean by 3D effects, e.g. reflection on complex topography etc. A reference could also be useful.
L50: here do you mean little information on discriminating the cloud phase?
L66: It would be really interesting if you could show here an example of the water and ice cloud jacobians for the 1.6 micron channel, possibly by comparing to those for the visible (and/or thermal IR) channels.
L84: How do you assess that the statement “are not very important” is true? Please add a reference to a paper where this is discussed and/or add a few explanations.
L97: Do you mean interpolate the reflectance for the specific value of the albedo at a given location? If so you should expand your text here as you discuss this interpolation only later in the paper.
L133: How many hidden layers, and did you test the effects of having more or fewer layers? how did you initialize the weights?
L136: What distribution did you use for the random numbers? If uniform, which intervals?
L138: How many epochs were used in the training?
Fig 2 caption: Please specify units of effective particle radius
L234 (“exceeds a threshold value of 1”): Is this ok also for low optical depth clouds? From Fig 3 there are quite a few profiles with log(tau_w) ~ 0.1. Did you test having a threshold dependent on cloud optical depth categories? And did you test the radiative effects of the use of different thresholds?
L235 (“where z_sfc is the height…”): do you mean here the height of the highest model level below the bottom cloud layer?
L244 (“exceeds a threshold value…”): I guess this at least partially answers my previous question. But I don't understand if this is done also for cth or only for ctp. And did you test other values (i.e. tau_t/5 etc.)
L255: It is not clear how these parameterizations are consistent with minimum and maximum values of IWP accepted by RTTOV
L262: If I understand correctly, please replace “full vertical profiles” with “full idealised vertical profiles
Fig 12 caption (“8 hidden layers with 25 nodes”): These numbers are inconsistent with those in the figure. Please correct the caption or the figure. Also, please check the total number of weights and biases is as stated.
Citation: https://doi.org/10.5194/egusphere-2023-353-RC1 -
AC1: 'Reply on RC1', Florian Baur, 31 Jul 2023
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-353/egusphere-2023-353-AC1-supplement.pdf
-
AC1: 'Reply on RC1', Florian Baur, 31 Jul 2023
-
RC2: 'Comment on egusphere-2023-353', Hartwig Deneke, 23 Jun 2023
My comments can be found in the supplemental PDF document.
-
AC2: 'Reply on RC2', Florian Baur, 31 Jul 2023
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-353/egusphere-2023-353-AC2-supplement.pdf
-
AC2: 'Reply on RC2', Florian Baur, 31 Jul 2023
Peer review completion
Journal article(s) based on this preprint
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
247 | 126 | 23 | 396 | 15 | 13 |
- HTML: 247
- PDF: 126
- XML: 23
- Total: 396
- BibTeX: 15
- EndNote: 13
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1
Leonhard Scheck
Christina Stumpf
Christina Köpken-Watts
Roland Potthast
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(2778 KB) - Metadata XML