| 01 Sep 2026
NEMO-GSL v1.0: An integrated hydrodynamic and ice-process model for a deep subarctic lake (Great Slave Lake, Canada)
Abstract. NEMO-GSL v1.0 is a three-dimensional hydrodynamic and ice configuration of Great Slave Lake (GSL), the deepest lake in North America and a central component of the Mackenzie River system, built on NEMO v5.0 with the SI3 ice module. The configuration targets conditions absent from ocean and temperate large-lake setups: weak, temperature-controlled stratification, seasonal ice cover exceeding seven months annually, and water levels governed by riverine throughflow. Simulation domain construction required a basin-wide bathymetric product, generated by Universal Kriging of digitized nautical charts, sparse soundings and crowd-sourced depths onto a 1 km grid with 81 vertical levels. Adaptations from NEMO defaults include generic length scale k–ω closure with the Canuto B stability function for weakly stratified freshwater, adjusted SI3 snow conductivity and ice albedo for freshwater ice under prolonged cover, and a modified river routine in which outflow is computed from mean lake level using monthly stage–discharge relationships, with a hydrostatic correction for the ice and snow load applied in winter. The configuration is forced by CaSR v3.2 atmospheric reanalysis and MESH-derived runoff. Evaluation over 1998–2001 against thermistor profiles at six stations, eddy-covariance evaporation, satellite-derived lake surface water temperature (LSWT) and ice phenology yields water-column RMSE of 2.08 °C, LSWT RMSE of 1.68 °C, mixed-layer depth (MLD) MAE of 6.42 m, seasonal evaporation totals within 0.25 % of observations, and ice freeze-up and break-up errors of 1 and 2 days, respectively. Forcing sensitivity experiments reveal wind forcing as the dominant control on simulated mixed layer depth. The bathymetric reconstruction methodology, ice–turbulence parameterization protocol, and river-forcing coupling strategy developed here are independent of GSL-specific inputs and are intended for reuse in other data-scarce, ice-covered lakes. Together, these developments are designed to establish a replicable workflow for subarctic lake modeling, extending NEMO’s application domain beyond the contexts of ocean and temperate large lakes.
This manuscript documents the configuration of the NEMO model for Great Slave Lake, Canada. Great Slave Lake has the interesting feature in that its main body is relatively shallow, while a complex region of islands and channels is extremely deep (by lake standards).
While the authors provide some justification for why they are reporting on configuring a deep ocean model for a medium-large sized lake with typical depths similar to on-shelf (mean depth of 41m has been reported). NEMO seems, a priori, to be the wrong tool for the job, and its use seems to have little a priori justification.
It is especially problematic that the authors present the problem of using NEMO for lakes as a “done deal”. I am aware of at least four (MIT gcm, ELCOM/AEM3D, ROMS/CROCO, FVCOM) different models in the modelling community for Canadian lakes, and my guess is that the application of these models, in terms of published papers, would outnumber those using NEMO by a factor of 10 to 1. Yet the statement made in the manuscript is that Dupont reported on applying NEMO to the Great Lakes, and then the issue is left there. I think there is a lot of rebalancing to do as far as the discussion is concerned.
The references do a nice job on many aspects of the modelling exercise (e.g. in setting why Great Slave Lake is interesting to study, in laying out large Arctic issue), but they fall short when it comes to modelling (to be fair there is discussion of the Umlauf et al papers). This seems very strange for a methods paper on modelling!
The authors state that parameterizations are vital for the sharp thermoclines in likes, but then provide little detail of why NEMO’s suite of parameterizations are the way to go. Why is a k-omega model appropriate? Why is something like the KPP scheme not appropriate when the thermal response is so surface trapped (figures 6 and 7). There are relevant papers discussed (The Umlauf et al papers) and what is around lines 205-210 is good. But I would want to know more on both the choice of model and how sensitive modelling is to the precise values chosen (i.e. the background values of eddy viscosity strike me as so low they are irrelevant; most dissipation will be numerical dissipation.
What is the cutoff for this model as far as horizontal resolution? The 1 km seems like a good start, but it’s not clear to me that all dynamic phenomena that is relevant to modeling is a priori hydrostatic. And what about vertical resolution during the winter when the dynamics is presumably quite sensitive?
I was also left wondering about the overall dynamics of the lake. How does transport within the shallow main portion compare to systematic exchange with the complex, deep arms? Surely the point of a modelling exercise like this one should, at least in part, have something to say about the actual dynamics in the lake (as opposed the statistics of error). I question reporting depth averaged circulation, given that the thermal stratification reported earlier is quite shallow.
What is the effect of using a thermodynamic only ice model? There seems to be a fair amount of detail (reported around line 220) in the model, but does that detail matter if the ice does not move? How sensitive is the model to the parameters?
What is known about the accuracy of the forcing product? Are there weather stations one can ground truth against? I get that you have to use reanalysis, but a bit more information would be useful on how good one expects it to be.
A different category of criticism is the over-reliance on RMSE. Of course RMSE is how one starts to measure error, but surely one cannot just stop there. Take Figure 6 for example; there are clear spatiotemporal differences that, at least to the eye, seem far more important than an overall RMSE value. I think this needs some discussion.
In summary, this is the start of something good. A lot of work was done, and the model will clearly be good for something. But the modelling context needs to be improved, and a subset of the above critiques need to be addressed (the authors can’t do everything, and I assume they will make judicious choices governed by what is doable).
Small points:
The manuscript does an admirable job describing the steps taken to produce a bathymetry, but the result (Fig. 4) is separated from the discussion by a large chunk of the article.
I question the need for an acronym for Great Slave Lake. GSL is generic sounding and the location is unique and deserves to be written out in full.
The wind rose in Fig 14 is pretty small at present. Could you fill the lower right space with the wind rose and put the legend in the space above?
The axis and axis label fonts in general are faint and small. I would increase size and perhaps make them bold.