the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Evaluating dust emission model performance using dichotomous satellite observations of dust emission
Abstract. Measurements of dust in the atmosphere have long been used to calibrate dust emission models. However, there is growing recognition that atmospheric dust confounds the magnitude and frequency of emission from dust sources and hides potential weaknesses in dust emission model formulation. In the satellite era, dichotomous (presence = 1 or absence = 0) observations of dust emission point sources (DPS) provide a valuable inventory of regional dust emission. We used these DPS data to develop an open and transparent framework to routinely evaluate dust emission model (development) performance using coincidence of simulated and observed dust emission (or lack of emission). To illustrate the utility of this framework, we evaluated the recently developed albedo-based dust emission model (AEM) which included the traditional entrainment threshold (u*ts) at the grain scale, fixed over space and static over time, with sediment supply infinite everywhere. For comparison with the dichotomous DPS data, we reduced the AEM simulations to its frequency of occurrence in which soil surface wind friction velocity (us*) exceeds the u*ts, P(us* > u*ts). We used a global collation of nine DPS datasets from established studies to describe the spatio-temporal variation of dust emission frequency. A total of 37,352 unique DPS locations were aggregated into 1,945 1° grid boxes to harmonise data across the studies which identified a total of 59,688 dust emissions. The DPS data alone revealed that dust emission does not usually recur at the same location, are rare (1.8 %) even in North Africa and the Middle East, indicative of extreme, large wind speed events. The AEM over-estimated the occurrence of dust emission by between 1 and 2 orders of magnitude. More diagnostically, the AEM simulations coincided with dichotomous observations ~71 % of the time but simulated dust emission ~27 % of the time when no dust emission was observed. Our analysis indicates that u*ts was typically too small, needed to vary over space and time, and at the grain-scale u*ts is incompatible with the us* scale (MODIS 500 m). During observed dust emission, us* was too small because wind speeds were too small and/or the wind speed scale (ERA5; 11 km) is incompatible with the us* scale. The absence of any limit to sediment supply caused the AEM to simulate dust emission whenever P (us* > u*ts), producing many false positives when and where wind speeds were frequently large. Dust emission model scaling needs to be reconciled and new parameterisations are required for u*ts and to restrict sediment supply varying over space and time. Whilst u*ts remains poorly constrained and unrealistic assumptions persist about sediment supply and availability, the DPS data provide a basis for the calibration of dust emission models for operational use. As dust emission models develop, these DPS data provide a consistent, reproducible, and valid framework for their routine evaluation and potential model optimisation. This work emphasises the growing recognition that dust emission models should not be evaluated against atmospheric dust.
- Preprint
(3014 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on gmd-2021-423', Anonymous Referee #1, 16 Feb 2022
In this paper the authors present an approach to verifying dust emission model performance by considering how the model(s) compares to satellite observations of dust emission point sources (DPS). For this purpose they consider dichotomous (presence or absence) comparisons of near-global DPS observations (from MODIS and SEVIRI, for North America, Northern and Southern Africa, the Middle East and Central Asia, and Australia) and simulated emissions from their albedo-based dust emission model (AEM). The model appears to overestimate the number of dust events by an order of magnitude, while the DPS database indicates that dust emission events are actually quite rare in space and time, considering the vast size of some of the deserts. The authors use this dichotomous information to understand the conditions under which the model performance is influenced by the representation of the wind fields, the assumption of infinite sediment supply, and the threshold wind friction velocity u*ts. There is value to this approach to validating dust emission models using observations that represent (as closely as possible) the dust emission events that actually occurred.
I am curious about this repeated claim in the manuscript that “The aim here is to demonstrate an alternative to comparing dust emission models to atmospheric dust” (line 79). As I see it, the satellite dust emission point sources are themselves interpretations of observations of atmospheric dust, back-tracking atmospheric dust in the satellite imagery for the purpose of identifying these emission points. Hence it seems to me that the authors are themselves still comparing the dust emission models to atmospheric dust. I suppose there is a distinction here between comparisons with the retrievals of atmospheric dust transport (implicitly or explicitly being criticised), as described by the dust AOD, and with the derivations of dust emission from atmospheric observations taken shortly afterwards (apparently a more robust method). Now I actually agree that it might be more robust to try to pinpoint the dust emission sources and to compare those with the dust emission model, but I think that it would very much be worthy of clarification and further discussion on the distinction between the two types of satellite observations of atmospheric dust which are being considered here. I think it would also be worthwhile to clarify the final sentence of the abstract (“… dust emission models should not be evaluated against atmospheric dust”) to indicate that this relates to the dust AOD, assuming that this is what you mean.
This points to a general criticism that I have, that while this does seem to me to be a worthwhile paper, describing a useful method of comparing models with dust emission sources identified by satellite sources, I do feel that it is let down at times by over-generalised statements such as the one described in the above paragraph. Another such statement is “These results demonstrate that there is no reasonable basis to calibrate model performance through an adjustment to a fixed global u*ts” (lines 482-484). If this is an implicit criticism of some model dust emission schemes, is this not something of a straw man argument? I would have thought that it has been quite widely accepted for many years that u*ts is dependent on the soil size distributions and roughness lengths, which vary worldwide. I suppose however that there is some value to confirming that this should not be done, as has been done here.
Specific comments
p.5, lines 156-157: a minor point, but as written this sentence implies to me that the red beam is associated with the 8.7 µm channel, while the green beam is associated with the 12.0 µm channel. It is of course the other way round.
p.7, Eq.6: presumably in these equations it is (1-As) multiplied by (1-Af). As written, this equation implies to me that no snow = no dust!
Figure 3: it might make sense here to fit the y-axis to 0-1 (as has been done for the x-axis, 0-0.6), it would make it slightly more intuitive to read.
p.13, line 392: “Taklamakan (Kazakhstan)”. Presumably you mean China.
Citation: https://doi.org/10.5194/gmd-2021-423-RC1 -
AC1: 'Comment on gmd-2021-423', Mark Hennen, 07 Mar 2022
Dear Reviewer, Thank you for taking the time to provide comments on our manuscript. We provide below responses to your comments / queries using indented, bullet points with a bold typeface.
In this paper the authors present an approach to verifying dust emission model performance by considering how the model(s) compares to satellite observations of dust emission point sources (DPS). For this purpose they consider dichotomous (presence or absence) comparisons of near-global DPS observations (from MODIS and SEVIRI, for North America, Northern and Southern Africa, the Middle East and Central Asia, and Australia) and simulated emissions from their albedo-based dust emission model (AEM). The model appears to overestimate the number of dust events by an order of magnitude, while the DPS database indicates that dust emission events are actually quite rare in space and time, considering the vast size of some of the deserts. The authors use this dichotomous information to understand the conditions under which the model performance is influenced by the representation of the wind fields, the assumption of infinite sediment supply, and the threshold wind friction velocity u*ts. There is value to this approach to validating dust emission models using observations that represent (as closely as possible) the dust emission events that actually occurred.
- We found your summary of our work to be consistent with our intent. Perhaps just one point of clarity. The frequency of occurrence evident in the dust emission point source (DPS) data is governed by the very large number of pixels where dust emission can occur but also their opportunity to occur in time. Both these conditions for actual dust emission are constrained by the entrainment threshold and the availability of sediment. Therefore, DPS data over space and time record dust emission for comparison with dust emission models.
I am curious about this repeated claim in the manuscript that “The aim here is to demonstrate an alternative to comparing dust emission models to atmospheric dust” (line 79). As I see it, the satellite dust emission point sources are themselves interpretations of observations of atmospheric dust, back-tracking atmospheric dust in the satellite imagery for the purpose of identifying these emission points. Hence it seems to me that the authors are themselves still comparing the dust emission models to atmospheric dust. I suppose there is a distinction here between comparisons with the retrievals of atmospheric dust transport (implicitly or explicitly being criticised), as described by the dust AOD, and with the derivations of dust emission from atmospheric observations taken shortly afterwards (apparently a more robust method).
- Whilst both approaches described here, and in the manuscript, have their origins in optical satellite remote sensing, each approach is measuring a property very different from each other. In the most straight-forward situation an image is used to pin-point the source of dust emission; that is a dust emission occurrence in space and time quantified by coordinates (x, y, z) in latitude, longitude and time (as shown in the figure below from Lee et al., 2009).
- “Fig. 1. A MODIS image (sensor: Aqua) of the region during the dust storm; image obtained from: http://visibleearth.nasa.gov/view_rec.php?id=19043. The image used has a pixel size of 250 m. Political boundaries, cities and points to identify dust sources were added by the authors. The source points were identified on an enlarged version of this image, with greater detail than shown here” (taken from Lee et al., 2009).
- In the example given above, manual inspection of the image permits the observer to identify only the point of origin (white circle) of a visible [atmospheric] dust plume. In contrast, measurements of Dust Optical Depth (DOD) are performed automatically, with limited opportunity to differentiate source of emission from adjacent pixels of transported dust. For that same image, DOD (above a threshold) reduces the information content to presence / absence of dust in the atmosphere, including areas downwind. These dichotomous values include dust emission, but also dust that may be suspended for days (depending on particle size and wind patterns). Consequently, dust emission source represents a diminutive contribution to all observed dust pixels, with the majority of DOD pixels representing transported atmospheric dust.
Now I actually agree that it might be more robust to try to pinpoint the dust emission sources and to compare those with the dust emission model, but I think that it would very much be worthy of clarification and further discussion on the distinction between the two types of satellite observations of atmospheric dust which are being considered here.
- This approach to pinpointing dust emission source has been used and has been evident in the literature for >10 years. With the support of the community working with these data, we have collated most of the published studies. In our manuscript, we used more than two pages to describe these DPS data and how they are retrieved. We then need to describe the innovation within a modelling framework to get the most value out of these data. Considering your perspective, in the revised manuscript we will consider how best in the space available to explain the methodology.
I think it would also be worthwhile to clarify the final sentence of the abstract (“… dust emission models should not be evaluated against atmospheric dust”) to indicate that this relates to the dust AOD, assuming that this is what you mean.
- Thanks for pointing that out. We will clarify that point in the revised manuscript. Note that when we pin-point the dust emission point source it is related to but not a description of dust in the atmosphere.
This points to a general criticism that I have, that while this does seem to me to be a worthwhile paper, describing a useful method of comparing models with dust emission sources identified by satellite sources, I do feel that it is let down at times by over-generalised statements such as the one described in the above paragraph. Another such statement is “These results demonstrate that there is no reasonable basis to calibrate model performance through an adjustment to a fixed global u*ts” (lines 482-484). If this is an implicit criticism of some model dust emission schemes, is this not something of a straw man argument? I would have thought that it has been quite widely accepted for many years that u*ts is dependent on the soil size distributions and roughness lengths, which vary worldwide. I suppose however that there is some value to confirming that this should not be done, as has been done here.
- Most, if not all, dust emission models assume that the entrainment threshold is fixed and static over time. This assumption in dust emission modelling has been evident since dust emission models were developed more than two decades ago. Our statement quoted above is evident from our results and it is a timely reminder that dust emission results and interpretations are not based on robust dust emission modelling but on calibration with atmospheric dust. With this new framework for understanding large scale dust emission patterns and timing we have new insight to tackle these long-standing assumptions.
Specific comments
p.5, lines 156-157: a minor point, but as written this sentence implies to me that the red beam is associated with the 8.7 µm channel, while the green beam is associated with the 12.0 µm channel. It is of course the other way round.
- Thanks, in the revised manuscript we will improve this description.
p.7, Eq.6: presumably in these equations it is (1-As) multiplied by (1-Af). As written, this equation implies to me that no snow = no dust!
- Thanks, in the revised manuscript we will improve this description.
Figure 3: it might make sense here to fit the y-axis to 0-1 (as has been done for the x-axis, 0-0.6), it would make it slightly more intuitive to read.
- Agreed, we will do that in the revised manuscript.
p.13, line 392: “Taklamakan (Kazakhstan)”. Presumably you mean China.
- Thanks, in the revised manuscript we will correct this mistake.
- References used in our responses
Lee et al. (2009) Land use / land cover and point sources of the 15 December 2003 dust storm in southwestern North America. Geomorphology Volume 105, Issues 1–2: 18-27.
Citation: https://doi.org/10.5194/gmd-2021-423-AC1
-
AC1: 'Comment on gmd-2021-423', Mark Hennen, 07 Mar 2022
-
RC2: 'Comment on gmd-2021-423', Anonymous Referee #2, 19 Jun 2022
Review of Hennen et al. Evaluating dust emission model performance using dichotomous
satellite observations of dust emissionThe paper proposes a framework for dust emission model evaluation based on dichotomous (presence=1 or absence=0) observations of dust emission point sources (DPS) derived from satellite data. To show the potential of the framework, a so-called albedo-based dust emission model (AEM) using a smooth entrainment threshold fixed over space and time is evaluated. The paper uses 9 DPS datasets using different sensors and approximations depending on the region. The results are that: 1) dust emission is rare (1.8 %) even in North Africa and the Middle East, which would indicate extreme, large wind speed events, 2) The AEM overestimates the occurrence of dust emission by between 1 and 2 orders of
magnitude. It is then concluded that 1) the smooth entrainment threshold is typically too small and needs to vary over space and time, 2) there is incompatibility between threshold and friction velocity scales, 3) False positives are linked to the absence of any limit in sediment supply in the model and therefore 4) new schemes are needed to the smooth threshold and account for restrictions in sediment supply. Finally, the authors state that the DPS data provide” a consistent, reproducible, and valid framework for the routine evaluation of dust emission models and potential model optimisation, and that the study emphasizes the “growing recognition that dust emission models should not be evaluated against atmospheric dust”.While the overall approach could represent a valuable complement to current evaluation capabilities of dust models, its current implementation and interpretation has, in my opinion, several conceptual flaws and limitations that very likely bias the conclusions of the paper and largely limit the applicability of the approach. In addition, in view of these limitations, statements such as “the study emphasizes the growing recognition that dust emission models should not be evaluated against atmospheric dust” are just not supported by evidence provided in the paper.
Below are my general comments/concerns. Based on them I do not recommend publication of this manuscript as I cannot see how those fundamental limitations and their impact on the conclusions and perspectives of the paper can be easily amended.
On the evaluation of models with observations of dust emission point sources vs atmospheric dust:
Satellites do not observe dust emission but atmospheric dust. Estimates of dust emission point sources are retrieved or inferred from atmospheric observations. In fact, the same applies to in situ measurements: emission cannot be observed directly and can only be inferred from airborne measurements. This is an important conceptual nuance, and it must be clear that the proposed framework relies on a DPS dataset, which infers emission point sources based on many assumptions and potentially important limitations as I will describe below.
Even if the implementation of the proposed evaluation approach would be sound, why dust emission models “should not be evaluated against atmospheric dust”? Why is it incompatible? The statement is just not justified, particularly given the limitations highlighted in my next comment. Evaluation efforts of dust models (with embedded emission and dust cycle) typically include a variety of observations (in situ, satellite, remote sensing) of different variables at different spatial and temporal scales, and all are very welcome and helpful to characterize the behavior of a dust model including its emission. Why not seeing different approaches as complementary? In any case, the statement is just an opinion and is basically not supported by the results of the paper.
On the use of DPS datasets to evaluate dust emission models:
At present there are well known limitations in the retrievals used to infer dust emission that are very likely strongly biasing the comparison with the dust emission model. Take for example SEVIRI Dust RGB product used over North Africa and the Middle East. It is well known that the product can detect the particularly high-concentration dust storms but fails in detecting thin, low level and/or low-medium concentration dust clouds/events, which can come from frequent low emission events that are widespread in North Africa and the Middle East (partly due to the high availability of saltators). This important limitation invalidates to a large extent the proposed dichotomous evaluation approach as dust emission from the model contains all type of dust emission events (us* > u*ts) and dust emission from the DPS is strongly biased towards high emission events, which makes the proposed framework currently inconsistent, and the conclusions likely flawed. For example, the overestimation of the occurrence of dust emission in the AEM by 1 – 2 orders of magnitude and the rarity of dust emission in the DPS for North Africa and the Middle East point towards a problem in this sense.
Another potential problem for the evaluation of global models is the inconsistency of the DPS among regions. Using MODIS in some regions and SEVIRI in others with their different sensitivities could further bias the conclusions on the behavior of a model in different regions. A nice exercise would be to compare a DPS in North Africa and the Middle East based on SEVIRI with another one based on MODIS, and see how the evaluation and the conclusions are impacted.
There are no easy solutions to circumvent the problem of the low-medium dust emission events. Also, this problem clearly evidences that quantitative AOD products (along with their quantified uncertainties) over sources regions can and should at least complement dust emission model evaluation efforts.
In addition to these inherent biases in the DPS and the associated comparison, the difference between simulated wind scale, and the DPS scale makes the interpretation of the results very complex. I acknowledge there is a section in the discussion about this problem but in my opinion the issue should already be considered in the basic design of the evaluation framework. In other words, models should be evaluated as much as possible at consistent spatiotemporal scales, otherwise conclusions can be fundamentally flawed.
All in all, the previous highlighted limitations are very likely biasing the results obtained and the derived conclusions. For example, the overestimation of the frequency of dust emission is partly attributed to the absence of any limit to sediment supply.
Citation: https://doi.org/10.5194/gmd-2021-423-RC2 - AC3: 'Reply on RC2', Adrian Chappell, 24 Jun 2022
Status: closed
-
RC1: 'Comment on gmd-2021-423', Anonymous Referee #1, 16 Feb 2022
In this paper the authors present an approach to verifying dust emission model performance by considering how the model(s) compares to satellite observations of dust emission point sources (DPS). For this purpose they consider dichotomous (presence or absence) comparisons of near-global DPS observations (from MODIS and SEVIRI, for North America, Northern and Southern Africa, the Middle East and Central Asia, and Australia) and simulated emissions from their albedo-based dust emission model (AEM). The model appears to overestimate the number of dust events by an order of magnitude, while the DPS database indicates that dust emission events are actually quite rare in space and time, considering the vast size of some of the deserts. The authors use this dichotomous information to understand the conditions under which the model performance is influenced by the representation of the wind fields, the assumption of infinite sediment supply, and the threshold wind friction velocity u*ts. There is value to this approach to validating dust emission models using observations that represent (as closely as possible) the dust emission events that actually occurred.
I am curious about this repeated claim in the manuscript that “The aim here is to demonstrate an alternative to comparing dust emission models to atmospheric dust” (line 79). As I see it, the satellite dust emission point sources are themselves interpretations of observations of atmospheric dust, back-tracking atmospheric dust in the satellite imagery for the purpose of identifying these emission points. Hence it seems to me that the authors are themselves still comparing the dust emission models to atmospheric dust. I suppose there is a distinction here between comparisons with the retrievals of atmospheric dust transport (implicitly or explicitly being criticised), as described by the dust AOD, and with the derivations of dust emission from atmospheric observations taken shortly afterwards (apparently a more robust method). Now I actually agree that it might be more robust to try to pinpoint the dust emission sources and to compare those with the dust emission model, but I think that it would very much be worthy of clarification and further discussion on the distinction between the two types of satellite observations of atmospheric dust which are being considered here. I think it would also be worthwhile to clarify the final sentence of the abstract (“… dust emission models should not be evaluated against atmospheric dust”) to indicate that this relates to the dust AOD, assuming that this is what you mean.
This points to a general criticism that I have, that while this does seem to me to be a worthwhile paper, describing a useful method of comparing models with dust emission sources identified by satellite sources, I do feel that it is let down at times by over-generalised statements such as the one described in the above paragraph. Another such statement is “These results demonstrate that there is no reasonable basis to calibrate model performance through an adjustment to a fixed global u*ts” (lines 482-484). If this is an implicit criticism of some model dust emission schemes, is this not something of a straw man argument? I would have thought that it has been quite widely accepted for many years that u*ts is dependent on the soil size distributions and roughness lengths, which vary worldwide. I suppose however that there is some value to confirming that this should not be done, as has been done here.
Specific comments
p.5, lines 156-157: a minor point, but as written this sentence implies to me that the red beam is associated with the 8.7 µm channel, while the green beam is associated with the 12.0 µm channel. It is of course the other way round.
p.7, Eq.6: presumably in these equations it is (1-As) multiplied by (1-Af). As written, this equation implies to me that no snow = no dust!
Figure 3: it might make sense here to fit the y-axis to 0-1 (as has been done for the x-axis, 0-0.6), it would make it slightly more intuitive to read.
p.13, line 392: “Taklamakan (Kazakhstan)”. Presumably you mean China.
Citation: https://doi.org/10.5194/gmd-2021-423-RC1 -
AC1: 'Comment on gmd-2021-423', Mark Hennen, 07 Mar 2022
Dear Reviewer, Thank you for taking the time to provide comments on our manuscript. We provide below responses to your comments / queries using indented, bullet points with a bold typeface.
In this paper the authors present an approach to verifying dust emission model performance by considering how the model(s) compares to satellite observations of dust emission point sources (DPS). For this purpose they consider dichotomous (presence or absence) comparisons of near-global DPS observations (from MODIS and SEVIRI, for North America, Northern and Southern Africa, the Middle East and Central Asia, and Australia) and simulated emissions from their albedo-based dust emission model (AEM). The model appears to overestimate the number of dust events by an order of magnitude, while the DPS database indicates that dust emission events are actually quite rare in space and time, considering the vast size of some of the deserts. The authors use this dichotomous information to understand the conditions under which the model performance is influenced by the representation of the wind fields, the assumption of infinite sediment supply, and the threshold wind friction velocity u*ts. There is value to this approach to validating dust emission models using observations that represent (as closely as possible) the dust emission events that actually occurred.
- We found your summary of our work to be consistent with our intent. Perhaps just one point of clarity. The frequency of occurrence evident in the dust emission point source (DPS) data is governed by the very large number of pixels where dust emission can occur but also their opportunity to occur in time. Both these conditions for actual dust emission are constrained by the entrainment threshold and the availability of sediment. Therefore, DPS data over space and time record dust emission for comparison with dust emission models.
I am curious about this repeated claim in the manuscript that “The aim here is to demonstrate an alternative to comparing dust emission models to atmospheric dust” (line 79). As I see it, the satellite dust emission point sources are themselves interpretations of observations of atmospheric dust, back-tracking atmospheric dust in the satellite imagery for the purpose of identifying these emission points. Hence it seems to me that the authors are themselves still comparing the dust emission models to atmospheric dust. I suppose there is a distinction here between comparisons with the retrievals of atmospheric dust transport (implicitly or explicitly being criticised), as described by the dust AOD, and with the derivations of dust emission from atmospheric observations taken shortly afterwards (apparently a more robust method).
- Whilst both approaches described here, and in the manuscript, have their origins in optical satellite remote sensing, each approach is measuring a property very different from each other. In the most straight-forward situation an image is used to pin-point the source of dust emission; that is a dust emission occurrence in space and time quantified by coordinates (x, y, z) in latitude, longitude and time (as shown in the figure below from Lee et al., 2009).
- “Fig. 1. A MODIS image (sensor: Aqua) of the region during the dust storm; image obtained from: http://visibleearth.nasa.gov/view_rec.php?id=19043. The image used has a pixel size of 250 m. Political boundaries, cities and points to identify dust sources were added by the authors. The source points were identified on an enlarged version of this image, with greater detail than shown here” (taken from Lee et al., 2009).
- In the example given above, manual inspection of the image permits the observer to identify only the point of origin (white circle) of a visible [atmospheric] dust plume. In contrast, measurements of Dust Optical Depth (DOD) are performed automatically, with limited opportunity to differentiate source of emission from adjacent pixels of transported dust. For that same image, DOD (above a threshold) reduces the information content to presence / absence of dust in the atmosphere, including areas downwind. These dichotomous values include dust emission, but also dust that may be suspended for days (depending on particle size and wind patterns). Consequently, dust emission source represents a diminutive contribution to all observed dust pixels, with the majority of DOD pixels representing transported atmospheric dust.
Now I actually agree that it might be more robust to try to pinpoint the dust emission sources and to compare those with the dust emission model, but I think that it would very much be worthy of clarification and further discussion on the distinction between the two types of satellite observations of atmospheric dust which are being considered here.
- This approach to pinpointing dust emission source has been used and has been evident in the literature for >10 years. With the support of the community working with these data, we have collated most of the published studies. In our manuscript, we used more than two pages to describe these DPS data and how they are retrieved. We then need to describe the innovation within a modelling framework to get the most value out of these data. Considering your perspective, in the revised manuscript we will consider how best in the space available to explain the methodology.
I think it would also be worthwhile to clarify the final sentence of the abstract (“… dust emission models should not be evaluated against atmospheric dust”) to indicate that this relates to the dust AOD, assuming that this is what you mean.
- Thanks for pointing that out. We will clarify that point in the revised manuscript. Note that when we pin-point the dust emission point source it is related to but not a description of dust in the atmosphere.
This points to a general criticism that I have, that while this does seem to me to be a worthwhile paper, describing a useful method of comparing models with dust emission sources identified by satellite sources, I do feel that it is let down at times by over-generalised statements such as the one described in the above paragraph. Another such statement is “These results demonstrate that there is no reasonable basis to calibrate model performance through an adjustment to a fixed global u*ts” (lines 482-484). If this is an implicit criticism of some model dust emission schemes, is this not something of a straw man argument? I would have thought that it has been quite widely accepted for many years that u*ts is dependent on the soil size distributions and roughness lengths, which vary worldwide. I suppose however that there is some value to confirming that this should not be done, as has been done here.
- Most, if not all, dust emission models assume that the entrainment threshold is fixed and static over time. This assumption in dust emission modelling has been evident since dust emission models were developed more than two decades ago. Our statement quoted above is evident from our results and it is a timely reminder that dust emission results and interpretations are not based on robust dust emission modelling but on calibration with atmospheric dust. With this new framework for understanding large scale dust emission patterns and timing we have new insight to tackle these long-standing assumptions.
Specific comments
p.5, lines 156-157: a minor point, but as written this sentence implies to me that the red beam is associated with the 8.7 µm channel, while the green beam is associated with the 12.0 µm channel. It is of course the other way round.
- Thanks, in the revised manuscript we will improve this description.
p.7, Eq.6: presumably in these equations it is (1-As) multiplied by (1-Af). As written, this equation implies to me that no snow = no dust!
- Thanks, in the revised manuscript we will improve this description.
Figure 3: it might make sense here to fit the y-axis to 0-1 (as has been done for the x-axis, 0-0.6), it would make it slightly more intuitive to read.
- Agreed, we will do that in the revised manuscript.
p.13, line 392: “Taklamakan (Kazakhstan)”. Presumably you mean China.
- Thanks, in the revised manuscript we will correct this mistake.
- References used in our responses
Lee et al. (2009) Land use / land cover and point sources of the 15 December 2003 dust storm in southwestern North America. Geomorphology Volume 105, Issues 1–2: 18-27.
Citation: https://doi.org/10.5194/gmd-2021-423-AC1
-
AC1: 'Comment on gmd-2021-423', Mark Hennen, 07 Mar 2022
-
RC2: 'Comment on gmd-2021-423', Anonymous Referee #2, 19 Jun 2022
Review of Hennen et al. Evaluating dust emission model performance using dichotomous
satellite observations of dust emissionThe paper proposes a framework for dust emission model evaluation based on dichotomous (presence=1 or absence=0) observations of dust emission point sources (DPS) derived from satellite data. To show the potential of the framework, a so-called albedo-based dust emission model (AEM) using a smooth entrainment threshold fixed over space and time is evaluated. The paper uses 9 DPS datasets using different sensors and approximations depending on the region. The results are that: 1) dust emission is rare (1.8 %) even in North Africa and the Middle East, which would indicate extreme, large wind speed events, 2) The AEM overestimates the occurrence of dust emission by between 1 and 2 orders of
magnitude. It is then concluded that 1) the smooth entrainment threshold is typically too small and needs to vary over space and time, 2) there is incompatibility between threshold and friction velocity scales, 3) False positives are linked to the absence of any limit in sediment supply in the model and therefore 4) new schemes are needed to the smooth threshold and account for restrictions in sediment supply. Finally, the authors state that the DPS data provide” a consistent, reproducible, and valid framework for the routine evaluation of dust emission models and potential model optimisation, and that the study emphasizes the “growing recognition that dust emission models should not be evaluated against atmospheric dust”.While the overall approach could represent a valuable complement to current evaluation capabilities of dust models, its current implementation and interpretation has, in my opinion, several conceptual flaws and limitations that very likely bias the conclusions of the paper and largely limit the applicability of the approach. In addition, in view of these limitations, statements such as “the study emphasizes the growing recognition that dust emission models should not be evaluated against atmospheric dust” are just not supported by evidence provided in the paper.
Below are my general comments/concerns. Based on them I do not recommend publication of this manuscript as I cannot see how those fundamental limitations and their impact on the conclusions and perspectives of the paper can be easily amended.
On the evaluation of models with observations of dust emission point sources vs atmospheric dust:
Satellites do not observe dust emission but atmospheric dust. Estimates of dust emission point sources are retrieved or inferred from atmospheric observations. In fact, the same applies to in situ measurements: emission cannot be observed directly and can only be inferred from airborne measurements. This is an important conceptual nuance, and it must be clear that the proposed framework relies on a DPS dataset, which infers emission point sources based on many assumptions and potentially important limitations as I will describe below.
Even if the implementation of the proposed evaluation approach would be sound, why dust emission models “should not be evaluated against atmospheric dust”? Why is it incompatible? The statement is just not justified, particularly given the limitations highlighted in my next comment. Evaluation efforts of dust models (with embedded emission and dust cycle) typically include a variety of observations (in situ, satellite, remote sensing) of different variables at different spatial and temporal scales, and all are very welcome and helpful to characterize the behavior of a dust model including its emission. Why not seeing different approaches as complementary? In any case, the statement is just an opinion and is basically not supported by the results of the paper.
On the use of DPS datasets to evaluate dust emission models:
At present there are well known limitations in the retrievals used to infer dust emission that are very likely strongly biasing the comparison with the dust emission model. Take for example SEVIRI Dust RGB product used over North Africa and the Middle East. It is well known that the product can detect the particularly high-concentration dust storms but fails in detecting thin, low level and/or low-medium concentration dust clouds/events, which can come from frequent low emission events that are widespread in North Africa and the Middle East (partly due to the high availability of saltators). This important limitation invalidates to a large extent the proposed dichotomous evaluation approach as dust emission from the model contains all type of dust emission events (us* > u*ts) and dust emission from the DPS is strongly biased towards high emission events, which makes the proposed framework currently inconsistent, and the conclusions likely flawed. For example, the overestimation of the occurrence of dust emission in the AEM by 1 – 2 orders of magnitude and the rarity of dust emission in the DPS for North Africa and the Middle East point towards a problem in this sense.
Another potential problem for the evaluation of global models is the inconsistency of the DPS among regions. Using MODIS in some regions and SEVIRI in others with their different sensitivities could further bias the conclusions on the behavior of a model in different regions. A nice exercise would be to compare a DPS in North Africa and the Middle East based on SEVIRI with another one based on MODIS, and see how the evaluation and the conclusions are impacted.
There are no easy solutions to circumvent the problem of the low-medium dust emission events. Also, this problem clearly evidences that quantitative AOD products (along with their quantified uncertainties) over sources regions can and should at least complement dust emission model evaluation efforts.
In addition to these inherent biases in the DPS and the associated comparison, the difference between simulated wind scale, and the DPS scale makes the interpretation of the results very complex. I acknowledge there is a section in the discussion about this problem but in my opinion the issue should already be considered in the basic design of the evaluation framework. In other words, models should be evaluated as much as possible at consistent spatiotemporal scales, otherwise conclusions can be fundamentally flawed.
All in all, the previous highlighted limitations are very likely biasing the results obtained and the derived conclusions. For example, the overestimation of the frequency of dust emission is partly attributed to the absence of any limit to sediment supply.
Citation: https://doi.org/10.5194/gmd-2021-423-RC2 - AC3: 'Reply on RC2', Adrian Chappell, 24 Jun 2022