the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 06 Dec 2024
Hyper-Resolution Land Surface Modeling for Farm-Scale Soil Moisture in India: Enhancing Simulations with Soil Vertical Heterogeneity
Abstract. Estimation of field-scale surface and rootzone soil moisture (SM) is crucial for agriculture water management. When ground observations are not available, Land Surface Models (LSMs) aid in reconstructing historical dynamics and providing predictions. However, they often run at coarse resolution (in the order of tens of kilometers), overlook subgrid processes (e.g., lateral flow), and thus underestimating the SM spatial heterogeneity. Considering this limitation, we applied the Noah-MP LSM with the HydroBlocks hyper-resolution modeling framework to estimate surface and rootzone SM at field scale (effective 30 meters resolution) for the first time in India. Recognizing the importance of rootzone processes for agriculture, the present study attempts to improve high-resolution rootzone SM simulations by incorporating vertical heterogeneity in soil properties into HydroBlocks using the SoilGrids global soil database. The analysis is carried out in Upper Bhima Basin (a subbasin of Krishna Basin) for 2020 with ERA5-Land meteorological forcing.
HydroBlocks simulations, configured with vertically homogeneous (VHom) and vertically heterogeneous (VHet) soil properties, were compared against GLEAM, ERA5-Land, SMAP-L3, and SMAP-L4, revealing temporal consistency (correlation between 0.76 and 0.94) and improved sub-grid (up to 0.2 m3m-3) and spatial variability (σθ), in particular VHet (σθ = 0.093 m3m-3) higher than VHom (σθ = 0.09 m3m-3). Both HydroBlocks configurations show reasonable performance against in situ SM observations, with VHet showing systematic improvement compared to VHom by reducing the bias in all sub surface layers and a higher correlation (0.60) than VHom (0.59) at deeper layer (0–60 cm). Finally, we performed a Sobol sensitivity analysis to investigate the seasonal sensitivity of soil on HydroBlocks (VHet) SM simulations for the first five soil layers (up to 1 meter depth). Results revealed that soil parameters interact more prominently in the surface layer and during monsoons. Soil porosity (MAXSMC), Brooks-Corey parameter (BB), and SM at wilting point (WLTSMC) are significant parameters across seasons. Their order of significance changes from surface to deeper layers; however, they remain consistent beyond 30 cm depth. This study finds that the hyper-resolution LSM with vertical soil heterogeneity can enhance small-scale SM simulations by accounting for varying parameter importance, interactions, and seasonal effects within the soil column.
- Preprint
(3898 KB) - Metadata XML
-
Supplement
(862 KB) - BibTeX
- EndNote
Status: open (until 17 Jan 2025)
-
RC1: 'Comment on hess-2024-339', Anonymous Referee #1, 13 Dec 2024
reply
General comments:
This study applies an existing high-resolution land surface model to a watershed in India and assesses its performance. The paper is very well-organized, and for the most part reasonably well-written, and the study is interesting and potentially highly valuable from a social-impact perspective. However, the manuscript has a lot of loose ends and some missed opportunities around (a) convincingly identifying the study’s context and motivation, (b) describing the technical methodology in sufficient detail to be justifiable and replicable, (c) possible quality issues with the physical reasonableness of some of the results, and (d) exploring implications and impacts of the outcomes. As a result, it is a little difficult to ascertain how novel and significant the research is, and in its current form, the manuscript feels like it's essentially a case study. Overall, the paper is promising but would, in my assessment, require significant revisions to be ready for publication in HESS. I hope the detailed comments below will be useful to the authors.
Detailed comments:
(1) The introduction could use some work, especially from a wider water resources community perspective. In particular, a few of the introductory section’s assertions motivating the modeling work described in this paper feel questionable and at times a little like a marketing pitch, yet also it also feels like the introduction underrepresents the larger potential socioeconomic significance of the work. Here are a few examples:
(1a) Lines 53-54, “Land surface Models (LSMs) have advantages over satellite and point scale observations by providing temporally consistent hydrologic estimates over a large extent” – this assertion is largely true, but it bears some additional explanation/justification/citation, perhaps especially re: satellite observations. Please provide some balance by (concisely) listing the pros and cons of each approach (e.g., in-situ measurements are direct observational data but cover only a miniscule point location, models have full spatial coverage but produce estimates based on theories and assumptions, etc) and perhaps also linkages between approaches (e.g., in-situ point-scale measurements are ground truth for remote sensing and models, etc), before presenting a clear rationale for a model-based approach in this specific application, along with a few relevant literature citations.
(1b) Lines 54-56, “Although LSMs are capable of accurate simulation of various land surface processes, traditional models are limited to macro scales (in the order of tens of kilometers), which are primarily intended to run synergistically with climate models” - it seems like there’s a lot to unpack here. The manuscript should provide a precise definition for what it actually means by a land surface model (LSM) and then place the assertion on lines 54-56 in context and ensure that it’s actually correct. For me, the first thing that comes to mind here is WRF-Hydro (the basis for the current generation of the United States NWS National Water Model), which as I understand it, includes the Noah-MP LSM within it and is routinely run over grid cells of less than kilometer. I don’t think WRF-Hydro is completely unique in this regard – I wonder whether things like some European ECMWF systems or the Canadian MESH model could be considered as falling into roughly the same category of high-res de facto LSM. The authors appear to be using the coarse LSMs employed in very large-scale climate studies as their only benchmark. Overall, the broader context of the HydroBlocks LSM used here, and its place within that context, seems a little unclear & requires better, more precise explanation to be meaningful, credible, and relevant to a wide hydrology & Earth system sciences readership.
(1c) Overall, the introductory section pays far less attention than it should to what seems like the most impactful aspect of the study – its potential to inform agricultural decision-making in poor small-plot farms (see also comment 15 below).
(2) Please briefly spell out exactly what’s meant here by subsurface lateral connectivity (line 66). Horizontal vadose-zone flow? Groundwater flow? This is (at least partially) explained later in the paper but a brief mention is warranted here.
(3) Section 2.2 – the authors may wish to make the reader’s job easier by stating up front that the purpose of these five data products is that they’re what the results of this study are being compared against. Also, the sentence on lines 149-151 is missing a verb.
(4) Section 2.3 – What about surface flows, which can be an important source of hydraulic connectivity at the 30 m spatial resolution considered? Is that considered negligible in the study? Perhaps that’s fine, but study assumptions, with justifications and limitations, ought to be briefly spelled out in this methodology section. There’s a lot of subject-matter overlap between this high-res LSM and many other hydrologic modeling approaches, and a wide water-resources readership will have questions about the approach and how it fits in more broadly in the hydrologic modeling literature. (Some key model assumptions and limitations are in fact very tersely listed near the end of the paper in lines 641-644; it might make sense to identify and justify these assumptions in the methodology section.)
(5) Lines 189-190, “HydroBlocks (Chaney et al., 2016, 2021) is a semi-distributed hyper-resolution LSM that clusters areas of hydrologic similarity into Hydrologic Response Units (HRU). The HRUs form the domain’s computing units and enable simulating land surface processes at an effective 30m spatial resolution.” – Again, to be meaningful for a wide hydrologic readership who may not be uniformly familiar with the HydroBlocks model used here, some additional words of explanation might be useful, as there are a few points here that seem a little unclear. The idea of discretizing a watershed into 30 m square grid cells is something one normally associates with a spatially fully distributed model, in contrast to the “semi-distributed” description used here. Do the authors call it semi-distributed because the cells are combined into GRUs? If so, is that consistent with standard terminology in the hydrologic community? Or does “semi-distributed” refer to the fact, appearing later in line 198, that the basin is discretized into 35 sub-watersheds? Or to something else? Overall, it feels like the manuscript’s discussion of how the model deals with basic questions of spatial scale would benefit from more precision, detail, and context, particularly given that the study’s claim to fame is all about spatial resolution.
(6) line 199 – please provide some brief explanation/justification for defining surface soil as 0-5 cm depth and the root zone as 0-100 cm depth. There’s nothing wrong with these choices, I think, as these are common and reasonable definitions – but they’re not the only ones, e.g., the root zone is also commonly defined as extending to 200 cm depth. Just a few words of explanation and a literature citation or two would go a long way here.
(7) line 220 – The acronym IMD has not been previously defined in the manuscript.
(8) lines 233-234 – how is this “upscaling” achieved? By a simple spatial average of HydroBlock results over the grid cell of the spatially coarser product being compared to? Or something else, given that (for example) HydroBlock grid cells may or may not align with the cell boundaries of spatially coarser models and remote sensing products? The manuscript returns to this question in lines 359-361 but remains somewhat vague there as well. In the interest of both manuscript clarity & study reproducibility, please briefly specify simple but important details like this.
(9) line 251 – to “save computational time,” only 1 of the 35 sub-watersheds is considered for sensitivity analysis. No rationale is given for how this watershed was selected. Was it randomly selected, as was the case for the 4 sub-watersheds selected for time-variation analysis (line 230)? Or was it singled out for being representative of the larger basin?
(10) The term “hyper-resolution” is used throughout the manuscript to describe and motivate this study, but the term doesn’t appear to have been specifically defined here. At what pixel size does a model become “hyper-resolution”? And is there a single definition for “hyper-resolution” in hydrology and hydrologic modeling? I think most would say it’s strongly context-dependent. For example, someone doing DHSVM modeling of a small research watershed, or MODFLOW modeling of a contaminated site, might view 30 m grid spacing as not being high-resolution at all, and even WRF-Hydro or other similar hydrologic models explicitly using established LSMs routinely use relatively fine meshes. The 30 m resolution is only “hyper-resolution” when specifically compared to LSMs implementations for climate studies at continental to global spatial scales (see also comment 1(b) above). And is grid cell size the only thing that determines a model’s net spatial resolution, or should other choices also count, like the spatial scale of the GRUs used (see also comment 5 above)?
(11) Are the root zone soil moistures in Section 3.1.2 precisely at 1.0 m depth, or averaged (or otherwise integrated) over the manuscript’s definition of the entire root zone (0-1.0 m depth)? Line 326 simply says “rootzone soil moisture (1 meter deep)” which seems vague. This vagueness appears to continue throughout the remainder of the manuscript.
(12) The HydroBlocks surface and root-zone soil moisture maps of Figures 5(a) and 6(a), compared alongside other products, are a punch line of the study, but there’s something strange here. In the PDF of the manuscript for review, the HydroBlocks modeling results exhibit two completely straight, extremely narrow lines of very low soil moisture, one trending NNE and another trending NE, which cut linearly across multiple sub-watershed boundaries and corresponding topographic divides. Physically, this seems completely unrealistic, and I see no discussion of it in the manuscript. Is this a major HydroBlocks modeling error? Or a visualization and mapping mistake? Or does the figure show something other than what its captions suggest it shows? Or am I missing the point, in which case it’s likely others will as well based on the information and graphics provided in the manuscript?
(13) Section 3.4 has some important material but seems a little tedious. The authors might wish to consider rewriting this section to focus on punch lines and move some of the details to an appendix or supplementary materials.
(14) Lines 627-629, “With improved subgrid heterogeneity of soil moisture simulations (Fig. 5(b) and Fig.6(b)), HydroBlocks, as a hyper-resolution LSM, can cater to the demands of field-scale agricultural applications” – This is a very interesting and valuable result, but the phrasing could be improved. The point of this hyper-resolution LSM is this: what’s sub-grid variability in a coarser model is captured by the hyper-resolution LSM’s finer grid. That is, HydroBlocks doesn’t have “improved subgrid heterogeneity of soil moisture simulations,” it has a higher-resolution grid and therefore improved fine-scale soil moisture simulations.
(15) The most interesting & impactful aspect of this study is, I think, its implications to small-plot farmers in India. It seems the point of all the detailed technical work presented here is to obtain plot-scale soil moisture information to inform agricultural practices among what might be a vulnerable population. However, aside from briefly raising that point as a general motivation at the very start and very end of the paper, the manuscript actually devotes almost no attention to it. I’d be interested in seeing some discussion of (a) whether and how a small-plot farmer in this region would actually use such high-resolution soil moisture information in their agricultural decision-making processes, and/or (b) how the retrospective model implemented here could be recast into a short- to medium-term soil moisture forecasting system. In particular, from an agriculture decision-making perspective, it seems (b) would be more useful than retrospective analyses or nowcasts, given that a small-plot farmer will in general already be acutely aware of current soil moisture conditions on their own farm without the benefit of a modeling exercise, but could potentially benefit tremendously from soil moisture forecasts that would facilitate planning ahead?
(16) It feels like another missed opportunity in this manuscript is that it didn't connect its modeling results to a very large, decades-long body of work around the continuum of spatial scales in hydrologic processes and how those related to modeling goals. Notwithstanding the odd straight lines in Figures 5(a) and 6(a) (see comment 12 above), there’s a really interesting comparison between the fine-scale HydroBlocks simulations in these figures, and the same HydroBlocks simulations upscaled to the resolution of other, coarser-grained soil moisture products in Figures 5(b) and 6(b). The sense one gets of what’s actually happening on the ground is substantially different between these two depictions of the same modeling results presented at differing spatial resolutions. Perhaps the authors could add further value to their modeling exercises by offering deeper comments on this? Blöschl and Sivaplan (1995), "Scale Issues in Hydrological Modelling: A Review," Hydrological Processes (9) 251-290 was a seminal paper on the topic - it would be a good starting point for the authors if they’re unfamiliar with these ideas, which continue to evolve (for a recent example, see for instance https://hepex.org.au/hepex-workshop-2023-forecasting-across-spatial-scales-and-time-horizons/. )
Citation: https://doi.org/10.5194/hess-2024-339-RC1
Data sets
In situ soil moisture observations for validation India Meteorology Department Agromet Division https://dsp.imdpune.gov.in/
SoilGrids 250 m soil data T. Hengl et al. https://soilgrids.org/
ERA5-Land soil moisture data in Copernicus Climate Change Service Climate Data store Joaquín Muñoz-Sabater, Emanuel Dutra, Anna Agustí-Panareda, Clément Albergel, Gabriele Arduini, Gianpaolo Balsamo, Souhail Boussetta, Margarita Choulga, Shaun Harrigan, Hans Hersbach, Brecht Martens, Diego G. Miralles, María Piles, Nemesio J. Rodríguez-Fernández, Ervin Zsoter, Carlo Buontempo, and Jean-Noël Thépaut https://cds.climate.copernicus.eu/datasets/reanalysis-era5-land?tab=overview
SMAP L3 Enhanced version 5 9 km surface soil moisture P. E. O'Neill et al. https://doi.org/10.5067/4DQ54OUIJ9DL
SMAP L4 9 km rootzone soil moisture R. Reichle et al. https://doi.org/10.5067/60HB8VIP2T8W
Global Land Evaporation Amsterdam Model (GLEAM) soil moisture Brecht Martens, Diego G. Miralles, Hans Lievens, Robin van der Schalie, Richard A. M. de Jeu, Diego Fernández-Prieto, Hylke E. Beck, Wouter A. Dorigo, and Niko E. C. Verhoest http://www.gleam.eu
Model code and software
HydroBlocks Noemi Vergopolan, Nathaniel W. Chaney, Peter Metcalfe, and Eric F. Wood https://github.com/chaneyn/HydroBlocks/tree/dev_noemi
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 105 | 21 | 2 | 128 | 32 | 0 | 1 |
- HTML: 105
- PDF: 21
- XML: 2
- Total: 128
- Supplement: 32
- BibTeX: 0
- EndNote: 1
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1