Watershed vulnerability assessment in the context of climate change, land use change, and water use change in the Nan River Basin, northern Thailand

The ecosystem's deviation from its natural reference condition is a function of its vulnerability and the magnitude of the stressors that affect it (Rasmussen et al. 2012; Tolkkinen et al. 2016; Arriagada et al. 2019). Watershed vulnerability is defined as the potential for future degradation of watershed processes and aquatic ecosystem health, arising from factors such as climate change, land use alterations, and shifts in water use (USEPA 2015). Degradation in watershed processes can severely compromise freshwater quality and availability, which are vital for human consumption, agriculture, and sustaining biodiversity. As watershed vulnerability increases, the risks of water scarcity, flooding, and habitat loss also intensify. Addressing watershed vulnerability is essential for sustaining reliable water supplies, protecting aquatic habitats, and enhancing the resilience of socio-economic systems, thereby securing long-term water security and ecological stability.

Understanding trends in climatic parameters such as air temperature and precipitation is crucial for assessing regional climate and the effects of climate change (Sharma et al. 2016; Dong et al. 2020b; Üneş & Kaya 2021). Climate change has inflicted significant and often irreversible damage on freshwater and terrestrial ecosystems, heightening watershed vulnerability and posing considerable challenges to achieving sustainable development, particularly in developing countries and areas prone to climate extremes (IPCC 2022).

Land use changes have profound effects on watershed health, significantly impacting water quantity and quality (Trang et al. 2017). Such changes can disrupt hydrological systems and exacerbate watershed vulnerability, as highlighted by Tebakari et al. (2018). The loss of natural land cover leads to increased soil erosion and sediment runoff, degrading water quality and aquatic habitats. This disruption reduces the watershed's ability to provide essential ecosystem services, including water purification, water regulation, groundwater recharge, erosion control, flood regulation, and habitat provision for wildlife.

Globally, the rapid increase in water withdrawal compared to population growth is projected to lead to a decline in water availability by 2050 as demand continues to rise (Boretti & Rosa 2019). Changes in water use patterns, such as increased withdrawals for agriculture, industry, or domestic purposes, can further stress water resources, leading to depletion, contamination, or conflicts over water allocation.

Moreover, alterations in maximum and minimum temperatures, precipitation patterns, land use, and water use collectively affect land suitability for various activities like agriculture and urban development. Conversely, changes in land use can influence local weather patterns and water availability, creating a complex interplay between climate, land use, and water management.

To manage the complex interactions of climate change, land use change, and water use, future projections are essential. These scenarios help assess watershed vulnerability and identify priorities for watershed protection and restoration. Utilizing projection tools is crucial for understanding environmental processes, projecting future trends, and formulating strategies for sustainable resource management and adaptation. Global climate models (GCMs) simulate future climate conditions, projecting temperature and precipitation patterns. The Dyna-CLUE model forecasts land use changes by considering population growth, urbanization trends, and policy interventions. Geographic information systems (GIS) analyze spatial data, mapping changes in land cover, the expansion of impervious areas, and the distribution of water resources within the basin and its sub-basins. These tools are employed in this study to understand the potential impacts on watershed vulnerability.

Identifying vulnerability indices for basins and their sub-basins is critical for addressing exposure to climate changes, impervious area changes, and water use projections. These indices quantify the extent of exposure to various stressors, including temperature and precipitation changes, expansion of impervious surfaces, and alterations in water use practices. Assessing vulnerability to these factors is essential for understanding their impact on watershed resources.

In monsoon regions, many populations depend heavily on freshwater for agriculture, water resources, industry, transport, and socio-economic activities (IPCC 2021). The Nan River Basin (NRB) in Northern Thailand contributes 25–40% of the annual flows in the Lower Chao Phraya River Basin, supporting key sectors including agriculture, manufacturing, services, and domestic water use (Chuenchum et al. 2017). Agriculture accounts for 35.1% of the basin's land use, with approximately 80% of water demand directed toward agricultural purposes (Jirasirichote et al. 2021; Petpongpan et al. 2021). The basin has experienced severe water scarcity due to both climatic and non-climatic drivers, with notable extreme floods and droughts occurring in 2011 and 2015 (Petpongpan et al. 2021; IPCC 2022).

Research on the NRB has investigated various aspects including the impacts of land use changes on water quality and flooding (Kaewmanee et al. 2024), spatial and temporal drought patterns and their agricultural impacts (Bastola et al. 2024), runoff and sediment yield prediction (Jirasirichote et al. 2021), and projections of hydro-climatic extreme events under climate change (Petpongpan et al. 2021). Most of these studies have focused on the Upper NRB (UNRB).

Given the NRB's history of hydro-climatic extremes, such as the severe floods and droughts in 2011 and 2015, and its essential role in sustaining life, agriculture, and economic activities, a comprehensive evaluation of the basin's vulnerability to future climate, land use, and water use changes is imperative. While previous studies have provided valuable insights, they often address specific drivers, such as land use changes or hydro-climatic extremes, without integrating the combined effects of climate change, land use change, and water use change at a basin-wide scale. Furthermore, limited attention has been given to the Lower Nan River Basin (LNRB), which faces distinct downstream challenges, including flooding, sedimentation, and water scarcity.

This study addresses these gaps by conducting a comprehensive watershed vulnerability assessment across the entire NRB, evaluating the hydrological and land use implications of these drivers. To achieve this, the study employs three key projection indicators: climate change, future land use changes (including impervious cover projections), and future water use changes. By assessing the vulnerability of sub-basins within the NRB to these stressors, the research facilitates targeted and localized interventions. The sub-basin-level analysis is critical for identifying specific vulnerabilities and prioritizing resources for adaptation and mitigation measures.

Moreover, this study provides a detailed understanding of how these impacts vary spatially and temporally, offering critical insights into watershed vulnerability and its implications for sustainable management. Adaptive management strategies are proposed to address these vulnerabilities, ensuring sustainable water resource management across the near future (NF), mid future (MF), and far future (FF) periods. These findings serve as a valuable resource for policymakers and stakeholders, enabling informed decision-making and proactive strategies to address future environmental challenges.

Based on the 2020 LULC map derived and consolidated from the Copernicus Climate Change Service, Climate Data Store (https://cds.climate.copernicus.eu), the basin encompasses an area of about 35,036.81 km². It consists of 52.83% agricultural land, 0.27% urban and built-up areas, 1.02% water bodies, 17.63% forest land, and 28.25% other types of land. Elevations within the basin range from 19 to 2,084 m above mean sea level. The basin is characterized by two distinct seasons namely the wet season, spanning from May to October, and the dry season, which persists from November to April of the subsequent year (Bastola et al. 2024).

The NRB is divided by the Sirikit Dam Reservoir into two portions namely Upper Nan River Basin (UNRB) and LNRB. The UNRB is characterized by mountainous terrain and forested areas, whereas the LNRB consists of fertile alluvial plains, making it suitable for agriculture and traversing 11 provinces, namely Kamphaeng Phet, Loei, Nakhon Sawan, Nan, Phayao, Phetchabun, Phichit, Phitsanulok, Phrae, Sukhothai, and Uttaradit (Chuenchum et al. 2017). This diversity underscores the NRB's vital role in the region's socio-economic and ecological landscape. Serving as a vital water resource, the basin significantly contributes to Gross Domestic Product (GDP) generation, with differing water demand and GDP levels across sectors; high water demand and GDP in agriculture, low water demand and GDP in manufacturing, and high water demand but low GDP in the service sector (Suttinon 2020).

This study employed eight CMIP6 GCM models and two SSP scenarios (SSP2-4.5 and SSP5-8.5) from Copernicus Climate Data Store (open, free, and unrestricted data), chosen based on data availability. Table 2 provides detailed information regarding these CMIP6 GCMs, including the developing institution, spatial resolution, and relevant references. Historical and future daily precipitation, and daily maximum and minimum temperatures were obtained from these models. CMIP6 models exhibit greater reliability for rainfall simulations and large-scale circulation, while also demonstrating higher quality in simulating regional climate across Southeast Asia (Khadka et al. 2021). Rather than relying on a single GCM, utilizing combinations of multiple GCMs (Acharya et al. 2014; Chokkavarapu & Mandla 2019) and scenarios, helps reduce uncertainty associated with results (Wilby 2006; Crosbie et al. 2011; Wang et al. 2016; Chokkavarapu & Mandla 2019).

The CMIP6 GCM models were validated using historical climate data from the TMD and RID. To eliminate uncertainties and minimize biases of these models' performance, this study applied the statistical approach specifically the Empirical Quantile Mapping (EQM) method, which will treat the gaps between observed and simulated results of daily precipitation, and maximum and minimum temperatures for the NRB. EQMs are non-parametric transformation methods and have been preferred over parametric approaches due to their demonstrated effectiveness in reducing biases from GCMs (Mishra et al. 2020). Several studies have highlighted the effectiveness of the EQM method in precipitation projection, attributing its success to frequency-based metrics that effectively address biases (Fang et al. 2015; Luo et al. 2018; Enayati et al. 2021; Brumatti et al. 2024).

The performance of the CMIP6 GCMs after bias correction were evaluated using statistical parameters which includes percent bias index (PBIAS), mean absolute error (MAE), coefficient of residual mass (CRM), modified Willmott d-index (d′), normalized root mean square error (NRMSE), and normalized standard deviation (NSTDV). The PBIAS calculates the relative difference in magnitude between GCM and observed data, where 0 means a perfect fit, negative values indicate under-prediction, and positive values denote over-prediction. The MAE quantifies the average difference between the observed values and the projected values within the dataset, where a value near to 0 implies unbiased prediction.

The CRM measures how much the model's predictions differ from the observed data, with a value of 0 indicating a perfect match, positive values mean the model tends to underestimate, and negative values indicate it tends to overestimate. The d′ metric, introduced by Legates & McCabe (1999), appropriately weights errors without inflating them by squared values, with a range from 0 to 1, where higher values indicate a better model fit.

The NRMSE, which normalizes RMSE by the data range, is a more suitable measure for evaluating interpolation errors in precipitation due to its spatial variability (Otto 2019; Konca-Kedzierska et al. 2023), with a value of 0 representing a perfect fit. The NSTDV is a dimensionless measure in comparing variability across different climate variables. It is calculated by ratio of the standard deviation of the GCM-derived predictor variables over observed standard deviation of the corresponding variable, in which 1 signifies a perfect fit. Table 3 summarizes the statistical parameters used in evaluating the performance of the bias correction.

After evaluating the eight GCMs, the top-performing models were selected to analyze future climatic conditions. The outputs from these models were synthesized to create a unified projection ensemble. An ensemble modeling approach, utilized in numerous studies of future climate change, combines the outputs of multiple GCMs to provide a more robust and comprehensive projection of future climate conditions (Parker 2013). The study by Nontikansak et al. (2022) emphasized the utilization of ensemble GCMs from CMIP6 under SSP2-4.5 and SSP5-8.5 scenarios to extrapolate future rainfall, effectively reducing uncertainty levels. These findings align with the approach adopted in this study, further enhancing the robustness of projections for the NRB.

The annual land use maps of the NRB from 2009 to 2020 were acquired from the global land cover datasets available on the Copernicus Climate Data Store. These maps classified seventeen land use types, which were subsequently aggregated into five categories: agricultural land, urban and built-up land, water bodies, forest land, and other lands. The 2009 map was selected as the baseline map from where the simulations will start. 2016, 2018, and 2020 maps were utilized for validation against the simulated maps of the respective years.

Verburg & Overmars (2009) introduced Dyna-CLUE, a hybrid land use change model that integrates top-down and bottom-up dynamics in land use modeling. Various land use and land change studies in Thailand have satisfactorily utilized the Dyna-CLUE model, demonstrating its effectiveness in diverse contexts, including Southeast Asian mountain regions and tropical areas. This success is attributed to the model's complexity, spatial allocation algorithm, and moderate data requirements (Trisurat et al. 2010; Lippe et al. 2017; Shrestha et al. 2018; Shrestha et al. 2020; Waiyasusri & Wetchayont 2020; Phuaphae et al. 2021; Waiyasusri & Chotpantarat 2022).

This study utilized Dyna-CLUE model (an open-source software) to project future land use change as illustrated in Figure 2. The Dyna-CLUE model comprises of a non-spatial demand module and spatial allocation module (Park et al. 2011; Zhang et al. 2015). The non-spatial demand module focuses on the rate of land use change during a specific time, while the spatial allocation module addresses the probable classes of land use change in each grid within the specified period (Shrestha et al. 2018). Key factors considered in the non-spatial demand module include the conversion matrix (Supplementary material, Appendix A), conversion elasticity (Supplementary material, Appendix B), and land use requirements. Sixteen land use drivers have been identified as the primary factors influencing land use changes in the NRB. These drivers are categorized into two groups: physical drivers and social drivers. Physical drivers encompass elevation, slope, aspect, rainfall, and soil types, while social drivers include distance from rivers, distance from roads, distance from the capital city, and population density. These driving factors were selected based on their well-established influence on land use change dynamics, as demonstrated in similar studies (Buhay Bucton et al. 2022; Pinsri et al. 2022), where physical and social factors have been shown to play a significant role in shaping urban expansion and forest conservation trends.

The Statistical Package for the Social Sciences (SPSS) software was employed to conduct binary logistic regression analysis for each land use type in relation to the 16 independent variables. The Relative Operating Characteristic (ROC) was used to compare the map of actual change to maps of modeled suitability for land cover change, following the methodology outlined by Pontius & Schneider (2001). An ROC value exceeding 0.50 indicates the model's effectiveness in distinguishing between classes or making predictions. The higher the ROC value, the more accurate and reliable the model is in its predictions, with 1.00 representing perfect accuracy in classification or prediction.

Two land use change scenarios are considered in this study namely business-as-usual (BAU), and rapid urbanization (RU). The BAU scenario is modeled based on historical land use trends and current policies. Historical land use data are analyzed to project future changes by applying trend analysis techniques that extend past patterns into future projections. This scenario incorporates the Thailand's 20-year National Strategic Plan (2018–2037) and National Forest Policy, which project an increase of forest land by at least 20% by 2037, aiming to reach 30% of total land by 2100. Additionally, this scenario predicts a continuous expansion of urban and built-up areas, a decline in agricultural and other land categories, and no change in water body extent. Projected land use maps for the BAU scenario are compared with baseline data to ensure consistency with historical trends. Quantitative analysis is conducted to validate that these projections align with historical patterns and policy impacts.

The RU scenario projects significant and rapid expansion of urban and built-up area from the current trends and future development plans, considering economic growth. This scenario predicts a slower rate of growth of forested land growth while assuming no change in the extent of water bodies. Projected land use maps for the RU scenario are examined and compared with baseline data to assess the extent of urban expansion and shifts in other land categories. Across both scenarios, spatial policy restricts land conversion within protected areas. This constraint is integrated into the modeling process to ensure compliance with existing policies. The projected land use outputs under both scenarios are used to identify and calculate the projected impervious land cover, which helps in assessing the impact of each scenario on land cover and impervious surfaces.

Projection of future water use includes sectors such as domestic, industrial, and agricultural as illustrated in Figure 2. The projection of future agricultural water use relies on baseline data sourced from the RID, which includes information on crop water requirement (in MCM/Ha), total area equipped for irrigation (in Ha), and duration of irrigation (in days). Utilizing this data, the agricultural water use is calculated, and then normalized by dividing it by the agricultural land areas during the baseline period, resulting in the agricultural water use rate (in MCM/year/Ha). This rate is subsequently multiplied by the projected agricultural land area derived from the land use change projection conducted using Dyna-CLUE, as discussed in Section 2.3, to estimate future agricultural water use. The scenarios under consideration involve estimating agricultural water use within BAU and RU, with land use as the dominant factor. These land use projections provide a previously unseen view into potential water use futures (Wilson et al. 2016). Projections from CMIP6 GCM models for climate variables were not incorporated in future agricultural water use.

The future industrial water use is derived from the total water use by industries and the number of industries during the baseline period, sourced from the Department of Industrial Works and the NSO, respectively. The ratio of total water use by industries to the number of industries during the baseline period determines the industrial water use rate (MCM/year/number of industries). This rate is then multiplied by the projected number of industries for the future period, resulting in future industrial water use.

Further, normalization technique is applied in this study to standardize the change metric values for precipitation and temperature changes, alterations in impervious area, and variations in water use across all sub-basins. Normalization of change metric values to a range of 0–1 enables the representation of both increases and decreases in metric values within a standardized scale. This approach helps assess whether changes in precipitation, temperature, impervious area, and water use are within expected ranges or deviate significantly from historical patterns, providing a standardized framework for comparing change metrics across different variables, sub-basins, and time periods. Through normalization, values near 0 suggest minimal to no change, while those near 1 denote substantial change, whether an increase or decrease, depending on the variable. Once the metric change values have been normalized to a range from 0 to 1, a 1-to-4-point scale is then utilized on these normalized values to determine the integrated watershed vulnerability with respect to stressor exposure.

The study conducted by Ahn & Kim (2019) evaluated the watershed vulnerability in terms of its exposure to stressors and revealed that a vulnerability score of 0.70 indicates a relatively high level of vulnerability concerning changes in impervious land cover. Additionally, a vulnerability score of 0.87 is considered high in relation to both impervious land cover and precipitation changes. Conversely, a vulnerability score ranging from 0.20 to 0.28 suggests low vulnerability concerning differences in domestic, industrial, and agricultural water use. Table 4 provides a description of the 1- to 4-point range of reference values for watershed vulnerability concerning future climate change, impervious area change, and water use change metrics in this study.

EQM bias correction serves as a post-processing method applied to GCMs to enhance their alignment with observed data, thus refining the accuracy of climate projections. While EQM effectively mitigates historical biases compared to observations, its effectiveness and performance were evaluated using performance parameters, such as PBIAS, MAE, CRM, d′, NRMSE, and NSTDV. The bias correction in this study led to significant improvements in model performance, consistent with the findings of Mendez et al. (2020), who reported that EQM marginally outperforms other bias correction methods by significantly reducing systematic biases in GCM outputs. Moreover, EQM ensures that corrected values remain within the range of observed data used during hindcasting, minimizing the risk of unrealistic extrapolation beyond historical limits (Boé et al. 2007; Gudmundsson et al. 2012; Cooper 2019). These improvements are reflected in the enhanced statistical performance metrics observed in this study (Table 5 and Supplementary material, Appendices C and D), underscoring the robustness and reliability of EQM for bias correction in climate projections.

Before bias correction, precipitation, maximum, and minimum temperature statistics varied significantly across stations, with PBIAS ranging from −75.32% to 56.75% for precipitation, −4.40% to 23.76% for maximum temperature, and −26.57% to 20.50% for minimum temperature. After correction, all metrics showed improvements, indicating enhanced model performance and suggesting that the correction method effectively mitigated historical biases. Notably, the NorESM2-MM model achieved a perfect PBIAS value of 0.00. A PBIAS within ±10% is considered very good (Barbosa et al. 2019; Carlos Mendoza et al. 2021).

MAE values for precipitation, maximum, and minimum temperatures before bias correction ranged from −2.12 to 2.23 mm, −7.81 to 1.47 mm, and 4.24 to 5.27 mm, respectively. After correction, the MAE values showed a marked reduction, suggesting effective bias correction. An MAE value near 0.00 indicates an unbiased prediction, while a low MAE reflects a strong fit between the variables (Kaur & Kaur 2023).

Before bias correction, CRM values ranged from −0.75 to 0.57 for precipitation, −0.04 to 0.24 for maximum temperature, and 0.27 to 0.20 for minimum temperature. Significant improvements were observed after applying bias correction. Most GCMs across the stations showed CRM values of 0.00 for precipitation, indicating a perfect fit between observed and bias-corrected data. However, a few GCMs displayed CRM values of −0.01, −0.02, and −0.03, suggesting slight overestimation. For maximum and minimum temperatures, CRM values across all GCM stations consistently remained at 0.00, further indicating the effectiveness of the applied bias correction.

The d′ values ranged from −17.67 to 19.69 for precipitation, −22.53 to 85.33 for maximum temperature, and −61.50 to 59.74 for minimum temperature prior to bias correction. Post-correction, these values showed improvement, indicating a strengthened agreement between model predictions and observed data for these variables.

The NRMSE values for precipitation, maximum temperature, and minimum temperature before bias correction ranged from 0.02 to 0.11, 0.02 to 0.37, and 0.00 to 0.27, respectively. Improvements in NRMSE values, ranging from 0.00 to 0.14 after bias correction, were observed, indicating the correction method effectively mitigated historical biases. An NRMSE value below 0.2 indicates better agreement between the gridded and observed data, with fewer residual errors (Joseph et al. 2020).

The NSTDV values, on the other hand, ranged from 0.41 to 1.36 for precipitation, 1.00 to 2.13 for maximum temperature, and 0.87 to 1.56 for minimum temperature prior to bias correction. After bias correction, the NSTDV values improved, demonstrating excellent agreement between the bias-corrected data and observed data for all GCMs across all stations. These improvements in statistical parameters highlight the robustness of EQM in addressing and enhancing GCM biases.

The GCM ranking system used in this study is derived from the earlier study by Pinsri et al. (2022), wherein rankings are determined by performance assessed through statistical parameters. The top-performing GCMs for precipitation, maximum temperature, and minimum temperature include CMCC-ESM2, NorESM2-MM, EC-Earth3-Veg-LR, EC-Earth3-CC, CNRM-CM6-1, and MRI-ESM2-0. These refined GCMs were selected to project the future climate of NRB, demonstrating strong performance.

The improvements in statistical parameters underscore the robustness of EQM in addressing and enhancing GCM biases. Additionally, the performance evaluation and ranking of the GCMs not only identify the most suitable models for future climate projections but also establish a strong foundation for analyzing relative changes in future scenarios.

Under both SSP2-4.5 and SSP5-8.5 scenarios, the average annual maximum temperature is projected to rise consistently, with the most significant increase observed in the FF period. The annual maximum temperature will increase by 0.42 °C in NF, 1.08 °C in MF, and 1.68 °C in FF under SSP2-4.5. A slightly higher increase of 0.46 °C in NF under SSP5-8.5 compared to SSP2-4.5, with a continuous upward trend observed at 1.51 °C in MF and 2.87 °C in FF. An upward trend is also evident in the average annual minimum temperature across all periods under both scenarios, with a more pronounced increase compared to the change in maximum temperature. Under SSP2-4.5, the minimum temperature is likely to rise by 0.51 °C in NF, 1.20 °C in MF, and 1.80 °C in FF. A greater increase is likely under SSP5-8.5, by 0.58 °C in NF, 1.66 °C in MF, and 3.22 °C in FF. The study of Promping & Tingsanchali (2022) in the UNRB align closely with the outcomes of this study, demonstrating a rising trajectory in projected maximum and minimum temperatures under the RCP4.5 and RCP8.5 scenarios within the CMIP5 framework (model preceding CMIP6). Consistent with these findings, this study's projections also corresponds with Foyhirun & Promping (2021), who reported temperature increases of approximately 2–3 °C in the UNRB under RCP4.5 and RCP8.5.

These projected changes in climate variables highlight the critical need to analyze specific metrics for the NRB, facilitating a more detailed understanding of future climate impacts.

Table 6 presents the metrics for future climate change, including precipitation, maximum temperature, and minimum temperature, across the NF, MF, and FF periods under SSP2-4.5 and SSP5-8.5 scenarios in the NRB, calculated using Equations (10) and (11). The trend of increasing precipitation and temperatures is evident across all periods, from NF to MF to FF, under both scenarios. There is a consistent rise in the average annual minimum temperature across all periods in both scenarios, showing a more significant increase than the change observed in maximum temperature. These projected increases in both maximum and minimum temperatures highlight a significant warming trend, which is crucial for understanding future climate impacts on the NRB. This study's findings align with Bastola et al. (2024), indicating that the NRB will experience wetter wet seasons, drier dry seasons, and rising temperatures. Similarly, Kiguchi et al. (2021) highlighted that managing water resources in Thailand is difficult due to the distinct contrast between the wet and dry seasons. They further emphasized that climate change will intensify these seasonal extremes, making water resource management even more challenging.

The projected increase in precipitation and temperatures across all periods and scenarios indicates a changing climate for the entire basin, offering a broad perspective of future conditions. However, this assessment must be supplemented with an evaluation of the specific vulnerabilities and responses of each sub-basin, as climate change impacts can vary significantly across the watershed. Assessing these vulnerabilities is vital for tailoring effective adaptation and mitigation strategies. Additionally, understanding potential land use changes under varying scenarios is essential for comprehensive watershed management, ensuring that these strategies are aligned with both climatic and land use shifts.

The Dyna-CLUE model was used to project land use changes, while the SPSS statistical software was employed for binary logistic regression analysis, a crucial step preceding data processing in Dyna-CLUE. Previous studies (Verburg et al. 2002; Trisurat et al. 2010; Shrestha et al. 2020) have demonstrated the use of logistic regression to identify significant relationships between driving factors and land use locations, validating results with ROC values. In this study, the analysis revealed significant relationships between the constant values of the driving factors in the regression model and the respective land use locations, with ROC values ranging between 0.761 and 0.962 for the five land use types, indicating good accuracy in prediction.

The land use model for the NRB was validated by simulating land use patterns for 2016, 2018, and 2020 using the 2009 land use map. Comparison with observed land use maps for the same years was conducted, and verification error was quantified using Kappa statistical analysis (K). Supplementary material, Appendix E displays validation results, comparing observed and simulated land use. Overall accuracy exceeds 90%, with Kappa coefficients of 0.89, 0.88, and 0.87 for 2016, 2018, and 2020, respectively, indicating strong agreement. A Kappa index exceeding 0.70 signifies a high level of agreement between the simulated and observed land use changes (Shrestha et al. 2018; Ghimire et al. 2021; Buhay Bucton et al. 2022; Pinsri et al. 2022).

The binary logistic regression analysis and model validation validate the Dyna-CLUE model's accuracy in predicting land use changes, with a high level of agreement between simulated and observed data demonstrating its reliability for future projections. Having validated the model's accuracy, projecting future land use changes across various scenarios will offer insights into potential transformations in the basin's landscape.

In both scenarios (Supplementary material, Appendix H), agricultural land exhibits a decreasing trend from the 2020 baseline, with the BAU scenario showing a more significant decline compared to RU. Specifically, under BAU, the rates of decrease are 3.16% (NF), 8.27% (MF), and 12.69% (FF), while under RU, the rates are 2.02% (NF), 4.78% (MF), and 7.62% (FF). Similarly, other land types also decrease, with RU showing lower rates of change compared to BAU: 0.03% (NF), 0.17% (MF), and 0.33% (FF) for RU, and 0.09% (NF), 1.40% (MF), and 4.23% (FF) for BAU.

Conversely, an increase in relative change from the 2020 baseline period is projected for urban and built-up land in both the BAU and RU scenarios, with RU demonstrating higher rates compared to BAU: 1.83% to 7.23% in RU and 0.18% to 1.12% in BAU. These findings suggest a more rapid pace of urbanization and development under the RU scenario compared to the BAU scenario. This result aligns with the findings of Promping & Tingsanchali (2022) in the UNRB, which reported a 5.27% increase in built-up area under the economic scenario and a 3.99% increase under the conservation scenario, illustrating similar trends of urban expansion under varying land use scenarios.

Forest land increases more significantly in the BAU scenario, with increments ranging from 3.08% (NF) to 9.24% (MF), and 15.81% (FF), compared to the RU scenario, which shows lower increases of 0.21% (NF), 0.45% (MF), and 0.72% (FF). While the rate of increase in forest land is higher in the BAU scenario compared to the RU scenario, both scenarios demonstrate the positive impact of Thailand's National Strategic Plan and National Forest Policy.

The projected land use changes under both BAU and RU scenarios reveal distinct trends in the NRB, with significant implications for urbanization, agricultural practices, and forest conservation. As these transformations unfold, they impact the expansion of impervious areas, which are critical for assessing the basin's future watershed vulnerability. Assessing the metrics of impervious area changes and their potential impact on the basin's hydrology is crucial, as it underscores the importance of understanding how significant landscape changes, particularly impervious area expansion, affect the basin.

The future impervious area change in the NRB was derived from land use projections under both BAU and RU scenarios. The projected impervious land area change metric for 2020 to 2100 (Table 7) is calculated using Equation (12). Impervious land is projected to expand significantly from 2020 to 2100, with BAU showing a change metric increase from 0.69 (NF), to 1.75 (MF), and 4.32 (FF), and RU showing an even higher escalation from 7.03 (NF), to 17.31 (MF), and 27.73 (FF). In both scenarios, there has been a noticeable shift toward impervious land in the regions bordering the protected zones.

The expansion of impervious areas in the FF period covers 1.39% and 7.49% of the total land area of the NRB under BAU and RU scenarios, respectively. When the proportion of impervious surfaces in a region reaches 10–15%, it results in a notable deterioration of river water quality and an increase in common pollutants like nitrogen, phosphorus, and heavy metals, ultimately causing a decline in the overall health of water bodies (Klein 1979; Brabec et al. 2002; Tasdighi et al. 2017; Dong et al. 2020a; Cheng et al. 2022). While the impervious land area in both scenarios remains below the threshold of 10–15% of the total basin area, this does not negate the vulnerability of sub-basins within the NRB to increased imperviousness. The imperative to identify and assess the vulnerability of each sub-basin to future impervious area changes underscores the necessity of tailored strategies addressing specific impacts, vulnerabilities, and stakeholder needs within the NRB.

The projected increase in impervious areas under both scenarios highlights significant shifts in land use that impact watershed vulnerability. Understanding these changes is crucial for developing tailored strategies to manage the specific challenges posed by impervious area expansion in the NRB. As these changes progress, examining how future water use patterns will change is essential for assessing their implications on the basin's hydrology and overall management.

The population projection utilizing the logistic growth method indicates a declining trend in the future population of the NRB (Supplementary material, Appendix I). Population declines in the North and Northeast regions are expected due to low fertility and out-migration, with the North projected to experience earlier declines than the Northeast based on migration patterns from 1985 to 1990 (Guest & Jones 1996). Future domestic water use is projected to decrease alongside declining population across future periods. The baseline annual domestic water use of 4.16 MCM is projected to decrease to 3.89 MCM/year (NF), 3.74 MCM/year (MF), and 3.61 MCM/year (FF), representing a decrease of 6.37% (NF), 9.95% (MF), and 13.26% (FF).

The future industrial water use is projected to increase in all periods, rising from a baseline of 0.62 MCM/year to 0.95 MCM/year in NF, 1.44 MCM/year in MF, and 1.93 MCM/year in the FF. This represents an increase in water use of 53.18%, 132.95%, and 212.73% in the NF, MF, and FF periods, respectively, compared to the baseline period. Agricultural water use exceeds domestic and industrial use, emphasizing agriculture's importance in the basin. In the BAU scenario, agricultural water use is projected to decline from 54.44 MCM/year to 41.35 MCM/year in the FF period, a 24.04% decrease. In the RU scenario, the decline is less pronounced, with agricultural water use decreasing to 46.58 MCM/year in the FF period, representing a 14.44% decrease from the baseline.

The future water use change metric is calculated using Equation (13). Results shown in Table 8 indicate a decreasing future water use change of 0.0542 in NF, 0.1318 in MF, and 0.2082 in FF from the baseline period under the BAU scenario. Similarly, under the RU scenario, there is a decreasing future water use change of 0.0342 in NF, 0.0766 in MF, and 0.1199 in FF. The decline in water use implies reduced water use within the NRB across the designated future timeframes, encompassing both the BAU and RU scenarios. While this overall decrease in water use may suggest reduced pressure on resources, it does not automatically indicate low watershed vulnerability. Understanding changes at the sub-basin level is crucial to assess localized risks and develop targeted management strategies. These water use changes need to be considered alongside climate change and land use shifts. This will lead into the assessment of sub-basin vulnerabilities, examining how these water use trends, combined with climate and land use changes, affect the NRB's overall watershed health and resilience.

Overall, the NRB shows high vulnerability to climate change in the future, indicating susceptibility to adverse impacts from shifts in climate variables. The increasing vulnerability to climate change highlights the significant risks facing the NRB, emphasizing the need to assess how changes in impervious areas might further impact sub-basin vulnerabilities and offer a comprehensive view of watershed resilience.

The analysis of future watershed vulnerability to impervious area changes indicates that while most sub-basins exhibit low vulnerability during the NF period, this shifts to increased vulnerability in the MF and FF periods, particularly under the RU scenario. This highlights the need to evaluate how impervious area changes interact with other stressors, emphasizing the importance of understanding their combined effects on watershed vulnerability. This insight is crucial for assessing how future water use changes will impact sub-basin vulnerabilities.

Vulnerability to domestic water use remains consistently high to very high across all sub-basins in the NF, MF, and FF periods, with exceptions in sub-basin 17 showing medium vulnerability in the MF period and low vulnerability in the FF period. Some sub-basins, particularly in the upper part of the NRB, experience a slight decrease in vulnerability of domestic water use, corresponding to a trend of decreasing population projections. Sub-basins 11 and 12 are identified as the most vulnerable to future domestic water use change across these three periods, followed by sub-basins 13, 14, and 15.

The vulnerability of the sub-basins to industrial water use increases from the NF to MF to FF periods. In the NF period, all sub-basins show low vulnerability, while in the MF period, most sub-basins exhibit medium to high vulnerability, except for sub-basins 11, 12, 16, and 17 with low vulnerability. In the FF period, sub-basins 1–10 are projected to be the most vulnerable to future industrial water use change, each with an industrial water use change metric of 1.0.

Overall, the vulnerability to future water use change across agricultural, domestic, and industrial sectors in most sub-basins is projected to range from medium to very high under both BAU (SSP2-4.5) and RU (SSP5-8.5) scenarios in the NF, and Under SSP2-4.5, sub-basin 14 exhibits the most vulnerability to water use variations in the NF period, while sub-basin 13 is the most vulnerable in both the MF and FF periods. Under SSP5-8.5, sub-basin 14 consistently emerges as the most vulnerable across all periods, with sub-basin 13 also being the most vulnerable during the MF period. In both scenarios, sub-basin 17 consistently exhibits the lowest vulnerability to future changes in water use during the FF period.

Based on the results, vulnerability to water use changes shows high risks in the agricultural and domestic sectors, with an increasing trend in industrial water use across periods. Notably, domestic water use remains consistently high in vulnerability. These findings highlight the significant risks posed by water use changes and emphasize the need to incorporate these insights with vulnerabilities identified in climate change and impervious area changes. This comprehensive integration of factors will underscore the overall watershed vulnerabilities.

Similar to SSP2-4.5, the overall watershed vulnerability index of sub-basins under SSP5-8.5 in the NF period shows low vulnerability, ranging from 0.00 to 0.20. In the MF period, sub-basins exhibit medium to high vulnerability indices, except for sub-basins 1 and 17, which demonstrate low vulnerability. In the FF period under the SSP5-8.5 scenario, most sub-basins indicate high to very high vulnerabilities, except for sub-basin 17, which exhibits a medium overall vulnerability index.

Overall, the watershed vulnerability to climate change, impervious area change, and water use change in the NF period is low under both SSP2-4.5 and SSP5-8.5 scenarios. The vulnerability index score increases from NF to MF to FF period under both scenarios. Under SSP2-4.5, sub-basin 14 consistently emerges as the most vulnerable during the NF and MF periods, with sub-basin 13 also being the most vulnerable during the MF period and sub-basin 10 during the FF period. Sub-basin 10 exhibits the highest vulnerability index with an overall vulnerability index of 1.0, followed by sub-basin 12 with a vulnerability index of 0.92, indicating very high vulnerability. Under SSP5-8.5, sub-basin 14 consistently emerges as the most vulnerable during the NF and FF periods, while sub-basin 4 is the most vulnerable during the MF period. Sub-basin 14 has an overall vulnerability index of 1.0 during the FF period, followed by sub-basins 4 and 5 with vulnerability indices of 0.98 and 0.96, respectively.

Sub-basin 14 includes Dan Sai (Loei), Chat Trakan, Nakhon Thai, Phrom Phiram, Wang Thong, Wat Bot (Phitsanulok), and Phichai, Thong Saen Khan (Uttaradit). Chuenchum et al. (2017) highlighted the increasing water demand in the LNRB, including the areas of Uttaradit and Phitsanulok provinces, driven by agriculture and intensified by climate change and economic development. These areas, supported by extensive irrigation systems, are projected to experience significant water deficits. This aligns with the findings of this study, where sub-basin 14 exhibits high vulnerability due to agricultural water demand, particularly during the NF (SSP2-4.5) and NF, MF, and FF periods (SSP5-8.5). As indicated in Supplementary material, Appendix K, the vulnerability of this sub-basin to climate change is very high, with a vulnerability score of 0.76 in the FF under SSP5-8.5. The challenges in managing water resources during the dry season, as noted by Chuenchum et al. (2017), underscore the urgent need for targeted water management strategies in these vulnerable areas.

Given the escalating vulnerability indices across the NF to MF to FF periods in both SSP2-4.5 and SSP5-8.5 scenarios, it is imperative to prioritize proactive measures for future protection, focusing on implementing targeted interventions in sub-basins with the highest vulnerability indices, to mitigate the potential impacts of climate change, impervious area change, and water use change. This comprehensive assessment integrates all key projection indicators, providing a complete view of watershed vulnerability.

The NRB faces significant challenges due to projected changes in climate, land use, and water use across various sectors. These changes necessitate adaptive management strategies to ensure sustainable water resource management over the NF, MF, and FF periods. Based on this study's findings, specific policies and strategies focusing on water conservation in agriculture and industry are outlined. By integrating climate projections, land use dynamics, and water use variations, these strategies aim to enhance resilience and support the long-term sustainability of the NRB's water resources.

The National Water Resources Management Strategies (2015–2026) and the 20-Year Master Plan on Water Resources Management (2018–2037) (http://www.onwr.go.th/) provide a structured framework for managing Thailand's water resources. These frameworks align with several adaptive strategies proposed in this study, which also support the objectives of Thailand's 20-Year National Strategic Plan (2018–2037). The NRB's adaptive strategies contribute to key national priorities, including water security for agriculture and industry (Strategy 2), flood management (Strategy 3), water quality improvement (Strategy 4), watershed conservation (Strategy 5), governance and decision-making (Strategy 6), and ensuring access to safe water through conservation and quality management (Strategy 1). These connections underscore the importance of integrating climate adaptation, land use planning, and sustainable water management into national and regional policies.

Farmers in the NRB have adopted various adaptation strategies, including adjusting farming calendars, practicing crop rotation, and enhancing irrigation systems (Arunrat et al. 2017). As climate variability increases, the adoption of advanced water-saving technologies becomes even more critical in ensuring agricultural sustainability. Techniques such as drip irrigation, precision farming, and Smart Irrigation Systems (SISs) are crucial for addressing these challenges by minimizing evaporation and runoff while ensuring efficient water application. These short-term adaptive measures not only enhance water use efficiency but also contribute to long-term water conservation efforts in the region. These approaches align with recommendations from Lakhiar et al. (2024) and Koontanakulvong (2024), emphasizing the early adoption and integration of innovative irrigation systems to build resilience.

Implementing such strategies during the NF period will enhance both short-term agricultural productivity and long-term resource efficiency, ensuring resilience against projected climate variability. Given the anticipated decrease in agricultural water use, these adaptive measures will help prevent unnecessary water consumption while supporting sustainable agricultural development. By adopting these strategies and establishing foundational policies in this period will lay the groundwork for effective water governance in later stages.

Beyond agriculture, initial industrial water conservation policies will also be crucial. The Thai government's 20-year water resource management plan includes expanding wastewater treatment facilities (Kanchanapiya & Tantisattayakul 2022) as a key strategy for addressing early-stage industrial water demands. Given the projected 53.18% increase in industrial water use, policies should focus on introducing incentives for industries to adopt water recycling and reuse technologies. These early interventions will set the stage for more comprehensive regulatory frameworks in the MF period.

One of the key adaptive strategies during the MF period is maintaining forest cover and controlling urban expansion. To achieve this, it is imperative to strengthen and implement Thailand's National Strategic Plan and National Forest Policy, which promote sustainable land use planning. As urban and built-up land use expands, enforcing land use policies will be essential in balancing development with environmental conservation. Additionally, establishing water user groups will facilitate knowledge transfer and empower farmers to make informed decisions, effectively utilizing water information systems, as highlighted by Koontanakulvong (2024). Strengthening community-based water management initiatives will ensure that localized adaptation strategies remain effective.

As industrial activities expand, leading to a projected 132.95% increase in water use during the MF period, implementing stricter regulations on both water use, and quality is essential. This period will focus on the active enforcement of policies developed in the NF period, ensuring compliance through monitoring and industry accountability. Many wastewater treatment facilities in Thailand remain non-upgraded and lack effective monitoring for optimal water reuse (Kanchanapiya & Tantisattayakul 2022). Mandating industry-wide adoption of advanced water conservation technologies will be critical in reducing the growing industrial water footprint. Strengthening public-private partnerships at the regional level will facilitate wastewater treatment upgrades and industrial best practices.

Enforcing the existing 20-year water resource management plan while developing extended long-term strategies is crucial for ensuring sustainable water governance during the FF period. These plans should integrate climate change projections, land use changes, and shifting water use patterns, while also adapting policies to the projected decline in agricultural and domestic water use and the continued rise in industrial demand. Additionally, strategies must include incentives for industries to continue innovating in water-saving technologies, while supporting the agricultural sector's transition from short-term irrigation efficiency (NF period) to drought-resistant crops and sustainable land management practices. This period will also require regulatory adjustments based on long-term industrial expansion trends, ensuring that policies remain relevant and adaptable.

To ensure the long-term success of water management strategies, it is crucial to establish national-level collaborative workspaces for water agencies, policymakers, and industry stakeholders. This broader-scale engagement will facilitate the exchange of advanced technological solutions, ensuring that water management practices remain adaptable to emerging challenges. Given the increasing complexity of water resource management in the FF period, fostering multi-sectoral collaboration will be essential in bridging gaps between policy, research, and implementation. Public-private partnerships should evolve beyond regional collaborations (MF period) to nationwide policy coordination (FF period), ensuring that water governance aligns with national economic and environmental objectives.

Engaging community stakeholders through integrated feedback mechanisms will ensure that policies remain inclusive and adaptive. Moreover, integrating digital platforms and decision-support systems will enable real-time data sharing, streamlining coordinated decision-making across multiple sectors. This long-term digital integration will strengthen adaptive capacity and support a sustainable transition toward smart water governance in the FF period.

This study evaluated the watershed vulnerability of the NRB in terms of future climate change, land use change, and water use change. Quantile mapping bias correction and statistical evaluation using performance parameters such as PBIAS, MAE, CRM, d′, NRMSE, and NSTDV were employed to identify the top performing CMIP6 GCMs. Utilizing the top six CMIP6 GCM models and two SSP scenarios (SSP2-4.5 and SSP5-8.5), projections indicate an overall increase in annual precipitation, maximum temperature, and minimum temperature across NF, MF, and FF periods under both SSP scenarios. Dyna-CLUE model was employed for land use change projections, achieving good accuracy in prediction with ROC values ranging between 0.761 and 0.962 for various land use types. Simulated maps showed strong agreement with observed land use patterns, with Kappa coefficients ranging from 0.87 to 0.89. These findings indicate a significant expansion of urban and built-up land, particularly under the RU scenario, as well as a notable increase in forest land across all future periods. Agricultural land and other land types are projected to decrease, with a higher rate of decrease observed in the BAU scenario during the FF period. Projected shifts in water use patterns indicate decreasing trends in agricultural and domestic water use, alongside an expected increase in industrial water use. These changes are attributed to shifts in land use patterns and population dynamics, with agricultural water use closely correlated with changes in agricultural land. Consequently, overall water use across all sectors is projected to decrease under both scenarios.

Further evaluation on the vulnerability of sub-basins within the NRB to these future changes was conducted in this study. Combining future climate change, impervious area change, and water use change, the overall vulnerability of the 17 sub-basins within the NRB remains at a low level during the NF under both SSP2-4.5 and SSP5-8.5 scenarios. This indicates that from 2020 to 2040, the combined effects of these factors are not expected to significantly increase vulnerability across the entire basin. However, the watershed vulnerability to these factors increases in the MF and FF periods. This suggests that as time progresses to 2100 and the effects of climate change, impervious area alteration, and shifts in water use intensify, the vulnerability of the watershed as a whole becomes more pronounced. By 2100, sub-basins 10 and 12 will face the highest vulnerability to climate change, impervious area expansion, and shifts in water use under SSP2-4.5. This highlights the need for targeted interventions in these areas, as they will experience heightened risks from the combined effects of future changes. Under SSP5-8.5, nine sub-basins are projected to have very high vulnerability, with sub-basins 4, 5, and 14 being the most vulnerable. This implies that these areas may experience similar challenges and require targeted interventions to reduce vulnerability.

To address future challenges, implementing advanced water-saving technologies and promoting policies for industrial water recycling and reuse will be critical in the NF period. In the MF period, strengthening sustainable land use planning and enforcing regulations on industrial water use and quality will be essential. For the FF period, developing long-term, adaptive water management strategies and fostering continuous learning and multi-sector collaboration will ensure resilience and sustainability. In summary, this study reveals significant future changes in climate, land use, and water use patterns in the NRB, with future projections assessing and identifying the vulnerability of sub-basins to future climate change, impervious area change, and water use change. This study's insights are crucial for prioritizing resources and guiding proactive planning and management strategies to tackle future challenges effectively.

This study marks a significant milestone in understanding watershed vulnerability within the NRB, providing valuable insights into future changes in climate, land use, and water use. However, to address the complexities of watershed vulnerability more comprehensively, future research should adopt an integrated approach. This approach should encompass ecological, economic, social, cultural, and policy factors to better assess and enhance resilience and adaptive capacity. This holistic strategy not only addresses underlying vulnerabilities but also bolsters adaptive capacity across multiple dimensions, thereby enhancing resilience. Such an approach is essential for effectively addressing the complexity of these impending changes and devising sustainable solutions.

The authors would like to acknowledge the Thai Meteorological Department (TMD), Royal Irrigation Department (RID), National Statistical Office (NSO), Department of Industrial Works (DIW), Provincial Waterworks Authority (PWA) of Thailand, and Climate Copernicus (European Union's Earth Observation Programme) for providing data accessible via their websites and digital formats for this research study. The authors also extend their gratitude to the reviewers and editors for their valuable comments, which have significantly enriched the quality of this paper with their expertise.

The first author would like to extend appreciation to the University of Science and Technology of Southern Philippines (USTP) for funding her study under the Faculty Development Program.

All relevant data are available from an online repository or repositories.

• Copernicus Climate Change Service, Climate Data Store (https://cds.climate.copernicus.eu) – for LULC maps, CMIP6 GCM models, and two SSP scenarios.

• APHRODITE (http://aphrodite.st.hirosaki-u.ac.jp/) – for missing precipitation and temperature values.

Additional datasets used in this study were obtained from the following government institutions of Thailand:

• Thai Meteorological Department (TMD)

• Royal Irrigation Department (RID)

• National Statistical Office (NSO)

• Department of Industrial Works (DIW)

• Provincial Waterworks Authority (PWA)

The authors declare there is no conflict.