Coupled Simulation of the Water–Food–Energy–Ecology System Under Extreme Drought Events: A Case Study of Beijing–Tianjin–Hebei, China

Abstract

The Beijing–Tianjin–Hebei (BTH) region is one of China’s most water-scarce yet economically vital areas, facing increasing challenges due to climate change and intensive human activities. This study develops an integrated Water–Food–Energy–Ecology (WFEE) simulation and regulation model to assess the system’s stability under coordinated development scenarios and extreme climate stress. A 500-year precipitation series was reconstructed using historical drought and flood records combined with wavelet analysis and machine learning models (Random Forest and Support Vector Regression). Results show that during the reconstructed historical megadrought (1633–1647), with average precipitation anomalies reaching −20% to −27%, leading to a regional water shortage rate of 16.9%, food self-sufficiency as low as 44.7%, and a critical reduction in ecological river discharge. Under future recommended scenario with enhanced water conservation, reclaimed water reuse, and expanded inter-basin transfers, the region could maintain a water shortage rate of 2.6%, achieve 69.3% food self-sufficiency, and support ecological water demand. However, long-term water resource degradation could still reduce food self-sufficiency to 62.9% and ecological outflows by 20%. The findings provide insights into adaptive water management, highlight the vulnerability of highly coupled systems to prolonged droughts, and support regional policy decisions on resilience-oriented water infrastructure planning.

1. Introduction

Water, food, and energy are three fundamental material bases and strategic resources essential for the survival and development of human society [1]. In recent years, intensified global climate variability, rapid urbanization, and improvements in living standards have significantly increased the demand for these resources across nations. Simultaneously, challenges such as soil erosion, excessive agricultural expansion, vegetation degradation, and water resource depletion have led to the progressive deterioration of aquatic ecosystems at the watershed scale [2,3,4]. Additionally, precipitation regime changes, stronger coastal rainfall–wind dependence, and rise of extreme weather events such as atmospheric rivers further intensify hydrological uncertainties and regional vulnerability [5,6,7]. Food and energy production, as well as ecological protection, increasingly compete for limited water resources. The interactions among water, food, energy, and ecological systems are inherently complex, as the advancement of one subsystem often occurs at the expense of others. Collectively, these four interdependent sectors constitute the Water–Food–Energy–Ecology (WFEE) nexus.

Within regional WFEE systems, the escalating demands from individual subsystems are expected to intensify due to continued socioeconomic development. Understanding the mutual constraints and interactions among these subsystems from a nexus perspective has become an urgent and intricate challenge—particularly in densely populated or ecologically fragile regions. As such, developing sustainable management strategies for WFEE systems has gained increasing global attention. For example, in California’s Central Valley, extensive groundwater abstraction for irrigation has caused water table declines exceeding 100 m in some areas over the past five decades, triggering severe ecological consequences. Furthermore, the energy required for groundwater pumping in the region has risen by approximately 15% over the past 20 years due to declining water tables [8,9]. Similarly, China, with its large population and rapid economic growth, faces mounting pressure to balance water allocation, food and energy production, and long-term ecological security [10,11].

Numerous studies have examined the WFEE nexus in China using diverse analytical frameworks. For instance, Wang et al. [12] applied a coupling coordination degree model and spatial analysis to explore national-level water–energy–food interactions, revealing that the composite performance of the energy and food subsystems generally surpasses that of the water subsystem. Li et al. [13] proposed a tri-dimensional evaluation framework—stability, coordination, and sustainability—for the WFEE system in Liaoning Province and assessed symbiotic security risks using Copula functions, showing overall improvements in risk coordination. Zhang et al. [14] used a modified SWAT model to simulate groundwater level changes in the Hai River Basin and assess the impacts of seasonal fallow policies on regional water–food–energy linkages. Li et al. [15] developed a system dynamics model of the Beijing WFEE system to project future peak demands, emphasizing the resource alleviation effects of Xiong’an New Area development. Sun et al. [10] introduced a competition–synergy evolutionary model to characterize WFEE interactions across China, revealing that water–food relationships are mostly competitive nationwide, while energy–food interactions are synergistic in the east and competitive in the west. Wang et al. [16] employed a Bayesian network to evaluate WFEE supply–demand risks in the Beijing–Tianjin–Hebei region, identifying water and food as critical sources of vulnerability. Wen et al. [17] modeled the WFEE nexus in Daqing using system dynamics and demonstrated that industrial restructuring and resource conservation could enhance system security and mitigate water pollution. Recent work has also highlighted how ecological restoration and microbial nitrogen cycling in water-scarce rivers contribute to ecosystem resilience and affect nutrient dynamics across seasons [18,19]. Yue and Guo [20] proposed an integrated optimization model incorporating the WFE–environment nexus for sustainable agricultural development, supporting trade-offs among economic, resource, and environmental goals.

Beyond China, WFEE-related studies have been conducted globally [21,22,23,24,25]. Giampietro et al. [26] applied the Multi-Scale Integrated Analysis of Societal and Ecosystem Metabolism (MUSIASEM) to assess WFEE issues in Mauritius, Punjab (India), and South Africa. Hermann et al. [27] employed the CLEWS (Climate, Land, Energy, and Water Strategies) framework to show how agricultural policy in Burkina Faso significantly influences energy use. El Gafy et al. [28] constructed a system dynamics model (SD-WFEN) in Egypt and proposed a Water–Food–Energy Nexus Index (WFENI) to quantify interdependencies. Hülsmann et al. [29] emphasized the importance of embedding ecological knowledge in WFEE frameworks, noting ecology’s role in balancing subsystems. Recent studies have introduced isotopic datasets and new indicators, such as moisture source complexity and atmospheric water capturability, which improve understanding of regional hydrological cycles and support WFEE planning in arid regions [30,31,32,33]. Heard et al. [34] uncovered the existence of long-distance telecoupling within the water–food–energy–urban nexus, highlighting cities’ reliance on external resource flows and the resulting transboundary environmental effects. Melo et al. [35] argued that forest landscape restoration enhances WFEE security and aligns with the Sustainable Development Goals, calling for the integration of forest–ecosystem interactions into the nexus framework.

While previous studies have made progress in analyzing the coupling of the WFEE system, most adopt static frameworks or focus solely on pairwise interactions among subsystems, limiting their ability to dynamically simulate cross-sectoral feedback mechanisms under compound environmental and socioeconomic pressures. Notably, few studies have integrated long-term historical reconstructions with modeling approaches to systematically assess the sustainability and potential risks of such systems under extreme events, such as severe droughts or reductions in inter-basin water transfers.

In recent years, the increasing availability of tree-ring records, paleoclimatic data, and historical archives has enabled high-resolution reconstructions of past climate events, offering valuable inputs for system modeling. For instance, Chen et al. [36] reconstructed a 531-year precipitation series in the Huashan region of Shaanxi using tree-ring data from pine trees, revealing drought processes in North China linked to ENSO events. Gergis et al. [37] employed 12 paleoclimate proxies to reconstruct 206 years of rainfall variability in southeastern Australia, identifying the 1997–2009 “Big Dry” as an extreme event in a long-term context. Zhao et al. [38] reconstructed 800 years of flood and drought severity in Wuzhou, Guangxi, from documentary records, providing insights into the frequency of hydroclimatic extremes. Zheng et al. [39] used “Rain and Snow Records by Village” archives to reconstruct 300 years of seasonal precipitation across 17 stations in North China, highlighting the temporal linkages between extreme hydrological events and El Niño episodes. These studies demonstrate that long-term historical reconstructions not only reveal the periodicity and drivers of extreme events but also offer a solid foundation for future scenario modeling and risk forecasting.

Achieving coordinated development across WFEE sectors is essential for sustainability. This challenge aligns closely with the United Nations’ 2030 Agenda for Sustainable Development [40], especially SDG 1 (no poverty), SDG 2 (zero hunger), SDG 6 (clean water and sanitation), SDG 7 (affordable and clean energy), and SDG 11 (sustainable cities and communities)—all of which are intrinsically connected to the WFEE system. Meanwhile, global climate change has intensified the frequency and magnitude of extreme weather events, including droughts and floods [41,42,43]. As one of northern China’s most densely populated and economically developed regions, the Beijing–Tianjin–Hebei (BTH) region is particularly vulnerable to climate-induced pressures [44,45,46,47], as shown by global-scale drought footprint tracking and the increasing frequency of hydrological droughts in semi-arid regions. It serves as a representative case of resource–development conflict, with multiple severe droughts and floods recorded in recent decades [48,49,50], posing urgent challenges for integrated water resource management, food security, energy stability, and ecological integrity.

To address these challenges, this study develops an integrated simulation and regulation model of the WFEE system for the BTH region. The model holistically incorporates water dynamics, crop production, energy consumption, and ecological sustainability to analyze coordination patterns and identify critical bottlenecks across multiple development scenarios. It further introduces historical reconstruction of climate extremes—particularly extreme droughts—and forward-looking scenario analysis to assess the vulnerabilities induced by reduced inter-basin transfers and climate-driven water resource degradation. By integrating system simulation, risk quantification, and adaptive policy evaluation, this research seeks to provide robust scientific evidence and actionable insights to support high-quality development and multi-objective coordination in water-stressed regions.

2. Materials and Methods

2.1. Study Area

The Beijing–Tianjin–Hebei (BTH) region is located in the North China Plain, covering an area of approximately 216,000 square kilometers. It encompasses three separate governmental units: Beijing municipality, Tianjin municipality, and Hebei Province, as shown in Figure 1. Despite possessing less than 1% of the nation’s water resources and only 2.3% of its land area, the BTH region contributes to 6.1% of national grain production, supports 7.8% of the national population, and accounts for 8.5% of the country’s GDP. However, the WFEE in the BTH region faces significant challenges, including substantial pressure on water resource-carrying capacity, declining food security assurance capabilities, severe ecological degradation, and reduced energy supply security. In 2020, the self-sufficiency rates for water resources (calculated as total water resources divided by total water use), food (total food production divided by total food demand), and energy (total energy production divided by total energy consumption) in the BTH region were 74%, 77%, and 27%, respectively. Meanwhile, the rate of water resource exploitation and utilization—defined as the rate of surface and groundwater supply to total water resources—reached 88%. (Data sources: Beijing, Tianjin, and Hebei Water Resources Bulletins; China Water Resources Statistical Yearbook; Beijing, Tianjin, and Hebei Statistical Yearbooks; China Statistical Yearbook).

2.2. Historical Drought Reconstruction Method

Although China has experienced several severe droughts since the founding of the People’s Republic, the associated losses and societal impacts remain relatively limited compared to the extreme drought events recorded in historical periods. For instance, previous studies have shown that several megadroughts during the Ming and Qing dynasties had profound consequences for state governance and social stability. These include the severe drought in northern China in the late fifteenth century, which contributed to fiscal exhaustion and institutional transformation during the Ming Dynasty [51]; the major drought from 1637 to 1643, which accelerated the collapse of the Ming regime [52]; and multiple drought cycles during the Qing Dynasty that triggered widespread famine, social unrest, and ecological crises [53]. The destructive impacts of these historical droughts far exceeded those of their modern counterparts.

To assess the security risks faced by the WFEE system in the BTH region under extreme climate conditions, this study reconstructs precipitation patterns during historically severe drought periods through a quantitative approach. Recognizing that drought and flood grades recorded in traditional historical documents can serve as proxies for annual precipitation variability—and that precipitation time series often exhibit periodic characteristics—this study integrates historical drought–flood records from the Atlas of Drought and Flood Distribution in China over the Past Five Centuries [54,55,56] with modern-era measured precipitation data to reconstruct the region’s precipitation history over the past 500 years.

First, the Morlet wavelet transform was applied to identify dominant periodic components in annual precipitation series from selected climate stations during the period 1961–2020. Based on the identified cycles, a suitable sliding window was defined for model input construction. Next, two supervised machine learning models—random forest (RF) and support vector regression (SVR)—were trained using annual precipitation measurements paired with corresponding historical drought–flood grades. Model performance was evaluated using root mean square error (RMSE) and mean absolute error (MAE) to determine the optimal algorithm for precipitation reconstruction. Finally, the selected model was applied to the historical drought–flood grade sequences from 1470 to 1960, enabling the reconstruction of annual precipitation over the past five centuries and facilitating the quantitative characterization of anomalies associated with historically extreme drought events.

2.2.1. Historical Drought and Flood Data

Based on the Atlas of Drought and Flood Distribution in China over the Past Five Centuries, this study selected seven cities within the BTH region—Beijing, Tianjin, Shijiazhuang, Tangshan, Baoding, Cangzhou, and Handan—which possess long-term historical drought–flood records. Corresponding observed annual precipitation data from 1961 to 2020 were obtained from the National Meteorological Data Center of China, and the spatial distribution of these stations is shown in Figure 1. The atlas classifies annual drought–flood conditions into five discrete grades based on historical climate chronicles, local gazetteers, and disaster records, as evaluated by climate experts: 2 for extreme drought, 1 for moderate drought, 0 for normal, −1 for moderate flood, and −2 for extreme flood. Using these records, a complete set of annual drought–flood grades from 1470 to 2020 was compiled for the seven selected cities.

2.2.2. Precipitation Periodicity Analysis

To identify the dominant periodic structures in regional precipitation variability, this study employed the Morlet Continuous Wavelet Transform (CWT) [57,58] to perform time–frequency analysis on annual precipitation series from multiple climate stations over the period 1961–2020. Wavelet transform is a widely used multi-scale analysis technique for non-stationary signals, allowing simultaneous examination of both local time and frequency characteristics. It is particularly suitable for identifying periodic patterns in hydrometeorological time series.

By applying the CWT to the annual precipitation series, the energy distribution across different scales was obtained, enabling the detection of dominant cycles. The CWT is defined as

$$W\left(a,b\right)={\int }_{-\infty }^{+\infty }x\left(t\right)\cdot {\psi }^{*}\left({\displaystyle \frac{t-b}{a}}\right)dt$$

(1)

where $x\left(t\right)$ is original time series; $\psi \left(t\right)$ is mother wavelet function; a is scale parameter, corresponding to frequency variations; b is translation parameter, corresponding to time variations; ${\psi }^{\mathrm{*}}$ is complex conjugate of the wavelet; and $W\left(a,b\right)$ is wavelet coefficient at scale a and time b, representing the localized energy of the signal.

In this study, the Morlet wavelet was selected as the mother wavelet, defined as

$$\psi \left(t\right)={\pi }^{-1/4}\cdot e^{i{\omega }_{0}t}\cdot e^{-t^2/2}$$

(2)

where ${\omega }_0$ is the central frequency, typically set to 6 to ensure a good balance between time and frequency localization.

2.2.3. Machine Learning Model Development and Evaluation

To reconstruct historical annual precipitation, two supervised machine learning algorithms—random forest (RF) and support vector regression (SVR)—were employed and compared. Both models were trained using a sliding-window approach, with sequences of drought–flood grades as inputs and annual precipitation as the target variable. This allowed the construction of nonlinear regression models that learn the relationship between historical hydroclimatic conditions and precipitation levels. RF [59,60,61] is an ensemble learning method that aggregates predictions from multiple decision trees, offering high robustness and resistance to overfitting. SVR [62,63], based on kernel functions and statistical learning theory, maps input features into a high-dimensional space to model complex nonlinear relationships. To ensure generalization and avoid overfitting, five-fold cross-validation was conducted, with 80% of the data used for training and 20% for validation in each fold. The average performance across folds was used for model comparison.

Two evaluation metrics were used to assess model performance: root mean square error (RMSE) and mean absolute error (MAE) [64]. Given the strong interannual variability and presence of extreme events in precipitation data, MAE reflects the average prediction deviation and indicates model performance under typical conditions, while RMSE is more sensitive to large errors, providing insights into the model’s capability to capture extreme drought or flood years.

The metrics are calculated as follows:

$$\mathrm{RMSE}=\sqrt{{\displaystyle \frac{1}{n}}\sum _{i=1}^n{\left(\widehat{y_i}-y_i\right)}^2}$$

(3)

$$\mathrm{MAE}={\displaystyle \frac{1}{n}}\sum _{i=1}^n\left|y_i-\widehat{y_i}\right|$$

(4)

where $y_i$ is the observed precipitation in year iii and $\widehat{y_i}$ is the model-predicted precipitation.

2.2.4. Historical Precipitation Series Reconstruction Method

Based on the Morlet wavelet analysis of annual precipitation data from 1961 to 2020 at seven representative cities, dominant precipitation cycles were identified and used to define the length of the input sliding window for machine learning models. For example, if the dominant periodicity was identified as five years, a five-year sliding window was used to construct the input feature vector for precipitation prediction. The input–output relationship for model training can be expressed as

$$X_t=\{D_t,D_{t-1},D_{t-2},D_{t-3},D_{t-4}\}$$

(5)

$$P_t=f\left(X_t\right)$$

(6)

where $D_t$ denotes the drought–flood grade in year t; $X_t$ is the feature vector composed of drought–flood grades in the previous n years (determined by wavelet periodicity); $P_t$ is the predicted precipitation in year t; and $f$() represents the machine learning model (RF or SVR) trained in the previous step.

The trained models were then applied to the historical drought–flood grade sequences from 1470 to 1960. For each year, the corresponding sliding window of drought–flood grades was input into the model to estimate annual precipitation. This process yielded a continuous reconstruction of the annual precipitation series for the BTH region over the past 500 years. The reconstructed precipitation series provides a quantitative basis for identifying representative historical extreme drought events and serves as critical boundary input data for simulating extreme scenarios within the WFEE-coupled system model.

2.3. WFEE Simulation and Regulation Model

In this study, the WFEE simulation and regulation model developed based on GWAS [65,66]. This model comprises five major components: a water cycle module, a water resource allocation module, a food production module, an energy consumption module, and a socioeconomic water demand forecasting module. The water cycle and water resource allocation modules simulate the natural and societal water cycles to coordinate the regulation of local and transferred water resources, aiming to achieve water security and ecological health by simulating both the water resource system and the ecosystem. The socioeconomic water demand forecasting module provides data on domestic and industrial water demands for the model. The food production module supplies information on agricultural water requirements and the status of food self-sufficiency rates. The energy consumption module calculates the energy usage throughout the entire process of water abstraction, supply, utilization, and wastewater treatment in the societal water cycle. The architecture of the model is illustrated in Figure 2.

2.3.1. Natural and Societal Water Cycle Simulation

The natural water cycle encompasses several key processes, including surface runoff, interflow, groundwater flow, evaporation, transpiration, and river channel convergence. In this study, the simulation of the natural water cycle is accomplished using the water cycle module embedded within the GWAS model. The societal water cycle, which includes water intake, water allocation, water consumption, and wastewater discharge, is simulated in this study utilizing the water resource allocation module also embedded within the GWAS model [67,68].

2.3.2. Socioeconomic Water Demand Forecasting Module

The domestic water demand includes water requirements for residential use and the tertiary sector. Residential water use is calculated based on per capita daily water consumption, while water demand for the tertiary sector is estimated using GDP and water quota standards.

$${W\_demand}_{urban}=365\times P_{n,urban}\times N_{urban}/1000/(1-{\eta }_p)$$

(7)

$${W\_demand}_{rural}=365\times P_{n,rural}\times N_{rural}/1000$$

(8)

$${W\_demand}_{third}={GDP}_{third}\times Q_{third}/(1-{\eta }_p)$$

(9)

$${W\_demand}_{dom}={W\_demand}_{urban}+{W\_demand}_{rural}+{W\_demand}_{third}$$

(10)

where ${W\_demand}_{dom}$ represents total domestic water demand (m³); ${W\_demand}_{urban}$ represents urban residential water demand (m³); ${W\_demand}_{rural}$ represents rural residential water demand (m³); ${W\_demand}_{third}$ represents water demand for the tertiary sector (m³); $P_{n,urban}$, $P_{n,rural}$ represent the urban and rural populations, respectively; $N_{urban}$, $N_{rural}$ represent the per capita daily water consumption in urban and rural areas, measured in liters per person per day (L/person·day); ${\eta }_p$ represents the urban water supply network leakage rate; and ${GDP}_{third}$, $Q_{third}$ represent the added value of the tertiary sector (in 10,000 yuan) and its corresponding water quota (m³/10,000 yuan), respectively.

2.3.3. Food Production Module

Variations in evapotranspiration (ET) are the best variable for reflecting crop yield levels [69]. Therefore, this study employs a crop water production function model [70,71] to simulate regional food production under different irrigation and rainfall conditions. The calculation formula is as follows:

$${\displaystyle \frac{Y_{max}-Y_{act}}{Y_{max}}}=K_y\left({\displaystyle \frac{{ET}_{max}-{ET}_{act}}{{ET}_{max}}}\right)$$

(11)

where $Y_{max}$ represents the maximum crop yield (tons/hectare); $Y_{act}$ represents the actual crop yield (tons/hectare); $K_y$ represents the sensitivity factor of crop yield to water deficit; ${ET}_{max}$ represents the potential evapotranspiration of the crop (mm); and ${ET}_{act}$ represents the actual evapotranspiration of the crop (mm).

For the potential evapotranspiration (${ET}_{max}$), the calculation formula is

$${ET}_{max}=K_c·{ET}_0$$

(12)

where $K_c$ represents the crop coefficient and ${ET}_0$ represents the reference crop evapotranspiration (mm), calculated using the Penman–Monteith equation recommended by FAO.

For the actual evapotranspiration (${ET}_{act}$), the calculation formula is

$${ET}_{act}=P_{eff}+{\displaystyle \frac{M}{A_c}}·\theta$$

(13)

where $P_{eff}$ represents the effective precipitation (mm), calculated based on references [72]; M represents the volume of irrigation water withdrawn (m³); $A_c$ represents the irrigated area of the crop (10⁻¹ hectares); and $\theta$ represents the irrigation efficiency coefficient.

2.3.4. Energy Consumption Module

The energy consumption throughout the entire societal water cycle encompasses energy used in water intake, water supply, water usage, and wastewater treatment and discharge. The specific calculation process and formulas are illustrated in Figure 3 below [73,74].

2.4. Study Data and Scenario Design

The modeling and scenario simulation of the Water–Food–Energy–Ecology (WFEE) system in this study required a comprehensive dataset, including spatial data, meteorological and hydrological data, socioeconomic and agricultural statistics, energy consumption data, and projections of future water demand and climate conditions.

(1) Spatial Data: Spatial datasets include a digital elevation model (DEM) and land use classification maps, sourced from NASA (https://www.earthdata.nasa.gov/topics/land-surface/digital-elevation-terrain-model-dem, accessed on 3 June 2025) and the Resource and Environmental Sciences Data Center of the Chinese Academy of Sciences (https://www.resdc.cn, accessed on 3 June 2025).

(2) Meteorological and Hydrological Data: Historical precipitation, evapotranspiration, and surface and groundwater resource data were obtained from the National Meteorological Data Center (https://data.cma.cn/, accessed on 3 June 2025) and the Third Water Resources Survey and Assessment of the Hai River Basin. Future climate projections (2021–2035) were derived from the GFDL-ESM2M model (https://www.gfdl.noaa.gov, accessed on 3 June 2025) under two Representative Concentration Pathways: RCP6.0 (representing the recommended scenario) and RCP4.5 (representing the water degradation scenario) [75].

(3) Socioeconomic and Agricultural Data: Data on population, GDP, industrial output, crop production, cultivated area, and sectoral water quotas were collected from the statistical yearbooks and water resources bulletins of the Beijing–Tianjin–Hebei region (Beijing, Tianjin, Hebei Water Resources Bulletin; China Water Resources Statistical Yearbook; Beijing, Tianjin, Hebei Statistical Yearbook; China Statistical Yearbook). Future water demand projections for the year 2035 range from 24.71 to 29.69 billion m³, depending on assumptions regarding economic growth rate, implementation of water-saving measures, and target food self-sufficiency levels [76].

(4) Energy Consumption Data: Energy consumption data cover the full social water cycle, including water abstraction, distribution, usage, and wastewater treatment. These were sourced from local statistical yearbooks (China Energy Statistical Yearbook, China Statistical Yearbook, and China Rural Statistical Yearbook) and relevant literature for the Beijing, Tianjin, and Hebei areas [73,74].

(5) Scenario Configuration: The future scenario framework is structured along three dimensions: water demand, ecological protection, and water supply capacity. Demand side includes domestic, industrial, agricultural, and environmental flow requirements. Key variables include moderate economic growth, an intensive water-saving strategy, and a food self-sufficiency target of 70%. Ecological protection includes targets for river and lake ecological restoration and groundwater level recovery. Supply side encompasses increased reuse of unconventional water resources (reclaimed water reuse rate of 45%), expansion of the South-to-North Water Diversion Project (SNWDP) (Central Route Phase I: 4.95 billion m³, Phase II: 1.5 billion m³; Eastern Route Phase II: 2.54 billion m³), and the Luan River transfer scheme (not supplying Tianjin during non-deficit years). This comprehensive configuration is referred to as the recommended scenario, which serves as the baseline for subsequent simulations [77].

3. Results and Discussion

3.1. Historical Extreme Drought Reenactment Results

3.1.1. Historical Drought Patterns in the BTH Region

Annual drought–flood grades from 1470 to 2020 were extracted for seven representative cities (

Table 1. Statistical Summary of Drought and Flood Grades in Various Cities.

Drought and Flood Grades	Beijing	Tianjin	Tangshan	Baoding	Cangzhou	Shijiazhuang	Handan
−2	48	46	49	54	54	57	48
−1	110	141	136	106	149	122	125
0	186	212	199	213	166	179	169
1	152	117	126	135	128	134	155
2	52	32	38	40	45	56	51

and Figure A1). A regional average time series was derived to illustrate the overall drought–flood variability (Figure 4). Drought events (moderate plus extreme) occurred slightly more frequently than flood events, indicating a persistent long-term exposure of the regional water system to drought stress. Normal years (grade 0) were relatively uncommon, further supporting the classification of the BTH region as a typical drought–flood transition zone characterized by high interannual hydroclimatic variability.

Spatially, western cities such as Beijing and Shijiazhuang experienced a higher frequency of drought events, likely due to their location in the semi-arid transition zone of the warm temperate climate belt. In contrast, eastern coastal cities such as Tianjin, Tangshan, and Cangzhou experienced more frequent flood events, driven by summer monsoons and convective storms originating over the Bohai Sea. Central plain cities like Baoding exhibited a more balanced distribution of drought and flood events but often experienced prolonged droughts, reflecting limited regional water storage and heightened sensitivity to drought variability. In summary, the historical drought–flood landscape of the BTH region is characterized by the pattern: more droughts than floods overall, with clear spatial heterogeneity.

3.1.2. Precipitation Periodicity Characteristics

To capture the periodic structure of precipitation variability, Morlet continuous wavelet transform was applied to annual precipitation time series for the seven representative cities over the period 1961–2020. As shown in Figure 5, both the wavelet power spectra and the corresponding wavelet variance curves are presented for each city. The wavelet variance curves provide a quantitative basis for identifying dominant periodicities, and the main and secondary cycles are visually highlighted using dashed lines. The results reveal prominent short-term cycles (approximately 4–6 years) with strong consistency across most locations, indicating that interannual precipitation variability in the region is influenced by large-scale climate oscillations. In addition, mid-term periodicities (8–14 years) were identified in several inland cities—such as Beijing, Baoding, and Shijiazhuang—suggesting potential links to decadal-scale climate modes. These findings indicate that precipitation variability in the BTH region is shaped by a combination of short- and medium-scale climate drivers.

From a spatial perspective, coastal cities (e.g., Tianjin, Tangshan, and Cangzhou) displayed more concentrated short-period wavelet energy. In contrast, inland and piedmont cities (e.g., Beijing, Baoding, and Handan) exhibited more complex wavelet structures, with both short- and mid-term cycles likely resulting from topographic and continental climatic influences. The temporal evolution of dominant precipitation cycles shows strong non-stationarity, with the intensity and presence of cycles changing over time. These results highlight the importance of accounting for time-varying features in both historical reconstruction and future scenario modeling.

Based on the results of the wavelet periodicity analysis, different sliding window lengths were adopted in the machine learning models to better capture local precipitation dynamics: a five-year window was used for cities such as Beijing and Baoding; six years for Tianjin, Shijiazhuang, and Handan; and four years for Tangshan and Cangzhou.

3.1.3. Construction and Evaluation of Machine Learning Models

Using annual precipitation and corresponding drought–flood grade data from 1961 to 2020, two machine learning models—RF and SVR—were trained and validated using five-fold cross-validation. Results are presented in Figure 6, and evaluation metrics are summarized in Table 1. Both models exhibited strong predictive performance across most cities, with predicted values closely following the ideal line (y = x), suggesting good linear consistency. The RF model showed enhanced responsiveness to high-precipitation scenarios, demonstrating superior nonlinear fitting capacity, though it tended to overestimate in cities such as Baoding and Cangzhou. The SVR model produced smoother predictions, especially in low-to-moderate precipitation ranges, and effectively reduced the influence of outliers. Performance varied by city. Beijing, Tianjin, and Tangshan showed excellent model fits, while cities like Cangzhou and Shijiazhuang exhibited more pronounced deviations, which underestimated several high-precipitation years (points below the line).

The comparative performance of RF and SVR models across cities in

Table 2. Comparative performance of RF and SVR models across cities.

Model		Beijing	Tianjin	Tangshan	Baoding	Cangzhou	Shijiazhuang	Handan
RF	RMSE	91.047	68.018	67.667	89.764	93.163	98.675	107.123
	MAE	70.099	59.973	56.528	69.927	60.467	68.073	82.504
SVR	RMSE	85.465	79.69	63.88	126.91	107.79	114.305	108.264
	MAE	60.929	65.862	56.984	95.758	86.35	99.664	72.363
Best Model		SVR	RF	SVR	RF	RF	RF	SVR

indicates that the RF model outperforms the SVR model in most cities, achieving lower RMSE and MAE values, particularly in Tianjin, Baoding, Cangzhou, and Shijiazhuang, reflecting its superior overall accuracy and robustness. Notably, RF achieved the best performance in Baoding. Although SVR showed slightly better performance in specific cities such as Beijing and Tangshan, its prediction errors were significantly higher in others, indicating weaker adaptability to highly variable datasets. In Handan, the two models performed comparably. RF yielded a slightly lower RMSE (107.12 vs. 108.26), suggesting a marginal advantage in capturing extreme values. However, SVR produced a notably lower MAE (72.36 vs. 82.50), reflecting better average predictive stability across most sample points. Therefore, SVR was selected as the preferred model for reconstructing historical precipitation in Handan.

3.1.4. Historical Precipitation Series Reconstruction Result

To capture long-term hydroclimatic variability and detect extreme drought events across the BTH region, annual precipitation anomaly series were reconstructed for seven representative cities (Beijing, Tianjin, Tangshan, Baoding, Cangzhou, Shijiazhuang, and Handan) from 1475 to 2020. Using optimal machine learning models trained on historical drought–flood grade data, the annual precipitation, and precipitation anomaly ($PA={\displaystyle \frac{P_i-\overline{P}}{\overline{P}}}\times 100\%$, where $P_i$ is annual precipitation in year i and $\overline{P}$ is the long-term mean precipitation) were estimated.

Figure 7 presents the anomaly series for each city (a–g) and for the regional average (h). Years with above-average precipitation are shaded in blue (flood years), and below-average years are shaded in red (drought years). The severe continuous drought events were defined as periods meeting two criteria: ≥5 consecutive drought years and average anomaly (PA) < −15% [78]. These events are highlighted by black rectangles. Labels denote the drought period and corresponding average precipitation anomaly.

A total of more than 50 continuous droughts were identified across the seven cities, with a particularly notable concentration during the Chongzhen period (1633–1644). This event is consistently reflected in all seven city series as a prolonged and severe drought, with average anomalies reaching −20% to −27% (e.g., −25% in Beijing, −27% in Tianjin, −21% in Tangshan, −25% in Baoding and Cangzhou, −23% in Shijiazhuang, and −20% in Handan). This confirms historical documentation that identifies the Chongzhen Drought as one of the most devastating in Chinese history, associated with widespread famine and political crisis [52,79]. In addition to the Chongzhen Drought, other significant historical droughts [36,39,53,80,81,82] are also captured: Early Qing drought clusters (e.g., 1677–1683 in Baoding, Shijiazhuang; 1741–1749 in multiple cities); 1876–1878 megadrought, visible in Tangshan and surrounding areas; and 1920s–1930s droughts, notably 1925–1932 in Baoding, 1926–1930 in Shijiazhuang, and 1933–1938 in Handan. A recent modern multi-year drought (1997–2007) occurred across almost all cities, with regional average anomaly reaching −17%.

The regional composite (Figure 7h) highlights the two longest and most severe droughts: the Chongzhen Drought (1633–1644) and the modern drought (1997–2007). These results demonstrate the model’s capability to reproduce historically documented events and uncover patterns of drought frequency, severity, and geographic spread. They also reinforce the necessity of incorporating long-term hydroclimatic memory into current WFEE system risk management and adaptation planning. Given that the recent 1997–2007 drought has already caused substantial ecological and socioeconomic stress in the BTH region [83], we further consider a hypothetical scenario: if a historical drought as severe and prolonged as the Chongzhen Drought were to recur today, what level of systemic threat would it pose under current climatic and resource conditions?

3.2. Extreme Climate Events and the Security of the WFEE System in BTH

3.2.1. Coupled Simulation Results Under the Future Recommended Scenario

From the perspective of the water resources system, as illustrated in Figure 8, Beijing and Tianjin are projected to experience no significant water supply deficits. Other cities within the region are expected to maintain water supply shortages below 5%. The overall water shortage rate for the region is estimated at 2.6%. In terms of interannual variability, water scarcity is negligible in wet years. In contrast, during dry years, the total regional shortage rate may rise to approximately 8%. Nonetheless, from a holistic viewpoint, projected regional water security is fundamentally ensured.

From the perspective of the food system, as shown in Figure 9a, the average annual food self-sufficiency rate in the BTH region is 69.3%, with interannual fluctuations ranging from 57.0% to 81.8%. Under current constraints of agricultural resources and ecological capacity, this level is generally regarded as acceptable. In recent years, due to strict limitations on agricultural water use—driven by groundwater overexploitation and ecological degradation—the region’s overall food self-sufficiency has declined to approximately 77%. Given the context of rapid urbanization, particularly in Beijing and Tianjin, where arable land and water resources are extremely limited, local food production capacity is further constrained, and the self-sufficiency rate is expected to continue its downward trend.

Against this backdrop, setting a threshold of 70% for food self-sufficiency represents a realistic balance between local production potential and the national grain circulation strategy. On one hand, Beijing and Tianjin, as megacities, have high population densities and a large proportion of non-agricultural land use, with very limited resources available for grain production. Their current food self-sufficiency rates are only 4.7% and 37.6%, respectively—substantially lower than that of Hebei Province (94.1%)—highlighting significant spatial heterogeneity within the region [84,85]. On the other hand, at the national scale, China has established a stable grain distribution pattern from major producing areas to major consuming areas. Through interprovincial grain transfer mechanisms, the basic supply of key consumption zones can be effectively secured while reducing overreliance on local natural resources [86]. This approach not only supports national food security, but also provides a practical foundation for addressing groundwater overexploitation and restoring river–lake ecosystems [87,88].

From Figure 9b (WI: water intake; WS: water supply; WTD: wastewater treatment and discharge; WU-D: domestic water usage; WU-I: industrial water usage; TEC: total energy consumption) on social water cycle energy consumption, the total energy consumption for the social water cycle is 149.77 billion kWh. Interannual energy consumption has gradually increased from 130.7 billion kWh to 158.3 billion kWh. The energy consumption for water intake shows an initial increase followed by a decline, with a multi-year average of 1.8 billion kWh during 2021–2035. In contrast, the energy consumption for water supply exhibits a steady upward trend, rising from 6.6 billion kWh in 2021 to 7.9 billion kWh in 2035, with a multi-year average of 7.5 billion kWh. Wastewater treatment and discharge has the lowest energy demand; although it shows a slight increasing trend, its average energy consumption remains at only 1.0 billion kWh over the same period. Of the total energy consumption, water use accounts for more than 90% of the energy used in the social water cycle, with domestic water use being the primary contributor. This indicates that the increase in domestic water use is the main driver behind the rise in energy consumption within the social water cycle.

From Figure 9c on river discharge to sea, the long-term average (during the projected period of 2021–2035) discharge to sea is 3.26 billion m³, which matches the 15-year average (from 2006 to 2020) for the BTH region. The interannual variation ranges from 2.08 billion m³ to 7.18 billion m³.

3.2.2. Risk Assessment Under Historical Extreme Drought Reenactment

To assess the potential impacts of historically extreme drought events on the WFEE system, precipitation data from 1633 to 1647—representing a 15-year megadrought centered on the Chongzhen Drought—were used to drive the WFEE model for the years 2021 to 2035. The simulation assumes that current land use, infrastructure, and socioeconomic conditions remain constant. The year 2021 was selected as the starting point because comprehensive and consistent baseline data from 2020 were available to initialize the system.

Under this historical drought scenario, average annual surface water availability drops to 5.09 billion m³, with the driest year falling below 4.0 billion m³. The results show severe stress across the WFEE system (Figure 10): the average total water shortage rate rises to 16.9%, and agricultural shortages escalate to 30.3%. In the three driest years, the total water shortage rate exceeds 25%, while the domestic and industrial sectors face deficits of 5.8–6.3%.

Agricultural water use is particularly constrained, with less than 55% of irrigation demand met. As a result, the regional food self-sufficiency rate falls to an average of 44.7% and dips below 40% in the most severe years. To compensate, the region would likely rely on increased groundwater overdraft or expanded grain imports, both of which raise sustainability concerns. From an ecological standpoint, freshwater discharge to the Bohai Sea declines to an average of 1.3 billion m³ per year and below 0.9 billion m³ during peak drought years—posing substantial risks to estuarine and wetland ecosystems.

This reenactment simulation provides quantitative evidence for identifying system vulnerabilities and stress thresholds, offering a model-based rationale for drought adaptation strategies. While some measures may appear intuitive, the results confirm their necessity and clarify their relative effectiveness under historically extreme conditions. Specifically, the reconstruction highlights the urgent need for a multi-pronged drought mitigation strategy, grounded in dynamic feedbacks within the WFEE system. Recommended measures include: (1) Enhanced buffering capacity through improved surface and groundwater storage, such as reservoir expansion, managed aquifer recharge, strategic well conservation, and ecological restoration. These actions are essential to bolster water availability and allocation flexibility during multi-year droughts. (2) Augmented inter-basin water transfers, particularly scaling up emergency allocations through the South-to-North Water Diversion Project (SNWDP) in dry years, to safeguard critical domestic and industrial supply. In such extreme scenarios, temporary ecological trade-offs may be considered, provided minimum environmental flows are preserved. (3) Integrated physical and virtual water management, including the strategic use of virtual water (e.g., grain imports) and a gradual transition toward less water-intensive cropping structures, especially in water-stressed subregions like Hebei. These measures strengthen resilience from both the supply and demand sides.

3.2.3. Risk Assessment of Drought and Future Water Resource Degradation in Inter-Basin Transfer Source Area

With increasing reliance on inter-basin water transfers, the BTH region’s water security is becoming highly sensitive to the hydrological reliability of water source areas [89,90]. One of the most critical infrastructures in this context is the South-to-North Water Diversion Project (SNWDP)—China’s largest water transfer system, designed to alleviate water scarcity in northern China by redirecting water from the Yangtze River Basin.

The Middle Route of the SNWDP, completed in 2014, transfers water from the Danjiangkou Reservoir on the Han River (a major tributary of the Yangtze) northward through a 1432 km-long canal to supply water to Henan, Hebei, Tianjin, and Beijing. This route now provides over 70% of Beijing’s municipal water supply, as shown in Figure 11. Given this dependence, understanding the vulnerability of the Danjiangkou Reservoir is crucial for assessing regional resilience under extreme drought scenarios.

According to its official operational principle, “power generation is subordinate to diversion, diversion to ecology, and ecology to flood safety,” the amount of water available for transfer is constrained by multiple regulatory priorities. Based on historical hydrological records (1956–2017) and official planning documents [91], the annual transferable water volume has declined from an average of 9.54 billion m³ (1956–1998) to 8.45 billion m³ (1999–2017), indicating increasing water stress in the source region.

To assess more extreme droughts, the period from 1993–2007 was selected as a representative multi-year dry sequence. During this time, the average transferable volume dropped to 7.88 billion m³—about 20% below planned levels. The WFEE coupled simulation and regulation model was driven by two distinct input datasets: the transferable water volume series from 1993 to 2007, representing a prolonged drought period in the Danjiangkou Reservoir region (SNWDP drought scenario), and the projected water availability under the water resource degradation climate scenario, indicating potential long-term water resource degradation.

As shown in Figure 12, the historical drought sequence leads to a moderate increase in water scarcity, with the regional shortage rate rising to 4.7%. While domestic and industrial sectors remain largely unaffected due to local supply substitution, agricultural deficits rise to 9.5%, and food self-sufficiency falls to 66.5%. Energy use remains stable, but freshwater discharge to the sea drops to 3.00 billion m³—0.26 billion m³ below the baseline. Under the water degradation scenario, impacts are more severe. Regional water shortage reaches 5.3%, agricultural deficits hit 11.0%, and peak-year shortages approach 19.9%. Food self-sufficiency drops further to 62.9%. Although energy use remains stable, freshwater discharge to the sea declines to 2.60 billion m³—0.66 billion m³ lower than the recommended scenario.

These results highlight that although the Danjiangkou Reservoir maintains a relatively robust supply capacity and the BTH region possesses a degree of local water source complementarity, prolonged drought in the source area primarily affects agriculture, with manageable consequences. Emergency groundwater abstraction may temporarily mitigate these deficits. However, climate-induced long-term water degradation poses greater systemic threats to supply security, food resilience, and ecological integrity.

Given the limited local water availability and the exhaustion of both demand-side and local supply-side potential in the BTH region, expanding inter-basin water transfers emerges as a critical adaptive strategy. Considering the lower probability of historical megadrought recurrence, but higher likelihood of long-term water stress from future degradation and source-area droughts, it is recommended that the South-to-North Water Diversion Project’s Eastern and Middle Route follow-up projects plan for an annual transfer volume in the range of 4.6–6.0 billion m³ to compensate for anticipated deficits and bolster system resilience.

3.3. Limitations and Future Research

Despite the utility of the WFEE model in simulating system responses under extreme drought scenarios, several limitations remain that point toward future development directions.

First, the ecological module currently characterizes environmental impacts solely through river discharge to the sea, serving as a proxy for freshwater availability to estuarine and wetland systems. While this indicator is generally adequate for assessing ecological stress in a regional water balance context, it lacks the capacity to capture the cascade effects of drought on biodiversity, habitat degradation, or ecosystem service loss. Future studies should enhance the ecological component by incorporating multi-dimensional indicators, such as ecological flow thresholds, habitat suitability indices, or ecosystem service valuation, potentially through coupling with ecological process models.

Second, the energy consumption module quantifies only the energy required for water intake, water supply, water usage, wastewater treatment, and discharge within the social water cycle. It does not consider the reverse water demand from energy production activities, such as thermal power generation or coal processing. This one-directional representation overlooks the bidirectional feedback within the water–energy nexus. Future research should integrate energy-sector water demands to improve systemic coupling and enhance scenario-based policy analysis.

Third, water demand projections are currently based on a linear relationship with GDP, which reflects average historical water-use intensities. While sufficient for general trend estimation, this approach does not account for potential nonlinear shifts due to industrial upgrading, technological breakthroughs, or policy-induced efficiency gains. Incorporating nonlinear forecasting models or scenario-driven demand pathways would improve predictive robustness under varying socioeconomic trajectories.

Fourth, only two machine learning models (RF and SVR) were employed to reconstruct historical precipitation. Although both demonstrated strong predictive performance and were suitable for the study’s objectives, time series-specific algorithms, such as XGBoost, LSTM, or attention-based deep learning models, were not tested. Future model development should explore these advanced architectures and compare their performance to enhance model adaptability and accuracy.

Fifth, the proposed strategy of “temporary ecological compromise during drought periods” reflects real-world emergency trade-offs but lacks a clearly defined minimum ecological flow threshold. Without this, such strategies may risk unintended ecological degradation. Future applications of the WFEE model should incorporate science-based flow safeguards to ensure environmental integrity under stress.

In sum, while the present model is sufficient for scenario-based system simulation and strategic evaluation, future research should aim for more holistic coupling, richer ecological representation, nonlinear demand modeling, robust algorithm validation, and ecologically grounded risk controls to support high-resolution planning under compound climate and resource pressures.

3.4. Potential Impacts of Expanded Inter-Basin Transfers Under the SNWDP

This study, from the perspective of water demand in the BTH region, proposes a potential expansion of the SNWDP Middle and Eastern Route follow-up projects, aiming to support the coupled development of water, food, energy, and ecology systems under drought stress. Since the initial operation of the SNWDP Middle Route in 2014, the project has brought substantial benefits to the receiving areas. The project has significantly improved drinking water safety and reduced groundwater over-extraction, with the groundwater table in Beijing notably recovering in recent years [92].

Nevertheless, expanding the SNWDP is not without trade-offs. As a national-level strategic infrastructure project, the SNWDP requires massive capital investment and affects water availability in both source and receiving regions. Historical records indicate that the initial investment in the Middle and Eastern Routes accounted for approximately 1–4% of China’s GDP in 2002, the year of construction initiation [93]. Future expansions will require upgrades to infrastructure standards, potentially increasing engineering and operational costs.

Moreover, the water price of transferred water is relatively high. According to initial pricing policies from the National Development and Reform Commission, the composite water price of the Middle Route is RMB 0.97/m³ for Hebei, 2.16/m³ for Tianjin, and 2.33/m³ for Beijing [94]. While these prices fall within the range of residential affordability (generally RMB 4.74–5.00/m³) [95], they significantly exceed current agricultural water prices in Hebei (RMB 0.30–0.48/m³) [96]. This pricing gap suggests that, without substantial subsidies or reform, transferred water may be economically unviable for large-scale agricultural use, posing a risk that projected demand will not translate into effective demand.

In addition, inter-basin transfers reduce the available water in the donor basins. The Middle Route currently diverts roughly 35% of the total runoff from the Han River Basin [97], which may impair local ecological conditions and supply security. Studies by Hu et al. [98] indicate a potential long-term decline in precipitation over the Danjiangkou Reservoir catchment, which could further constrain available transfer volumes. Similarly, the Eastern Route draws from the lower Yangtze River, and reducing freshwater discharge into the estuary could exacerbate saltwater intrusion and other ecological risks [99].

Therefore, any proposal to increase annual transfer volumes to 4.6–6.0 billion m³ must be carefully evaluated through a comprehensive cost-benefit analysis that considers engineering investment, operating expenses, regional fiscal sustainability, ecological risks, and the economic efficiency of alternative strategies such as virtual water trade (e.g., food imports). Balanced decision-making should reflect the interconnected nature of water supply, food security, ecosystem health, and long-term financial viability.

4. Conclusions

This study addresses the challenges of increasing water scarcity, complex inter-system coupling, and escalating climate risks in the BTH region by developing an integrated WFEE simulation and regulation model. The model incorporates modules for hydrological simulation, sectoral water demand forecasting, crop yield estimation, and energy consumption accounting to evaluate system behavior and vulnerabilities under various scenarios. The main conclusions are as follows:

(1) A historical precipitation reconstruction framework was established for the BTH region over the past 500 years. By integrating the Atlas of Drought and Flood Distribution in China for the Last 500 Years with observed rainfall data and combining Morlet wavelet-based periodicity analysis and machine learning models (RF and SVR), a continuous annual precipitation series from 1470 to 2020 was successfully reconstructed. The Chongzhen Drought (1633–1644) event is consistently reflected in all seven city series as a prolonged and severe drought, with average anomalies reaching −20% to −27% (e.g., −25% in Beijing, −27% in Tianjin, −21% in Tangshan, −25% in Baoding and Cangzhou, −23% in Shijiazhuang, and −20% in Handan).

(2) Under a coordinated development scenario, which includes enhanced water conservation, increased reclaimed water utilization (up to 45%), and expansion of South-to-North Water Transfer volumes, the region can achieve near-equilibrium in water supply and demand (average annual deficit rate of 2.6%). Groundwater extraction and recharge can reach a sustainable balance, the food self-sufficiency rate can stabilize at 69.3%, and ecological water demand can be met. The total energy consumption of the social water cycle is estimated at 149.8 TWh, indicating strong systemic coordination and sustainable potential.

(3) Reoccurrence of historical megadroughts would severely threaten WFEE system security. Simulation using reconstructed precipitation from the 1633–1647 drought period shows a surge in regional water deficit to 16.9%, with agricultural water shortages exceeding 30%, food self-sufficiency dropping to 44.7%, and freshwater discharge to the sea falling below 0.9 billion m³. These results underscore the system’s vulnerability to prolonged compound droughts and the necessity for unconventional adaptive measures.

(4) Droughts in inter-basin water source areas and long-term water degradation pose systemic risks. Simulations under sustained drought in the Han River Basin (with transferable water reduced to 7.88 billion m³) and under future water resource degradation scenarios indicate an increase in BTH water scarcity to over 5%. Agricultural water use becomes constrained, food self-sufficiency drops to 62.9%, and ecological water conditions further deteriorate—highlighting the system’s growing dependence on external water sources and its increasing fragility.

(5) To enhance drought resilience and climate adaptability, this study recommends expanding the annual inter-basin water transfer capacity via the Eastern and Middle Routes of the SNWDP to 46–60 billion m³. This measure would help stabilize the WFEE system under extreme climate stress, ensuring water security, food supply, ecological restoration, and low-carbon development in the BTH region.