This paper introduces CHANS-SD-YRB v1.0, a system dynamics CHANS model of the Yellow River Basin and operates at provincial (human processes) and sub-basin (natural processes) scales. The manuscript reads well and a supplement with more elaborate model descriptions. By coupling the human and natural processes within a regional CHANS framework, this model description paper lays the foundations for policy-oriented modelling for the Yellow River Basin. Overall, the model is an ambitious synthesis, but several methodological choices and transparency gaps need some tightening. Find my specific comments and suggestions below.

1. Introduction

The first two paragraphs of the introduction goes back and forth between regional and global CHANS research and modelling, which breaks the flow and is a little disorientating. It would be clearer if CHANS research and modelling is first introduced more generally. After introducing the main models, I suggest that the authors identify the overrepresentation of global scale CHANS models. And then bring in the distinction and need for more regional scale modelling, which situates this work within.

L66-69: The sentence on CHANS theories and models applies to both regional and global scale. I suggest moving this above, prior to the delineation between global vs. regional CHANS models for a better flow. 

L70-71: I suggest rephrasing it as “growing regional environmental and societal pressures”

L75-81: The sentence mixes model classes (e.g. IAMs, iESMs) and specific methodologies (e.g., system dynamics), which makes it slightly misleading – since the SD models can also be considered IAMs. I recommend revising this sentence to clearly distinguish model classes from methodologies and examples.

For instance: “Numerous integrated modelling approaches have been developed to simulate human–natural interactions at the global scale. These include system dynamics-based integrated assessment models (IAMs), such as ANEMI (Breach & Simonovic, 2021), FeliX (Rydzak et al., 2013; Ye et al., 2024), and FRIDA (Rajah et al., 2025); process-based and optimization-based IAMs (Vaidyanathan, 2021), which dominate traditional policy scenario analysis; integrated Earth system models (iESMs) (Jain et al., 2022), which couple socio-economic and biophysical components within Earth system frameworks; and Earth system models with synchronously coupled human components, such as E3SM-GCAM (Di Vittorio et al., 2025).”

2. Model description

L162: The citation provided for the sentence is a model description paper of a specific SD model. Given that the sentence is on SD method more generally, I would suggest references to more foundational works (e.g., Forrester, 1968; Richardson, 2011)

1. Forrester, J. W.: Principles of systems, Pegasus Communications, Inc., Waltham, MA, 1968.
2. Richardson, G. P.: Reflections on the foundations of system dynamics: Foundations of System Dynamics, Syst. Dyn. Rev., 27, 219–243, https://doi.org/10.1002/sdr.462, 2011.

2.2.1 Population

In Figure 3, the only direct cross-sectoral feedback is between the Economy and Population, through life expectancy. Total fertility is kept exogenous to the model, yet in most SD models fertility rate tends to be endogenized with a function from GDP per capita – assumption being that fertility rate is negatively correlated with higher literacy levels and access to contraceptives etc. What is the justification for excluding this relationship in the model?

Migration rate is also depicted as an exogenous input to the model. In the model, this appears to be an endogenous variable as a function of GDP difference between national and YSB, which scales the exogenous parameter, migration rate other effects. Also, the migration rate other effects is a time-dependent parameter (lookup table) that increases migration slightly between 2000 and 2005, sharp rise by 2010, and then a decline by 2015. Thereafter, this parameter is held constant. How was this determined? How does this affect future projections of migration? The modelling choices and assumptions here should be clearly documented in the model description.  

L237-238: Births, deaths, and migration are “determined by a series of social and economic factors” is very vague. What and how are these factors conceptualised and represented in the model? Given that life expectancy is endogenous and part of the key inter-sectoral feedback, I would expect more descriptions of how life expectancy is modelled.

2.2.2. Economy

L267-268: Capital elasticities are obtained from T21-China model. Given that they are time-varying, it would useful for readers to understand the assumptions in that model for deriving the shape. The lookup table stops at 2020. Is the parameter value then held constant for future projections? If so, is that a valid assumption for projecting future production?

L264: TFP is an important yet endogenous variable in the model, which should also be documented as an equation in the manuscript given that it includes several effects (infrastructure, education, health). The elasticities for these effects, unlike labour and capital, are constant parameters in the model. How were these elasticities determined? Education (avg years of schooling) is derived from census yearbooks for the historical period, but what is the strategy for future projections? Infrastructure is also an exogenous variable, however, the data and its source does not seem to be documented in the manuscript or the supplement.

L270: Crop and livestock prices are exogenous inputs from China statistical yearbook up to 2020. Are prices then held constant for the future period?

Within the Economy sector of the model, there is internal feedback between GDP and the Capital stocks through private and government savings as a fraction of GDP. I recommend including this link in Figure 3 with a brief description in the manuscript. Currently, in the supplement, the formulations surrounding private and governmental investments are not fully described. There are several lookup tables here (e.g., investment share tables, saving share of income table), which would benefit from some elaboration in the supplement at the very least.

2.2.3. Energy

Most of the endogenous variables in the energy sector are formulated as linear functions of GDP per capita. Though this was simply stated as such in the manuscript and supplement, there is no justification as to why linear relationships were chosen. More concerning, is the use of IF THEN ELSE functions to specify two different sets of constants and slopes before and after 2010 within a few of these relationships. These assumptions are not documented either.

2.2.4. Food

L293-297: Per capita meat demand is derived from a nonlinear regression of GDP per capita and historical per capita production. In the model, there is an IF THEN ELSE function which specifies a different relationship for time 1981. Why is that so?

L303-305: Crop yields are influenced by precipitation, temperature and CO2 concentration from the Climate sector with a reference to the supplement. The supplement, however, does not describe the influence of CO2 concentrations. There is also a mismatch between the model and the supplement creating confusions. In the model, there are two different variables: ‘YRB food production’ and ‘province crop production’ determined by ‘province yield’ and ‘province yield future’ respectively. Precipitation and temperature influences only ‘province yield’ whereas CO2 concentration influences ‘province yield future’. The supplement does not sufficiently describe this formulations for me to understand the conceptualization here. Please clarify.

Food demand formulations (e.g., S34 and S36) rely on exogenous inputs for daily requirements and dietary proportions from the historical period. How is future demand projected beyond the historical period?

2.2.8. Land

The transfer matrix stock is used to calculate the ‘final historical transfer matrix’ in the model. There is a separate ‘future transfer matrix’ variable with multiple subscripts having the value 0 while others simply take the final historical transfer matrix value. This warrants elaboration in the supplement, since it’s not clear how land use changes are determined beyond data replication for the historical period.

3 Model validation and application

I appreciate that the authors have provided the scenario specification in the supplement S5 along with Table 2 that lists some of the data sources and assumptions for future baseline of key variables. I understand that there are over a 100 exogenous variables in the model. However, it is quite disorienting that there is an excel file for time-varying data inputs and several other time-varying inputs hidden within lookup tables (i.e., data points that are interpolated). And, as alluded to in my above comments in the model description, it is unclear how many of these time-varying inputs are formulated for the future baseline (most stop at historical period). I strongly recommend that an additional supplement file is included for ALL exogenous time-varying inputs, including those hidden in table functions. For clarity, the file should list all the historical data points used to drive model behaviour along with a column for data source, followed by all the future baseline data points with another column for data source/assumptions.

Table S2 lists quite a number of exogenous inputs. I would like for a distinction between exogenous variables (time-varying) and parameters (constants). While the above file should contain all the historical and future data points for exogenous variables, another file or table should be provided for all model parameters along with the estimated value and data source (literature or model calibration). Having these files would considerably improve the transparency of the model – especially for readers unfamiliar with Vensim software.

The main text for description of simulation results should explicitly refer to figures to guide the reader.

The authors have emphasized high correlations for historical fits. However I am slightly skeptical since the current approach risks overfitting: several relationships are fitted to historical to the historical period being validated and many exogenous time series drive the dynamics of the model. To build confidence in the behavioural validity, the authors should use out-of-sample validation: model fit to a subset of historical time series (e.g. 1981 - 2005) and use the remains sample points to validate the model outputs.

Sensitivity analyses to quantify parametric uncertainty is also a validation standard in system dynamics. No such analysis appears in the manuscript or supplement. 

L436: What are some plausible explanations for Shanxi exhibiting lower accuracy?

L504: What is the red line? I do not see a red line in Figure 14(a) land area. Why does cropland area “need” to stay above the red line?

4 Discussion 

L635-638: Although the model is presented as a CHANS framework, the Climate sector functions as an exogenous driver rather than a dynamically coupled component. The absence of climate feedbacks limits the scope of coupling. The model cannot represent second-order feedback dynamics such as how afforestation or carbon emissions might alter local hydrology or climate, critical for CHANS theory. Therefore, the authors should explicitly acknowledge this limitation and avoid overstating its scope, especially in L597-597 where they state the “model’s comprehensive coupling broadens its scope of application.” They should also discuss implications of climate feedback omissions for scenario analysis.

The authors have justified using CMIP6 data instead of observed records to ensure temporal consistency, but they do not discuss the potential bias introduced by replacing observed data. In the model, it appears that the authors have considered SSP1.26, 2.45 and 5.85. Showing and discussing the sensitivity of results to alternative climate pathways could help demonstrate robustness.

Sang et al develops a regional system dynamics model to represent coupled human natural systems in the Yellow River Basin, integrating population, economy, energy, food, water, land, carbon, and climate processes to explore historical dynamics and future baseline projections. Overall it looks good, with extensive explaination of the model structure and some illustrative results. However, my main concern is that the human dimension, especially the energy sector, is relatively weakly specified to support the strong claims made about cross sector dynamics in this coupled framework. Because energy use is tightly linked with economic activity, water demand, food production, and carbon emissions, a simplified or poorly validated energy representation risks propagating bias across the entire system. So it’s unclear to me whether the projected interactions among energy, water, food, and carbon are internally consistent or robust, especially in the future projections that are central to the paper’s conclusions.

A few specific comments:

First, A key issue is the system boundary of the energy sector. Although an energy module is included, the model appears to explicitly represent only electricity, while other major energy uses, particularly in industry, are treated in a highly aggregated manner. This is problematic because non electric energy use and structural change in industry are major drivers of both emissions and water demand. In addition, important indirect effects are not clearly represented. For example, electrification trends such as electric vehicle deployment would increase electricity demand while reducing oil consumption, yet such substitution dynamics are not discussed. Similarly, changes in industrial energy structure could feed back to economic output and water use, but it is unclear how these interactions are captured. Given the stated goal of representing coupled human natural systems, it is important for the authors to clarify how these cross sector linkages are treated and how sensitive the results are to the simplified energy boundary.

Second, the description of the energy sector in Section 2.2.3 is unclear. The statement that “coal, oil, and gas consumption are derived from linear relationships between historical sectoral GDP” and corresponding consumption is ambiguous. It is not clear whether this refers to total final energy consumption or only to fuels used for electricity generation. More importantly, a linear extrapolation based on historical GDP trends is unlikely to be appropriate for long term projections under ongoing structural change, efficiency improvements, and decarbonization. The manuscript also does not clearly explain how total energy demand evolves over time or how energy intensity changes are represented. Besides, electricity generation shares are imposed exogenously based on Li et al. 2024, which makes the future pathway highly deterministic. The underlying policy assumptions and narratives are not clearly described. So Figure 16 would benefit from validation against recent historical data before 2025 and comparison with ranges reported in the literature. In particular, the emerging nuclear capacity around 2030 requires stronger justification in terms of feasibility and policy assumptions in China’s context.

Third, the paper emphasizes the regional and spatially explicit nature of the CHANS SD YRB model, but this advantage is not fully demonstrated in the results. Although Figure 1 highlights spatial resolution across provinces and sub basins, many subsequent figures are conceptual diagrams rather than quantitative outputs. Model validation for energy variables ends in 2020, and maybe there are more recent to validate, such as 2021-2024? It would be important to show how near term projections after 2020 compare with observations, particularly for electricity generation and fuel consumption. Also, author may consider presenting and validating provincial level energy results beyond more aggregated models.

Lastly, there are several clarity issues that should be addressed. The meaning of “fire” in the energy sector in Figure 2 is unclear. In addition, the term “per capital” appears repeatedly in the text and figures and should be corrected to “per capita” throughout the manuscript.

This manuscript develops a regional coupled human–natural systems model for the Yellow River basin (CHANS SD-YRB) using a System Dynamics framework that integrates 10 interacting sectors at annual, provincial, and sub-basin scales while explicitly conserving hydrological connectivity among sub-basins. The model reproduces historical dynamics with strong reported agreement and is then used for baseline projections under continued policies and SSP245. The key novelty is the basin-specific, multi-sector CHANS platform that couples socio-economic dynamics with water, sediment, land, carbon, and climate processes in a single connected sub-basin framework, enabling diagnosis of cross-sector feedbacks and trade-offs relevant for long-term sustainability planning. However, some revisions are required before the manuscript can be considered for publication.

Major:

Could the authors clarify the key advantages of this study (and the CHANS SD-YRB model) compared with Sang et al. (2025)?

Figure 4: The model appears not to include a price response to market changes. Under scenarios with carbon pricing, this assumption may miss some impacts of the carbon price. Could the authors discuss this limitation?

Lines 284: Electricity generation shares are treated as exogenous and sourced from the yearbook. Do these shares account for future changes? If not, given that the SSP2 baseline is above 6.0 W/m²  while this study uses SSP245 (which still implies future climate policy), would assuming fixed shares remain reasonable?

Minor:

Line 92: Could the authors please double-check the citation format here?

Line 178: Could the authors clarify how these variables are used as proxies for disaggregation?

Equations (5) and (6): Could the authors explain why the functions are specified in this form?

Lines 467: Could the authors elaborate on why SSP245 aligns more closely with the climate trends in the YRB?