the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 13 Jan 2026
Evaluating Modifications to Tiedtke Cumulus Parameterization for Improving Summer Precipitation Forecasts in the Nested Grid of Taiwan Global Forecast System (TGFS v1.1)
Abstract. The Taiwan Global Forecast System (TGFS), which is a global numerical weather prediction system equipped with a nested regional domain, has become operational at Taiwan’s Central Weather Administration since September 2023. To improve summertime precipitation forecasts in the 4.8-km-resolution nested simulations of TGFS, numerous modifications to the Tiedtke cumulus parameterization are made. First, scale-aware treatments to the convective mass flux and trigger condition are applied in the Tiedtke scheme to enhance the role of grid-scale microphysics processes, which effectively improves the pattern and location of afternoon thunderstorm precipitation. Based on the scale-aware Tiedtke version, further changes are made to reduce the precipitation biases. Constraints on convective updrafts are modified to reduce the occurrence and intensity of convection. The definition of convective cloud tops is revised to reduce their height to a more reasonable level. Additionally, the convective trigger function, entrainment, and detrainment rates are assumed to be more sensitive to environmental relative humidity, resulting in a further reduction in precipitation. The combination of these modifications successfully improves the precipitation forecast skill for summer rainfall events.
-
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
(2236 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(2236 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2025-5324', Peter Bechtold, 26 Jan 2026
-
AC1: 'Reply on RC1', Chang-Hung Lin, 12 Feb 2026
The comment was uploaded in the form of a supplement
-
RC5: 'Reply on AC1', Peter Bechtold, 14 Feb 2026
I am glad that the authors responded to all my comments and performed complementary evaluations
Citation: https://doi.org/10.5194/egusphere-2025-5324-RC5 -
AC5: 'Reply on RC5', Chang-Hung Lin, 03 Apr 2026
The authors thank Dr. Peter Bechtold for the valuable feedback and are pleased that our complementary evaluations have addressed all of the comments.
Citation: https://doi.org/10.5194/egusphere-2025-5324-AC5
-
AC5: 'Reply on RC5', Chang-Hung Lin, 03 Apr 2026
-
RC5: 'Reply on AC1', Peter Bechtold, 14 Feb 2026
-
AC1: 'Reply on RC1', Chang-Hung Lin, 12 Feb 2026
-
RC2: 'Comment on egusphere-2025-5324', Anonymous Referee #2, 30 Jan 2026
To improve summer precipitation forecasts in the Taiwan Global Forecast System (TGFS), authors modified the Tiedtke cumulus parameterization with scale-aware treatments and adjusted convection constraints. These updates, including revised cloud-top definitions and increased sensitivity to humidity, effectively reduced precipitation biases and improved the accuracy of afternoon thunderstorm patterns. The manuscript is well-written and offers valuable insights for enhancing the convection parameterization scheme. I find it suitable for publication, but recommend that the authors address following points.
- As the authors note, the enhanced contribution from the convection scheme consumes more environmental instability through subgrid-scale processes, resulting in the suppression of grid-scale convective updrafts and a corresponding reduction in precipitation intensity. This is clearly demonstrated by the scale-aware adjustment experiment (adj_SCA in Table 2), where a larger cloud-base mass flux increases the sub-grid scale convection and consequently reduces the overestimated total precipitation, compared to the SCA experiment (Figs. 6 and 7). Conversely, the CUP, TOP, and CRH experiments (Table 2) were intended to reduce the sub-grid scale convection intensity by extending the adjustment time scale, lowering the convective cloud top, and increasing entrainment as a function of environmental RH. These changes were expected to increase total precipitation due to reduced suppression of environmental instability. However, as the authors note, CUP, TOP, and CRH appear to further reduce the total precipitation (Fig. 6). The authors must address this inconsistency between the strength of sub-grid scale convection and the total precipitation by providing the ratio of sub-grid scale precipitation to total precipitation for the CUP, TOP, and CRH experiments, compared with the adj_SCA experiment.
- Is the increased widespread light rain in the CRH mainly from the sub-grid scale precipitation or from the grid-scale precipitation? If it is mainly from the latter, the increased sub-grid scale convection intensity may help to reduce the widespread light rain. Therefore, it would be valuable for the authors to present a sensitivity analysis of the widespread light rain in response to varying sub-grid scale convection intensity.
Citation: https://doi.org/10.5194/egusphere-2025-5324-RC2 -
AC2: 'Reply on RC2', Chang-Hung Lin, 12 Feb 2026
The comment was uploaded in the form of a supplement
-
RC4: 'Reply on AC2', Anonymous Referee #2, 12 Feb 2026
The points I raised in the first review were well addressed. I accept the paper for publication.
Citation: https://doi.org/10.5194/egusphere-2025-5324-RC4 -
AC6: 'Reply on RC4', Chang-Hung Lin, 03 Apr 2026
The authors thank Reviewer #2 for the positive feedback and for recommending the acceptance of our manuscript.
Citation: https://doi.org/10.5194/egusphere-2025-5324-AC6
-
AC6: 'Reply on RC4', Chang-Hung Lin, 03 Apr 2026
-
RC4: 'Reply on AC2', Anonymous Referee #2, 12 Feb 2026
-
RC3: 'Comment on egusphere-2025-5324', Anonymous Referee #3, 30 Jan 2026
The manuscript is well written and provides sufficient detail regarding the changes the authors made to the Tiedtke scheme in TGFS v1.1, as well as the experimental design used to demonstrate the positive effects of these changes. Most of the modifications to the Tiedtke scheme are derived from other schemes. From a dynamical perspective, these modifications are not entirely independent in their effects on controlling precipitation strength associated with unresolved convection.
For readers like me, it is important that the manuscript include discussions that clarify which changes are most effective at alleviating precipitation forecast biases. I recommend that the authors revise the manuscript to expand Figure 3 by incorporating results from all experiments summarized in Table 2 for the afternoon thunderstorm case, as in Figure 6 for the southwesterly flow case.
Given the partition of resolved and unresolved air motion within the TGFS numerics, the modified Tiedtke scheme—due to its suitability for gray-zone resolutions—should represent the unresolved convective mass flux relative to the resolved counterpart in a manner that diminishes as grid size decreases. Therefore, I also recommend that the revised manuscript include a discussion, supported by numerical evidence if necessary, addressing whether the overestimate of precipitation shown in Figure 4 for the SCA experiment can be alleviated solely by reducing the mass flux scaling factor. This discussion is essential to help readers determine whether the other changes, aside from the mass flux scaling, are equally important in making the modified Tiedtke scheme scale-aware.Citation: https://doi.org/10.5194/egusphere-2025-5324-RC3 - AC3: 'Reply on RC3', Chang-Hung Lin, 03 Apr 2026
-
RC6: 'Comment on egusphere-2025-5324', J. M. Piriou, 24 Feb 2026
In this study, the authors start with a 2016 version of the Tiedtke-Bechtold (TDK) scheme, which they modify significantly to meet the needs of Taiwanese forecasters, in particular to reduce precipitation and improve its localization in a 5 km grid model. This work updates TDK based on other published work.
More specific comments:
- It is interesting, as you do, to validate the modifications by stacking them on top of each other: ORI, SCA, …, CRH.
- Figure 4 shows the temporal sequence of precipitation, with precipitation forecasts showing a 4 hours lead. It would be interesting to plot the diurnal cycle of precipitation over a period of one month to see how much of this time lead is a systematic flaw in the convection scheme.
- Section 3.4 (from line 171) “Dependency on environmental relative humidity for convection triggering, entrainment, and detrainment” also aims to reduce intense precipitation (see line 172). The revised entrainment values are based on ... Bechtold 2008, which is not recent, especially since this work develops a 2016 version of the MPAS of the Tiedtke scheme, which is more recent. It seems that your work would have been easier if you had based it on a more recent version of the Tiedtke scheme.
- In section 3.1, starting at line 125, “Scale-aware parameterization,” the authors do not describe how they handle mesh dependency, but simply cite publications. It would be good to briefly describe the process with a few simple equations, so that this article would be more “stand alone.”
- In section 3.3, “Definition of convective cloud top,” by changing the threshold from 0 to 10^-8 kg/kg, the height of convective clouds is significantly reduced and the intensity of precipitation decreases (see line 170). This surprising result needs to be explained: what physical mechanisms make the TDK scheme so sensitive to this threshold? Could you explain?
- In section 5.2, “Southwesterly flow case,” it is surprising that you used 48- to 60-hour lead time, because in an operational context, 24- to 36-hour lead time are more important for decision-making. Is there a reason for this? Similarly, in Figure 11, you show precipitation between 96 and 108 hours, which sounds like a methodological problem, because if we are interested in such distant lead times, we should also show the 24-36 hour, 48-60 hours (which you have done), and 72-84 hours, which would give an idea of the spread linked to the lead time itself.
- You mention in line 106 that you are using the Bechtold et al. 2014 approach, which delays the diurnal cycle by modifying the CAPE based on criteria related to atmospheric boundary layer turbulence. Our experience with the Tiedtke scheme indicates that this delay, which is formulated differently over land and over ocean, tends to favor rainfall over ocean versus rainfall over land, which is particularly clear on precipitation fields near the coasts. You could also look into this for your future studies.
This work is interesting. I recommend its publication.
Citation: https://doi.org/10.5194/egusphere-2025-5324-RC6 - AC4: 'Reply on RC6', Chang-Hung Lin, 03 Apr 2026
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2025-5324', Peter Bechtold, 26 Jan 2026
The authors present modifications of the Tiedtke convection scheme and apply and evaluate these in the 4 km nested runs over Taiwan. I like the presentation of the manuscript as the modifications are presented and evaluated step by step for case studies followed by a general evaluation, the manuscript is also well written in general. I also broadly agree with these modifications (see below) and the discussion of the convective features related to model deficiencies.
specific comments:
-SCA: yes the mass flux scaling should definitely be applied before applying the CFL criterium. Could you mention what model time step you use? Also in the CFL criterium computation it might be better to use for very small time steps eg dp/(g*max(dt,300))
- cloud top criterium: yes the convective cloud top height is definitely overestimated in the original version, you might also want to test an additional criterium on the cloud top height that has been introduced in the German weather service (DWD) and ECMWF, ie for the updraught to continue in the vertical dT/dz<-3.e-3 and Buo>-2
- entrainment modification: Here I find it less convincing, you say it further reduces "convection" but the results show more light precipitation indicating more widespread light convection. Indeed, when looking at your formula it appears that for z=100m you get entrainment rates of 10-3 m-1 which is similar to the original, but for z=500 m (a typical cloud base height) it is 2x10-4 which is already much smaller than the original (Note als you say d1 is non-dimensional but it has units m^-1).....??? At leats in your conclusions you seem not to have adopted this version for operations.
You also discuss in detail - a discussion which is welcome - reasons for having precipitation bands sometimes too far offshore compared to observations. Then you mention at the end that this problem is even more widespread in winter over a relatively warm sea. Indeed, this problem is shared by all global modelling centres employing a convective parametrization. The reasons are as you broadly mentioned the interaction between heating and the circulation (forming a quasi stationnary near coast circulation, having updraught and downdraught in the same grid cell and the lack of advection. We could largely address this problem in the upcoming operational version at ECMWF (cycle 50r1, April 2026) by handing over a significant amount of convective precipitation to the large-scale cloud scheme where it is advected and evaporated (no publication available)
Citation: https://doi.org/10.5194/egusphere-2025-5324-RC1 -
AC1: 'Reply on RC1', Chang-Hung Lin, 12 Feb 2026
The comment was uploaded in the form of a supplement
-
RC5: 'Reply on AC1', Peter Bechtold, 14 Feb 2026
I am glad that the authors responded to all my comments and performed complementary evaluations
Citation: https://doi.org/10.5194/egusphere-2025-5324-RC5 -
AC5: 'Reply on RC5', Chang-Hung Lin, 03 Apr 2026
The authors thank Dr. Peter Bechtold for the valuable feedback and are pleased that our complementary evaluations have addressed all of the comments.
Citation: https://doi.org/10.5194/egusphere-2025-5324-AC5
-
AC5: 'Reply on RC5', Chang-Hung Lin, 03 Apr 2026
-
RC5: 'Reply on AC1', Peter Bechtold, 14 Feb 2026
-
AC1: 'Reply on RC1', Chang-Hung Lin, 12 Feb 2026
-
RC2: 'Comment on egusphere-2025-5324', Anonymous Referee #2, 30 Jan 2026
To improve summer precipitation forecasts in the Taiwan Global Forecast System (TGFS), authors modified the Tiedtke cumulus parameterization with scale-aware treatments and adjusted convection constraints. These updates, including revised cloud-top definitions and increased sensitivity to humidity, effectively reduced precipitation biases and improved the accuracy of afternoon thunderstorm patterns. The manuscript is well-written and offers valuable insights for enhancing the convection parameterization scheme. I find it suitable for publication, but recommend that the authors address following points.
- As the authors note, the enhanced contribution from the convection scheme consumes more environmental instability through subgrid-scale processes, resulting in the suppression of grid-scale convective updrafts and a corresponding reduction in precipitation intensity. This is clearly demonstrated by the scale-aware adjustment experiment (adj_SCA in Table 2), where a larger cloud-base mass flux increases the sub-grid scale convection and consequently reduces the overestimated total precipitation, compared to the SCA experiment (Figs. 6 and 7). Conversely, the CUP, TOP, and CRH experiments (Table 2) were intended to reduce the sub-grid scale convection intensity by extending the adjustment time scale, lowering the convective cloud top, and increasing entrainment as a function of environmental RH. These changes were expected to increase total precipitation due to reduced suppression of environmental instability. However, as the authors note, CUP, TOP, and CRH appear to further reduce the total precipitation (Fig. 6). The authors must address this inconsistency between the strength of sub-grid scale convection and the total precipitation by providing the ratio of sub-grid scale precipitation to total precipitation for the CUP, TOP, and CRH experiments, compared with the adj_SCA experiment.
- Is the increased widespread light rain in the CRH mainly from the sub-grid scale precipitation or from the grid-scale precipitation? If it is mainly from the latter, the increased sub-grid scale convection intensity may help to reduce the widespread light rain. Therefore, it would be valuable for the authors to present a sensitivity analysis of the widespread light rain in response to varying sub-grid scale convection intensity.
Citation: https://doi.org/10.5194/egusphere-2025-5324-RC2 -
AC2: 'Reply on RC2', Chang-Hung Lin, 12 Feb 2026
The comment was uploaded in the form of a supplement
-
RC4: 'Reply on AC2', Anonymous Referee #2, 12 Feb 2026
The points I raised in the first review were well addressed. I accept the paper for publication.
Citation: https://doi.org/10.5194/egusphere-2025-5324-RC4 -
AC6: 'Reply on RC4', Chang-Hung Lin, 03 Apr 2026
The authors thank Reviewer #2 for the positive feedback and for recommending the acceptance of our manuscript.
Citation: https://doi.org/10.5194/egusphere-2025-5324-AC6
-
AC6: 'Reply on RC4', Chang-Hung Lin, 03 Apr 2026
-
RC4: 'Reply on AC2', Anonymous Referee #2, 12 Feb 2026
-
RC3: 'Comment on egusphere-2025-5324', Anonymous Referee #3, 30 Jan 2026
The manuscript is well written and provides sufficient detail regarding the changes the authors made to the Tiedtke scheme in TGFS v1.1, as well as the experimental design used to demonstrate the positive effects of these changes. Most of the modifications to the Tiedtke scheme are derived from other schemes. From a dynamical perspective, these modifications are not entirely independent in their effects on controlling precipitation strength associated with unresolved convection.
For readers like me, it is important that the manuscript include discussions that clarify which changes are most effective at alleviating precipitation forecast biases. I recommend that the authors revise the manuscript to expand Figure 3 by incorporating results from all experiments summarized in Table 2 for the afternoon thunderstorm case, as in Figure 6 for the southwesterly flow case.
Given the partition of resolved and unresolved air motion within the TGFS numerics, the modified Tiedtke scheme—due to its suitability for gray-zone resolutions—should represent the unresolved convective mass flux relative to the resolved counterpart in a manner that diminishes as grid size decreases. Therefore, I also recommend that the revised manuscript include a discussion, supported by numerical evidence if necessary, addressing whether the overestimate of precipitation shown in Figure 4 for the SCA experiment can be alleviated solely by reducing the mass flux scaling factor. This discussion is essential to help readers determine whether the other changes, aside from the mass flux scaling, are equally important in making the modified Tiedtke scheme scale-aware.Citation: https://doi.org/10.5194/egusphere-2025-5324-RC3 - AC3: 'Reply on RC3', Chang-Hung Lin, 03 Apr 2026
-
RC6: 'Comment on egusphere-2025-5324', J. M. Piriou, 24 Feb 2026
In this study, the authors start with a 2016 version of the Tiedtke-Bechtold (TDK) scheme, which they modify significantly to meet the needs of Taiwanese forecasters, in particular to reduce precipitation and improve its localization in a 5 km grid model. This work updates TDK based on other published work.
More specific comments:
- It is interesting, as you do, to validate the modifications by stacking them on top of each other: ORI, SCA, …, CRH.
- Figure 4 shows the temporal sequence of precipitation, with precipitation forecasts showing a 4 hours lead. It would be interesting to plot the diurnal cycle of precipitation over a period of one month to see how much of this time lead is a systematic flaw in the convection scheme.
- Section 3.4 (from line 171) “Dependency on environmental relative humidity for convection triggering, entrainment, and detrainment” also aims to reduce intense precipitation (see line 172). The revised entrainment values are based on ... Bechtold 2008, which is not recent, especially since this work develops a 2016 version of the MPAS of the Tiedtke scheme, which is more recent. It seems that your work would have been easier if you had based it on a more recent version of the Tiedtke scheme.
- In section 3.1, starting at line 125, “Scale-aware parameterization,” the authors do not describe how they handle mesh dependency, but simply cite publications. It would be good to briefly describe the process with a few simple equations, so that this article would be more “stand alone.”
- In section 3.3, “Definition of convective cloud top,” by changing the threshold from 0 to 10^-8 kg/kg, the height of convective clouds is significantly reduced and the intensity of precipitation decreases (see line 170). This surprising result needs to be explained: what physical mechanisms make the TDK scheme so sensitive to this threshold? Could you explain?
- In section 5.2, “Southwesterly flow case,” it is surprising that you used 48- to 60-hour lead time, because in an operational context, 24- to 36-hour lead time are more important for decision-making. Is there a reason for this? Similarly, in Figure 11, you show precipitation between 96 and 108 hours, which sounds like a methodological problem, because if we are interested in such distant lead times, we should also show the 24-36 hour, 48-60 hours (which you have done), and 72-84 hours, which would give an idea of the spread linked to the lead time itself.
- You mention in line 106 that you are using the Bechtold et al. 2014 approach, which delays the diurnal cycle by modifying the CAPE based on criteria related to atmospheric boundary layer turbulence. Our experience with the Tiedtke scheme indicates that this delay, which is formulated differently over land and over ocean, tends to favor rainfall over ocean versus rainfall over land, which is particularly clear on precipitation fields near the coasts. You could also look into this for your future studies.
This work is interesting. I recommend its publication.
Citation: https://doi.org/10.5194/egusphere-2025-5324-RC6 - AC4: 'Reply on RC6', Chang-Hung Lin, 03 Apr 2026
Peer review completion
Journal article(s) based on this preprint
Model code and software
CWA TGFS v1.1 model The Unified Forecast System https://doi.org/10.5281/zenodo.17452036
CWA TGFS v1.1_CRH model The Unified Forecast System https://doi.org/10.5281/zenodo.17452254
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 1,547 | 765 | 198 | 2,510 | 93 | 123 |
- HTML: 1,547
- PDF: 765
- XML: 198
- Total: 2,510
- BibTeX: 93
- EndNote: 123
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Chang-Hung Lin
Guo-Yuan Lien
Ling-Feng Hsiao
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(2236 KB) - Metadata XML
The authors present modifications of the Tiedtke convection scheme and apply and evaluate these in the 4 km nested runs over Taiwan. I like the presentation of the manuscript as the modifications are presented and evaluated step by step for case studies followed by a general evaluation, the manuscript is also well written in general. I also broadly agree with these modifications (see below) and the discussion of the convective features related to model deficiencies.
specific comments:
-SCA: yes the mass flux scaling should definitely be applied before applying the CFL criterium. Could you mention what model time step you use? Also in the CFL criterium computation it might be better to use for very small time steps eg dp/(g*max(dt,300))
- cloud top criterium: yes the convective cloud top height is definitely overestimated in the original version, you might also want to test an additional criterium on the cloud top height that has been introduced in the German weather service (DWD) and ECMWF, ie for the updraught to continue in the vertical dT/dz<-3.e-3 and Buo>-2
- entrainment modification: Here I find it less convincing, you say it further reduces "convection" but the results show more light precipitation indicating more widespread light convection. Indeed, when looking at your formula it appears that for z=100m you get entrainment rates of 10-3 m-1 which is similar to the original, but for z=500 m (a typical cloud base height) it is 2x10-4 which is already much smaller than the original (Note als you say d1 is non-dimensional but it has units m^-1).....??? At leats in your conclusions you seem not to have adopted this version for operations.
You also discuss in detail - a discussion which is welcome - reasons for having precipitation bands sometimes too far offshore compared to observations. Then you mention at the end that this problem is even more widespread in winter over a relatively warm sea. Indeed, this problem is shared by all global modelling centres employing a convective parametrization. The reasons are as you broadly mentioned the interaction between heating and the circulation (forming a quasi stationnary near coast circulation, having updraught and downdraught in the same grid cell and the lack of advection. We could largely address this problem in the upcoming operational version at ECMWF (cycle 50r1, April 2026) by handing over a significant amount of convective precipitation to the large-scale cloud scheme where it is advected and evaporated (no publication available)