Comparing drivers of hydrological shifts across regions: the case of southern Australia

Campion, Nyree; Fowler, Keirnan; Turner, Margot; Hall, Joel

doi:10.5194/egusphere-2026-378

Preprints

https://doi.org/10.5194/egusphere-2026-378

Preprints

16 Feb 2026

| 16 Feb 2026

Comparing drivers of hydrological shifts across regions: the case of southern Australia

Nyree Campion, Keirnan Fowler, Margot Turner, and Joel Hall

Abstract. Several regions globally have recently experienced persistent shifts in the relationship between rainfall and runoff, triggered by multi-annual drought. These regions are climatically diverse; however, few assessments have yet been undertaken to draw parallels (if any) between the processes responsible. We present a comparative analysis of these hydrological shifts between south-west Australia and south-east Australia, two regions separated by over 2,700 km (~1,700 miles). We apply existing methods based on Hidden Markov modelling to characterise shifts in rainfall-runoff relationships in 254 catchments in Eastern and 54 in Western Australia. Of the catchments analysed, 51 % of Eastern and 63 % of Western catchments displayed a movement away from the historical rainfall-runoff relationship to one of reduced flow generation following a multi-year period of drier climate. The reduced flow state persisted in 31 % of catchments in Eastern Australia despite a return to near-normal climatic conditions after multi-year drought, whereas in Western Australia neither the climate nor the flow states have returned to earlier norms (i.e. nearly all shifted catchments have stayed shifted). Interestingly, some catchment characteristics that were correlated with shifts in one region were anticorrelated in the other, possibly indicative of different causative processes. For example, in Western Australia the shifted catchments are typically those that have not been cleared for agriculture and thus retain forest coverage; the opposite is true in Eastern Australia. We suggest a possible link to pre-existing trends in groundwater for cleared catchments, where those in Western Australia may have been experiencing rising groundwater levels due to clearing occurring recently (mid-1900s) relative to Eastern Australia (late-1800s). These findings suggest the importance of land use history when considering changes in rainfall-runoff relationship. We recommend further comparative studies be conducted to synthesise understanding across geographies and better inform water planning decisions under climate change.

Received: 23 Jan 2026 – Discussion started: 16 Feb 2026

Competing interests: At least one of the (co-)authors is a member of the editorial board of Hydrology and Earth System Sciences.

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

Nyree Campion, Keirnan Fowler, Margot Turner, and Joel Hall

Status: final response (author comments only)

RC1: 'Comment on egusphere-2026-378', Anonymous Referee #1, 03 Mar 2026

Review of “Comparing drivers of hydrological shifts across regions: the case of southern Australia” by Campion and co-authors

General comment

Campion and co-authors present a comparative analysis of characteristics and causes of hydrological shifts following drying in southern Australia, by means of Hidden Markov modelling and a correlation analysis. They further put their findings in the context of groundwater and land-use changes, for which they provide some evidence from available – sparse - data. In so doing, they illustrate widespread shifts in rainfall-runoff (R-R) relationships in both south-west and -east Australia during the Millennium drought around the beginning of the century, but different possible causes for such shifts (evapotranspiration sustainment by high groundwater tables in forested catchments in the south-west, and low groundwater tables in cleared catchments in the south-east). The topic is timely and of interest to HESS readership, given the increasing evidence of worldwide changes in catchment behaviour under drying, our still poor understanding of their causes, and the socio-ecological implications of such changes. While the authors rely on well-established methods to investigate the problem at hand and standard statistical analyses, they present some interesting new insights on the topic, and I commend the manuscript for its comparative nature and the amount of data – from different sources – brought together. I believe that this can be considered suitable for publication in HESS after a round of revisions to address my comments below, mostly related to clarification on data consistency and the proposed interpretation of results.

Major comments

1. Streamflow dataset consistency

I think more information on this point is needed. The authors thoroughly describe how they selected catchments according to their hydroclimate, but they fall short in the rationale for selecting catchments according to data availability, if any. They disclaim that they analyse a varying number of catchments over the years (caption of Fig. 4) and some infilling for missing data (L138–141). But have thresholds on maximum missing data allowed and minimum number of years with available streamflow observations been set? I ask the authors to consider these points and clarify them in the text. Also, adding the number of analysed catchments over the years in Fig. 4 would be appreciated.

2. Correlation analysis

The authors explore correlation between 118 (!!) catchment properties and the magnitude of shifts in R-R relationships, but they only report results in Fig. 7 and in the text for a bunch of them. I see that reporting all correlation values, even in an appendix or supplementary, may be overwhelming, but I wonder whether the authors could find a way to summarize them, for instance by reporting the maximum correlation for property type (e.g., climate/land-cover).

3. Groundwater data and proposed explanation for hydrological shifts

In the discussion, the authors introduce data from 9 groundwater wells to support their hypothesis of diverging causes for hydrological shifts in southern Australia, together with information on land-use changes. While I believe introducing the groundwater data in the discussion only is a legitimate choice, I recommend adding more information on this dataset (e.g., in Sect. 2 or in an appendix as done for the land-use datasets). How were the groundwater wells selected? I would assume for data availability, but this is not stated. Where are the wells located? Showing their location in a map with respect to the catchments under study would be interesting to me. Are the groundwater wells potentially influenced by human activities? Are data before 1990 available? I believe that data even from just a few wells before the 1990s would be highly beneficial to support the hypothesis of still increasing groundwater tables at the onset of the Millennium drought because of land clearing in the 1940s. If no data is available, could the authors better discuss how this hypothesis fits in the broader groundwater literature and previous works on shifts in south-western Australia during the Millennium drought because of decreases in groundwater contribution (e.g., Kinal & Stoneman, 2012 and others already cited in the introduction)? An alternative (or add-on) I see to the use of data before 1990 would be showing also groundwater data for catchments without (or with low) clearing in south-western Australia, if available. Also, is the reported rainfall decline since the 70s similar in western and eastern Australia? Adding information on this should be rather straightforward for the authors from the dataset they currently use and it would support more strongly the point of differences between the two regions because of differences in land-use history. In summary, I believe the authors could better work out their proposed mechanistic explanation of causes in shifts in south-western Australia showing some additional data, if available, and adding some more discussion.

4. Language

The paper is well written, even though I found it at times slightly difficult to follow, in two directions in particular. First, the authors refer quite often to previous works for their datasets and methods (e.g., Sect. 2.2). I believe providing some more details on the main characteristics of the datasets and methods used, rather than just referring to previous papers, would help the readers. Second, in the discussion Sect. 3.2 the main points are not so easy to grasp (also because of the many interactions in place). I suggest the authors streamlining this part. Maybe a summary conceptual figure could help? Not required, just as an idea that I leave the authors to decide on.

Minor comments

5. L11–12 and L87, as a reader not particularly familiar with this geographical area, I would appreciate here some information on catchment differences between the two regions, other than the distance one to another only, to better appreciate the variety of the case study.

6. Abstract, I would suggest clarifying the study period and when the reported shifts occurred.

7. L118, could the authors please quantify this? Also, could focusing on arid catchments (aridity as defined by the authors less than 1) be maybe more informative?

8. Fig. 1, the text in panels b-d is rather small to read, I wonder whether a different arrangement of this figure could help in making these panels bigger and thus allow to increase font size. Also, I would find useful information on land use here as well, given the central role in the manuscript.

9. L162–163, could the authors please provide more details on how the magnitude and timing of onset of hydrological shifts are estimated? This would ease the readers in following the work without having to overly rely on previous works in my view (see comment #3).

10. L186, why was the year 1991 chosen?

11. L198–200, I assume these numbers relate to Fig. 3, which I would suggest placing before current Fig. 2.

12. L220–222, isn’t this because the normal state is defined as the flow in 1975 (L178)?

13. L320–321, what exactly the authors did here is not totally clear to me. I believe that with this sentence they mean that they consider wells with data over the common period 1990–2024 and they use different y axes, however, I would suggest clarifying the data pre-processing, if any.

14. L363–369, could the authors actually show these differences in e.g. maps? It seems to me that they have all the required information from their dataset anyway.

15. Fig. 9, what do the abbreviations on the y axis stand for?

Technical corrections

16. Sect. 3 is repeated twice.

References

Kinal, J., & Stoneman, G. L. (2012). Disconnection of groundwater from surface water causes a fundamental change in hydrology in a forested catchment in south-western Australia. Journal of Hydrology, 472–473, 14–24. https://doi.org/10.1016/j.jhydrol.2012.09.013

Citation: https://doi.org/10.5194/egusphere-2026-378-RC1

AC1: 'Reply on RC1', Nyree Campion, 24 Jun 2026

Response to Anonymous Referee Comment 1

We thank anonymous referee 1 for their thoughtful comments and suggestions, which will improve the manuscript substantially, particularly with respect to transparency of data and analyses. Please see below for our replies to each of the detailed comments.

	Comment	Response
1	Streamflow dataset consistency
	I think more information on this point is needed. The authors thoroughly describe how they selected catchments according to their hydroclimate, but they fall short in the rationale for selecting catchments according to data availability, if any. They disclaim that they analyse a varying number of catchments over the years (caption of Fig. 4) and some infilling for missing data (L138–141). But have thresholds on maximum missing data allowed and minimum number of years with available streamflow observations been set? I ask the authors to consider these points and clarify them in the text.	The catchments selected for analysis were restricted to CAMELS-AUS v2 (Fowler et al., 2024) catchments, which themselves are Hydrological Reference Stations (Turner, 2012). This limits selection to gauging stations which have experienced less than 10% change in land use over the period of record upstream of the station, and to those which are missing less than 5% of data over the period of record. This will be further clarified in the manuscript. In addition, we will add text to clarify the minimum data requirements for a catchment to be included, as requested.
	Also, adding the number of analysed catchments over the years in Fig. 4 would be appreciated.	Figure 4.0 will be updated to include the number of catchments analysed for each year.
2	Correlation analysis
	The authors explore correlation between 118 (!!) catchment properties and the magnitude of shifts in R-R relationships, but they only report results in Fig. 7 and in the text for a bunch of them. I see that reporting all correlation values, even in an appendix or supplementary, may be overwhelming, but I wonder whether the authors could find a way to summarize them, for instance by reporting the maximum correlation for property type (e.g., climate/land-cover).	We will include additional information within the appendix on the correlation values computed. This may be represented as a table or a correlogram, with the aim of showcasing metrics which have demonstrated significant and strong correlation.
3	Groundwater data and proposed explanation for hydrological shifts
	In the discussion, the authors introduce data from 9 groundwater wells to support their hypothesis of diverging causes for hydrological shifts in southern Australia, together with information on land-use changes. While I believe introducing the groundwater data in the discussion only is a legitimate choice, I recommend adding more information on this dataset (e.g., in Sect. 2 or in an appendix as done for the land-use datasets). How were the groundwater wells selected? I would assume for data availability, but this is not stated.	The groundwater wells were selected based on availability, period of record, and location relative to catchments included in the analysis. The manuscript will be updated to include further information on these criteria and the relevant data sources.
	Where are the wells located? Showing their location in a map with respect to the catchments under study would be interesting to me.	A map will be included in the appendix showing the well location with respect to the catchments analysed.
	Are the groundwater wells potentially influenced by human activities?	We agree that this is an important consideration and recognise that the groundwater wells may be influenced by human activities. All bores selected in the Western Australian study region are within aquifers that are proclaimed, and groundwater is used for irrigated agriculture, water supply, industry and urban spaces, amongst other users. As such, the groundwater levels in each of these bores may be influenced by abstraction in addition to climate and land clearing. The impact of abstraction is predicted to be minimal for some of the bores; however, we will revise our limitations section to acknowledge this more explicitly. We will also provide further details in the appendix, where available, on the likely degree of abstraction influence on the selected bores.
	Are data before 1990 available? I believe that data even from just a few wells before the 1990s would be highly beneficial to support the hypothesis of still increasing groundwater tables at the onset of the Millennium drought because of land clearing in the 1940s.	We agree this would be useful. We have done an initial search for older bores and found nothing extending further back than 1986. We will confer with another colleague in a different department who we believe might have knowledge of older data, and if this is successful then we will update the plots accordingly.
	If no data is available, could the authors better discuss how this hypothesis fits in the broader groundwater literature and previous works on shifts in south-western Australia during the Millennium drought because of decreases in groundwater contribution (e.g., Kinal & Stoneman, 2012 and others already cited in the introduction)?	Thank you for the suggestion, and we will certainly expand this discussion with reference to Kinal and Stoneman (2012), Hughes et al. (2012) and others cited earlier in the paper.
	An alternative (or add-on) I see to the use of data before 1990 would be showing also groundwater data for catchments without (or with low) clearing in south-western Australia, if available.	Figure 9 depicts groundwater levels for forested catchments (those which have experienced little to no clearing). We will revise the figure caption to clarify this.
	Also, is the reported rainfall decline since the 70s similar in western and eastern Australia? Adding information on this should be rather straightforward for the authors from the dataset they currently use and it would support more strongly the point of differences between the two regions because of differences in land-use history. In summary, I believe the authors could better work out their proposed mechanistic explanation of causes in shifts in south-western Australia showing some additional data, if available, and adding some more discussion.	We will revise the discussion to provide further detail on the rainfall decline in South-West WA and Eastern Australia, differences in the hydroclimate trends and how the key climate drivers differ between regions. We will support this discussion with a timeseries of rainfall aggregated across our study catchments, separated by study region.
4	Language
	The paper is well written, even though I found it at times slightly difficult to follow, in two directions in particular. First, the authors refer quite often to previous works for their datasets and methods (e.g., Sect. 2.2). I believe providing some more details on the main characteristics of the datasets and methods used, rather than just referring to previous papers, would help the readers.	We will provide further information on the hydrometeorology and groundwater datasets used within this project, so that the reader is not forced to return to the previous papers before they can interpret the present paper.
	Second, in the discussion Sect. 3.2 the main points are not so easy to grasp (also because of the many interactions in place). I suggest the authors streamlining this part. Maybe a summary conceptual figure could help? Not required, just as an idea that I leave the authors to decide on.	Thanks for noting that this is optional. We will trial some possible options for a conceptual figure to clarify the proposed drivers.
	Minor Comments
5	L11–12 and L87, as a reader not particularly familiar with this geographical area, I would appreciate here some information on catchment differences between the two regions, other than the distance one to another only, to better appreciate the variety of the case study.	We will incorporate information on catchment differences and similarities earlier in the manuscript.
6	Abstract, I would suggest clarifying the study period and when the reported shifts occurred.	The abstract will be updated to include further clarification on the study period and the timing of reported shifts.
7	L118, could the authors please quantify this?	We will include the number of catchments with long-term aridity less than 0.50 in Western Australia.
	Also, could focusing on arid catchments (aridity as defined by the authors less than 1) be maybe more informative?	We agree that grouping catchments by climatic indices could reduce heterogeneity; however, we think a wholesale restructuring would be unhelpful for the aims of this study. Our main objective is to compare hydrological shifts across the two selected regions, and one of the key findings is that the relationship between hydrological shifts and forest cover differs between them. Because forest cover and climate covary to some extent, restructuring the analysis primarily by climate class could obscure this important contrast. We recognise that there is no plot that shows covariance between forest cover and aridity, so we propose this addition to the manuscript.
8	Fig. 1, the text in panels b-d is rather small to read, I wonder whether a different arrangement of this figure could help in making these panels bigger and thus allow to increase font size.	We will revise Figure 1 to better represent key information.
	Also, I would find useful information on land use here as well, given the central role in the manuscript.	In the revision of Figure 1, we will include an additional map panel presenting information on forest coverage or land use more generally.
9	L162–163, could the authors please provide more details on how the magnitude and timing of onset of hydrological shifts are estimated? This would ease the readers in following the work without having to overly rely on previous works in my view (see comment #3).	Hidden Markov Modelling (HydroState; Peterson et al., 2021) detects hydrological shifts by treating runoff as moving between hidden states (normal, low, etc.) that have different rainfall–runoff relationships. The timing of onset is estimated from when the most probable hidden state changes, and the magnitude is estimated from the difference between observed runoff and modelled normal-state runoff, weighted by the probability that the catchment is in a shifted state. We will revise the manuscript to make this aspect of the methods more accessible.
10	L186, why was the year 1991 chosen?	The period between 1991 and 1996 was selected as the reference period, due to it being prior to widespread hydrological shifts in both regions. 1991 was specifically due to data availability. We will clarify this in the text.
11	L198–200, I assume these numbers relate to Fig. 3, which I would suggest placing before current Fig. 2.	These numbers do relate to Figure 3, as such we will revise figure arrangement to better follow the text.
12	L220–222, isn’t this because the normal state is defined as the flow in 1975 (L178)?	The method determines flow state objectively based on data alone. All flow transitions are relative, and the method allows for a “high” flow state. Thus, if a later year had been chosen as “normal”, the same sequence of transitions would have been returned by the method, but instead of it being from “normal” to “low” it would have been from “high” to “normal”. We will update the manuscript to make this clearer.
13	L320–321, what exactly the authors did here is not totally clear to me. I believe that with this sentence they mean that they consider wells with data over the common period 1990–2024 and they use different y axes, however, I would suggest clarifying the data pre-processing, if any.	“mTOC” refers to metres below top of casing, with a greater negative value indicating a further distance from the top of the casing. We have ensured that the x and y axes have the same range (i.e. 10 metre depth and dates from 1990 to 2024) such that the difference in magnitude of water level movement can more clearly be observed. This was intended to show that groundwater levels in Western Australian catchments increase upwards of 9m, while those in Victoria only vary around 2-4m. We will clarify this in the figure and the surrounding text.
14	L363–369, could the authors actually show these differences in e.g. maps? It seems to me that they have all the required information from their dataset anyway.	We will produce maps for inclusion in the appendix which depict the spatial variation of elevation, slope and multi-resolution valley bottom flatness (MrVBF), a topographic index which quantifies the extent of depositional locations.
15	Fig. 9, what do the abbreviations on the y axis stand for?	“mAHD” refers to metres relative to the Australian Height Datum, whereas “mTOC” refers to metres below top of casing. We will revise the figure and caption to ensure consistency.
16	Sect. 3 is repeated twice.	We will revise the numbering of sections.

References

Fowler, K., Zhang, Z., and Hou, X.: CAMELS-AUS v2: updated hydrometeorological timeseries and landscape attributes for an enlarged set of catchments in Australia, Zenodo [data set], https://doi.org/10.5281/zenodo.12575680, 2025.

Hughes, J. D., Petrone, K. C., and Silberstein, R. P.: Drought, groundwater storage and stream flow decline in southwestern Australia, Geophys. Res. Lett., 39, L03408, https://doi.org/10.1029/2011GL050797, 2012.

Kinal, J., and Stoneman, G. L.: Disconnection of groundwater from surface water causes a fundamental change in hydrology in a forested catchment in south-western Australia, J. Hydrol., 472, 14-24, https://doi.org/10.1016/j.jhydrol.2012.09.013, 2012.

Turner, M.: Hydrologic reference station selection guidelines. Melbourne, VIC: Bureau of Meteorology, Australia.

Citation: https://doi.org/10.5194/egusphere-2026-378-AC1

RC2: 'Comment on egusphere-2026-378', Anonymous Referee #2, 15 May 2026

Reviewing Comparing drivers of Hydrological shifts across regions: the case of southern Australia. HESS

The authors explore the dynamic of the rainfall/runoff relationships by comparing a number of catchments in two diverse southern Australian areas. In both cases, more than half of the catchments have exhibited a flow reduction compared to previous conditions, with some catchments in the east and none in the west returning to previous conditions. To explain this behaviour, the authors explore the role of groundwater and the influence of LULC change on groundwater. The manuscript is grammatically well written, but it may benefit from more proofreading. Logically, the cohesion and coherence of the paper spark some concerns, as groundwater sounds a bit like a posterior addition to try explaining the model results better. Although it is treated properly in the introduction, it does not give the same feeling throughout the paper. The model setup and other methodological choices need to be clarified better. The paper would benefit from an improved discussion section, especially 3.2. Figures 1, 4, 5, and 6 need improvements before being acceptable for publication. Only after thoroughly answering all comments can the paper be considered for publication.

A list of major comments and concerns:

The main concern is related to the comparability of the two areas in South Australia that are not climatically linked. This becomes especially relevant for the temporal analysis within the first research question, where different climatic drivers are acting. I don't fully get the rationale for the selection of these two areas, especially when you look deeper, and in the supplementary information, you say "...the intent with the selection of eastern catchments was to ensure they were not too different from western...", which I don't see as a very solid choice. Can you justify your selection better? At the moment, the conclusions drawn from the analysis might be hindered by the different nature of these areas. Would it not be better to restructure the analysis, grouping the catchment by some climatic indices?
Can you discuss how the Hidden Markov model deals with different population sizes, and, more generally, the uncertainty of the model? You have a ratio of 5 to 1 catchment east to west. I think the method is still fine as long as you acknowledge a wider uncertainty in the west. An assessment of the model uncertainty would also be positively seen.
I could not see anything about soil types in your discussion, while in the CAMELS-AUS v2 dataset, soils are present (as in your supplementary material). Please elaborate on the effects of soil type on your analysis, and probably add a sentence about this in the limitations.
You suggest groundwater to partially explain the cause of the different behaviour from east to west. To check the importance of this, you should differentiate catchments with a deep groundwater level, where groundwater is disconnected from the surface water, from others with high surface water/ groundwater interaction. In this way, you can test your assumptions. Have you tried anything like this?
In the introduction section, references need to be better distributed. At present, there are several claims not appropriately backed up by the associated reference.
Figure 1 is not particularly clear. The background colours confound the main message of where the red-marked catchments are. Figure 3 does this in a much better way. It is not clear what the endpoints of the arrow are (Perth to Melbourne?). Is the change in precipitation expressed in %?
I recall the model was introduced by Peterson in 2021. However, can you improve the model description in this paper to avoid flipping too much between papers? How was the model run (i.e. each catchment individually)? Did you produce different annual models per catchment, as in Petersen et al. 2021, accounting for effects such as temporal autocorrelation or lags with climatic inputs?
How will the choice of the annual data affect the variability due to the seasonality of precipitation? This might be particularly relevant for catchments exhibiting low synchronicity between climatic drivers (such as the high PET in summer, high rainfall in winter or the Mediterranean-like climate in Western and South Australia). Is it what you are showing in Figure A1? Can you elaborate on this?
Line 221-222 and Figure 4. You mention a change in the rainfall-runoff relationship, but Figure 4 only shows flows. Better to try quantifying rather than saying "clear peak". Are these “two part” related, or is something missing? How did you select the groups in Figure 4?
The conclusion section is pretty thin and needs improvement to convey the message better.

Minor comments

Line 136, 150, 290, etc…. Consider not starting the sentence with these circumlocutions.
Line 165. Why did you Box-Cox transform and then use a gamma distribution? The gamma distribution could have fitted the raw data already, but once you transformed wasn't it better to go for a normal distribution?
Line 182-184, can you elaborate on how you chose 1997? Was it by your method or another? In the second case, you might need a citation.
Line 210. Avoid using non-quantifiable adverbs.
Line 210b. "To the east" of what?
Line 233. A bit confusing.
Figure 5. I'm having some trouble understanding the x-axis. Isn't it just catchments, rather than proportion? The lines 229-235 do not do a good job either. To help me understand it better, can you, for example, explain what drives these drops in the proportion of areas forested for the shifted catchments of WA, like the one around x=0.58?
Fig 6. The text is disproportionated compared to the rest of the figure.
How did you choose the results presented in Figure 7? Why are you only talking about 3 metrics out of 118? Wouldn't it better to present a table rather than a figure?
Can you elaborate/speculate about the sentence at lines 284-285?
Section 3.2 is cumbersome and does not feel like making the point.
Line 303. Instead of "proposing" trends in groundwater levels, it would be better to present a couple of case studies to back up your suggestion.
Section 3.2.1 Can you say land clearing is a cause of what you want to explore if there is such a large time lag between the cause and the state of equilibrium? Can't something else also playing a hidden effect here?
Line 371. What is "our" east region?
Line 382. Again "our". Now the paper sounds like a study of a group of the Eastern part of Australia to which some catchments have to be added in the west to try explaining something.

Citation: https://doi.org/10.5194/egusphere-2026-378-RC2

AC2: 'Reply on RC2', Nyree Campion, 24 Jun 2026

Response to Anonymous Referee 2 Comment

We thank anonymous referee 2 for their careful and constructive review. Their comments have helped us to identify several areas where the manuscript required clearer explanation and stronger presentation, particularly with respect to methodological choices, discussion structure, and the communication of key findings. Please see below for our responses to each comment.

	Comment	Response
	Major comments
1	The main concern is related to the comparability of the two areas in South Australia that are not climatically linked. This becomes especially relevant for the temporal analysis within the first research question, where different climatic drivers are acting. I don't fully get the rationale for the selection of these two areas, especially when you look deeper, and in the supplementary information, you say "...the intent with the selection of eastern catchments was to ensure they were not too different from western...", which I don't see as a very solid choice. Can you justify your selection better? At the moment, the conclusions drawn from the analysis might be hindered by the different nature of these areas.	Thank you, catchment selection is a legitimate concern. Below is our reasoning, and we will update the manuscript to ensure this logic is more prominent (e.g. at L110). As per the manuscript L83, "Despite the proliferation of studies examining shifts in rainfall-runoff relationship most studies are focussed on their own region, and comparisons of underlying processes across regions are rare". This is the gap we are trying to fill, but any such attempt will be subject to hydroclimatic contrasts across regions—such differences are unavoidable. The two regions here present a rare case where such differences can be lessened via catchments selection, as follows: The eastern region is geographically spread, and CAMELS-AUS v2 provides over 350 catchments to choose from; however, many of these are so far north that the seasonality of precipitation switches to become summer dominated (Fig. A1). In contrast, the western region is geographically compact, and all catchments exhibit winter-dominated precipitation. The two sets of catchments are made more similar by limiting the eastern set using a measure of seasonality, as described in Appendix A.
	Would it not be better to restructure the analysis, grouping the catchment by some climatic indices?	We propose to add a new plot in response to this comment (see end of next paragraph), but we think a wholesale restructure may be unhelpful, as explained below. We concede that this type of restructuring could control for the influence of variables. However, it would also obscure some of the interesting findings. In particular, both regions contain both forested and cleared catchments, and an interesting finding is the correlated/anticorrelated relationship with hydrological shifts, i.e. "in Western Australia the shifted catchments are typically those that have not been cleared for agriculture and thus retain forest coverage; the opposite is true in Eastern Australia" (abstract). If we were to restructure by climate, we would miss this important finding because the forested/cleared pattern covaries with climate, to some extent. Overall, we feel we have been transparent about the influence of different factors (e.g. Figure 7). However, on reflection, we realise that there is no plot that shows the reader the above observation (i.e. covariance between forest cover and aridity) so we propose to add a plot to the manuscript to show this.
2	Can you discuss how the Hidden Markov model deals with different population sizes, and, more generally, the uncertainty of the model? You have a ratio of 5 to 1 catchment east to west. I think the method is still fine as long as you acknowledge a wider uncertainty in the west.	The Hidden Markov Model is run for each catchment individually, as such the ratio of catchments in the study regions does not bias the identification of flow state. However, we agree that the smaller sample size in Western Australia (54 catchments, versus 254 in the east) leads to greater uncertainty in the regional summaries and comparisons, and we will revise the manuscript to make this clearer.
	An assessment of the model uncertainty would also be positively seen.	We will update the appendix to include the AIC associated with the model developed for each catchment, in addition to plots demonstrating the probability of catchments being in a given state for a given year.
3	I could not see anything about soil types in your discussion, while in the CAMELS-AUS v2 dataset, soils are present (as in your supplementary material). Please elaborate on the effects of soil type on your analysis, and probably add a sentence about this in the limitations.	We will include further information on the correlation values in the appendix (as requested by Anonymous Referee 1), which will include results pertaining to soil type.
4	You suggest groundwater to partially explain the cause of the different behaviour from east to west. To check the importance of this, you should differentiate catchments with a deep groundwater level, where groundwater is disconnected from the surface water, from others with high surface water/ groundwater interaction. In this way, you can test your assumptions. Have you tried anything like this?	We acknowledge the additional strength which considering groundwater levels would contribute to the proposed mechanisms of non-stationarity, however, due to a lack in available groundwater data within our study regions, this check is not feasible.
5	In the introduction section, references need to be better distributed. At present, there are several claims not appropriately backed up by the associated reference.	We will review the introduction, attempt to identify the unsupported claims, and revise the introduction to ensure adequate distribution of references.
6	Figure 1 is not particularly clear. The background colours confound the main message of where the red-marked catchments are. Figure 3 does this in a much better way. It is not clear what the endpoints of the arrow are (Perth to Melbourne?).	We will revise Figure 1 for clarity, also considering the feedback from Anonymous Referee 1, ensuring to label the arrow endpoints with Perth and Melbourne.
	Is the change in precipitation expressed in %?	Change in precipitation is presented as a proportion, not a percentage. The figure caption provides this detail.
7	I recall the model was introduced by Peterson in 2021. However, can you improve the model description in this paper to avoid flipping too much between papers?	We will update the methodology to include more information on the model.
	How was the model run (i.e. each catchment individually)?	The model is run for each catchment individually, using daily data aggregated to an annual timestep. We will update the manuscript to clarify this.
	Did you produce different annual models per catchment, as in Petersen et al. 2021, accounting for effects such as temporal autocorrelation or lags with climatic inputs?	We did produce different models as part of an initial testing phase, and not all of this is in the paper for the sake of brevity. For example, we settled on the Box-Cox transformation of observed data with lambda being greater than or equal to 0, after testing lambda fixed at zero in a subset of catchments (not shown). Due to the high computing power required to test all possible model combinations available in HydroState across all of the catchments, we did not test options such as allowing auto-correlation. Regardless, we are relatively confident in our results due to the similarities with Peterson et al. (2021) (compare Peterson Fig. 2B with our Fig. 3). The different models produced for each catchment varied only in their state complexity, with the retained model selected based on the Akaike Information Criterion (AIC).
8	How will the choice of the annual data affect the variability due to the seasonality of precipitation? This might be particularly relevant for catchments exhibiting low synchronicity between climatic drivers (such as the high PET in summer, high rainfall in winter or the Mediterranean-like climate in Western and South Australia).	Yes, it’s true that asynchronicity of climate drivers is a key consideration in these Mediterranean catchments. The use of annual data is unavoidable because the method adopted here (along with nearly all other methods in this research area) uses annualised data as inputs. This is a limitation that future research should aim to fix, and we will add this to the recommendations, in addition to ensure the point is clearer in the methods.
	Is it what you are showing in Figure A1? Can you elaborate on this?	See also the notes further up, under comments 1 and 2. Yes, Figure A1 is intended as a site-selection tool to ensure similarity between the selected catchments in Eastern Australia, when compared to those in Western Australia. Specifically, we used the moisture index seasonality metric of Knoben et al. (2018) and excluded eastern catchments with values below approximately 1.15, which tended to remove summer-dominant regimes and retain catchments with winter-dominant or more comparable year-round seasonality.
9	Line 221-222 and Figure 4. You mention a change in the rainfall-runoff relationship, but Figure 4 only shows flows.	Figure 4 presents the proportion of catchments in a given flow state for a given year, as determined by the Hidden Markov Modelling. The flow state itself is an indication of the rainfall-runoff relationship, i.e. a low flow state is where there is lower flow than expected for a given rainfall. This is broadly consistent with language used in previous studies such as Peterson et al. (2021).
	Better to try quantifying rather than saying "clear peak".	We will update the results to be more specific, namely, to note that the proportion of catchments increases from 0.4 in 2005 to 0.5 in 2009 before reducing to 0.25 two years later.
	Are these “two part” related, or is something missing?	We request that the reviewer please be more specific as to the meaning of this question.
	How did you select the groups in Figure 4?	If the groups being referred to are the flow states, the method implemented objectively determines this – indeed, this is the key output of the model. If the groups being referred to are each panel of the figure (top and bottom), these separate the catchments by the study regions being compared in this paper – those in Eastern Australia and those in Western Australia.
10	The conclusion section is pretty thin and needs improvement to convey the message better.	We will revise the conclusion to add more strength to the messaging.
	Minor Comments
1	Line 136, 150, 290, etc…. Consider not starting the sentence with these circumlocutions.	We will revise the manuscript accordingly.
2	Line 165. Why did you Box-Cox transform and then use a gamma distribution? The gamma distribution could have fitted the raw data already, but once you transformed wasn't it better to go for a normal distribution?	We recognise the value of adding further method details. In the HydroState framework, the transformation applied to streamflow and the choice of emission/error distribution are specified as separate components of the model. In our implementation, we used Qhat.boxcox together with QhatModel.homo.gamma.linear, meaning that runoff was first Box-Cox transformed and the Hidden Markov Model was then fit using a gamma observation model. This combination is explicitly permitted within HydroState, which allows Box-Cox, log, Burbidge or no transformation to be combined with gamma, normal or truncated normal error distributions. The manuscript will be updated to include this information.
3	Line 182-184, can you elaborate on how you chose 1997? Was it by your method or another? In the second case, you might need a citation.	We selected the year 1997 based on the results which led to the development of Figure 4, i.e. 1997 being the year where catchments in the Eastern Australian study region exhibited changes to the rainfall-runoff relationship and is also before the Western region catchments exhibited significant shifts. The manuscript will be revised for further clarity.
4	Line 210. Avoid using non-quantifiable adverbs.	We will revise the sentence to read “In Western Australia, hydrological shifts were concentrated along the coast and towards the south of the region (Fig. 3).”
5	Line 210b. "To the east" of what?	We refer to the catchments in Eastern Australia.
6	Line 233. A bit confusing.	We will add words so that it reads “Thus, the pattern in the west is reversed in the east, shown by the red line (not shifted) being below and above the blue line (shifted) in the west and east regions, respectively.”
7	Figure 5. I'm having some trouble understanding the x-axis. Isn't it just catchments, rather than proportion?	While you are correct in that each bar in the top 4 panels does represent an individual catchment, the x-axis has been revised to reflect proportion, meaning the number of catchments divided by the total number of catchments in a particular study sub-section (i.e. catchments in Eastern Australia which exhibited shifts). Otherwise, it would not have been possible to compare the distribution of forest coverage between different sample sizes (bottom panels).
	The lines 229-235 do not do a good job either. To help me understand it better, can you, for example, explain what drives these drops in the proportion of areas forested for the shifted catchments of WA, like the one around x=0.58?	The intention with this figure was to depict the deviation in distribution of forest coverage, i.e. that shifted catchments in Western Australia typically have high forest coverage, whereas shifted catchments in Eastern Australia typically have less forest coverage (when compared to non-shifted catchments). The top 4 panels represent catchment land use within each category (Eastern vs Western Australia, and Shifted vs Not Shifted), in descending order of forest coverage.
8	Fig 6. The text is disproportionated compared to the rest of the figure.	We will revise the “Estimated magnitude of shift due to non-stationarity (%)” font size to better match the remainder of the figure.
9	How did you choose the results presented in Figure 7?	We selected the 3 metrics presented in Figure 7.0 as those which showed the strongest correlation (both positive and negative) across both regions. The manuscript will be updated for clarity.
	Why are you only talking about 3 metrics out of 118?	The text preceding the figure discusses a selection of other metrics. We intend to include further information on the correlation values in the appendix (as requested by Anonymous Referee 1).
	Wouldn't it better to present a table rather than a figure?	We believe the figure to be a useful representation of the extent of scatter for the given metrics. Further information provided can be presented in a tabular format.
10	Can you elaborate/speculate about the sentence at lines 284-285?	We will add text to say that this may be indicative that this transition of climatic condition did not initiate a change in dominant processes—at least, not to the extent seen later after 2000.
11	Section 3.2 is cumbersome and does not feel like making the point.	We will revise Section 3.2 to be more succinct and direct.
12	Line 303. Instead of "proposing" trends in groundwater levels, it would be better to present a couple of case studies to back up your suggestion.	We present a selection of case studies in Figure 8, displaying trends in groundwater. We intended Line 303 to act as an introduction to the hypothesis, with further evidence, detail and case studies from the literature provided in the proceeding sections. We will review these sections to better highlight any relevant studies.
13	Section 3.2.1 Can you say land clearing is a cause of what you want to explore if there is such a large time lag between the cause and the state of equilibrium? Can't something else also playing a hidden effect here?	Slow aquifer response can be seen in regions with low recharge rates and/or deep water tables (Allison et al., 1990), meaning previous extensive land clearing is a plausible explanation for the continued rise in standing water level. We do not deny other factors may also be influential in the observed behaviour; however, we intend to highlight the importance of groundwater mechanisms in persistent shifts in the rainfall-runoff relationship.
14	Line 371. What is "our" east region?	We refer to the catchments in Eastern Australia selected for comparison with those in Western Australia.
15	Line 382. Again "our". Now the paper sounds like a study of a group of the Eastern part of Australia to which some catchments have to be added in the west to try explaining something.	The study is presented as a comparison between Eastern and Western Australia; in aiming to reduce repetition, we have opted for alternative referential terms. If Anonymous Referee 2 rejects the use of “our” in the transcript, we will revise wording to reduce possessive language.

References

Allison, G. B., Cook, P. G., Barnett, S. R., Walker, G. R., Jolly, I. D., & Hughes, M. W.: Land clearance and river salinisation in the western Murray Basin, Australia, J. Hydrol., 119, 1-20, https://doi.org/10.1016/0022-1694(90)90030-2, 1990.

Knoben, W. J., Woods, R. A., and Freer, J. E.: A quantitative hydrological climate classification evaluated with independent streamflow data, Water Resour. Res., 54, 5088-5109, https://doi.org/10.1029/2018WR022913, 2018.

Peterson, T. J., Saft, M., Peel, M. C., and John, A.: Watersheds may not recover from drought, Science, 372, 745-749, https://doi.org/10.1126/science.abd5085, 2021.

Citation: https://doi.org/10.5194/egusphere-2026-378-AC2

Nyree Campion, Keirnan Fowler, Margot Turner, and Joel Hall

Viewed

Total article views: 1,539 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
853	560	126	1,539	82	87

HTML: 853
PDF: 560
XML: 126
Total: 1,539
BibTeX: 82
EndNote: 87

Views and downloads (calculated since 16 Feb 2026)

Month	HTML	PDF	XML	Total
Feb 2026	448	220	42	710
Mar 2026	269	203	67	539
Apr 2026	71	66	7	144
May 2026	40	51	4	95
Jun 2026	25	20	6	51

Cumulative views and downloads (calculated since 16 Feb 2026)

Month	HTML	PDF	XML	Total
Feb 2026	448	220	42	710
Mar 2026	269	203	67	539
Apr 2026	71	66	7	144
May 2026	40	51	4	95
Jun 2026	25	20	6	51

Viewed (geographical distribution)

Total article views: 1,537 (including HTML, PDF, and XML) Thereof 1,537 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 25 Jun 2026

Short summary

Globally, many river systems have seen less flow for the same rainfall after periods of long drought. This study investigates this behaviour in over 300 catchments across the south-east and south-west of Australia. Through comparison we infer possible underlying causes, expanding on current studies limited to single locations. Over half the catchments studied displayed a drop in flow. We suggest the influence of pre-existing trends in groundwater and highlight the importance of land use history.


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