the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 19 Jul 2024
PyGLDA: a fine-scale Python-based Global Land Data Assimilation system for integrating satellite gravity data into hydrological models
Abstract. Data Assimilation (DA) of time-variable satellite gravity observations, e.g., from the Gravity Recovery and Climate Experiment (GRACE), GRACE-Follow On (GRACE-FO) and future gravity missions, can be applied to constrain the vertical sum of water storage simulations of Global Hydrological Models (GHMs). However, the state-of-the-art DA of these measured Terrestrial Water Storage (TWS) changes into models is often performed regionally, and if globally, at low spatial resolution. This choice is made to handle the considerably high computational demands of DA, and to avoid numerical problems, e.g., instabilities related to the inversion of covariance matrices. To fully exploit the potential of satellite gravity observations and the high spatial resolution of GHMs, we developed a Python-based open-source PyGLDA system that allows performing DA globally at a fine scale with high numerical efficiency. The main novelties of the system include (i) implementing a globe-scale patch-wise DA via domain localization and neighbouring-weighted global aggregation and (2) its great compatibility between basin-scale and grid-scale DAs. This PyGLDA system represents a considerable functional advancement on previous implementations with wide and flexible options offered to allow for various user-specific studies. The modular structure of PyGLDA provides users with various possibilities to interact with (and add/remove) individual water storage compartments, change the representation of observations, and, therefore, the ability to choose different GHMs. In this paper, we present a full description of this system and its application for the Danube River Basin as a regional case study and through a global DA. The DA demonstrations are performed using the monthly TWS fields of GRACE (2002–2010) and the W3RA water balance model at 0.1-degree/daily spatial-temporal resolution.
- Preprint
(4819 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on gmd-2024-125', Anonymous Referee #1, 25 Jul 2024
General comments:
This study presents an open-source global data assimilation (DA) system for integrating GRACE/GRACE-FO total water storage into Global Hydrological Models. The study achieves a previously unseen high spatial resolution, with the potential to increase the use of GRACE/GRACE-FO data in the hydrological modelling community. The study is well within the scope of GMD and a relevant contribution, particularly as it provides a reproducible and transparent approach. The authors have done a thorough effort to test that their implementation can be used at the targeted levels and with consistency.
The paper could benefit from a proof-reading and more importantly some clarifications – quite a few terms are treated as self-explanatory which they are not for non-GRACE experts, who might still benefit from using the tools and models developed and kindly made available by the authors.
While the data processing and DA setups are quite thorough, the GHM used is still quite fundamental to the DA and clearer details on its limitations and structure could be beneficial both to understanding some of the results, but also to identify where additional experiments and developments are warranted. I think the paper would gain in strength and impact if it were to include some model validation and analysis of the hydrological consistency.
Specific comments:
l. 28-30 I would revise how you classify the models – most of these are rainfall-runoff and conceptual which is also the case of a water balance model. In general we talk more of conceptual versus physically based and whether a model is lumped, distributed or semi-distributed (e.g. SWAT with Hydrological Response Units). The question can also be how many processes are lumped/conceptualized and how simple the model structure is.
l. 71 spelling of project name
2.2 Which up- or down-scaling routine are you using?
2.3 How are the parameters down-scaled? Is there any risk of non-linearities at the boundaries between pixels or basins?
2.4 Please explain more clearly what ITSG-2018 is and who is developing it? In 2.4 you only present the three official data centers, and it would be good with a bit more background information on ITSG-2018 and why you prefer an unofficial provider (and whether that is available to any user). Why is a better quality expected with ITSG-2018?
Figure 1. Several of the subparts are difficult to read even when zooming in on a screen (figure in b, boxes in the ensemble output) – I recommend that you either remove the text or increase the font size.
Would refining the grid size not increase your susceptibility to lateral redistribution? Does W3RA include any kind of routing component?
l.186 spelling of Gaussian
l. 189 verb missing in sentence
Table 1 – please give more details on the corrections which use acronyms or simple symbols, as well as what each correction does.
l. 206 Please clarify – do you mean that if the required number of ensembles is run, each individual ensemble is independent of each other or is that the case no matter the number of ensembles?
l. 214 What would be a recommended number of ensembles?
Figure 3 You are assuming no inter-correlation within the tiles but there is clearly hydrological correlation between several neighboring tiles crossing large river basins – can you clarify how that is not an issue?
3.3.2 How do you determine the width of the transition zone? Are there large differences where the average might not make sense? Have you tested whether your assumption of reasonability of the inverse-distance weight is correct?
l 325 This we would expect – the GW has much longer memory than the surface components.
l. 334 Please clarify “the phase shift has been compromised” – where is it happening in the model?
l. 377 Is the water balance also maintained?
l. 439 Have you considered using GRDC discharge data for validation?
Citation: https://doi.org/10.5194/gmd-2024-125-RC1 -
RC2: 'Comment on gmd-2024-125', Anonymous Referee #2, 20 Jan 2025
The manuscripts describes a Python-based Global Land Data Assimilation
(PyGLDA) software for integrating the estimates of Terrestrial Water Storage
(TWS) variations from satellite gravimetry data into a Global Hydrological Model
(GHM). The former data are provided by GRACE and GRACE Follow-on (GFO) satellite
missions. As far as the hydrological model is concerned, the authors have
adapted W3RA model (line 84). Technically, the Ensemble Kalman filter
(EnKF) and an Ensemble Kalman Smoother (EnKS) can be used as
data assimilation tools (line 90).In general, the manuscript provides a sufficiently detailed description
of the developed software. Nevertheless, there are a number of points
which remain unclear. In particular, some statements in the abstract
seem to be inconsistent with the main text. This makes an impression that
the actual status of the developed software is much less advanced
than the abstract claims.I would like to highlight the following uncertain points:
1. In the abstract, the authors mention that "choice is made ...
to avoid numerical problems, e.g., instabilities related to the
inversion of covariance matrices" (lines 5-6). At the same time,
they admit at the end of the manuscript that "Potential numerical
instability introduced by the inverse of the covariance matrix ...
has not been found yet so far in our study. Therefore, in the current
version, none of the covariance localization techniques has been integrated
into PyGLDA, and consequently, users may face the potential risk of instability
when the gridded DA is configured to run at a smaller scale" (lines 466-470).
These two statements seem to contradict to each other.2. The abstract claims that W3RA hydrological model
is used just to demonstrate the performance of the developed data
assimilation software (lines 15-16). At the same time, the manuscript
itself does not go beyond W3RA. Moreover, the concluding section mentions
that "PyGLDA is being developed to allow for more candidate GHMs, other
than W3RA" (lines 454-455). Thus, it remain unclear to what extent
the developed software in the present form is suited to assimilate
GRACE/GFO data into other hydrological models.3. The authors mention that the adopted hydrological model W3RA
ignores lateral water redistribution between grid cells
(lines 174-175). This sounds odd. I cannot imagine a
hydrological model that ignores streams and rivers.
Perhaps, the authors mean the absence of underground
water redistribution?4. In the description of data assimilation procedure
(Sect. 3.2 and further; also, Fig. 2), the authors seem to use the
term "ensemble" instead of the term "ensemble member".
This makes the description not very clear.5. At many places in the main text, the authors address correlations
(or covariances), but do not explain whether they mean signal
correlations or error correlations.6. The authors mention that they use a multiplicative error
to disturb, among other, forcing fields (e.g., lines 312-314).
Furthermore, they mention that the forcing fields
include the daily precipitation, the maximal/minimal temperature
and the surface solar radiation (lines 123-124). On the other hand,
appendix A admits that the only disturbed forcing field is
precipitation: "In PyGLDA, we consider multiplicative errors
for precipitation, thus q = 1" (lines 499-500). This is inconsistent.7. The statement cited in the previous item implies that no disturbances
are applied when the nominal precipitation is zero and no disturbances
are applied at all to the other forcing fields. This seems to be sub-optimal,
since it may result in an insufficient variability of forcing fields
(and, therefore, of the state vectors) among the ensemble members.In addition, the manuscript suffers from numerous textual deficiencies.
Among others, there are many very long sentences,
which definitely should be split into 2-3 shorter ones.
Occasionally, the manuscript makes an impression that
the authors have never taken time to read the text they wrote.
In the enclosed annotated manuscript, I left a number of
specific suggestions concerning an improvement of the text.
Nevertheless, I encourage the authors also to carefully revise
the manuscript themselves.I recommend to publish the manuscript provided that the authors
update it in line with my remarks.
Data sets
The auxiliary dataset to drive our global data assimilation system Fan Yang https://doi.org/10.5281/zenodo.12206756
Model code and software
Source code for our data assimilation system Fan Yang https://github.com/AAUGeodesyGroup/PyGLDA
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 523 | 177 | 15 | 715 | 26 | 13 |
- HTML: 523
- PDF: 177
- XML: 15
- Total: 715
- BibTeX: 26
- EndNote: 13
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1