Satellites reveal hotspots of global river extent change

Rivers are among the most diverse, dynamic, and productive ecosystems on Earth. River flow regimes are constantly changing, but characterizing and understanding such changes have been challenging from a long-term and global perspective. By analyzing water extent variations observed from four-decade Landsat imagery, we here provide a global attribution of the recent changes in river regime to morphological dynamics (e.g., channel shifting and anabranching), expansion induced by new dams, and hydrological signals of widening and narrowing. Morphological dynamics prevailed in ~20% of the global river area. Booming reservoir constructions, mostly skewed in Asia and South America, contributed to ~32% of the river widening. The remaining hydrological signals were characterized by contrasting hotspots, including prominent river widening in alpine and pan-Arctic regions and narrowing in the arid/semi-arid continental interiors, driven by varying trends in climate forcing, cryospheric response to warming, and human water management. Our findings suggest that the recent river extent dynamics diverge based on hydroclimate and socio-economic conditions, and besides reflecting ongoing morphodynamical processes, river extent changes show close connections with external forcings, including climate change and anthropogenic interference.

A list of potential drivers is established, but it is still very difficult to hierarch them. One of the reasons is linked to the fact that PI/PD and PGS are analysed separately so that it is difficult to separate what is liked to the geomorphic functioning (e.g., channel shifting) and what is linked to real changes (narrowing vs widening due to hydrology or morphological modifications).
I provided a lot of comments on the two manuscripts that should help the authors to understand the main key points I listed above.
Reviewer #2 (Remarks to the Author): The manuscript presents an interesting analysis of river inundation extent changes from a global perspective using satellite images and correlating tendencies with climate change (precipitation changes, meltwater and global aridity), anthropic factors (relating to nightlight to identify developed countries, river embankments or dam construction and other human interventions) and geomorphological factors.
I consider that this manuscript has a very high value and is very appropriate to be published nonetheless I have some suggestions or comments.
The analysis compares the mean river inundation situation of the period 1984-1999 (15 years) to that of 2000-2018 (18 years). Therefore it can not be stated that the analysis is about "global river changes of the past nearly 40 years". It is just of the last 18 years, as the starting value is that of the mean extent of whole sub-period of 1984-1999, so it is compared with that of the last 18 years, so that is the period for which we are assessing if it has been a widening, narrowing or stabilization of inundation. I do not see clearly explained how OCI value is obtained (-100 to 100). Some specific comments: Line 55. Maybe it could be find a more appropriate reference instead of "19". Lines 114-115. The percentages do not correspond to the absolute data, I might be wrong but it is 14.7% instead of 18.9% and 5.5% instead of 7.1% according to each "n". Line 365. "..., we chosen ..." substitute with "..., we chose..."     Figure S12. I would keep the same scale for the graphs (b and c). b ranges from 0 to 0.6 and c from 0 to 0.2. In order to be comparable it should be the same one.
This gives the idea of more increasing than decreasing. Figure S16. Legend it is confusing because of the the negative symbol and the hyphen, try to separate more. Table S1. If I am not wrong, the table caption has a mistake: "..., and significantly decreased (PSI) inundation frequency.." it is "increased".
Reviewer #3 (Remarks to the Author): I read this paper with great interest -it is a very ambitious exercise in remote sensing, leveraging a number of tools and global datasets to deliver a summary picture of change in major rivers around the world. It seems to be a natural outgrowth of earlier work by Allen and Pavlevsky (2018), introducing a temporal dimension to the large-scale mapping of river width and inundation.
I found that while the paper has produced some very nice plots and summaries of global inundation trends, both the introduction and conclusion did not distinguish on what is new and novel in this work and focus reader attention on a specific question. Clearly the paper shows a strong methodological advance, but I found that this does not clearly emerge as the focus of the paper. Many of the patterns identified in the results are consistent with previous works, and indeed most of the insights are confirmed by appeal to existing studies and literature. These broad trends are largely understood, although this is a novel way of detecting and displaying these global patterns.
I found there were a few key problems with the presentation of the work: (1) There are a multitude of processes responsible for changes in river width, but only the broadest of reasons are addressed. The result is a fairly broad and diffuse statement on human interference and climate change. There is not a focused conclusion that emerges from this.
(2) Error is not addressed in the main paper, but is largely relegated to (brief) treatment in the supplementary material and previous work. What are the chances of clear and continuous coverage of each river each year? How are the seasonal effects (including winter ice, nival conditions, and veg changes) controlled for? The ratio of width and depth for various rivers will have a strong influence on the nature of inundation, so some rivers are far more likely to show variability. The examples provided in Figure 3b-e are not subtle ones. Without at least a few examples of the robustness of the method -how often the technique gets it right (local verification), particularly in smaller systems. I should think it would be of interest to the readership to better understand the precision of the technique, and what thresholds can be found at the tipping points between 'stable' and 'decreasing', for instance. Given that error is not clearly demarcated or addressed in this remote sensing study, I do not know how much confidence to invest in the work that follows.
(3) the magnitude of change is relative, and the implications of these changes are difficult to convey. In some regions, variability is cyclical and not unexpected. In others, the change is profound and irreversible. The percentage of moderate versus 'significant' change is not given much context. From this perspective, the paper seems unable to offer new and nuanced insights into the nature of decadal river change. Despite this problem, the authors have done a good job of reviewing existing literatures to confirm and contextualize the findings, showing that results are consistent with current understandings of global river change.
My overall sense is that the major contribution in this work is the methodology behind it, and not the patterns of channel change summary reported in the core paper. As someone with an interest in patterns of river change, I found much more interesting material in the appendix and supplementary sections than the broad and fuzzy generalizations in the main work. The tremendous effort that has gone into the geospatial change detection is much better reflected in the tools and techniques, rather than the description of changes, which are broadly known, but have not previously been summarized with this kind of detailed, spatially-explicit method.
I believe that the work would be better tailored for a premiere remote sensing journal, where the readership will better appreciate the innovation that is represented here. I would encourage the authors to merge the core and supplementary material for a more expansive treatment of the technique, better delineation of a 'research question', explanation of some of the errors/pitfalls encountered, and deeper perspective on the implications of change in the various large river systems. In its abbreviated form, I think you are doing something of a disservice to your efforts in algorithm development and application.

REVIEWER COMMENTS
Reviewer #1 (Remarks to the Author): The MS entitled "Satellite-observed global river channel changes driven by anthropogenic and climate forcings" submitted for publication in Nature Communications explores changes in current water extent along fluvial network worldwide between mid-1980s to 2018 and factors controlling them. If the descriptive approach is convincing and really interesting, the causal approach, still exploratory to assess the "main" drivers, is difficult to follow and not convincing at all, except the part on dams. Indicated drivers are mainly interpretations and not facts and it is very difficult to really assess what is true and not true, and to hierarch these potential drivers.
Response: Thanks for the in-depth comments providing clear merits and demerits evaluation. We carefully considered the problems pointed out by the reviewer and performed further studies to improve the analysis, particularly regarding the hierarchy of different types of change and the potential drivers.
One of the main problem is linked to the fact that the explanatory variables are not clear (or introduced clearly), making the strategy to detect potential causes fairly weak. What the variables can say should be a bit more detailed to provide a stronger basement and state potential hypothesis to be tested. The selected variables do show changes in current water extent (not inundation) over the period but also channel behaviour (shifting vs non-shifting). Water extent provides an information on current flow conditions (not inundation) and can inform on changes in hydrological or morphological signal (narrowing = lower current flows or regulation/embankment/diversion ; widening = reservoir or higher current flows -interbasin transfer for example) AND channel morphodynamics (not change)… liked to channel shifting within the floodplains corresponding to specific multiple-thread channels (braided or anabranching/anastomosing) or free-meandering/wandering channels mainly (please note the morphodynamics can change as well… rivers can be less and less shifting or more and more). This is not clear in the title nor in the abstract. The keyissue is how to separate channel pattern types (stable and unstable) from real hydrological and morphological changes. If this is not done correctly, the analysis of potential drivers of change is weak. This is particularly true with the case of stable channels related to developed countries. Most of temperate and/or lowland systems over 40 years are stable independently of any infrastructures. The Amazone or the Congo are good examples. The widening related to new reservoirs is convincing because it is fairly easy to locate new dams and their reservoirs. The problem is much trickier with the potential climate change effect and the paragraph L203 -220 is not convincing at all. Again, there is a need to remove what is linked to the river behavior (there are many braided shifting rivers in Tibet and also new reservoirs) to explore this issue (what about west Africa or south Europe, why north Canada is not concerned?).
Response: We sincerely appreciate your incisive discussions about the indication of variables and the associated scenarios of water extent changes. The insightful comments and constructive suggestions inspired us to solve the key issue -how to separate different types of signals in the detected water extent changes. This is a challenging issue on a global scale, and and we are lacking literature reference to do so. However, we did find that river pattern shifting and morphological changes were evident on the OCI map. We developed an expert-based protocol streamlining the manual interpretation of morphological dynamics and employed a machine learning approach to identify morphological changes globally (Fig. S2, S3, S4; Fig. 1). As stated in the summary above, we separated different types of signals, morphological changes, dam-related river expansion and hydrological signals. Regarding morphological changes, we understand that they can be natural evolutions or can change as well (more or less shifting); however, we cannot make further identifications or conclusions based on the time span of current observations. In other words, our results represent an observation of global river morphological changes from 2000-2018 compared to 1984-1999. The pattern of morphological changes and potential driving factors were discussed in the first section of the results. The concerning points are addressed as follows.
Explanatory variables: According to the suggestion, we briefly described the explanatory variables and the indication in the introduction and in more details in the first section of the Methods. Specifically, we introduced how the inundation frequency changes were calculated and the meaning of the different classes. We stated potential hypothesis on the pattern of frequency changes regarding morphological dynamics, dams, and hydrological signals.

Climate change effect (paragraph L203 -220):
The updated analysis focuses on the hydrological signal after removing morphological changes and dam-related water expansion.
Obvious widening in the Tibetan Plateau and eastern Siberian can be observed on the map (Fig.3,   Fig.5). We performed hotspot analysis to present the most prominent hydrological signal globally and provided a detailed analysis of changes of climate variables in these regions (Fig. 6).
Abstract should be adjusted to better understand what is observed, what are the potential interpretations, what are the most important changes or channel behaviors.
Response: Thanks for the suggestion. We revised the abstract to better reflect the supported findings (facts) and the implications.
Terms need to be fixed. Inundation or braided index are not the appropriate terms and this can provide confusions and misunderstanding in terms of drivers (see some of my comments on the drafts).
Response: We agree with this suggestion. We avoided the usage of 'Inundation extent' or 'braided index' and replaced with "river water extent", "river flow regimes", "number of channels" instead.
Thanks for the detailed comments. We made corrections accordingly.
The number of available images (observations) between the first period and the second is significantly different (roughly it doubled over the period). It means the probability to observe more extreme conditions A list of potential drivers is established, but it is still very difficult to hierarch them. One of the reasons is linked to the fact that PI/PD and PGS are analysed separately so that it is difficult to separate what is liked to the geomorphic functioning (e.g., channel shifting) and what is linked to real changes (narrowing vs widening due to hydrology or morphological modifications).
Response: We agree that separate analysis of PGS/PI/PD leads to fuzzy impressions without a clear picture of the overall change patterns. Therefore, in addition to separating different types of river changes, we modified our analysis and presentation logic. Regarding the hydrological signal, we first presented the general patterns (which signal (increase, decrease, stability) is dominant in different regions/basins, Fig. 3, Fig. 4), and then focused on the most prominent changes (increase/decrease, Fig.5) and lastly the pattern of river stability (Fig. 8).
I provided a lot of comments on the two manuscripts that should help the authors to understand the main key points I listed above.
Response: Sincere thanks for the detailed comments. We reviewed them carefully and made corrections accordingly.
Some specific comments in the Manuscript: Line25-26: "almost 40% of the global river area experienced obvious changes…" ---it would be good to indicate the total river length studied because it is 40% of a river network that is far from being the whole one due to the rough resolution of landsat images.
Response: Thanks for pointing out the confusing details. We primarily used river area rather than river length in the statistics. To be clear, we explicitly explained the metric of 'river area' in the first section of the Methods part, and gave the exact statistics (2,097,799 km by length and 769,390 km 2 by area) in the abstract and the main text. Our analysis is primarily built on the Landsat-based global river network (SWORD) and observations. We agree that the resolution of Landsat images limits the ability of a more complete river network mapping (e.g, river width < 30 m, Allen & Pavelsky, 2018).
Line27: "and evolving water policies" ---not clear / explicit Response: This statement is removed here and explained later. To be clear, we specified 'water abstraction' and 'water use management to replace the 'evolving water policies'.
Line29: "particularly the Tibetan Plateau and eastern Siberia" ---reservoirs are also important in this part of the world Response: We revised our analysis for a separate analysis of different types of signals. With a focus on the hydrological signal, we excluded new reservoir-type river reaches and morphological dynamics from the analysis. The hotspots of river extent changes still include the Tibetan Plateau and eastern Siberia, which means that river widening is a strong signal for the two regions. The sentences are modified to clarify the new clues of analysis.
Line33: "…confirming the expected services of river infrastructure including levees and embankments" --Response: Regarding this concern, we rephrased the argument to reduce the confidence. Although we do not provide direct evidence for each case, current observations and studies support this argument.

' River widths maintain relatively high stability in developed regions
Line70: "…inundation changes…" ---not inundation. "flow channel" Response: The term has been modified according to the suggestion.
Line72: "…of two major state-of-the-art surface water databases." ---not only because to make your interpretation as strong as possible to explain your cause-effect relationship you use other information. The paper describes and provides potential explanations Response: We revised this sentence to highlight the method of deriving global river extent changes, and the analysis (cause-effect relationship) were summarized at the end of the introduction.

Line81: "…dynamics of rivers…" ---unclear
Response: This sentence was revised to improve clarity. "To retrieve river extents we implemented…" Line84: "…probability of changes…" ---what kind of changes do you observe: change of the channel position in the floodplain, change in water extent due to hydrological or morphological changes. Clear need to explain properly what you can potentially see with this indicator and identify potential hypothesis you test.

Response: This is a good suggestion. We elaborated the indication of the used metrics (frequency changes vs. water extent changes) and the associated potential hypothesis to be tested. A detailed description is also supplemented in the Methods section. "For rivers, positive (negative) values on the OCI map indicate increase (decrease) of water flow frequency, informing changes in river extent which can be associated with hydrological signals (flow conditions: widening/higher or narrowing/lower) or morphological changes in river platforms."
Line88-89: need to be clarified

Response: Detailed description of how OCI changes are related to the water extent changes and river morphological dynamics is provided in the Methods section (Statistical variables and the indication).
Line89-90: "pattern of river channel evolution" ---unclear again Response: Here, we mean the type and pattern of river extent changes (widening or narrowing, or morphological dynamics). The statement was revised for clarity.
Line96: "river channel changes" ---changes in flow extent along the channel network Response: The suggested term is to the point. Throughout the manuscript, we updated the term 'river channel changes' to 'river extent changes' or 'river flow regime changes'.
Line97: "…inundation…" ---unclear again. Do you explore "mean flow" conditions with your indicator or "extreme flows -minimum and maximum" including inundation. This must be said here to clarify what is observed.

Response: According to the statistical method of the OCI database, the indicator (OCI value, namely the frequency changes) refers to the comparisons of mean flow conditions in normal cases
(observations were not biased toward certain seasons in each period). The statement was revised accordingly.
Line97: "…638,887 km 2 of river area…" ---river length would be more meaningful. What is the frequency of the water extent changes in this total area?! Response: We provided statistics of different classes of frequency changes for the global river area.
Line108: "…river channel changes…" ---patterns of flooding? What it is observed here is not really a "channel change" Response: The term 'river channel changes' was misused and we corrected the usage accordingly.
Line127: "river channel" ---delete Response: The term 'river channel changes' was corrected to 'river extent changes.
Line129: "…increased inundation frequency than decreased" ---it is very confusing because you mixt different things. Net decrease, net increase and a balance in the case of channel shifting... and considering separately increase and decrease only does not really help to understand what's going on mixing effective hydrological and morphological changes and channel behaviors.
Response: We agree that the mixture of morphological dynamics in the analysis will disturb the understanding of the hydrological signal. In the revision, we devoted a lot of time and effort to separate the morphological signal, which has never been done on a global scale. By identifying and separating different types of river extent changes, we enabled a clear focus on the hydrological signal here. Given that widening and narrowing signals probably coexist in a single basin due to high variability of river changes, we also considered the net increase/decrease metrics in our revised analysis to evaluate the magnitude of the primary signal. See the section titled 'Patterns in mega-

basins' for a detailed revision of all changes.
Line130-131: "…top three in percentage of increase (63.0%, 43.0%, and 46.5%, respectively)…" ---Typically they do not widen, they just move in their floodplain Response: After revision (with morphological signal separated), the analysis here is mainly about the hydrological signal.
Line154: "…changes…" ---change in water area extent and river behavior Response: This statement was removed in the revised manuscript. We took this advice and revised related expression in other parts of the manuscript.
Line154: "…first-order attributions…" ---how do you consider they are first-order? Can you provide the river length concerned by each of them? You may state explicit hypothesis and show how you can validate them.

Response: We have revised the logic of the results according to the suggestions. After revision, we retrieved hotspots of river changes (excluding dam-related changes and morphological dynamics)
and provided analysis of the causes.

Line155: "…significant increases in inundation frequency…" ---water extent
Response: Thanks for the detailed advice. We carefully reviewed the phrase 'inundation frequency' and made revisions accordingly.
Line165: "…11,404 km 2 of the intensified inundation area." ---a table summerizing the contribution of the different potential drivers and the unexplained part would help to better hierarch them.
Response: Thanks for the suggestion. Accordingly, we recognize dam-related river expansion as a particular form of river extent changes and analyzed it separately. In the revised analysis of this issue, we provided information about the total area of new reservoir-type river reaches, the percentage of expanded flow area, and its contribution to river widening signal on different spatial scales.
Line166: "…46.4% of the total river area increase…" ---it is really high?! Response: We are sorry for the confusing point here. Here 46.4% is the ratio of dam-related expanded river extent to the total expanded river extent globally. The expanded river extent refers to the part of river areas marked as significantly increased frequency (OCI > 75%). The high ratio (46.4%) indeed reflects the remarkable contribution of new dams to the global river widening signal.
In our revised analysis, we refined our analysis of reservoirs (new-reservoir inventory intersecting SWORD river network) and excluded basins dominated by morphological dynamics in the statics of global widening signal. The updated ratio (31.9%) still support new dams as a major contributor to global river widening.
Response: Here we originally mean that improved water use management helped the recovery of normal water flow in the region. This section was revised and merged in our analysis of the human impact on river extent changes.

Response: We provided detailed definition of no-flow cutoff in the main text here. 'The high percentage of increased flow coverage in this region is essentially a recovery from no-flow cutoff (Less than 1 m 3 /s runoff at the Lijin hydrological station situated at the lowest reach of the Yellow River).'
Line195: "…improved during the early 21 st century" ---why is it so? Explanation fairly unclear Response: We provided more detailed explanations by summarizing the main levers the local authorities used to improve water use management in the Yellow River basin. Related studies are also cited for further references of this topic.
Line210-211: "…the percentages of increased inundation extents are 67.5%, 39.2%, and 51.0%, respectively" ---you need to remove what is linked to the river behavior (many braided shifting rivers in Tibet and also new reservoirs) to explore this issue.
Response: Thanks for the notes. In the updated manuscript, we provided separated analysis of morphological behavior and hydrological signal.
Line214: "…precipitation increasing at 11.62, 40.32, and 25.19 mm/year" ---This is a total annual increase but it has nothing to do with flooding that is related to maximum rainfall events or ice-melt Response: We agree that extreme precipitation events can cause flooding events and consequently impact on the water frequency changes. However, the frequency change on the OCI map is the mean difference of the overall frequency between the two epochs. Therefore, individual events have limited influences on the long-term statistics. In other words, we inspected that the OCI map reflects comparisons of mean situations in two periods, and comparison of the overall changes of climate variables is appropriate. We can expect water diversion (also related to damming) can play a significant role.
Response: We agree that groundwater and river network are different things, but we politely disagree that they are independent. In many regions, groundwater interchanges with water in rivers and lakes through base flow, particularly in arid regions (simple illustration at https://books.gwproject.org/groundwater-in-our-water-cycle/chapter/groundwater-connection-with-streams/). Line279: "…levees to constrain lateral floods" ---we do not focus here on flooding or inundation but on current flow area. This can't be a driver at all.

Excessive abstraction of groundwater can lead to declined groundwater level, inverted stream
Response: We are sorry for the misleading information. Here we mean that the levees were built for the purpose of constraining lateral floods. We rephrased the sentence to avoid such confusion.
Line287: "…bank erosions, thus substantially stabilizing river…" ---dikes and bank protections are (or can be) two distinct infrastructures. Very unclear this point. We may hypothesis that in developed countries, we should expect to have fairly more shifting rivers in the piedmont areas due to bank protections to prevent bank erosion... but you need first to locate where are potential shifting rivers and which part of them do not shift anymore. At this stage, the demonstration is far from being convincing.
Response: We agree with your opinion that the function of man-built river infrastructures should be evaluated with more rigorous reasoning and evidence. However, a rigorous justification here can be impractical due to limited observations and understanding of which and how rivers evolve in response to natural and human forcings. As suggested, historical observations (several decades or hundreds long) are required to show the behavior of rivers before and after setting of infrastructures, but availability of such observations on a global scale is the key issue unsolved. Here we showed the pattern of river stability related to the development level of human societies from a perspective of region scales, and it may be impossible to go to the details for individual rivers or reaches as observations (e.g., detailed river infrastructures) are incomplete. The demonstration here offers a simple and straightforward explanation for the function of river embankment infrastructures, as they normally consolidate the river banks, mitigating the erosion and sedimentation processes and therefore leading to stable river flow regimes. We rephrased the sentences to clarify the simplified assumption here.
Line297: "…can be characterized by four major metrics…" ---different things are mixed here. You may clarify if you want to analyse river pattern (planform types -multiple vs single thread, sinuosity ...) and the potential drivers explaining these patterns (e.g. slope) Response: The analysis was updated according to the suggestion. In characterizing river pattern, we employed two metrics in the classification: the number of channels and sinuosity. Line299-300: "…where rivers have high sinuosity and steep hydraulic gradients…" ---implicitely here, you are looking for a specific highly shifting channel patterns : the free meandering patterns. This may be said more explicitely. Another highly shifting pattern, the most important one, is the braided pattern usually located a bit upstream.

We devoted a framework to interpret such signals on the basis of the new river database and the OCI map showing river extent changes.
Line304-305: "stable channels are associated with high braiding index and wide rivers, which are the general characteristics of midstream or downstream river…" ---not braided... multiple-thread.
Braided systems are located in mountain plains or piedmonts, always upstream of the shifting meandering systems Response: Thanks for pointing out the details. The analysis was revised in the updated manuscript.

Line381-382: "…homologous pairs of months between two epochs." ---(reviewer only highlighted)
Response: This statement means that frequency changes were compared between homologous months, and then averaged among pairs. Line387: "…inundation extents" ---current flow extent may be more appropriate than inundation.
Inundation may concern an area that is usually dry and can be for a short period of time inundated. What we see here is different. When the channel moves, it abandons some areas and occupies others. What we see then is the current flow channel that change its position. Nothing to do with inundation.
Response: Thanks for the detailed explanation and the suggestion. We reviewed the use of 'inundation extents' through the manuscript and revised where necessary to assure the appropriate usage. We agree that river extent or river flow regime is more appropriate than inundation extents.

Some specific comments in the Supplementary Text:
Line28: "…or migration downstream" ---usually when sediment supply is reduced, channel migration is also reduced, both are linked. This hypothesis downstream reservoirs is not tested Response: Thanks for notifying this point. In this section, we talk about some straightforward scenarios but not facts. We understand concerns about the untested hypothesis. Thus, we revised the description in this section (moved to the methods part).

Response: We understand concerns about untested hypothesis. The related description was weakened.
Line35-36: "…for flood prevention, which generally leads to reinforced stability of the inundation area (widths)…" ---This can't be studied with the data. There is a big confusion with the term inundation.

Response: We rephrased the sentence and particularly the term 'inundation' here and elsewhere.
Line60-64: unclear why this is here.
Response: These statements were removed to appropriate places in the main text.

Reviewer #2 (Remarks to the Author):
The manuscript presents an interesting analysis of river inundation extent changes from a global perspective using satellite images and correlating tendencies with climate change (precipitation changes, meltwater and global aridity), anthropic factors (relating to nightlight to identify developed countries, river embankments or dam construction and other human interventions) and geomorphological factors.
Response: Thank you so much for the summary and commendation. We carefully evaluated all the comments below and made a thorough revision to further improve the clarity and rigorism of the manuscript.
I consider that this manuscript has a very high value and is very appropriate to be published nonetheless I have some suggestions or comments.

Response: Thanks again. We have carefully addressed each of these suggestions and comments.
The analysis compares the mean river inundation situation of the period 1984-1999 (15 years) to that of 2000-2018 (18 years). Therefore it can not be stated that the analysis is about "global river changes of the past nearly 40 years". It is just of the last 18 years, as the starting value is that of the mean extent of whole sub-period of 1984-1999, so it is compared with that of the last 18 years, so that is the period for which we are assessing if it has been a widening, narrowing or stabilization of inundation.

Response: This is a good point, and we agree our analysis is about decadal changes in river extents.
We revised the statements in the abstract, introduction, and conclusion sections (from 'the past nearly 40 years' to 'the early 21st century or 'during 2000-2018') to avoid confusion.
I do not see clearly explained how OCI value is obtained (-100 to 100).

Response: We are sorry for the missing details. We revised the Methods section about the method of deriving OCI value. 'The OCI map was created by averaging all surface water occurrence differences derived from homologous pairs of months between the two epochs (16 March 1984-31 December 1999 and 1 January 2000-31 December 2018) from the GSW database'.
Some specific comments: Line 55. Maybe it could be find a more appropriate reference instead of "19".

Response: Thanks for the suggestion. We removed this reference and replaced it with an appropriate one (Best 2019) to support the influences of human impact on river systems.
Lines 114-115. The percentages do not correspond to the absolute data, I might be wrong but it is 14.7% instead of 18.9% and 5.5% instead of 7.1% according to each "n".
Response: We are sorry for the confusing point. The percentage is the statistics by river area not by basin count, but only the number of basins was given. We updated the statistics here (note that we updated the river network data and the analysis method), and explicitly specified if the percentage is by area or by count in the main text.
Figure S12. I would keep the same scale for the graphs (b and c). b ranges from 0 to 0.6 and c from 0 to 0.2. In order to be comparable it should be the same one. This gives the idea of more increasing than decreasing.
Response: This is a good idea. We updated the Fig. 4   We avoided such issues in the revised figure.
Table S1. If I am not wrong, the table caption has a mistake: "..., and significantly decreased (PSI) inundation frequency.." it is "increased".

Response: Sorry for the mistake. It should be 'significantly increased'.
I read this paper with great interest -it is a very ambitious exercise in remote sensing, leveraging a number of tools and global datasets to deliver a summary picture of change in major rivers around the world. It seems to be a natural outgrowth of earlier work by Allen and Pavlevsky (2018), introducing a temporal dimension to the large-scale mapping of river width and inundation. I found that while the paper has produced some very nice plots and summaries of global inundation trends, both the introduction and conclusion did not distinguish on what is new and novel in this work and focus reader attention on a specific question. Clearly the paper shows a strong methodological advance, but I found that this does not clearly emerge as the focus of the paper. Many of the patterns identified in the results are consistent with previous works, and indeed most of the insights are confirmed by appeal to existing studies and literature. These broad trends are largely understood, although this is a novel way of detecting and displaying these global patterns.

Response
Response: We sincerely appreciate your constructive feedback on the presentation of this work. We highly agree that we should appropriately clarify what is new in this work regarding both methodologies and insights into the scientific question. Accordingly, we modified the abstract, introduction and conclusions sections as well as new supporting figures to highlight our contributions in terms of methodologies and findings. Specifically, we emphasized that our results are based on a new framework (Fig. S1, S2, S3) for detecting and interpreting decadal river extent changes. The key methodological points include how to do the statistics on appropriate spatial scales and interpret the types of changes related to dams, morphological dynamics, and hydrological signals. In terms of findings, we argued that a global map of river morphological dynamics and a global quantification of new reservoir-type river reaches are novel contributions to the literature. Regarding the river widening and narrowing patterns, most of them have been confirmed or mentioned in previous works, but we provide the first satellite-based evidence of regional river widening in the Tibetan Plateau and eastern Siberia, which has only been indicated or speculated in previous studies. We agree that this work emerges a novel way of detecting and displaying global patterns of river flow regime changes. We believe it contributes to an improved understanding of river changes in the scenario of climate change and intensified human interferences and spatially-explicit guidance for better prioritizing future river protection for sustainable development.
I found there were a few key problems with the presentation of the work: (1) There are a multitude of processes responsible for changes in river width, but only the broadest of reasons are addressed. The result is a fairly broad and diffuse statement on human interference and climate change. There is not a focused conclusion that emerges from this.
Response: We agree that river extent changes are results of complicated processes driven by different forcings, including those of climate and anthropogenic, internal or external. To better hierarch multiple potential drivers, we devoted further efforts to classify different types of river changes, namely morphological dynamics, dam-related widening, and hydrological signals. Our analysis was therefore considerably refined in the revised manuscript, with a separate analysis of these different types of changes and the first-order attributions. We notably strengthened the presentation of the pattern of hydrological signals with hotspot analysis and clear evidence of climate change by referring to multiple climate datasets. We acknowledge the limitation that only the broadest reasons for these changes were addressed, as our focus was on global and broad scaled where river width changes exhibited high spatial variability. Besides, there is limited understanding regarding how river morphology evolves naturally or in response to external forcings. To answer the question, combined in-site and laboratory observations as well as numerical modeling approaches, are required, which is beyond the scope of this study.
(2) Error is not addressed in the main paper, but is largely relegated to ( 'stable' and 'decreasing', for instance. Given that error is not clearly demarcated or addressed in this remote sensing study, I do not know how much confidence to invest in the work that follows. Response: We appreciate that the reviewer has considered many aspects of the uncertainties, which helped us deal with the tough issue. We, therefore, conducted further work to address these questions in the last section of the Methods part (Uncertainties). We addressed the uncertainties in three aspects: the robustness of the OCI statistics, the temporal coverage and the possible seasonal effect of Landsat observations, and how the statistics are related to in-situ river observations. First, the robustness of frequency change patterns is illustrated from cases of different hydrological regimes/regions (Fig. S19). This confirms our expectations as the algorithm for water detection has been highly developed, and the OCI statistics took account of seasonality ( (Fig. S20). Note that the number of observations in different months can differ, but this difference was generally consistent for a specific region over the years.
Seasonal effects were also controlled as the frequency statistics (OCI) were compared in homologous pairs of months. Third, discharge measurements at gauging stations were used to evaluate the reliability of our statistics (Fig. S21). The results (Fig. S21) show an overall good correlation between decadal net changes in river extent and relative decadal annual discharge changes, given the differences in spatial scales (point-based versus region-based).

About river stability:
We agree that 'The ratio of width and depth for various rivers will have a strong influence on the nature of inundation, so some rivers are far more likely to show variability'. This is also validated from our results that about half of the studied river reaches showed relative stability in flow extents, regardless of the types of river changes. Our work was essentially to detect where prominent changes occurred, the types of the changes, and the potential main drivers. Because currently, we know little information on the geometrics of rivers (such as the ratio of width and depth) globally, we cannot offer an in-depth analysis of the stability. We only provide a perspective on the stability from the general large-scale patterns but not the details of specific rivers or reaches.
Additionally, we clarify that ' Figure 3b Fig. 3, and identify the dominant signal. We proposed using the relative magnitude of net increase, which is the difference between areas of increase and decrease divided by the total river area ((increased area -decreased area)/total river area) to measure the dominance of different signals. This metric acts as a normalized variable for comparing changes in different regions.
(3) the magnitude of change is relative, and the implications of these changes are difficult to convey. In some regions, variability is cyclical and not unexpected. In others, the change is profound and irreversible.
The percentage of moderate versus 'significant' change is not given much context. From this perspective, the paper seems unable to offer new and nuanced insights into the nature of decadal river change. Despite this problem, the authors have done a good job of reviewing existing literatures to confirm and contextualize the findings, showing that results are consistent with current understandings of global river change.
Response: Thanks for the insightful comments on the complexity of river changes. We provided specifications on the context of the used metrics and hypothesis in the introduction and methods part.
Particularly, we review the 'significant' change as very high confidence of changes in water extents and flow conditions compared to 'moderate' changes, considering the variation and observation limitations. We set up a hypothesis that river flow regime changes, either of morphological dynamics or hydrological signals, can be detected by analyzing the pattern of frequency changes. Furthermore, we systemically considered different types of river extent changes and separated them using machine learning. The results present a global map of river morphological changes, a global assessment of reservoirs on river flow regime changes for the first time and a consistent map of river widening or narrowing signals that occurred in the early decades of the 21st century. We expect such information helps improve understanding of the pattern of decadal river water changes and contributes to better guidance for river protection. We acknowledge that future studies are needed to understand how rivers respond to climate and human forcings.
My overall sense is that the major contribution in this work is the methodology behind it, and not the patterns of channel change summary reported in the core paper. As someone with an interest in patterns of river change, I found much more interesting material in the appendix and supplementary sections than the broad and fuzzy generalizations in the main work. The tremendous effort that has gone into the geospatial change detection is much better reflected in the tools and techniques, rather than the description of changes, which are broadly known, but have not previously been summarized with this kind of detailed, spatiallyexplicit method.
Response: Thanks for pointing out the issue with detailed explanations. We seriously considered this problem and revised the logic and the core contents to avoid a total bias toward technical issues.
Specifically, considerable changes have been made in describing the patterns to avoid broad and fuzzy generalizations. We supplemented detailed information and various supporting statistics to be as specific and clear as possible. As you suggested that there is interesting material in the appendix and supplementary sections, we synthesized information that could be interesting and helpful for understanding river changes into the main figures. However, as you suggest that the methodology is novel and should be highlighted, we have to achieve a balance between explaining the methods, hypothesis, and uncertainties and describing the patterns and findings. We enriched methodological information in the Methods part and put supporting figures and tables in the supplementary file.
I believe that the work would be better tailored for a premiere remote sensing journal, where the readership will better appreciate the innovation that is represented here. I would encourage the authors to merge the core and supplementary material for a more expansive treatment of the technique, better delineation of a 'research question', explanation of some of the errors/pitfalls encountered, and deeper perspective on the implications of change in the various large river systems. In its abbreviated form, I think you are doing something of a disservice to your efforts in algorithm development and application.
Response: We appreciate that you advocate the methodology highly. But generally, we believe this work appeals to a broad audience interested in river dynamics in response to climate change and direct human impacts. Although the methodology offers a new way of examining global river changes, the results clearly demonstrate the merits of the methodology by presenting an overall picture of different types of river extent change signals and the global pattern in recent decades. We understand that the previous manuscript presentation was weak in offering a clear and in-depth analysis of the results, so the findings seemed ambiguous and general. Therefore, we made substantial changes in the logic and contents, with efforts to identify different types of river extent changes and the hierarchy of multiple drivers more systematically. According to the suggestions, we also reorganized the materials in the supplementary files and methods for a clear and expansive

Dear reviewers,
Thank you very much for your help in reviewing and improving this manuscript. With full consideration of your concerning points and constructive comments, we performed a thorough revision to improve the analysis and presentation of this study. This takes longer than expected, and thanks for the extended time which allows for the completion of a better-quality revision. Here we explain and summarize the major changes below.

Logics of the main changes:
One key issue raised by the reviewers is that our previous analysis did not hierarch multiple potential drivers of river extent change, which leads to a too general or fuzzy description of the river change patterns. In the original manuscript, we discussed river changes of increase, decrease, and stability separately and organized them into five different stories. In response to the reviewer's suggestions, we now hierarch multiple potential drivers by separating different types of river flow regime changes: morphological dynamics, river expansion due to new dams, and hydrological signals (impacted by climate change and/or other ways of human activities). The challenge of this question lies in identifying the morphological dynamics that result in river extent changes, which require detailed information on the river morphological attributes. Therefore, we reprocessed all river statistical extent and improved the techniques of generating river extents with the informative attributes (e.g., meandering length, sinuosity, number of channels) according to the SWORD river database (the river network mostly inherits from the GRWL database). Finally, the schematic chart of the revised data processing and analysis is presented in Fig. S2 in the supplementary file.
Our analysis assumes that 1) morphological dynamics represent a particular style of river extent changes that are predominated by channel shifting rather than hydrological signals such as water extent widening or narrowing, 2) dam-related river expansion reflects notable river widening due to direct human intervention, 3) the remaining information reflects the general hydrological signal (widening/narrowing). Therefore, the logic of the results is changed from five stories to three parts. To interpret the hydrological signal, we used a new metric, the relative magnitude of net increased river extents, to analyze the dominant signal and retrieve hotspots of river changes. The original three stories of river changes (widening in the third-pole and pan-artic, narrowing in arid regions of interior continents, and the Yellow River case) are reorganized into the third-part -hotspot analysis. In addition, we conducted a systemic climate change analysis for the eight hotspots to discuss the potential forcings. Due to updates with SWORD and the hierarchical analysis, the specific statistics of river changes are somehow different. Nevertheless, the main results and arguments remain similar in general. The renewed results explicitly show where the widening/narrowing signal has been prominent and what the potential role of climate change and human interferences is. Given revision in methods and analysis, as well as comments from the reviewers, we modified the title to better reflect the main idea of this work.
Details of these changes are specified as follows.

(1) New information about morphological dynamics
This new information identifies river basins (Level-9 in the HydroBASINS dataset) where significant river extent changes reflect morphological dynamics (migration or shifting of meandering/multi-thread channels). Such signals can be distinguished based on the OCI map and the morphological attributes. We developed an expert-based guidance to manually interpreted this type of signal (Fig. S3, S4) for training samples and optimized three widelyused machine learning approaches for the global classification.

(2) Separated evaluation of the dam-related river expansion
We retrieved new reservoir-type river reaches by using an updated global reservoir inventory. The results separately evaluate the dam-related river expansion in the second part.
We focus on where new dams were built and to what extent the dams impacted river water extent changes.

(3) Hotspot analysis of the hydrological signal on river extent changes
In addition to a general description of the pattern of river extent changes, we performed further hotspot analysis to identify the location and extent of the change signal. Given the high spatial variability of such a signal, we focused on the four largest river-widening hotspots and the four largest narrowing hotspots. Different climatic datasets were used to analyze the general trends of precipitation and evaporation changes in these hotspots. The climate analysis generally explains the contrasting pattern of river extent changes reflected in the two types of hotspots, except the Yellow River basin, where improvement of water use policy probably played a critical role. We argue that the increasing human water extraction could have aggravated the drying-related river narrowing in the four negative hotspots.

(4) Uncertainty analysis
To address the reviewers' concerns about the OCI-based statistics' reliability, we addressed the algorithm robustness, the seasonal effect, and the uneven coverage of the Landsat observations in a new section in the Methods. We compared different sources of data on a couple of typical cases to evaluate the robustness of the Landsat-based river change patterns, and demonstrated that seasonal effects had been primarily mitigated in the statistics algorithm.
Finally, we compared in-situ discharge observations with our statistics to show that the metric used in this study is a generally reliable indicator of the hydrological signal. We also discussed that uncertainties are mainly sourced from the temporal unevenness of Landsat observations which means temporal ambiguity for a few local regions.

(6) Language and others
We would like to thank the reviewers for the careful reviewing and detailed comments.
We corrected presentation problems and grammatical errors, e.g., region names and use of some terms and phrases. In addition, new references were added to support some of the new arguments. We also included the latest literature relevant to this work for comparison.
We attach the detailed item-by-item response to all comments and suggestions for the evaluation.

Sincerely
Corresponding author and other co-authors

REVIEWERS' COMMENTS
Reviewer #1 (Remarks to the Author): As reviewer #1, I reviewed this MS for this second run. I am pleased to say that the authors carefully took into account the different suggestions. We have now a very impressive MS with clear messages. THe different drivers are convincingly introduced and separated and the take-home messages are straighforward. I am very pleased to recommand it for publication. I have only very minor comments/suggestions in the joined annoted files.
Reviewer #3 (Remarks to the Author): The authors are commended for the considerable revision work. The paper has become more focused, and the process distinctions are very helpful. While I'm still not entirely clear on precisely what is new in the paper, the three process categories (R,M,H) are new, and provide a helpful lens for interpreting the width changes. The statistics on river widening and narrowing have been reported in other work, though perhaps not with the detail and process distinction that is offered here. I do question whether one can neatly partition these three without overlap, but I would a agree there is probably a suitably dominant category for any given river. The important point is made (l.147-148) that there can be multiple processes at work. I think the identification of the global hotspots of change is the most notable part of the work. In my review of the earlier version, I pointed to the problem of fairly broad and diffuse statements about river change -I think the hotspots component could be a central focusing research question, and it should perhaps be duly reflected in the title. I would say the abstract does not effectively capture this contribution (the conclusion does this better, to some extent), and it could be sharpened quite a bit to reflect this focus, and capture reader attention. While the broadscale quantification of river change is impressive, there is important context and purpose here that makes this more than an accounting exercise.
The M-Type section does not come away with any new, tangible conclusions (as stated [l.153], it would be difficult to do so!). The importance here is mainly partitioning this population from the larger dataset -if more room was needed for the hotspot discussion, this could be tightened a bit. I find the link between stable morphologies and socio-economic development (l.344ff) a bit.. fraught, as built-up areas are generally sited far from more dynamic (i.e. steepland, mountain front) rivers and fans with higher sediment supply and more active meander dynamics. The countries in the ranking with the most stable rivers (Fig 8) are also the ones in old, flat and stable cratonic settings (excepting Japan, of course!). But indeed, embankments do keep things static, as well.
The figures are very nicely done, and convey all the key messages effectively. I have not thoroughly checked the text, but I have listed a few issues, below. With some further streamlining, I expect this paper will be of great interest to Nature Geoscience readership. l.28 "water occurrence changes" -suggest "changes in water extent" l.37 "could be likely driven" -suggest "could likely be driven" l.38, 45, 56 simply, "interference" (no s) l.49 "Rivers are one of.." l.56 -when a river becomes a reservoir, is it 'widened', or has it become a lake (i.e. not a river anymore)? l.74-75 "..changes exclusively in global rivers.." Why exclusively? l.126 "On this type of river basins.." remove s, or make it "these types of river basins" l.128 Are these necessarily flow regime changes, or could they be evolving morphologies? l.138 Similarly, is there necessarily decreased channel stability, or is that how the river evolves? l.154 in-situ, or on-site. l.235 "outpowering that of a decrease" -reword. l.303 "hotspot (d)"