The UN's first Sustainable Development Goal (SDG) aims to end poverty in all its forms by 2030. This necessitates a comprehensive approach considering economic, environmental, social, and resource factors, moving beyond a single economic indicator to a multidimensional poverty index (MPI). While the Millennium Development Goals (MDGs) achieved significant progress, many countries still lag in multifaceted social indicators. China's success in poverty reduction, lifting over 700 million people out of poverty using targeted poverty alleviation strategies and poverty alleviation funds (PAFs), provides a valuable case study. However, future planning requires scenario analyses incorporating MPIs to inform policy decisions. This study addresses key questions: the global status of multidimensional poverty before the SDGs, future SSP trends in various regions (China, US, India, Middle East/Africa), and the future trend of MPI in China influenced by PAFs-SSPs. Measuring poverty accurately requires considering regional differences, spatial characteristics, and poverty drivers, hence the use of MPI, which considers health, education, and living standards. Nighttime light satellite imagery (NLS) offers an efficient method for monitoring poverty, especially in data-scarce regions. The study utilizes various machine learning techniques and multisource data (NLS, MODIS, statistical data) to analyze China's MPI from 2000-2017, predicting future trends under different shared socioeconomic pathways (SSPs).

Existing literature highlights the shift from single-dimensional to multidimensional poverty measures, emphasizing the need for comprehensive indices like the MPI developed by the Oxford Poverty and Human Development Initiative and the UNDP. Studies have explored the use of nighttime light satellite imagery (NLS) and machine learning for poverty monitoring and prediction, with examples including Li et al.'s (2019) work on identifying high-poverty counties in China using DMSP/OLS data and Yu et al.'s (2017) use of NPP/VIIRS data. The Shared Socioeconomic Pathways (SSPs) framework, with its five scenarios (SSP1-SSP5), provides a basis for future scenario modeling, though its application to MPI prediction is relatively limited. This research builds upon previous work by integrating MPI, PAFs, and SSPs using a data-driven model incorporating machine learning and Monte Carlo simulation to create a more robust and comprehensive analysis.

The study employs a hybrid model combining machine learning and a multidimensional weighted learning framework. The MPI weight is determined using a combination of entropy weight and Monte Carlo simulation, dynamically adjusting the weights based on the accuracy of identifying poverty alleviation counties. Data normalization and standardization are performed using equations (1) and (2). Information entropy is calculated (equation 3), and Monte Carlo simulation determines the index weights (equation 4). The relationship between MPI and nighttime light intensity is modeled using various machine learning methods (support vector machines, logistic regression, decision trees, random forest) to predict MPI from incomplete datasets (equation 6 and 7). The complex relationship between MPI and PAFs is defined by equation (8), learned from historical data using least squares fitting. PAFs-SSPs are modeled using Monte Carlo simulation, considering PAFs boundaries and growth rates in different provinces. Five SSP scenarios (SSP1-SSP5) are simulated, each with specific PAF allocation strategies reflecting different development priorities (equations 9-18). The study uses multisource data (NLS, MODIS, statistical data) to construct the MPI for 2369 counties in China from 1998-2020, encompassing environmental, resource, economic, and social aspects. A random forest model is used for MPI prediction due to data incompleteness. The G* natural fracture method is used for county classification. The Monte Carlo simulation incorporates uncertainty through three layers: total PAFs, random noise, and alteration of weight in different county aspects (SSP5).

The analysis reveals significant spatiotemporal variations in China's MPI from 1998 to 2020. The study found that the random forest method showed the best performance in predicting MPI, compared to other machine learning methods. Comparing the MPI of four regions (China, the US, India, and Middle East/Africa) under different SSPs, the findings showed that China's MPI growth rate is higher than developing countries and the US. This suggests that China and the US might have similar MPIs in the future. The rapid growth rate of China's MPI underscores the effectiveness of its poverty alleviation policies. A comparison with India and the Middle East/Africa reveals that these regions require at least 10-14 more years to eliminate poverty compared to China's 2020 achievement. Analyzing the spatial distribution of MPI under SSP1 in 2012, 2025, 2035, and 2050 shows a significant reduction in extreme and general poverty. The Monte Carlo simulation (107 simulations) indicates that the economic dimension weight is highest (0.467 ± 0.033), followed by social welfare, resources, and environment. Simulations across different SSPs demonstrate varying trajectories of poverty reduction. SSP1 achieves a balance between development and regional equity, while SSP2 leads to increased regional disparities. SSP3 minimizes regional differences but requires high PAF investment. SSP4 achieves the highest overall development but with high regional inequality. SSP5 minimizes PAFs but at the cost of environmental damage. The analysis also reveals that SSP1 is the most suitable pathway for China to achieve the SDGs. The analysis of the future investment in PAFs and the distribution of MPI indicates that SSP1, SSP4, and SSP5 can reach the developed county goal in 2025, with SSP1 showing the minimum difference in regional balance. SSP2 shows the best overall development level, but with intensified regional differences.

The findings highlight the importance of data-driven, multidimensional approaches to poverty alleviation. The integrated model incorporating PAFs, MPIs, and SSPs offers a valuable tool for long-term planning and policy formulation. The success of China's targeted poverty alleviation strategy underscores the potential for replicating this approach in other developing countries. The significant differences in MPI trajectories under different SSPs emphasize the importance of policy choices in achieving SDG targets. The study's framework can inform the design of targeted interventions and resource allocation strategies to address regional disparities and achieve sustainable poverty reduction. The relative success of SSP1 in balancing development and equity provides valuable insights for policy makers. The study also highlights the impact of external factors, such as the COVID-19 pandemic, on global poverty reduction efforts.

This study presents a novel data-driven model for predicting multidimensional poverty trajectories under various development scenarios, using China as a case study. The findings demonstrate the crucial role of poverty alleviation funds and highlight the importance of adopting an integrated strategy that considers multiple dimensions of development. The proposed model framework, combining machine learning, spatial statistics, and scenario analysis, offers a valuable tool for policymakers in designing effective poverty alleviation strategies. Future research could focus on incorporating additional data sources, refining the model to account for uncertainty, and applying the framework to other developing countries.

The study's future tendency predictions for MPI have some limitations due to data availability. At the national level, only two factors (economy and environment) were considered due to data limitations. At the county level analysis, village-level data was lacking. Future work should aim to incorporate more data to improve the model's accuracy and extend the analysis to lower geographical levels.