Next Article in Journal
Convolutional Neural Network with Spatial-Variant Convolution Kernel
Next Article in Special Issue
Correlation Analysis between Air Temperature and MODIS Land Surface Temperature and Prediction of Air Temperature Using TensorFlow Long Short-Term Memory for the Period of Occurrence of Cold and Heat Waves
Previous Article in Journal
A Method for Extracting Some Key Terrain Features from Shaded Relief of Digital Terrain Models
Previous Article in Special Issue
A New and Simplified Approach for Estimating the Daily River Discharge of the Tibetan Plateau Using Satellite Precipitation: An Initial Study on the Upper Brahmaputra River
 
 
Letter
Peer-Review Record

Discharge Estimation Using Harmonized Landsat and Sentinel-2 Product: Case Studies in the Murray Darling Basin

Remote Sens. 2020, 12(17), 2810; https://doi.org/10.3390/rs12172810
by Zhuolin Shi 1,2, Yun Chen 3, Qihang Liu 1,2 and Chang Huang 1,2,4,*
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Remote Sens. 2020, 12(17), 2810; https://doi.org/10.3390/rs12172810
Submission received: 23 June 2020 / Revised: 22 August 2020 / Accepted: 26 August 2020 / Published: 30 August 2020
(This article belongs to the Special Issue Remote Sensing for Streamflow Simulation)

Round 1

Reviewer 1 Report

In generall the application of this known method to Sentinel-2 has some merit. The study design does however not allow too many conclusions.

Some of your conclusions are not supported by your results. Using just 1 year for calibration and 1 year for validation is rather week. Do therefore not oversell your results.

Remove unsupported statements or where possible extent the methods and results part to support them.

English language editing is required in many places.

Detailed comments:

Abstract

Line 16 :  A representative one is to -> An example is the (or: For example, the use of …)

Line 19 :  fitting a linear regression -> fitting a linear function OR applying a linear regression

Line 72: the NIR band is the easy required input of C/M method, -> rephrase

Figure 1: partially lacks features required for maps (grid, scale (on right))

Line 113: flow -> discharge (also later on)

Line 135: return period -> repeat interval / revisit interval

Line 136 – 142: It somehow is clear what you mean, but it needs rephrasing

Figure 3: explain C, V and M in the caption

Figure 4: Explain C etc. in caption, add scale; where is the gauge located

Line 223: ‘Successfully utilizing Harmonized Landsat and Sentinel-2 product here suggests’ ->rephrase

Line 237: ‘our study implies that a regression model established for a single water year is robust enough for modelling discharges for other time periods, ‘, you only proof that for one year, what is not sufficient. Rephrase this or validate this for more years.

Line 238: extend -> extended

Line 264 in Discussion: ‘We conducted a rough multiple M …’, either drop this or include it properly in the methods and results section

292: proves -> demonstrates (you do not really proof it, 1 year is not sufficient)

293: ‘Further, it demonstrates that the C/M method can be applied to completely ungauged basins ..’ This statement is not supported by the results, just speculation based on discussion of a different study. Remove this.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

General comment:

The work deals with the application of a well know method (C/M method) for discharge estimation to the recent Harmonized Landsat and Sentinel-2 (HLS) product. Despite the methodology is not new, the idea is interesting and has value. The manuscript is well structured and presented. I have some comments regarding the solidity of the investigation. As a general statement, I believe the work might be suitable for publication after some modifications.

Main comments:

L107: river widths are 60 and 90 m, while in the abstract you claim to investigate the possibility to estimate the river discharge for river having width ranging in the interval 30-100 m. The same is recall at line 277. How can you say that? Please, clarify this apparent inconsistency.

L158-L163: the methodology adopted to identify M point is clear. However, this requires the availability of recorded discharge values with which calculate the correlation. This is clearly a limit of the methodology, since it restricts the possibility to apply it only at those locations where you already get data.

What are the available options to overcome this limit? Have you tried to investigate if M points are characterized by specific characteristics (e.g., range of reflectance, limited variability within the year, or other proxies)? This would definitely improve the breath of your investigation.

Section 2.2.1: this section appears to me too synthetic. I am in line with the concept of directing the readers to other sources that provide details on the applied methodology (Tarpanelli et al. in this case), however, the information reported here is very limited. A reader should get the main concept and then decide whether to investigate further or not. I believe some additional explanations on C/M approach are useful (maybe with a methodological sketch?).    

L197: Isn’t the opposite? The interpolation line (the one you are going to use to estimate Q based on C/M value) will give you discharge lower than what you observed. The same for peaks. This is also evident looking at Figure 6a.

Results: I fear two case studies represents a limited set of scenario to provide general consideration on the applied methodology. Would it possible to extend the application to additional locations, maybe in different context? If not, this limitation should be considered and emphasized in the document.

Discussion: the discussion is interesting but rather general and does not provide a deeper consideration of the possible reasons behind the difference you got on the performance at the two locations. Do you have any idea regarding the reasons of such difference? Does it depend only on river width? Is it related to discharge variability at the locations? Again, the limited number of locations adopted for the test represent a limit, avoiding the opportunity to have more examples for such considerations.

L261 –on: this second approach should be anticipated in the methodology section. Results coming from its application need to be better described and additional detailed provided. Also, the sentence “Overall, we consider the  multiple M strategy improves the stability by possibly sacrificing some accuracy” need clarification. How did you evaluated the stability? Is there a quantitative evaluation behind this statement? Looking at the performance indexes you have shown there are no elements sustaining the adoption of the multiple M approach.

L272: how would you define a “serious” cloud issue? Also, how many more images did you used in your case by moving from single M to multiple M approach?

 

Minor comments:

L38-39: please reword the sentence. It seems to me the verb is missing.

L107: how representative is the river width extracted from Google? I guess the width might vary in relation to the timing of the image. May be in these cases river widths are quite constant in time and easy to detect, however, it might be useful to spot available river widths dataset (e.g., Frasson et al., 2019; Yang et al., 2019; Andreadis et al., 2013).

Also, why did you need to extract the width if they are gauging stations?

L133: how many images did you use at the end for the two periods (modelling and validation)? Please, specify.

L150: what is the extent of the ROI? How did you define it?

L151: “certain distance to the gauge”. Please, specify what you mean for “certain distance”.

L175: I would consider rewording the sentence relative to “excellent” RRMSE. It’s correct, but I feel like the terminology is not really appropriate.

Figure 4: it would be interesting to see the variability of p value in panel e) and f). Maybe increasing the number of classes?

L226: “accuracy according to [34].”. This is not very informative. Please explain with plain text what you meant here.

L249-250: “it was suggested…[]”. I am lost. Where was it? Is it a result of your investigation? Or, does it come from previous studies? Please, explain better the concept.

L293-294: please remove or reword this sentence. In this study, I do not think you demonstrate what you state here.

 

Additional references

Frasson, P. R. D. M., Pavelsky, T. M., Fonstad, M. A., Beighley, R. E., & Yang, X. (2019). Global Relationships Between River Width , Slope , Catchment Area , Meander Wavelength , Sinuosity , and Discharge Geophysical Research Letters. Geophysical Research Letters, 46, 3252–3262. https://doi.org/10.1029/2019GL082027

Yang, X., Pavelsky, T. M., Allen, G. H., & Donchyts, G. (2019). RivWidthCloud : An Automated Google Earth Engine Algorithm for River Width Extraction From Remotely Sensed Imagery. IEEE Geoscience and Remote Sensing Letters, 1–5.

Andreadis, K. M., Schumann, G. J. P., & Pavelsky, T. (2013). A simple global river bankfull width and depth database. Water Resources Research, 49(10), 7164–7168. https://doi.org/10.1002/wrcr.20440

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

line 72 : of -> for the
line 114 : , used -> were used
line 139 : Flowchart of methodology used -> A flowchart detailing the used inputs and methods ...
, which -> . It
line 142 : the estimation formula -> the empirical model
line 143 : discharges -> discharge
line 179 : put everything after ';' in ()
line 180 : representing RMSE value normalized by observed discharge, which indicates higher modelling accuracy with lower value ->

representing the RMSE value normalized by the observed discharge, which indicates higher modelling accuracy with a lower value

line 237: The successful use of Harmonized Landsat and Sentinel-2 product in this study implies -> The results based on the use of a harmonized Landsat and Sentinel-2 product in this study suggest ...

line 266: The performances of discharge estimations would be different possibly due to river width, floodplain, etc. ->
The performance of discharge estimation may vary due to river width, floodplain patterns, etc.

line 287: which may possibly due to many parameters such as river width, floodplain ->
which may result from river width or floodplain specific patterns.

Author Response

Comments and Suggestions for Authors

line 72: of -> for the

line 114: , used -> were used

line 139: Flowchart of methodology used -> A flowchart detailing the used inputs and methods ...

, which -> . It

line 142: the estimation formula -> the empirical model

line 143: discharges -> discharge

Response: Thanks for your careful review. These have been revised.

 

line 179: put everything after ';' in ()

line 180: representing RMSE value normalized by observed discharge, which indicates higher modelling accuracy with lower value -> representing the RMSE value normalized by the observed discharge, which indicates higher modelling accuracy with a lower value

Response: These sentences have been reorganized.

 

line 237: The successful use of Harmonized Landsat and Sentinel-2 product in this study implies -> The results based on the use of a harmonized Landsat and Sentinel-2 product in this study suggest ...

line 266: The performances of discharge estimations would be different possibly due to river width, floodplain, etc. ->

The performance of discharge estimation may vary due to river width, floodplain patterns, etc.

line 287: which may possibly due to many parameters such as river width, floodplain ->

which may result from river width or floodplain specific patterns.

Response: Thanks for giving us these patient and detailed comments. We have revised all of them as suggested.

 

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.


Round 1

Reviewer 1 Report

MDPI Remote Sensing

# 759199

Discharge Estimation Using Harmonized Landsat and Sentinel-2 Product: Case Studies in the Murray Darling Basin

Zhuolin Shi, Yun Chen, Qihang Liu and Chang Huang

The classification in 3 categories as described in lines 41-49 is not very clear. Many other references do exist on this issue and I wonder where you place them? For example:

https://www.sciencedirect.com/science/article/abs/pii/S0034425718305637

This one seems close to your Category 1 but without local data. So where do you place it?

https://www.sciencedirect.com/science/article/pii/S0022169418300805

This one seems close to Category 2 (since using models) but without requiring bathymetric data. So is it Category 3?

https://agupubs.onlinelibrary.wiley.com/doi/full/10.1002/2015WR018434

All these algo are in which category? Category 3? They do not use bathymetry data, but they use a mean annual discharge (From Water Balance Model data)

Please improve you bib review and if you propose a classification it must be more rigorous than what you propose here. If your classification is based on the data used, please tell what these data are precisely, and then the fact that they use models or not is another criteria. In your 3rd Category you say “only remotely sensed data which is sensitive to surface water dynamics”. Please be more precise, you mean water elevation and water width? Water slope? etc

Line 99: you use ML/Day unit. This unit is used only in Australia. I guess your paper targets the international audience. Please use metric units (m3/s).

You should give more details about the C/M method. Even if you refer to Tarpanelli et al, you can give more details in section 2.2.1. For the readers not accustomed to this method, and to the HLS date you can give more details, and illustrate. You talk about the “ratio of typical land and water reflectance from NIR band” (line 124). What does this means? Each pixel has such value in these 2 categories. Can you illustrate with some data? C and M are what? One pixels? Several pixels? For C it seems to be a series of pixels and then you average some values for them. For M it seems this is a unique pixel. You definitely have to better explain these C, M pixels selection and meaning. People from remote sensing community may know it, but your targeted audience is also hydrologist.

Lines 155-157: why your sum is over i for RMSE and over t for NSE? Do you have a regular time step of 3 days? It should be the same, or define the link between the 2 (t=i.DT , with DT=3 days?)

Line 245: “Further, the C/M method could be applied to completely ungauged basins if their hydrologically similar basins with flow observations could be found.” Well not proved at all … And your sentence is not clear. What do you mean by hydrologically similar … This point is very important since as you say you need in situ obs for your methodology. So what is the use of the method since you have these in situ obs ?

As a conclusion in addition to minor modifications you really have to clarify the methodology about C and M points.

Author Response

  1. The classification in 3 categories as described in lines 41-49 is not very clear. Many other references do exist on this issue and I wonder where you place them? For example:

https://www.sciencedirect.com/science/article/abs/pii/S0034425718305637

This one seems close to your Category 1 but without local data. So where do you place it?

https://www.sciencedirect.com/science/article/pii/S0022169418300805

This one seems close to Category 2 (since using models) but without requiring bathymetric data. So is it Category 3?

https://agupubs.onlinelibrary.wiley.com/doi/full/10.1002/2015WR018434

All these algo are in which category? Category 3? They do not use bathymetry data, but they use a mean annual discharge (From Water Balance Model data)

Please improve you bib review and if you propose a classification it must be more rigorous than what you propose here. If your classification is based on the data used, please tell what these data are precisely, and then the fact that they use models or not is another criteria. In your 3rd Category you say “only remotely sensed data which is sensitive to surface water dynamics”. Please be more precise, you mean water elevation and water width? Water slope? etc

 

Response:

Thanks for this advice. It is understood that there are numerous discharge estimation methods that have been proposed so far. It is a little bit too ambitious for us to categorize all of them into three categories. What we are trying to do here is to provide a roughly classification to these methods according to their general assumptions and input data requirements. The first category corresponds to the traditional and old school statistical methods, represented by the rating curve method. The second category corresponds to those physically based hydrological or hydraulic modeling methods, which were widely used by hydrologists. We tend to put all the recent remote sensing driven methods into the third category. We are aware that the upcoming SWOT mission would inject huge fresh vitalities into this category, because it will provide direct measurements of surface water slope, effective river width, and water level, which are essential parameters for estimating discharge variation. For now, many pre-SWOT studies have been conducted, some using AirSWOT data, and some using multi-source remote sensing data including altimetry data to synthesize SWOT-type observations for discharge estimation. Therefore, we would categorize these three literatures (Oubanas et al, 2018; Kim et al, 2019; Durand et al, 2016) into category three. C/M method is just another type of remote sensing driven method that relies on the correlation between discharge variation and surface water extent dynamics, which can be simply reflected by common optical or microwave imagery. But it also has disadvantages such as difficulties in selecting C and M pixels, and requirement of modelling discharge observations. We have revised this part by rephrasing these categories and supplementing several references, see as below. It has to be noted that there are still many relevant studies that have not been mentioned here, besides, some studies may not be able to precisely classified into any of these categories.

“Overall, there are generally three categories of methods for discharge estimation using remote sensing data [6]. The first category correlates remotely sensed water levels or inundation areas acquired at or near a gauge station with simultaneously collected ground data. This kind of methods is usually called rating curve, such as [7]. The second category uses physically based hydrological or hydraulic modelling based on river topographic information. Remotely sensed river width or water level were generally used as one of the input parameters, or reference for calibrating the models. One example of this category is [8]. Methods in these two categories can be difficult to operate due to complex model computation and/or shortage of input data including river bathymetry information. The third category is remote sensing driven methods that use remotely sensed data as the major inputs to estimate discharge based on simple hydraulic geometry in the river channels (such as [9-13]). The upcoming Surface Water and Ocean Topography (SWOT) satellite mission is expected to provide a great innovation into this category, due to its ability of direct measurements of surface water slope, effective river width, and water level [3]. Until now, many pre-SWOT studies have been conducted for discharge estimation, some using airborne calibration and validation instrument for SWOT (AirSWOT) [4], and some using multi-source remote sensing data (including altimetry data in particular) to synthesize SWOT-type observations [14-16]. Another type of method in the third category is the C/M method developed by Brakenridge et al [17], taking advantage of the correlation between river discharge and surface water extent dynamics. While the surface water extent dynamics can be easily captured by many conventional sensors, this C/M method has the advantage of low data requirements. Its original principle is …”

 

References:

Oubanas, H.; Gejadze, I.; Malaterre, P. O.; Mercier, F. River discharge estimation from synthetic SWOT-type observations using variational data assimilation and the full Saint-Venant hydraulic model. J. Hydrol. 2018, 559, 638–647. (https://doi:10.1016/j.jhydrol.2018.02.004).

Kim, D.; Yu, H.; Lee, H.; Beighley, E.; Durand, M.; Alsdorf, D. E.; Hwang, E. Ensemble learning regression for estimating river discharges using satellite altimetry data: Central Congo River as a Test-bed. Remote Sens. Environ. 2019, 221, 741–755.

(https://doi:10.1016/j.rse.2018.12.010).

Durand, M.; Gleason, C. J.; Garambois, P. A.; Bjerklie, D.; Smith, L. C.; Roux, H.; Rodriguez, E.; Bates, P. D.; Pavelsky, T. M.; Monnier, J.; Chen, X.; Di Baldassarre, G.; Fiset, J.-M.; Flipo, N.; Frasson, R. P. D. M.; Fulton, J.; Goutal, N.; Hossain, F.; Humphries, E.; Minear, J. T.; Mukolwe, M. M.; Neal, J. C.; Ricci, S.; Sanders, B. F.; Schumann, G.; Schubert, J. E.; Vilmin, L. An intercomparison of remote sensing river discharge estimation algorithms from measurements of river height, width, and slope. Water Resour. Res. 2016, 52(6), 4527-4549. (https://doi.org/10.1002/2015WR018434).

 

 

  1. Line 99: you use ML/Day unit. This unit is used only in Australia. I guess your paper targets the international audience. Please use metric units (m3/s).

Response:

Thanks, we have already converted ML/Day to m3/s (1 ML/day = 0.01157 m3/s), including figure2, 5, 6, table2 and other related expressions in the manuscript.

 

  1. You should give more details about the C/M method. Even if you refer to Tarpanelli et al, you can give more details in section 2.2.1. For the readers not accustomed to this method, and to the HLS date you can give more details, and illustrate. You talk about the “ratio of typical land and water reflectance from NIR band” (line 124). What does this means? Each pixel has such value in these 2 categories. Can you illustrate with some data? C and M are what? One pixels? Several pixels? For C it seems to be a series of pixels and then you average some values for them. For M it seems this is a unique pixel. You definitely have to better explain these C, M pixels selection and meaning. People from remote sensing community may know it, but your targeted audience is also hydrologist.

Response:

“ratio of typical land and water reflectance from NIR band” is revised as “ratio of reflectance from NIR band between one stable land pixel for calibration (C) and one pixel within river for measurement (M)”.

 

Both C and M are originally defined as one unique pixel to stand for a typical land pixel never inundated even in flood and a pixel within river whose reflectance is most sensitive to surface water extent dynamics. With the development of C/M method, value of one C pixel on the land was later revised to be the average value of multiple land pixels to increase the stability of C observations. In this study, C stands for the average reflectance value of multiple land pixels that meet our criteria, M is a single pixel carefully selected based on the correlation coefficient r. Detailed descriptions have been supplemented to section 2.2.1 to explain the C/M method more clearly as below.

 

“The original principle of C/M modelling is to take advantage of the ratio of reflectance from NIR band between one stable land pixel for calibration (C) and one pixel within river for measurement (M), and establish the relationship between the ratio and the observed river discharge. In this study, C pixel was revised to be the average value of multiple land pixels in order to increase the stability of C observations. M remains to be the reflectance of a single pixel.”

 

Regarding to the HLS data, the following description has been added to section 2.1.2.

 

“The HLS products are based on a set of algorithms to obtain seamless products from both sensors (OLI and MSI), including atmospheric correction, cloud and cloud-shadow masking, spatial co-registration and common gridding, bidirectional reflectance distribution function normalization and spectral bandpass adjustment.”

 

 

 

  1. Lines 155-157: why your sum is over i for RMSE and over t for NSE? Do you have a regular time step of 3 days? It should be the same, or define the link between the 2 (t=i.DT , with DT=3 days?)

Response:

Thanks. They should be the same, both represent the tth observation. There is no such definition as regular time step here. Corresponding observations were extracted from daily series according to the date of remote sensing data. We have revised equation (1).

 

  1. Line 245: “Further, the C/M method could be applied to completely ungauged basins if their hydrologically similar basins with flow observations could be found.” Well not proved at all … And your sentence is not clear. What do you mean by hydrologically similar … This point is very important since as you say you need in situ obs for your methodology. So what is the use of the method since you have these in situ obs ?

Response:

Indeed, the C/M method has this drawback of requiring some in situ observations to establish the model. But we can still see its merit in two aspects. First, as was demonstrated in our study, we used only one year of observed discharge data to establish the model, and the model performs well for the other year. This means that we can reconstruct a long time series of discharge from time series remote sensing data through modelling with limited observations. This can be helpful for basins that have abandoned gauges which have collected some observations to be continuously monitored by satellites. This can also be helpful for some basins if we can set up some low cost discharge observation devices to collect some data for establishing the C/M models. Second, for those remote basins that are hard to access, a general solution is to apply hydrological analogy to gauge basins or easily accessed basins. This means that we can use the observed discharge of their hydrologically similar basins for establishing the C/M models (Li et al. 2014; Li et al. 2019). Regarding the hydrologically similar, a paragraph as below has been supplemented to the Discussion Part.

 

“Besides, based on the theory of hydrological analogy [36], basins are hydrologically similar if they have similar geographical conditions, climate and water sources. The river discharge characteristic of a gauged basin can be applied to estimate the discharge of a hydrologically similar ungauged basin. Based on this theory, Li et al [24] presented a similarity coefficient of multiplying basin area and multiyear average precipitation, which was then combined with the C/M model at a gauged basin to obtain discharge estimation at its hydrologically similar basin. This confirms the applicability of C/M method in ungauged basins, as long as we can find a hydrologically similar basin of them that has discharge observations.”

 

References:

Li, F.; Zhang, Y.; Xu, Z.; Liu, C.; Zhou, Y.; Liu, W. Runoff predictions in ungauged catchments in southeast Tibetan Plateau. J. Hydrol. 2014, 511, 28–38.

(https://doi:10.1016/j.jhydrol.2014.01.014).

Li, H.J.; Li, H.Y.; Wang, J.; Hao, X.H. Extending the ability of Near‐Infrared images to monitor small river discharge on the northeastern Tibetan Plateau. Water Resour. Res. 2019, 55(11), 8404-8421. (https://doi.org/10.1029/2018WR023808).

 

As a conclusion in addition to minor modifications you really have to clarify the methodology about C and M points.

Response:

Thank you very much for your valuable comments and suggestions. According to them, we have revised the literature review in Introduction, clarified the selection process of C and M, and further explained the application of C/M method in ungauged basins. We have also made some other minor revisions to make this manuscript read more fluently. We hope this version would resolve your concerns about our paper.

Reviewer 2 Report

Overall, the article is written in good quality with solid logic. I plan to accept the article ONLY AFTER the following issues are clearly addressed.

The article has the originality issue because this study is a simple implementation of the existing C/M method with the additon of new Sentinel-2 data. Therefore, the introduction must be more thorough and literature reviews including the recently published papers on the topic.

The presented methodology requires the flow discharge value to be measured to develop the C-M relationship, which is hard to achieve at the remote basins that are hard to access. Authors may want to add a chapter discussing specifically how to overcome this inherent methodological drawback.

I am not sure whether the metric of N-S coefficient is really helpful because the satellite images are not always available at a uniform frequency due to the clouds while N-S is meaningful when the observation frequency is uniform. I suggest the metrics to be excluded from the manuscript.

Section 2.2.1: Did you use all available satellite images or selected ones? Please provide a clear logic and explanation on which images were used to develop the C-M relationship. In addition, I don't agree with the idea of choosing only one pixel to develop the C-M relationship because you cannot use the C-M relationship for the validation period if the pixel is hidden by the clouds. Please consider using multiple pixels with various C-M relationship and add as many point as possible for the validation. Here, I will be okay even if the result is not good, but I am more interested in whether you provide a quantified metric on the performance of the proposed methodology with multiple C-M pixels. You may want to compare the case of one CM-pixel and multiple C-M pixels.

 

 

Author Response

Comments and Suggestions for Authors

Overall, the article is written in good quality with solid logic. I plan to accept the article ONLY AFTER the following issues are clearly addressed.

Response:

Thanks for the helpful comments. We have taken them seriously and revised our manuscript accordingly. We hope this revised version would satisfy you and resolve your concerns.

 

  1. The article has the originality issue because this study is a simple implementation of the existing C/M method with the addition of new Sentinel-2 data. Therefore, the introduction must be more thorough and literature reviews including the recently published papers on the topic.

Response:

Indeed, this is an extend application of existing C/M method with a new and popular synthesized product (HLS) for Australia small rivers. While the original C/M method was proposed originally for using coarse resolution microwave data and MODIS data on large rivers, this study evaluated the feasibility of applying much higher resolution HLS data for estimating discharge on relatively small rivers. Since the HLS product has the advantages of high spatial resolution and high temporal resolution, our study provides a detailed investigation regarding the applicability of this product, and would promote its application in the field of river discharge estimation. We have revised the Introduction by reorganizing the literature review, and supplementing C/M related studies (Hou et al. 2018; Li et al. 2019) particularly as below.

 

“Overall, there are generally three categories of methods for discharge estimation using remote sensing data [6]. The first category correlates remotely sensed water levels or inundation areas acquired at or near a gauge station with simultaneously collected ground data. This kind of methods is usually called rating curve, such as [7]. The second category uses physically based hydrological or hydraulic modelling based on river topographic information. Remotely sensed river width or water level were generally used as one of the input parameters, or reference for calibrating the models. One example of this category is [8]. Methods in these two categories can be difficult to operate due to complex model computation and/or shortage of input data including river bathymetry information. The third category is remote sensing driven methods that use remotely sensed data as the major inputs to estimate discharge based on simple hydraulic geometry in the river channels (such as [9-13]). The upcoming Surface Water and Ocean Topography (SWOT) satellite mission is expected to provide a great innovation into this category, due to its ability of direct measurements of surface water slope, effective river width, and water level [3]. Until now, many pre-SWOT studies have been conducted for discharge estimation, some using airborne calibration and validation instrument for SWOT (AirSWOT) [4], and some using multi-source remote sensing data (including altimetry data in particular) to synthesize SWOT-type observations [14-16]. Another type of method in the third category is the C/M method developed by Brakenridge et al [17], taking advantage of the correlation between river discharge and surface water extent dynamics. While the surface water extent dynamics can be easily captured by many conventional sensors, this C/M method has the advantage of low data requirements. Its original principle is calculating the ratio between a stable land pixel for calibration (C) and a pixel within river for measurement (M), then fitting a linear regression between C/M series and observed discharge series to give discharge information at a site. The C/M method was first implemented using the Near Infrared (NIR) band of Moderate Resolution Imaging Spectroradiometer (MODIS), then implemented using brightness temperature of Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E) data [18], later published as Global Flood Detection System (GFDS) for global discharge estimation and flood detection (http://www.gdacs.org/flooddetection/overview.aspx), even for those ungauged and inaccessible rivers [19, 20]. Tarpanelli et al [21] further implemented this method and successfully estimated discharge for four reaches of the Po River in northern Italy based on carefully selected M and C pixels from MODIS NIR images. They then applied this method to the Niger-Benue River in 2017 [22] and achieved reasonable accuracy. Hou et al [23] tested C/M method using both MODIS and GFDS data across the Amazon Basin, a good performance of satellite-based river gauging was obtained to monitor the discharge.”

 

References:

Hou, J.W.; Dijk, A.; Renzullo, L.; Vertessy, R. Using modelled discharge to develop satellite-based

river gauging: a case study for the Amazon Basin. Hydrol. Earth Syst. Sci. 2018, 22, 6435–6448. (https://doi.org/10.5194/hess-22-6435-2018).

Li, H.J.; Li, H.Y.; Wang, J.; Hao, X.H. Extending the ability of Near‐Infrared images to monitor

small river discharge on the northeastern Tibetan Plateau. Water Resour. Res. 2019, 55(11), 8404-8421. (https://doi.org/10.1029/2018WR023808).

 

  1. The presented methodology requires the flow discharge value to be measured to develop the C-M relationship, which is hard to achieve at the remote basins that are hard to access. Authors may want to add a chapter discussing specifically how to overcome this inherent methodological drawback.

Response:

Thanks. Indeed, the C/M method has this drawback in its nature. But we can still see its merit in two aspects. First, as was demonstrated in our study, we used only one year of observed discharge data to establish the model, and the model performs well for the other year. This means that we can reconstruct a long time series of discharge from time series remote sensing data through modelling with limited observations. This can be helpful for basins that have abandoned gauges which have collected some observations to be continuously monitored by satellites. This can also be helpful for some basins if we can set up some low cost discharge observation devices to collect some data for establishing the C/M models. Second, for those remote basins that are hard to access, a general solution is to apply hydrological analogy to gauge basins or easily accessed basins. This means that we can use the observed discharge of their hydrologically similar basins for establishing the C/M models (Li et al. 2014; Li et al. 2019). A paragraph as below has been supplemented to the Discussion Section to enlighten this.

 

“Besides, based on the theory of hydrological analogy [36], basins are hydrologically similar if they have similar geographical conditions, climate and water sources. The river discharge characteristic of a gauged basin can be applied to estimate the discharge of a hydrologically similar ungauged basin. Based on this theory, Li et al [24] presented a similarity coefficient of multiplying basin area and multiyear average precipitation, which was then combined with the C/M model at a gauged basin to obtain discharge estimation at its hydrologically similar basin. This confirms the applicability of C/M method in ungauged basins, as long as we can find a hydrologically similar basin of them that has discharge observations.”

 

Li, F.; Zhang, Y.; Xu, Z.; Liu, C.; Zhou, Y.; Liu, W. Runoff predictions in ungauged catchments in

southeast Tibetan Plateau. J. Hydrol. 2014, 511, 28–38.

(https://doi:10.1016/j.jhydrol.2014.01.014).

Li, H.J.; Li, H.Y.; Wang, J.; Hao, X.H. Extending the ability of Near‐Infrared images to monitor

small river discharge on the northeastern Tibetan Plateau. Water Resour. Res. 2019, 55(11), 8404-8421. (https://doi.org/10.1029/2018WR023808).

 

  1. I am not sure whether the metric of N-S coefficient is really helpful because the satellite images are not always available at a uniform frequency due to the clouds while N-S is meaningful when the observation frequency is uniform. I suggest the metrics to be excluded from the manuscript.

Response:

Thanks for the advice. According to our understanding, for N-S coefficient, the time step at which the data are recorded appears to be an insignificant factor unless the sample size is small (McCuen et al. 2006). Here, we have exploited all the available high quality remote sensing observations for estimating discharge and pick all the corresponding gauge observations to calculate the N-S coefficient so that both of them have the same observation frequency. In this case, we would like to keep this index, considering it has been widely used to show the performance of discharge estimation models, such as Tarpanelli et al (2013) and Li et al. (2019). In the manuscript, we have clarified that the gauge observation was selected according to the satellite observation frequency, as below.

 

“Using observed discharge corresponding to remote sensing observations in water year 2018 as the reference, the performance of discharge estimation was evaluated by …”

 

References:

McCuen, R.; Knight, Z.; Cutter, A. Evaluation of the Nash–Sutcliffe Efficiency Index. Journal of

Hydrologic Engineering, 2006, 11(6). (https://10.1061/(ASCE)1084-0699(2006)11:6(597)).

Tarpanelli, A.; Brocca, L.; Lacava, T.; Melone, F.; Moramarco, T.; Faruolo, M.; Pergola, N.;

Tramutoli, V. Toward the estimation of river discharge variations using MODIS data in ungauged basins. Remote Sens. Environ. 2013, 136, 47-55.

(https://doi.org/10.1016/j.rse.2013.04.010).

Li, H.J.; Li, H.Y.; Wang, J.; Hao, X.H. Extending the ability of Near‐Infrared images to monitor

small river discharge on the northeastern Tibetan Plateau. Water Resour. Res. 2019, 55(11), 8404-8421. (https://doi.org/10.1029/2018WR023808).

 

  1. Section 2.2.1: Did you use all available satellite images or selected ones? Please provide a clear logic and explanation on which images were used to develop the C-M relationship. In addition, I don't agree with the idea of choosing only one pixel to develop the C-M relationship because you cannot use the C-M relationship for the validation period if the pixel is hidden by the clouds. Please consider using multiple pixels with various C-M relationship and add as many point as possible for the validation. Here, I will be okay even if the result is not good, but I am more interested in whether you provide a quantified metric on the performance of the proposed methodology with multiple C-M pixels. You may want to compare the case of one CM-pixel and multiple C-M pixels.

Response:

We first downloaded all available HLS images during the selected time period, and then checked each of them one by one to select those that have no cloud coverage in the study area. This has been further clarified in section 2.1.2. as below.

 

“All available images in the modelling and validation water years (2017 and 2018 respectively) were acquired and quality checked to ensure cloud free in the study area using visual inspection, resulting in about 65% of images being retained.”

 

In this study, only one single M pixel was selected and used to establish the discharge estimation model. Indeed, as what you said, using a single M pixel has the disadvantage of easily affected by cloud. We noticed that Li et al (2019) developed a Multiple Pixel Ratio (MPR) method, using ratio of mean value of multiple land pixels and multiple inundated pixels, to further improve the C/M method. In their study, a set of high correlation pixels within the river were selected and averaged as the M. Overall, this is a strategy that improves the stability by possibly sacrificing some accuracy.   We have supplemented a comparison between our one M pixel strategy and multiple M strategy, and added below content into the Discussion Section.

 

“Besides, as was noticed by Li et al [24], using a single M pixel to establish the model would sometimes lead to sample size reduction due to occasional cloud coverage on the pixel. They therefore proposed a multiple M strategy by selecting a set of candidate M pixels based on their r values, and taking their average reflectance as the virtual M pixel. We conducted a rough comparison about the single M strategy and multiple M strategy regarding to their modelling accuracy. Average value of multiple M pixels with Pearson’s r over 0.70 was used as the input of M pixel in the multiple M strategy. As was expected, different degrees of accuracy drop were observed when switching the single M strategy to multiple M strategy. For gauge 410130, R2 dropped from 0.81 to 0.78 (RMSE equals to 5.26 m3/s, RRMSE equals to 0.61, NSE equals to 0.01). For gauge 414200, R2 dropped from 0.77 to 0.64 (RMSE equals to 22.41 m3/s, RRMSE equals to 0.20, NSE equals to 0.64). Overall, we reckon the multiple M is a strategy that improves the stability by possibly sacrificing some accuracy. Therefore, we would suggest to choose proper strategy for different cases. If the cloud issue in the study area is serious, multiple M strategy may be a good choice to preserve as many remote sensing observations as possible, even though the accuracy might be slightly affected.”

 

References:

Li, H.J.; Li, H.Y.; Wang, J.; Hao, X.H. Extending the ability of Near‐Infrared images to monitor small river discharge on the northeastern Tibetan Plateau. Water Resour. Res. 2019, 55(11), 8404-8421. (https://doi.org/10.1029/2018WR023808).

 

Round 2

Reviewer 2 Report

I regret that I have to reject this article. The manuscript did not improve since my last suggestion. As I previously suggested and also as the authors agreed, the methodology of the article lacks the originality, so I suggested a way to improve the practicality of the methodology, which is to use multiple points to develope the C/M relationship, which was not performed in the revised manuscript. The low acquisition frequency of the passive sensors due to incremental weather condition is the major obstacle, and this paper must address this short coming. Otherwise, this paper is a simple repetition of the previous methodology.

Author Response

Comments and Suggestions for Authors

I regret that I have to reject this article. The manuscript did not improve since my last suggestion. As I previously suggested and also as the authors agreed, the methodology of the article lacks the originality, so I suggested a way to improve the practicality of the methodology, which is to use multiple points to develop the C/M relationship, which was not performed in the revised manuscript. The low acquisition frequency of the passive sensors due to incremental weather condition is the major obstacle, and this paper must address this short coming. Otherwise, this paper is a simple repetition of the previous methodology.

Response:

        Thank you for your comments that helped us greatly improved our manuscript. However, we respectfully disagree with your opinion this time. Indeed, we didn’t propose a new methodology here in this study, we simply extended the existing C/M method using the new and popular synthesized HLS product, and applied it to typical small rivers in the Murray-Darling River Basin. Since the HLS product has the advantages of both high spatial and temporal resolution, our study provides a detailed investigation regarding the applicability of this product, and would promote its application in the field of river discharge estimation. Moreover, we are in an era of multi-source data explosion, it is expected that more and more synthesized products like the HLS would be developed in the future. This study will serve as an exploratory study that proves the feasibility of using multisource synthesized datasets as the input for discharge estimation. Therefore, we believe this work is worth publishing even though we didn’t propose a new methodology.

        As was suggested by your comments in the first round, we did have investigated the strategy of using multiple M pixels. We agree it is a good strategy when the input images are frequently contaminated by cloud coverage. However, we decide not to revise our methodology thoroughly, according to the following considerations. First, the multiple M strategy has already been well documented by Li et al (2019), adopting this strategy would not further increase the novelty of our method. Second, it is quite clear that this strategy is an approach that improves the stability by possibly sacrificing some accuracy. While the multiple M strategy is a scheme for increasing the usable image count, using HLS product can also be considered as another scheme for increasing the usable image count by synthesizing multisource data. While our goal is to prove the applicability of HLS data for C/M method, keeping the single M strategy have the advantage of presenting ideal results so that we can focus on the performance of HLS data. Using the single M pixel in our study, we have actually retained a large proportion of usable images for both the modeling year and the validation year. Therefore, we believe it is not necessary to revise our framework thoroughly to switch to the multiple M strategy. However, we did have added a paragraph to compare and discuss both strategies. As was expected, the application of multiple M strategy would inevitably affect the accuracy. But if the cloud issue in the study area is serious, multiple M strategy may be a good choice to preserve as many remote sensing observations as possible, even though the accuracy might be slightly affected. We believe this is enough to illustrate this issue.

 

Li, H.J.; Li, H.Y.; Wang, J.; Hao, X.H. Extending the ability of Near‐Infrared images to monitor small river discharge on the northeastern Tibetan Plateau. Water Resour. Res. 2019, 55(11), 8404-8421.

(https://doi.org/10.1029/2018WR023808).

Back to TopTop