General comments

This paper presents the newly developed auto-calibration tool, EFDC-ACT, a R-package for auto calibrating the 3D hydrodynamic and water quality model, EFDC. The content and topic of this paper fits GMD’s scope. It presents a modelling tool that advances the 3D model hydrodynamic and water quality calibration process. The tool is original and novel, the data is reliable and transparent. The results of the model calibration with the new tool are shown to be satisfactory. The manuscript has clear structure, background, methods, and results were clearly presented, and discussion is constructive, with fluent and precise language.

However, the paper lacks little consistence, starting introduction and methods with heavy description of 3D Yuqiao Reservoir model, and its chronic development over the years, and equally heavy description of the data sources regarding hydrological years (dry, wet and normal years), which diverted the focus of the paper. While the result section and discussion section are focused on the EFDC-ACT performance, which is very good.

To fit GMD’s Development and technical paper scope, this study should focus mainly on the statistical and auto-calibration technics applied in 3D-hydrodyanmic model calibration practices and the novelty of this tool regarding this practices. The YRWQM (Yuqiao Reservior Water Quality Model) is a practice on EFDC model (a 3D hydrodynamic and water quality model), and the scientific questions of different hydrological years, and why it’s causing bias in calibrating WQ parameters such as chla, are good argument and background for a better results of re-calibration, and therefore support the evidence the newly developed tool is an advance, but should not be main focus.

Sectional comments: 

My proposed revision of paper is more sectional.

1. Introduction: the introduction part contained the main points of motivations of this study: difficulties in calibrating complex 3D hydrodynamic models (starting from line 35), and short data calibration creating equifinality problems in models. It also covered available auto calibration practices (PEST-EFDC, and PEST-Delft3D). The introduction should continue the path of analysis the current auto calibration methods and what’s lacking in current practices, fx. Not so fit for EFDC calibration. A closer look at EFDC’s parameters and calibration challenges will bring the focus near the initiation of EFDC-ACT development. I miss the overview of the EDDC model structure and its parameter sets, and specific challenges of calibrating such 3D complex model.

1. Materials and Methods: the most important part of this sections is 2.4, where the material of this study is the calibrated YRWQM from previous study, no need to present the study sites and data sources in detail. Methods also lacks part of benchmarking process, i.e. the former calibration of 2006,2007, which corresponding to the dry years in the re-calibrated period, should be a benchmarking point to compare the model calibrating performance. And the comparison of these two (the former YRWQM and the recalibrated YRWQM) model performances should not only include the calibration period, but also the validation period (the period where model parameters are not adjusted to data, to validate model’s ability of presenting the reality).
2. Results: the current structure is very clear and sound, and I like esp. figure 4, a more clear way to present model performance. However, contrary to the conventional model application studies, the parameter values for calibrated model and performance of the calibration should not be the focus. This paper is stressing the advances of EFCD-ACT, and comparisons of calibration process (time consumption, models sets estimation etc. ) and calibration performance (R2, RMSE, NSE, KGE, etc.), between the former YRWQM, and current YRWQM, should be addressed. The improvements and challenges regarding the new tool during application should be presented (section 3.2 and 3.3 should be presented in detail in my opinion).
3. Discussion: excellent. This part is the best regarding this paper’s purpose: presenting EFDC-ACT, and its technical and practices and performance. And its contribution to 3D model application practices.
4. Data and source code availability: EFDC-ACT src code is most available. However, observed data only contains the observation of the reservoir’s inflow, outflow, temperature, WSE. However, no data on water quality and chla, which is presented in the calibration. In order to reproduce the study, meteo forcing and other input data for EFDC model are also needed. Data on data.cma.cn is a mess to dig out the data for this study. If there’s concern of sharing these datasets or configure files of EFCD, should the authors address the reasons.

In general, I think this paper deserves publication, with its excellent work and unique study area. The EFDC-ACT tool deserves to get more attention in EFCD users. The study contained all the necessary components for a better presentation of EFDC-ACT, but should have some proper adjustment to fit GMD’s Development and technical paper scope, also for the purpose of showing the strength of this new tool.

Best of luck!

Review of “Recalibration of a three-dimensional water quality model with a newly developed autocalibration toolkit (EFDC-ACT v1.0.0): does the model calibrated with a wider hydrological variability become more robust?”

The authors presented a recalibration effort to improve model performance and discussed the impacts of using wider hydrological variability on the robustness of model performance. The authors also provided a R based optimization tool to automate the calibration of the EFDC model. This tool will improve the efficiencies in calibration of sophisticated models and is valuable to the scientific society. At the same time, the write up of the manuscript needs some improvements. Therefore, my recommendation is revision.

Below is a list of my detailed comments:

• More data and wider hydrological variability will improve the robustness of model performance is well-accepted. Logically, a model calibrated with wider hydrological variability is better. My recommendation is to focus on the quantitative improvement in this manuscript instead of asking the qualitative question. The title should be changed to something like “how much improvement will be achieved with a wider hydrological variability”.
• The correct way of conducting model calibration using different conditions (such as dry, normal, wet years) when long-term data are not available is to use one set of data to identify the parameter values, then using the identified parameter values to run other conditions. If the results in other conditions are good, we achieved a good model calibration. Otherwise, we go back to the first condition and identify another set of parameter value, and then again put them to other conditions. This process is iterative until model results from all conditions achieve good results. This manuscript lists totally different parameter values for dry, normal, wet years/ This needs to be fixed.
• The manual calibration process is indeed a time-consuming process. However, it is a necessary process for the modeler to learn if the developed model is a good representation of the real waterbody. Furthermore, modelers inexperienced or unfamiliar with the model should not be the ones to conduct such manual calibration. It should be noted that modelers inexperienced or unfamiliar with the model could falsely use the auto-calibration tool so that the identified parameter values are not in reasonable range and perfectly fall into the problem of obtaining good model error statistics values while using wrong parameters. Auto-calibration should be viewed as an efficient way to refine calibration after using learning the system with the manual calibration.
• The authors cited two modeling studies. One was used data from a decade observations and the other used data from two years. ”Cerco and Noel (2005) recalibrated the Chesapeake Bay model with a decade of observations resulting in a clear improvement in modeling primary production and light attenuation. Lung and Nice (2007) recalibrated the Patuxent Estuary model with two years of data and adequately modeled Chl a and DO concentrations.” The second example is not using long-term data.
• The authors mentioned “ The seasonal cycle of water temperatures was pronounced, with higher temperatures in summer and lower temperatures in winter.”. This is not sufficient to describe a successful model development. Any temperature simulation will generate such seasonal trend given the right weather conditions and inflow temperatures. It is suggested that details should be added such as if highest temperature and lowest temperature are captured by the model.
• The authors mentioned “The 11 primary parameters in the water quality model were listed in Table 1, eight of which governed algal 220 kinetics (CChl, PM, Keb, TM1,TM2, KHP, BMR, PRR), two parameters influenced phosphorus cycling (KRP, KDP), and one affected reaeration (REAC). Among these parameters we also found six of them to be sensitive (Table 1).” It is suggested that the process of choosing the 11 primary parameters should be provided in the manuscript.
• Figure 3 shows the model results after recalibration. The model results from the original calibration should be shown together so that readers can visually inspect the improvement of calibration.
• The values shown in Table 1 look like the recalibration replaced the algae from the spring growth algae to summer growth algae with much higher metabolism rates. Is there any algae study to support such change of algae type?