the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 08 Jan 2025
Improving JULES Soil Moisture Estimates through 4D-En-Var Hybrid Assimilation of COSMOS-UK Soil Moisture Observations
Abstract. Accurate soil moisture estimates are essential for effective management and operational planning in various applications, including flood and drought response. However, soil moisture values derived from land surface models often exhibit significant deviations from in-situ observations. Data assimilation combines model information with observations to enhance prediction accuracy. Previous studies have typically focused on either estimating the initial soil moisture state or optimizing Pedotransfer Function (PTF) constants, which link soil texture to the hydraulic properties of the land surface models. In contrast, in this study, we employ a novel approach by performing joint state-parameter assimilation for the JULES model. We optimized both the PTF constants and the initial soil moisture conditions simultaneously. Using Four-Dimensional Ensemble Variational hybrid data assimilation, we ingested field-scale soil moisture observations from the Cosmic-ray Soil Moisture Monitoring Network across 16 diverse sites in the UK. The results demonstrate that joint state-parameter assimilation significantly enhances the accuracy of soil moisture estimates, improving the average Kling Gupta Efficiency values from 0.33 to 0.72 across different soil characteristics. These findings indicate that our proposed joint state-parameter assimilation framework holds great potential for enhanced predictive accuracy in land surface models.
- Preprint
(7539 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 30 Apr 2025)
-
RC1: 'Comment on egusphere-2024-3980', Anonymous Referee #1, 04 Mar 2025
reply
General comments
The topic of the manuscript is important and of actual relevance. It deals with the assimilation of soil moisture values derived from cosmic-ray neutron sensing into a land surface model, and addresses improvements resulting from adapting the initial soil moisture state or adapting parameters within a set of pedotransfer functions or both. The manuscript is well-structured and well-written but in parts repetitive and vague in its description of its substance and potential advances. It lacks a reasonable consideration of its methodology in a way that gives full and clear picture to what has already been developed and published, and what is a new development in this study, accompanied by shortcomings in presenting and discussing results of the current study in respect to previous similar or possibly almost identical results. Though the topic itself offers potential to deeper insight and methodological advances, this potential has not truly been exploited. Some more details in the following.
Specific comments
- A big part of the methodology and the monitoring data set have been presented already in Cooper et al., 2021a (https://doi.org/10.5194/hess-25-2445-2021). The second scenario of the manuscript seems to be very similar to this work published by the manuscript’s second author as main author. It is using the same land surface model JULES, the same pedotransfer functions including the same starting parameter values, the same 16 cosmic-ray neutron locations out of the about 50 sites of the COSMOS-UK network, the same year for the data assimilation and the same following year for the forecast, and the same ensemble size. And the 4D-En-Var assimilation method outlined in the manuscript is depicted already in Pinnington et al. (2020), also including a combination of parameter vector and state vector, and seems to be part of LAVENDAR already together with JULES.
- While the study presented in the manuscript does go somewhat further by now also applying the data assimilation method to include the initial soil moisture state, it does not describe the methods and results of the manuscript on basis of the existing work but as an unclear mix with vague formulations of origin. These publications are cited but often rather as context. A clear distinction between existing work and own work is necessary. Maybe there are reasons to use exactly the same setting as before, though using another data set and approach could contribute more novelty, but this needs to be discussed and rectified. Also results should be discussed on basis of existing work, not in a diffuse way. To give one simple example, in the conclusions, second paragraph, the manuscript presents that KGE has improved from 0.33 to 0.66 for the parameter-only assimilation, which is identical to the statement in the conclusion of Cooper et al. (2021a) that ‘we see an improvement in the average KGE metric from 0.33 (range 0.10 to 0.69) before data assimilation to an average of 0.66 after data assimilation’. It should be referred to the previous result and made clear that this finding is identical to this previous result and why or to be discussed how it nevertheless may be different and why.
- Features of cosmic-ray neutron sensing are partly wrong and even contradictory within the manuscript. For example, in line 30 the horizontal footprint size is specified as ‘approximately 25 to 30 hectares’, in line 112 then ‘an area up to 120,000 m2’, which is only half of the former. Also, the depth specifics are not explained adequately and the fourth, deepest layer of the JULES model actually is not linked to the soil moisture observation by cosmic-ray neutron sensing at all, and the third layer likely only sometimes. And, the equations used for a weighted average of model layer soil moisture values to compare to the observed soil moisture (Cooper et al., 2021a, Pinnington et al., 2021) is a mere average accounting for the different layer thicknesses but not accounting for the strongly decaying weight with depth of cosmic-ray neutron sensing. Furthermore, the error estimate for the observations does not account for the relation between hourly values and a daily value for this cumulative measurement nor the Poisson distribution of its uncertainty instead of a Gaussian.
- In respect to the definition of the Cosby’s pedotransfer functions, there are also shortcomings. The manuscript refers to Cosby et al. (1984), but not everything presented is given there and seems neither developed within the manuscript’s study. Cosby et al. (1984) has reported linear relations between grain fractions and four hydraulic variables, but not the full mathematical equations as presented in 2.3. Therefore, a part of the earlier development seems to be missing. Marthews et al. (2014) could be cited directly in this respect, as one component. But further considerations would be helpful. And some discussion, why this set of pedotransfer functions? Only because they have been used in the similar preceding study (Cooper et al., 2021a)? And why not start with the parameters adjusted there already? How does it compare to other pedotransfer functions as used in other studies, etc.
Technical corrections, just some examples
- The title is full of abbreviations and unclear
- The introduction to monitoring of soil moisture starts with a general list of remote sensing sensors (and references) and rather outdated observation networks reported in 2006 and 2007. This part could be more to the point and up to date.
- In line 56 a reference is needed, as such and also for the claim to be more accurate.
- Line 175 It is a bit uncalled-for to first give p a time index and then declare that it is constant in time.
- Line 245 to 253 Replace Ne by the value chosen here (50), as mentioned anyway several times and as the other particular parameters are also specified as values.
- The references contain a large number of malformed doi links.
References here are named as cited in the manuscript.
Citation: https://doi.org/10.5194/egusphere-2024-3980-RC1
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 142 | 33 | 5 | 180 | 6 | 12 |
- HTML: 142
- PDF: 33
- XML: 5
- Total: 180
- BibTeX: 6
- EndNote: 12
Viewed (geographical distribution)
| Country | # | Views | % |
|---|---|---|---|
| United States of America | 1 | 52 | 32 |
| United Kingdom | 2 | 43 | 27 |
| India | 3 | 9 | 5 |
| China | 4 | 8 | 5 |
| Belgium | 5 | 5 | 3 |
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
- 52