the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 15 Jun 2026
A dense rain-gauge observation framework for advection-consistency diagnosis of urban convective rainfall
Abstract. Convective rainfall identification is commonly based on intensity thresholds, radar reflectivity patterns, or supervised classification methods. However, these approaches often lack an explicit connection to the physical constraints governing rainfall-field evolution. This study proposes a physics-informed definition of convective rainfall based on the continuity equation, interpreting convective rainfall as a spatiotemporal deviation from rainfall evolution consistent with horizontal advection under the rainfall continuity equation.
The analysis utilizes data from a high-density rain gauge network in Bangkok. Rainfall fields are reconstructed on a kilometer-scale grid, and the motion field is estimated using the Horn-Schunck optical flow method under no-flux boundary conditions. The rainfall field is then advected forward in time without including local source or sink terms. The residual between the observed field and the purely advected field is interpreted as representing non-advective processes. The framework is designed for dense urban rain-gauge networks, where direct surface rainfall measurements are available and quantitatively calibrated radar rainfall products are not available for the present analysis.
Residual-feature analyses indicate that the detected events exhibit properties consistent with convective rainfall, including spatial localization, rapid development, and a clear association with the diurnal cycle. The separation between convective and non-convective events in the residual feature space provides an empirical evaluation of the diagnostic behavior of the framework.
The results suggest that advection consistency provides an interpretable physics-based foundation for defining and detecting convective rainfall using rain gauge observations alone.
- Preprint
(6220 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 20 Jul 2026)
- RC1: 'Comment on egusphere-2026-1314', Jenna Ritvanen, 26 Jun 2026 reply
-
RC2: 'Comment on egusphere-2026-1314', Anonymous Referee #1, 05 Jul 2026
reply
The manuscript presents an approach to detect and identify convective rainfall based on relatively dense rain gauge networks. The methodology is applied in the Bangkok metropolitan area. I find the method is novel and can be of interest for the community. At the same time, the manuscript requires important work before it can be considered for publication. In particular, I am concerned from 1) oversimplification of some aspects and overselling of the connection between the used method and precipitation physics, 2) sub-optimal organisation of the manuscript and presentation, 3) evaluation and discussion present circular reasoning, 4) codes are not shared.
In its current form, the manuscript overly simplify some instances and makes a too strong connection between the used method and “physics”, leading the readers to believe that the identification of conviction is physically based — in practice it uses some assumptions related to the continuity equation but does not provide connection with the actual physical processes behind convective precipitation. This is fine per se, but presentation should be clear about it.
I come as second reviewer posting comments in this open discussion. Therefore, I will not repeat the aspects already raised by reviewer #1.
General comments
- While I understand the physical rationale behind the use of approach to identify convection, I think a stronger case should be made on it because all precipitation systems, not only convective, may be naturally associated with growth and decay. This means that spatial and temporal scales of the residuals become important, together with the size of the domain and the gauge density. These aspects are crucial for replicating the study and for using the method in other applications.
- Overall the link with physics of the proposed approach remains indirect, both because of the point above and because of the smoothed nature of the rain gauge information. I think the manuscript should use a more prudent phrasing when referring to physics and physics-based when presenting this method.
- There is some circular reasoning between the identification method and the evaluation, as the same variables (mainly advection residuals) are used for both identification and validation. More on this below.
- Neither codes nor data are made available so the study cannot be replicated. This is slightly concerning also considering that codes were drafted using AI tools. Also: what is the point of presenting a new method in an open access journal if then codes to use the method are not provided?
- A discussion of how the method should be (re-)parametrised for a new study area is missing. Which parts should be adapted? How?
Presentation
- In general, the organisation makes the paper very difficult to follow. I suggest to reduce the number of sections to a more usual structure (data, methods, results — and then subsections). I suggest to reorganise the methods to reduce the large number of repetitions (some are mentioned more in detail below).
- Section 2: the literature review seems only marginally relevant to the topics discussed in the paper. The section should be better focused and should only include relevant aspects.
- Verb tense in line 183 is different from the previous paragraphs.
- Line 27-35: I understand these lines were added following the first editorial screening. I feel the concepts are needed in the paper, but this is a bit too early in the manuscript. Also, currently it sounds more like discussion than real introduction. I suggest to move it a bit later, e.g. after the following paragraph in which methods to identify convection are mentioned, and to rephrase to make it more introductory.
- Lines 174-175 state “temporal alignment across stations to ensure consistency at the 15-min resolution throughout the study period” and lines 180-182 state “The dataset is temporally aligned such that each time step corresponds to a 15-min accumulation period, ensuring that rainfall fields can be compared consistently between consecutive time steps for motion estimation.” It sounds like a repetition. Or perhaps I misunderstood one of the two parts, in which case they should be better clarified.
- Lines 196-197 perhaps need to be supported by references
- The notation for u=(u,v) in eq. 3 is different from the one in section 5.3 V=(u,v).
- Section 4.1 is a repetition of sections 3.4 and Section 4.2 is also a repetition of things already said in section 3. It feels like section 4 did not add any new information. Can be removed?
- The introductory part of section 8 seems to repeat some methodological concepts that would better pertain the previous sections
- I suggest to move lines 456-459 to the next section and to merge it to the first line of the next section.
- Lines 521-522: verbatim repetition from the previous subsection
- Lines 630-631 give an important background motivation for the study that I think should be better placed in the introduction.
Technical aspects
- How dense should a gauge network be to use this approach successfully? How large/small should the study area be?
- Data QC: it should be better specify how the “physically plausible limits” for 15-min rainfall were determined and what these limits are.
- I expected section 4 to give details on how the rainfall fields were “smoothed”, but I did not get that info. This information should be provided before section 5 discusses the motion fields.
- Large ungauged areas - how is the smoothing affecting the results?
- It seems one of the assumptions for the method to work is that the study area (urban area) is much larger than the typical scale of convection. This is because it only identifies convection that grows within the study area and does not identify convective cells that are advected through the domain. If I understood correctly, this should be better clarified.
- Line 456-458: “Positive residuals… weakening or decay.” This is true only if the advection time is very short. In general, residuals are to be expected also when the advection field is not stationary in time. Also, residuals may appear following the fact that the station network is sparse with respect to the scale-size of convective cells.
- Some of the results discussed in 8.3 and 8.4 are naturally following the way convective storms are identified as described in 7.6 (notably, among others, the last sentence in lines 567-570). Perhaps it would be best to better emphasise what is independent from those choices and what is a consequence.
Citation: https://doi.org/10.5194/egusphere-2026-1314-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 6 | 3 | 0 | 9 | 0 | 0 |
- HTML: 6
- PDF: 3
- XML: 0
- Total: 9
- BibTeX: 0
- EndNote: 0
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Please see the attachment for the comments.