Improving predictions of land-atmosphere interactions based on a hybrid data assimilation and machine learning method
- 1State Key Laboratory of Earth Surface Processes and Resource Ecology, School of Natural Resources, Faculty of Geographical Science, Beijing Normal University, Beijing, China
- 2School of Environment and Sustainability, University of Saskatchewan, Saskatoon, SK, Canada
- 3National Center for Atmospheric Research, Boulder, CO, USA
- 4Institute of Urban Study, School of Environmental and Geographical Sciences (SEGS), Shanghai Normal University, Shanghai, China
- 5School of Geography and Tourism, Anhui Normal University, Wuhu, China
- 6Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, China
- 1State Key Laboratory of Earth Surface Processes and Resource Ecology, School of Natural Resources, Faculty of Geographical Science, Beijing Normal University, Beijing, China
- 2School of Environment and Sustainability, University of Saskatchewan, Saskatoon, SK, Canada
- 3National Center for Atmospheric Research, Boulder, CO, USA
- 4Institute of Urban Study, School of Environmental and Geographical Sciences (SEGS), Shanghai Normal University, Shanghai, China
- 5School of Geography and Tourism, Anhui Normal University, Wuhu, China
- 6Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, China
Abstract. The energy and moisture exchange between the land surface and atmospheric boundary layer plays a critical role in regional climate simulations. This paper implemented a hybrid data assimilation and machine learning framework (DA-ML method) into the Weather Research and Forecasting (WRF) model to optimize surface soil and vegetation conditions. The hybrid method can integrate remotely sensed leaf area index (LAI), multi-source soil moisture (SM) observations, and land surface models (LSMs) to accurately describe land surface states and fluxes. The performance of the hybrid method on the regional climate was evaluated in the Heihe River Basin (HRB), the second largest endorheic river basin in Northwest China. The findings indicate that the DA-ML method improved the estimation of evapotranspiration (ET) and generated a spatial distribution consistent with the ML-based watershed ET (ETMap). The WRF simulations overestimated (underestimated) the air temperature (specific humidity) in the vegetated areas of the HRB. In contrast, the estimated air temperature and specific humidity from WRF (DA-ML) agree well with the observations, especially in the midstream oasis. The DA-ML framework enhanced oasis-desert interactions by improving the soil and vegetation characteristics. The wetting and cooling effects and wind shield effects of the oasis were enhanced by the DA-ML. The wetting and cooling effect of the oasis can transfer water vapor to the surrounding desert, which benefits the oasis-desert ecosystem. The results show that the wetting and cooling effects only negligibly changed the local precipitation in the midstream oasis. However, upstream of the HRB, the integration of LAI and SM will induce water vapor intensification and promote precipitation, particularly on windward slopes.
Xinlei He et al.
Status: open (until 08 Feb 2023)
-
RC1: 'Comment on hess-2022-379', Anonymous Referee #1, 27 Dec 2022
reply
This study proposed a hybrid data assimilation and machine learning framework to integrate in-situ and remotely sensed-based soil moisture observations and remotely sensed leaf area index (LAI) into the Weather Research and Forecasting (WRF) model. The ensemble Kalman filter (EnKF) approach is used to update the leaf biomass and specific leaf area by assimilating the remotely sensed LAI. A machine learning surrogate model is used to integrate soil moisture profile observations and remote sensing soil moisture product to estimate the three-layer soil moisture. In general, the hybrid framework coupled with the WRF model can improve the simulation of air temperature, specific humidity, wind speed, and precipitation, etc. in the Heihe River basin (HRB). In addition, the hybrid model can highlight the oasis-desert effect and improve the simulation of regional wind speed and precipitation. These results contribute to understanding regional climate and land-atmosphere interactions in the HRB with an advanced WRF model. The entire manuscript meets the scope of this journal. However, several points in the manuscript need to be addressed. So I suggest a minor revision is needed before publication.
Major comments:
- The authors need to emphasize the advantages of the hybrid framework coupled with the WRF model compared to the previous Noah-MP model. This includes the innovative aspects of the study objectives, content, and results.
- Although the advantages of hybrid modeling are obvious, the authors still need to explain why ML methods were constructed to estimate soil moisture instead of directly assimilating SMAP soil moisture. In addition, the uncertainties in the estimation of soil moisture from the hybrid model need to be discussed.
Minor comments:
- How to match the spatial resolution of different datasets to the WRF system, for example, land cover data with a spatial resolution of 30 m, while WRF is set to 3 km.
- The MODIS LAI is the most widely used remote sensing product. Describe why GLASS LAI can be used for assimilation instead of using other products.
- The values of WRF (DA-ML) simulated LAI, 1.12, 1.05, 1.49, and 0.33, are obviously lower than the values drawn in Figure 2, especially at cropland. Another question is that the LAI of WRF (DA-ML) in Figure 2 is a little larger than the LAI of GLASS, not lower.
- If the horizontal coordinate in Figure 3 is Julian Day Number, its starting value should be clearly marked. Furthermore, after 200 days in the midstream, the simulation of soil moisture from the WRF (DA-ML) is hard to capture the observed peak values.
- Line 252: The reliability of ETMap should be described.
- Line 265: In the validation work, air temperature and specific humidity simulations and observed heights need to be listed.
- Figure 6 and 7: The standard deviation of the observations is missing at Hulugou station.
- Figures 9 and 10 show the Mean vertical profile of differences in air temperature and specific humidity between the WRF (DA-ML) and WRF (OL) during the growing season in 2015 in the midstream and downstream oasis. However, the vertical profile locations are unclear even though the rectangle has been marked in Figure 8. And I want to confirm the mean vertical profile of Figures 9 and 10 should be marked as a line or a rectangle area.
- Line 403: “The height is about 650hPa”. Can hPa be converted to m?
- Line 417: “Driven by background northerly winds, more watervapor fluxes from the midstream oasis region were carried to the upstream region”. This conclusion is hardly obtained from the Figure 13.
- The size of the horizontal and vertical coordinates in Figure 13b and c are too small.
-
RC2: 'Comment on hess-2022-379', Anonymous Referee #2, 08 Jan 2023
reply
This study investigates the performance of the advanced WRF model in the HRB based on the coupled data assimilation and machine learning framework. Also, the authors assessed the impact of the hybrid framework on near-surface air conditions and land-atmosphere interactions in this region. The paper is readable and easy to follow. However, the manuscript still needs moderate revisions. Please find my comments below:
(1) Although the description of the manuscript is clear, I am still confused about why the authors did not directly assimilate the Soil Moisture Active Passive (SMAP) soil moisture observations. The advantages of the machine learning-based soil moisture surrogate model need to be enhanced.(2) It is not clear which soil moisture data are used for training and which data are used for validation. Independent soil moisture validation data are required.
(3) Line 114: The observation elements of the automatic weather stations need to be briefly described.
(4) Line 132: ETMap is also uncertain and affected by assumptions. Explain why it can be used as a reference.
(5) Line 195: The structure diagram can provide a clear description of the coupled land-atmosphere framework.
(6) Line 202: The details of the statistical metrics need to be listed.
(7) In Figure 2, the WRF (DA-ML) LAI value is still higher than GLASS products in cropland. Please explain the specific reasons.
(8) Line 235: “The maps of estimated LAI and SM from the DA-ML method consistently resembled the rainfall, vegetation cover, irrigation event, and shallow groundwater table features”. Please explain why.
(9) In Figure 5, the spatial distribution of ET estimates from the WRF (OL) and WRF (DA-ML) is more consistent in the upstream of the HRB. The authors need to explain whether the improvement of the WRF (DA-ML) is related to the original performance of the WRF (OL). Compared to ETMap, ET estimated from the WRF (OL) is underestimated in the downstream oasis, please explain more.
-
RC3: 'Comment on hess-2022-379', Anonymous Referee #3, 08 Jan 2023
reply
This paper takes advantage of the opportunity provided by the abundance of data in the Heihe River basin to illustrate the importance of accurate soil moisture and LAI information for climate modeling in regions with highly heterogeneous land surfaces. The spatial and temporal variations of soil moisture and LAI in the WRF are realistically expressed by data assimilation and machin learning (DA+ML). After assimilating the state variables from observations or satellite remote sensing, both soil moisture content LAI values are increased, which then increases evapotranspriation in the model and futher reduces the air warming bias and dry bias in the simulation. The improved simulation shows more realistic oasis-desert boundary and the wind shield effect within the oasis. Overall, this is an excellent study in terms of capacity building that improves climate modelling through implementing detailed information of land characteristics in a basin with very complex underlying surfaces. Nevertheless, I think this paper can be organized better and some moderate revisions are required..
1. The scientific question to be answered is unclear. If the authors intend to answer a question of general interest, applying satellite data as an input to SM and LAI is understandable because they are globally accessible. However, the application of in situ soil moisture as an input, as done in this study, has no way to expand spatially. If the authors are trying to answer a scientific question specific to the Heihe Basin, the challenges of climate modeling in this basin should be addressed. In either case, it should be stated in the INTRODUCTION in the form of motivation.
2. The structure and presentation of the paper could be imporved. (1) 4.1 should be verification of data assimilation rather than validation of model simulations of LAI and SM. LAI and SM are essentially an input (through ML+DA). Their realistic representaiton in the model verifies the correctness of the implementation in the model, but it does not mean the model's simulation capability. (2) 4.2 For land-air interaction simulations, the key linkage is sensible heat, latent heat (or evapotranspiration), and momentum fluxes, and there is a lack of description of sensible heat and momentum fluxes. (3) 4.3 Specific humidity and air temperature, which has been presented properly. (4) 4.4 Wind speed and precipitation. This part currently lacks quantitative assessment and the results are not convincing. I am not surprised by this, because the improved representations to SM and LAI are local and there are other errors in the model itself. Therefore, it is very difficult to improve wind speed and precipitation quantitatively. I think it is acceptable to consider this part as sensitivity analysis rather than evaluation. If so, this subsection can be much shorten, e.g., deleting Figures 13b-c and Figures 14-15.
3. Suggest to revise the title. The work of this paper is not a prediction but a simulation; land-air interactions are not presented: it shows the response of the atmosphere to the change of the surface state, but does not present the influence of the atmosphere on the land.
4. L329-332: How mountain winds affect the climate in the oasis and how the cooling/wetting affects the air temperature and humidity aloft should be further clarified. Particularly, the height of 600hPa was chosen too arbitrarily. In the oasis and desert region, the influenced height is far lower than 600hPa. Later, the authors explained the phenomenon of warmer and drier aloft through horizontal advection between oasis and desert, but I guess the enhanced subsidence over the oasis in the WRF(DA-ML) is the cause.
5. In section 4.3, about the simulated wind speed difference between the two cases: when you update LAI, do you update the vegetation height (or roughness length and zero-plain displacement) in the WRF?
Minor comments:
In relevant figures, please indicate where is desert and where is oasis; otherwise, it is hard to understand what you are describing.
L179ï¼what is “the standardized soil texture”ï¼
L195-1963the WRF model and DA-ML method were coupled and run dynamically and consistently through the cycles of steps one and two.” What is the time interval of the cycles? This is critical information for applications.
L368: what you mean by "divergent wind direction". I can understand the whole sentence neither.
L390: In the upper atmosphere, air masses enhanced the background northerly winds (orange areas)? It is hard to understand.
L391-395, and some similar sentences: it is not the focus of this work to study the ecological effect, which has been discussed in many early studies. Please delete.
L396-397: You have not established the causality among these components.
L415-418: Figure 12 shows downslope wind, so how could it transfer water vapor from the oasis upslope. There are some similar issues (e.g. L424-425). The authors must be more cautious to draw a conclusion.
L 474: “resembled the rainfall, vegetation cover, irrigation event, and shallow groundwater table features.” Is it the conclusion of this study?
L476: You have not presented “the simulated seasonal cycles of air temperature”. Instead, you only give the seasonal mean!
-
RC4: 'Comment on hess-2022-379', Anonymous Referee #4, 08 Jan 2023
reply
This is a very well-written manuscript, and this reviewer enjoy reading it through. Some major comments as below:
- The wetting/cooling effect of the oasis is interpreted in this manuscript as WRF(DA-ML) - WRF(OL). Based on figure 9, this manuscript emphasizes that this oasis effect is related to irrigation and crop growth in the midstream. However, the wetting/colling effect of oasis is by itself there, no matter if the DA-ML framework is applied or not. As such, this reviewer found that this manuscript is lacking of certain indices to demonstrate physically the wetting/cooling effect of the oasis (for example, one can use the difference in air temperature, and relative humidity between (above) the oasis and the surrounding areas). And then you can check how this indicator will be impacted by DA-ML (e.g. Oasis_Indictor (DA-ML) - Oasis_Indictor (OL))
- The authors state that the wetting/cooling effects of the downstream oasis are due to the shallow groundwater and riparian forest growth. This reviewer can understand that the 'riparian forest growth' can be reflected via the LAI assimilation. However, it is not explicitly clear how the shallow groundwater kicks in here. Are the authors suggesting the assimilation of root zone SM could be used to reflect the effect of shallow groundwater? If that is the case, the author should demonstrate it is indeed the case using the root zone SM, groundwater table measurements, and Noah-MP GW table simulations.
- Although the SM surrogate model development has been published in another paper. This reviewer strongly suggested the author illustrate how these surrogate SM models were constructed with workflow/flowchart etc.
Please see attached some other minor comments.
Xinlei He et al.
Xinlei He et al.
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
381 | 102 | 15 | 498 | 4 | 5 |
- HTML: 381
- PDF: 102
- XML: 15
- Total: 498
- BibTeX: 4
- EndNote: 5
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1