Rainfall regimes, their transitions, and long-term changes during Indian summer monsoon

Raut, Bhupendra; Deshpande, Aditi; Kamble, Devyani; Ingle, Sandip; Naik, Parmeshwar; Walde, Shwetal; Kumar, P. Pradeep; Sen, Purnendranath

doi:10.5194/egusphere-2025-4585

Preprints

https://doi.org/10.5194/egusphere-2025-4585

Preprints

06 Nov 2025

| 06 Nov 2025

Rainfall regimes, their transitions, and long-term changes during Indian summer monsoon

Bhupendra Raut, Aditi Deshpande, Devyani Kamble, Sandip Ingle, Parmeshwar Naik, Shwetal Walde, P. Pradeep Kumar, and Purnendranath Sen

Abstract. We present a diagnostic framework and accompanying dataset of daily rainfall regimes during the Indian Summer Monsoon (ISM) for June–September 1961–2018. Using high-resolution (0.25°) daily rainfall and unsupervised k-means clustering, eleven objectively defined spatial rainfall patterns were identified and linked with characteristic low-level winds, sea-level pressure and moisture fields, separating different modes of active and break phases. The dataset provides (i) centroid rainfall patterns of each regime and (ii) daily cluster IDs, enabling reconstruction of the full temporal sequence of rainfall regimes and calculation of transition probabilities between states. Transition analysis confirms that break phases are the most persistent while monsoon depressions are more transient, mirroring observed synoptic life cycles. A decomposition of rainfall change between 1961–1989 and 1990–2018 shows that drying in Northeast India (∼9 %) is driven by fewer break periods with northeast-focused rainfall, whereas Gangetic Plain drying (∼7 %) is linked to both intensity and frequency changes. This regime-based approach provides a powerful diagnostic tool to examine synoptic drivers, long-term changes in rainfall intensity and frequency, regime transition dynamics, and to evaluate model representation of ISM variability, teleconnections and trend attribution.

Received: 30 Sep 2025 – Discussion started: 06 Nov 2025

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Preprint (PDF, 3003 KB)

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 (3003 KB)

Download & links

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Journal article(s) based on this preprint

07 Jul 2026

A regime-based diagnosis of transition probabilities and changes in frequency and intensity of Indian Summer Monsoon rainfall

Bhupendra A. Raut, Aditi Deshpande, Devyani Kamble, Sandip Ingle, Parmeshwar Naik, Shwetal Walde, P. Pradeep Kumar, and Purnendranath Sen

Weather Clim. Dynam., 7, 1173–1188, https://doi.org/10.5194/wcd-7-1173-2026,https://doi.org/10.5194/wcd-7-1173-2026, 2026

Short summary

Bhupendra Raut, Aditi Deshpande, Devyani Kamble, Sandip Ingle, Parmeshwar Naik, Shwetal Walde, P. Pradeep Kumar, and Purnendranath Sen

Interactive discussion

Status: closed

CC1:
'Comment on egusphere-2025-4585', Nima Zafarmomen, 03 Dec 2025
The manuscript presents a regime-based analysis of Indian Summer Monsoon (ISM) rainfall for JJAS 1961–2018, using:

IMD 0.25° daily rainfall

k-means clustering to identify 11 spatial rainfall regimes

NCEP–NCAR Reanalysis-1 to characterize circulation, pressure, and moisture anomalies for each cluster

Markov-chain style transition probabilities between regimes

A decomposition of rainfall change (1961–1989 vs 1990–2018) into frequency vs intensity contributions by regime and region.

The study aims to link objectively derived rainfall regimes to synoptic drivers (depressions, trough position, breaks, etc.), examine their transition dynamics, and use these regimes to interpret long-term regional rainfall changes (particularly over Thar & Kutch, Indo-Gangetic Plains, and Northeast India).
The manuscript would benefit from a clearer articulation of what is genuinely new compared to prior regime-based studies of the ISM (e.g., Straus 2022, Catto et al. 2012–style decompositions, previous clustering of rainfall vs winds). Please more explicitly contrast your k-means rainfall regimes with wind-based regimes and IMD active/break diagnostics, and state what new physical insight emerges.

The Elbow method and five initializations provide only limited evidence that 11 clusters are “optimal,” especially for high-dimensional rainfall fields. Please show more formal robustness tests (e.g., silhouette scores, explained variance, random subsampling, or clustering on sub-periods), and ideally a sensitivity example for k = 9 or 12.

The transition matrix is interpreted as a Markov chain, but daily rainfall regimes are highly autocorrelated and embedded in multi-day synoptic systems. Please discuss the validity and limitations of the Markov assumption, and consider including multi-day transitions (e.g., 2-day lag).

Several regimes are qualitatively associated with monsoon depressions, mid-tropospheric cyclones, and trough positions, but the linkage remains somewhat descriptive. Can you provide more objective metrics (e.g., depression track climatology, MTC detection, trough latitude index) to verify that specific clusters indeed correspond to those synoptic features?

I strongly recommend to consider "Analysis of historical global warming impacts on climatological trends for the partially gauged Hirmand river basin based on multiple data products and bias correction methods" in your paper.
Citation: https://doi.org/10.5194/egusphere-2025-4585-CC1
- AC3: 'Reply on CC1', Bhupendra Raut, 31 Mar 2026
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-4585/egusphere-2025-4585-AC3-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2025-4585-AC3
RC1:
'Comment on egusphere-2025-4585', Adway Mitra, 09 Dec 2025

The paper focuses on identifying spatial patterns of Indian Monsoon rainfall using K-means clustering. These patterns are identified with specific phases of Indian Monsoon, such as the break phase with low overall rainfall, and several other phases, each of which is characterized by rainfall over some particular region like the Indo-Gangetic Plane, Northeastern India or Kutch and Thar. It is argued that these 10-odd clusters represent the spatial distribution of rainfall on most of the days in the study period, it is assumed that these patterns follow Markovian dynamics, and transition probabilities are worked out with some physical justifications. The change in the frequency of each of the clusters/patterns and the intensity of rainfall associated with them is studied across the entire study period, and such change is considered as a marker of climate change. I have the following observations about this paper:
1) Originality - This is not the first work that aims to identify a set of "canonical" patterns of rainfall. MISO or Monsoon Intra-Seasonal Oscillations (Suhas et al, 2013, "An Indian monsoon intra-seasonal oscillations (MISO) index for real time monitoring and forecast verification") and the binary Random Fields (Mitra et al, 2019, "Spatio-temporal patterns of daily Indian summer monsoon rainfall ", Sharma et al, 2020, "Spatio-temporal relationships between rainfall and convective clouds during Indian monsoon through a discrete lens" ). The patterns identified here have significant similarity with these patterns, and hence some sort of qualitative and quantitative comparison is needed, along with a justification of why at all more patterns like this is needed.
2) Choice of K-means - It is not very clear what was considered as the feature vectors for clustering. Is it vectorized form of the rainfall values of each day? If so, we may be missing out on the spatial aspects due to vectorization (2D to 1D). Also, what is the nature of the space on which we are doing the clustering? What kinds of clusters are desired, and is K-means the ideal way to do it? What is the within-cluster variance and across-cluster separations?
3) What is the basis of considering the first-order Markovian dynamics of the clusters? What is the physical justification of it? How many days does each cluster persist and why? The kind of transition dynamics that has been estimated - how can it be justified using known processes (oscillations etc) associated with the Monsoon>
The paper may become suitable for acceptable and publication after these queries are answered and the paper is modified accordingly.

Citation: https://doi.org/10.5194/egusphere-2025-4585-RC1
- AC1: 'Reply on RC1', Bhupendra Raut, 31 Mar 2026
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-4585/egusphere-2025-4585-AC1-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2025-4585-AC1
RC2:
'Comment on egusphere-2025-4585', Anonymous Referee #2, 18 Jan 2026

Please find my comments in the attached file.

Citation: https://doi.org/10.5194/egusphere-2025-4585-RC2
- AC2: 'Reply on RC2', Bhupendra Raut, 31 Mar 2026
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-4585/egusphere-2025-4585-AC2-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2025-4585-AC2

Interactive discussion

Status: closed

CC1:
'Comment on egusphere-2025-4585', Nima Zafarmomen, 03 Dec 2025
The manuscript presents a regime-based analysis of Indian Summer Monsoon (ISM) rainfall for JJAS 1961–2018, using:

IMD 0.25° daily rainfall

k-means clustering to identify 11 spatial rainfall regimes

NCEP–NCAR Reanalysis-1 to characterize circulation, pressure, and moisture anomalies for each cluster

Markov-chain style transition probabilities between regimes

A decomposition of rainfall change (1961–1989 vs 1990–2018) into frequency vs intensity contributions by regime and region.

The study aims to link objectively derived rainfall regimes to synoptic drivers (depressions, trough position, breaks, etc.), examine their transition dynamics, and use these regimes to interpret long-term regional rainfall changes (particularly over Thar & Kutch, Indo-Gangetic Plains, and Northeast India).
The manuscript would benefit from a clearer articulation of what is genuinely new compared to prior regime-based studies of the ISM (e.g., Straus 2022, Catto et al. 2012–style decompositions, previous clustering of rainfall vs winds). Please more explicitly contrast your k-means rainfall regimes with wind-based regimes and IMD active/break diagnostics, and state what new physical insight emerges.

The Elbow method and five initializations provide only limited evidence that 11 clusters are “optimal,” especially for high-dimensional rainfall fields. Please show more formal robustness tests (e.g., silhouette scores, explained variance, random subsampling, or clustering on sub-periods), and ideally a sensitivity example for k = 9 or 12.

The transition matrix is interpreted as a Markov chain, but daily rainfall regimes are highly autocorrelated and embedded in multi-day synoptic systems. Please discuss the validity and limitations of the Markov assumption, and consider including multi-day transitions (e.g., 2-day lag).

Several regimes are qualitatively associated with monsoon depressions, mid-tropospheric cyclones, and trough positions, but the linkage remains somewhat descriptive. Can you provide more objective metrics (e.g., depression track climatology, MTC detection, trough latitude index) to verify that specific clusters indeed correspond to those synoptic features?

I strongly recommend to consider "Analysis of historical global warming impacts on climatological trends for the partially gauged Hirmand river basin based on multiple data products and bias correction methods" in your paper.
Citation: https://doi.org/10.5194/egusphere-2025-4585-CC1
- AC3: 'Reply on CC1', Bhupendra Raut, 31 Mar 2026
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-4585/egusphere-2025-4585-AC3-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2025-4585-AC3
RC1:
'Comment on egusphere-2025-4585', Adway Mitra, 09 Dec 2025

The paper focuses on identifying spatial patterns of Indian Monsoon rainfall using K-means clustering. These patterns are identified with specific phases of Indian Monsoon, such as the break phase with low overall rainfall, and several other phases, each of which is characterized by rainfall over some particular region like the Indo-Gangetic Plane, Northeastern India or Kutch and Thar. It is argued that these 10-odd clusters represent the spatial distribution of rainfall on most of the days in the study period, it is assumed that these patterns follow Markovian dynamics, and transition probabilities are worked out with some physical justifications. The change in the frequency of each of the clusters/patterns and the intensity of rainfall associated with them is studied across the entire study period, and such change is considered as a marker of climate change. I have the following observations about this paper:
1) Originality - This is not the first work that aims to identify a set of "canonical" patterns of rainfall. MISO or Monsoon Intra-Seasonal Oscillations (Suhas et al, 2013, "An Indian monsoon intra-seasonal oscillations (MISO) index for real time monitoring and forecast verification") and the binary Random Fields (Mitra et al, 2019, "Spatio-temporal patterns of daily Indian summer monsoon rainfall ", Sharma et al, 2020, "Spatio-temporal relationships between rainfall and convective clouds during Indian monsoon through a discrete lens" ). The patterns identified here have significant similarity with these patterns, and hence some sort of qualitative and quantitative comparison is needed, along with a justification of why at all more patterns like this is needed.
2) Choice of K-means - It is not very clear what was considered as the feature vectors for clustering. Is it vectorized form of the rainfall values of each day? If so, we may be missing out on the spatial aspects due to vectorization (2D to 1D). Also, what is the nature of the space on which we are doing the clustering? What kinds of clusters are desired, and is K-means the ideal way to do it? What is the within-cluster variance and across-cluster separations?
3) What is the basis of considering the first-order Markovian dynamics of the clusters? What is the physical justification of it? How many days does each cluster persist and why? The kind of transition dynamics that has been estimated - how can it be justified using known processes (oscillations etc) associated with the Monsoon>
The paper may become suitable for acceptable and publication after these queries are answered and the paper is modified accordingly.

Citation: https://doi.org/10.5194/egusphere-2025-4585-RC1
- AC1: 'Reply on RC1', Bhupendra Raut, 31 Mar 2026
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-4585/egusphere-2025-4585-AC1-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2025-4585-AC1
RC2:
'Comment on egusphere-2025-4585', Anonymous Referee #2, 18 Jan 2026

Please find my comments in the attached file.

Citation: https://doi.org/10.5194/egusphere-2025-4585-RC2
- AC2: 'Reply on RC2', Bhupendra Raut, 31 Mar 2026
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-4585/egusphere-2025-4585-AC2-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2025-4585-AC2

Peer review completion

AR – Author's response | RR – Referee report | ED – Editor decision | EF – Editorial file upload

AR by Bhupendra Raut on behalf of the Authors (31 Mar 2026) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (11 Apr 2026) by Nili Harnik

RR by Anonymous Referee #2 (26 Apr 2026)

RR by Anonymous Referee #1 (26 Apr 2026)

For final publication, the manuscript should be

accepted as is

accepted subject to technical corrections

accepted subject to minor revisions

reconsidered after major revisions

rejected

Were a revised manuscript to be sent for another round of reviews:

I would be willing to review the revised manuscript.

I would not be willing to review the revised manuscript.

Suggestions for revision or reasons for rejection

I am still not satisfied by the authors' responses to my previous comments about the choice of K-means clustering, and Markovian dynamics.

1) Regarding K-means, the authors have responded that "each grid point serves as an independent axis in this high-dimensional space", but I don't think that each grid can be considered independently of its neighbors in case of spatial data. In fact, this problem will be there not only for K-means, but also for the other clustering approaches mentioned by the authors. Of course, K-means will give some clustering result, but we do not know if the clusters found this way are "suitable" - for this the authors need to justify the results considering inter-class and intra-class distances, and Euclidean distance will not be very suitable in this case- the authors may need to consider a more suitable measure to compare spatial data.

2) In case onf Markovian Dynamics, the response by the authors does not provide the answer to my basic question - why should we consider a Markovian transition system at all in this case? What is the physical basis to justify the Markovian hypothesis that the cluster on day (t+1) depends only on the cluster on day (t), with some probability distribution? What should be the order of the Markov Chain? To answer this question, the authors may try to summarize the most prominent transition patterns ( including self-transitions), and associate them to some well-known dynamics of the monsoon. However, since these are well-known anyway, the authors must also justify what are the new transition patterns that this approach suggests, and what is the physical explanation of such new patterns?

Hide

RR by Anonymous Referee #3 (01 May 2026)

For final publication, the manuscript should be

accepted as is

accepted subject to technical corrections

accepted subject to minor revisions

reconsidered after major revisions

rejected

Were a revised manuscript to be sent for another round of reviews:

I would be willing to review the revised manuscript.

I would not be willing to review the revised manuscript.

Suggestions for revision or reasons for rejection

Review of “A regime-based diagnosis of transition probabilities and changes in frequency and intensity of Indian Summer Monsoon rainfall” by Raut et al.

Recommendation: Major revisions

The editor asked me to review this paper after previous rounds of revisions and the editor also asked me specific aspects to comment on.

This study investigates monsoon rains in India and tries to identify preferred regime states via k-means clustering. K-means is widely used in atmospheric sciences. I have some concerns about the author's use of k-means. Using gridded data has the problem of spatial autocorrelation which can affect the clustering. Typically one could do a EOF/PCA analysis to get independent data but that is probably not advisable for precipitation. So I would suggest including the longitudes and latitudes into the k-means analysis; maybe also including topography might be advantageous. That might improve the k-means clustering analysis.

Fig. 2 shows the WCSS but to me the change would occur at cluster number 4 or 5. Not 11. Later in the manuscript the authors also group the 11 clusters into 4. 4 clusters would be consistent in my view with Fig. 2. It would also be useful to see a silhouette plot.

The authors show that there is significant change in precipitation between the 1st and 2nd half of the data period. K-means assumes stationarity of the data. Are the clusters stable when k-means is applied to the 1st and 2nd half of data? Such an analysis would make the study also more interesting.

The authors then compute transition probabilities assuming the Markov property. It is necessary to show that the Markov property is valid. What does the partial autocorrelation function look like? Or a transition frequency analysis between t+1 and t-1 could be performed to see whether they are almost independent.

Hide

ED: Reconsider after major revisions (03 May 2026) by Nili Harnik

AR by Bhupendra Raut on behalf of the Authors (12 Jun 2026) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (18 Jun 2026) by Nili Harnik

RR by Anonymous Referee #3 (20 Jun 2026)

ED: Publish as is (20 Jun 2026) by Nili Harnik

AR by Bhupendra Raut on behalf of the Authors (21 Jun 2026) Manuscript

Journal article(s) based on this preprint

07 Jul 2026

A regime-based diagnosis of transition probabilities and changes in frequency and intensity of Indian Summer Monsoon rainfall

Bhupendra A. Raut, Aditi Deshpande, Devyani Kamble, Sandip Ingle, Parmeshwar Naik, Shwetal Walde, P. Pradeep Kumar, and Purnendranath Sen

Weather Clim. Dynam., 7, 1173–1188, https://doi.org/10.5194/wcd-7-1173-2026,https://doi.org/10.5194/wcd-7-1173-2026, 2026

Short summary

Bhupendra Raut, Aditi Deshpande, Devyani Kamble, Sandip Ingle, Parmeshwar Naik, Shwetal Walde, P. Pradeep Kumar, and Purnendranath Sen

Viewed

Total article views: 3,907 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
1,719	1,980	208	3,907	155	149

HTML: 1,719
PDF: 1,980
XML: 208
Total: 3,907
BibTeX: 155
EndNote: 149

Views and downloads (calculated since 06 Nov 2025)

Month	HTML	PDF	XML	Total
Nov 2025	560	338	60	958
Dec 2025	304	447	55	806
Jan 2026	355	328	50	733
Feb 2026	164	316	14	494
Mar 2026	214	276	17	507
Apr 2026	71	75	6	152
May 2026	26	77	4	107
Jun 2026	9	11	0	20
Jul 2026	14	107	2	123
Aug 2026	2	5	0	7

Cumulative views and downloads (calculated since 06 Nov 2025)

Month	HTML	PDF	XML	Total
Nov 2025	560	338	60	958
Dec 2025	304	447	55	806
Jan 2026	355	328	50	733
Feb 2026	164	316	14	494
Mar 2026	214	276	17	507
Apr 2026	71	75	6	152
May 2026	26	77	4	107
Jun 2026	9	11	0	20
Jul 2026	14	107	2	123
Aug 2026	2	5	0	7

Viewed (geographical distribution)

Total article views: 3,896 (including HTML, PDF, and XML) Thereof 3,896 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 10 Aug 2026

Download

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Preprint (3003 KB)
Metadata XML

Short summary

We demonstrated unsupervised clustering methodology and post-clustering analysis framework by analysing daily Indian summer monsoon rainfall to uncover recurring weather patterns and their transitions. We show breaks are prolonged, rain-bearing systems are transient, and decomposed the frequency and intensity effects in long term rainfall changes.


Total:	0
HTML:	0
PDF:	0
XML:	0