In this study, 31 provinces including the autonomous regions and municipalities in China were selected as the research object, while the data of Taiwan Province, Hong Kong and Macao Special Administrative Regions were temporarily missing and not within the scope of the study. This study used the road data, DEM data, NPP data, meteorological data, socio-economic data, and nighttime light data for the analysis of poverty. Among them, the road data and DEM data with a spatial resolution of 90 m were obtained from the Geographical Information Monitoring Cloud Platform, which can be used to calculate the proportion of roads, mean elevation, and mean slope in the counties. As for the NPP data, the MOD17A3H data with a 500 m resolution in 2018 were obtained from NASA’s MODIS product website (https://ladsweb.Nascom.Nasa.gov). The meteorological data were obtained from the China Meteorological Data Network (http://data.cma.cn/), which used the statistics of national meteorological stations to estimate multi-year mean precipitation of each county.

Socio-economic data mainly contained the population density of each county, number of middle school students, number of primary school students, number of employees in rural areas, total power of machinery, grain output, GDP per capita, net income of rural residents, regional fiscal revenue, savings deposit balance of residents, total retail sales of social marketing, urbanization rate, number of beds in health institutions, number of tourists, total tourism income, gross industrial production, number of industrial enterprises above designated size, the total output value of farming, forestry, animal husbandry, and fishery. The original socio-economic data were collected from statistical yearbooks and statistical bulletins of counties and cities, and the Moving Range method was used to normalize the original data and control the data within a range of [0, 1] to reduce dimensional influence.

Nighttime light data were obtained from NOAA’s National Geophysical Data Center (http://ngdc.noaa.gov/eog/download.html). This study first downloaded and estimated annual nighttime light data in 2018 by using mean value method to synthesize monthly data, and then used annual nighttime light data in 2016 to mask annual nighttime light data in 2018. Finally, the Shanghai nighttime light data was used as the maximum threshold to denoise for annual nighttime light data in 2018. In addition, all geographic data were processed by geometric correction, registration and projection transformation.

2.2.1 Estimation for the average nighttime light

The total nighttime light data or average nighttime light data can reflect the regional light characteristics (Yu et al., 2015). This study adopted the average nighttime light index (ANLI) to evaluate the poverty level of counties. The formulas for calculating ANLI are as follows:

where TNLI refers to the total nighttime light index in a region; DN_i is the radiation value of each pixel in the region; and n is the number of pixels in the region. The higher the ANLI, the lower the poverty level in the county, and vice versa.

2.2.2 Sustainable livelihoods framework

Multidimensional poverty identification aims to judge whether there is a spatial poverty trap by analyzing the spatial endowment of various types of capital, which covers economic, social, natural, ecological, and other aspects. The sustainable livelihoods approach (SLA) is a multidimensional analysis framework established by the British International Development Agency in 2000, which mainly includes five aspects of sustainable livelihoods capitals, namely economy, nature, human, material, and society. This study used the SLA to evaluate multidimensional poverty from the above-mentioned sustainable livelihoods capitals as well as environmental/background vulnerability (Figure 1), attempting to portray the balance of various types of livelihood capital. On the basis of previous studies (Liu and Xu, 2015; Liu and Xu, 2016; Zhou et al., 2018; Ding et al., 2020), 19 indicator systems in six dimensions were finally determined in this study by considering the national strategy requirements of comprehensive poverty alleviation and following the principles of data availability, dynamics, and relevance. Among them, although grain output can represent both nature capital and material capital, this study classified it as material capital for the commercial nature of grain by referring to relevant literature (Li et al., 2018; Pan et al., 2018). Similarly, nighttime light data can not only reflect the economic capital, but also characterize the state of social activities. In this study, it was classified as society capital due to the principle of balance in the number of indicators selected in each dimension (Feng et al., 2018; Luo et al., 2020). On this basis, the analytic hierarchy process and entropy method were used to estimate the index weight (Table 1).

The identification of multidimensional poverty was based on county multidimensional development index (CMDI). First, the scores of each dimension were calculated separately. Then the polygon area method was used to calculate the areas from each dimension as there was a certain correlation between the six dimensions. In this manner, the CMDI was finally obtained. The principle of this method is that it supposes the scores of the ith county in the six dimensions as a, b, c, d, e, and f (Figure 1), and the angle between any two dimensions is α (α = 360°/6), then the area of the county multidimensional development model is calculated as follows:

As the different sorting methods for the six dimensions would result in various areas, the mean value was calculated for various possible results. Therefore, the final CMDI is calculated as follows:

In this study, the CMDI can not only represent the multidimensional poverty level of the county, but also reveal its sustainability and risk resistance (Liu and Xu, 2016). The higher the CMDI, the lower the multidimensional poverty, the stronger the sustainability and the ability to resist risks, and the better the multidimensional development of the county.

The relative poverty identification method was referred to the experience of multidimensional poverty index classification in Mexico as well as the measurement standards of relative poverty rate in the UK (Sun and Xia, 2019; Wang and Feng, 2020). The specific method is as follows. First, 60% of the median of each indicator was selected to calculate the score of each dimension, which was used as a criterion of relative poverty assessment. If the score of a county was higher than the criterion, it meant that there was non-relative poverty in this dimension, and vice versa. Then, counties with poverty in any three or more dimensions were defined as counties with severe relative poverty, whereas those with any three or less dimensions were defined as counties with light relative poverty, and the remainder were defined as counties with non-relative poverty.

2.2.3 Coupling coordination model

The coupling model is used to express the complex relationship of mutual influence between different things, and the degree of coupling can judge the degree of interaction between two things, but cannot judge whether the two things have high-level promotion or low-level inhibition interactions. Nevertheless, the degree of coupling coordination can provide the level of coordination while judging the degree of interaction (Yang et al., 2020). Therefore, this study constructed a coupling coordination model to explore the poverty alleviation pathways of poverty-stricken counties as follows, which was based on the industry development index, education development index, tourism development index, agriculture development index, and CMDI.

where u₁ refers to the CMDI of each county; u₂ is the industry/education/tourism/agriculture development index of each county; C is the degree of coupling, reflecting the spatial correlation between u₁ and u₂; T is the comprehensive coordination index, representing the contribution of u₁ and u₂ to the degree of coupling coordination; λ and μ are the undetermined coefficients, both of which are set to 0.5 (their sum is 1) according to the research objective of this study; D is the degree of coupling coordination, which reflects the level of coordinated development between u₁ and u₂. A high degree of coupling coordination reflects that the industry/education/tourism/agriculture development and county development are in a coordinated relationship, in which they could promote each other (Li et al., 2019). A low degree of coupling coordination reflects a restraining relationship, in which one restricts the development of the other or they restrict each other’s development, leading to a “development trap”.

After measuring the degree of coupling coordination, this study used the K-means clustering method (Ji et al., 2017) to divide counties into three categories, and obtained the classifying standards of the coupling coordination type (Table 2) by combining K-means clustering results and the existing research (Rong et al., 2016; Feng and Li, 2020). Finally, the counties were divided into three types including coordinated development type, imbalance type and recession type.

Based on the classification of coupling coordination types, four coupling coordinated difference types were divided according to the actual situation of the counties and the development differences between counties and development levels of industry/education/tourism/ agriculture in the counties (Table 3) (Rong et al., 2016; Ji et al., 2017; Li et al., 2019; Feng et al., 2020). By analyzing the coupling coordination types and the coupling coordinated difference types at the county level, the poverty alleviation pathway appropriate to each poverty-stricken county in China was selected.

3.1.1 Multidimensional poverty identification based on CMDI

This study identified the multidimensional poverty counties in China based on CMDI, and used the natural break point method to divide counties into five levels, namely, extreme poverty, poverty, general, rich, and extremely rich (Duclos and Tiberti, 2016). The spatial distribution of CMDI multidimensional poverty counties was shown in Figure 2. The 611 counties identified as poverty areas or extreme poverty areas, and defined as counties with multidimensional poverty identified by CMDI accounted for 25% of the total number of the counties. These counties were mainly located in northwestern China, among which, extreme poverty areas were mainly located in Gansu, Yunnan, Xinjiang, and Sichuan in southwest and northwest of the country, being 26%, 36%, 12%, and 11%, respectively. And 40% of poverty areas were concentrated in Hebei, Shanxi, Tibet, and Sichuan. Most counties in China were identified as general areas, mainly distributed in the central and eastern parts. In addition, rich areas and extremely rich areas were mainly distributed in Beijing, Shanghai, Guangzhou, and Shenzhen and their radiated provincial capitals or prefecture-level cities.

To verify the accuracy of multidimensional poverty identification based on CMDI, this study compared the results with the 585 national poverty-stricken counties in 2018. The results showed that 79% of national poverty-stricken counties were identified as counties with multidimensional poverty, while 132 national poverty-stricken counties, mainly in Henan, Hubei, and Inner Mongolia, could not be identified, among which, 67 counties had achieved poverty alleviation by the end of 2018. The reason for the national poverty-stricken counties not being identified as counties with multidimensional poverty was that the national evaluation put more weight on absolute poverty of income, which means although these counties may have income in poverty level, the development level may be relatively high in other dimensions which made them have high potential for overcoming poverty. For example, the national poverty-stricken counties in Tibet had improved the development level through poverty alleviation, which made them achieve complete poverty alleviation by the end of 2019. In addition, the counties which were identified as counties with multidimensional poverty but not listed in the national statistics were mainly distributed in Hebei, Sichuan, Xinjiang, Tibet, and Yunnan. The discrepancy in Hebei, Sichuan, and Yunnan can be attributed to their generally low scores for the six dimensions, and the discrepancy in Xinjiang and Tibet may be due to the poor development in one dimension leading to low scores for the six dimensions.

3.1.2 Multidimensional poverty identification based on ANLI

Nighttime light data can represent the degree of human social and economic development to a certain extent because it contains multidimensional information such as population and economy. Therefore, ANLI can be used to analyze the multidimensional poverty situation in China, and the spatial distribution of ANLI multidimensional poverty classification was shown in Figure 3. It showed that the areas with extreme poverty and poverty were mainly distributed in northwestern China, including a total of 602 counties, defined as counties with multidimensional poverty in China and identified by ANLI with approximately 55% of the counties located in Sichuan, Inner Mongolia, Tibet, and Xinjiang. Among them, extreme poverty areas accounted for 85%, 64%, 49%, and 47% of all counties in Tibet, Qinghai, Xinjiang, and Inner Mongolia, respectively. Poverty areas were mainly distributed in southwest and northeast China, with Yunnan, Sichuan, and Heilongjiang accounting for 11%, 10%, and 9% of the total counties with multidimensional poverty. In addition, rich areas and extremely rich areas were mainly distributed in Beijing, Shanghai, Guangzhou, Shenzhen, and their radiated provincial capital cities or prefecture-level cities.

With the development of nighttime remote sensing, scholars have begun to use nighttime light data to identify poverty in China (Yu et al., 2015; Pan et al., 2018; Luo et al., 2020). To evaluate the reliability of identifying poverty by nighttime light data, this study compared the counties with multidimensional poverty identified by ANLI with 585 national poverty-stricken counties in China in 2018, and found that 371 national poverty-stricken counties could be identified by ANLI, accounting for 63% of the total number of national poverty-stricken counties. The unidentified counties were distributed all over the country, mainly in the southwest and central regions: approximately 20% in Yunnan and Guizhou, and approximately 40% in Hebei, Henan, Shaanxi, and Shanxi. The discrepancy could be mainly attributed to the high population density and overall high power consumption over the central region, which may result in a higher ANLI. Counties with multidimensional poverty identified by ANLI but not listed in the national statistics were mainly concentrated in Heilongjiang, Inner Mongolia, Xinjiang, and Sichuan. The reason for these areas identified as multidimensional poverty by ANLI was that Inner Mongolia and Xinjiang were wide and sparsely populated areas that had low mean electricity consumption, as well as the overall electricity consumption in Sichuan and Heilongjiang was low, which led to low ANLI.

Besides, high consistency was found between counties with multidimensional poverty identified by ANLI and those identified by CMDI. Although there were certain differences in spatial distribution, the overall situation was roughly the same. Rich areas and extremely rich areas were distributed in Beijing, Shanghai, Guangzhou, and their surrounding areas, while Sichuan, Xinjiang, Tibet, and Yunnan were identified as being more susceptible to poverty by the two methods. However, most counties in Tibet were identified as extreme poverty areas and most counties in Yunnan were identified as poverty areas according to the results of ANLI, which contradicted with the classification results of CMDI. The reason was that CMDI comprehensively considered the development of counties more systematically, which not only included the social factors but also other factors, such as economics and materialization. Although Tibet had a low ANLI due to its large area and sparse population, its economic and material conditions were relatively good, which resulted in most counties in Tibet not identified as extreme poverty by CMDI. On the contrary, Yunnan had a large population and high electricity consumption which led to relatively high ANLI, but the economic and material conditions were poor, resulting in identification of many counties in the province as extreme poverty by CMDI.

3.1.3 Relative poverty identification in China

Based on the relative poverty identification method, counties with relative poverty in China were divided into three types including light relative poverty, severe relative poverty and non-relative poverty (Figure 4). The results showed that 60% of counties in China were in a state of relative poverty, among which, about 47% of counties were identified as light relative poverty and distributed throughout the country. There were 302 counties that were identified as severe relative poverty areas and were mainly distributed in northwestern China including Xinjiang, Qinghai, Tibet, and Gansu. However, these areas also had some counties such as Turpan county in Xinjiang, Yinchuan city in Ningxia, Guyuan and Tianshui cities in Gansu identified as non-relative poverty, which had an overall relatively good level of development. Moreover, counties with non-relative poverty were mainly distributed in southeastern China, concentrated in the Yangtze River Delta urban agglomeration.

To analyze the relative poverty status of counties in China more intuitively, the counties with multidimensional poverty identified by CMDI and the national poverty-stricken counties were used to compare with the counties identified as relative poverty. The results showed that the counties identified as relative poverty were basically those which were identified as multidimensional poverty based on CMDI, and only eight counties with multidimensional poverty including Pingle and Pingguo counties in Guangxi, Luancheng county in Hebei, and Chun’an county in Zhejiang were not identified as counties with relative poverty. Approximately 31% of the counties without multidimensional poverty were identified as counties with relative poverty, of which 29 counties were identified as counties with severe relative poverty, mainly located in Sichuan, Yunnan, and Xinjiang. The reason for the discrepancy may be that the development level of these counties was low in all dimensions, and more than three dimensions were below the relative poverty line. However, the development level of each dimension of those counties was relatively balanced, leading to a relatively high CMDI, so those counties were identified as counties without multidimensional poverty. Compared with the national statistics, 41% and 57% of the national poverty-stricken counties were identified as counties with severe relative poverty and counties with light relative poverty, respectively. And it was obvious that the national poverty-stricken counties identified as counties with light relative poverty had higher potential to overcome poverty. In addition, 768 counties not listed in the national statistics were identified as counties with light relative poverty. Above all, it was shown that the problem of relative poverty had become increasingly prominent in China in addition to facing absolute poverty and multidimensional poverty. Therefore, the state and government are suggested to pay more attention to relative poverty, conduct in-depth research, and formulate corresponding poverty alleviation policies.

3.1.4 Spatial autocorrelation analysis of multidimensional poverty

Both ANLI and CMDI can represent the degree of multidimensional poverty to a certain extent. The higher the value, the lower the degree of poverty. In order to further explore the spatial distribution of counties with multidimensional poverty in China, this study used Moran’s I and LISA to conduct global autocorrelation analysis and local autocorrelation analysis. During the global autocorrelation analysis, the Moran’s I values of ANLI and CMDI were 0.413 and 0.651, respectively, both of which passed the 5% significance test, indicating that the multidimensional poverty levels in counties had strong spatial correlation. As for the local autocorrelation analysis, the counties were divided into four types including LL, LH, HH, and HL. Among them, the analysis on the counties with multidimensional poverty identified by ANLI (Figure 5a) showed that 601 counties with LL type, which represented an agglomeration with high level of poverty, were distributed across the country, reflecting that the counties with multidimensional poverty in China were in agglomeration. Furthermore, 112 counties with HH type, which represented an agglomeration with low level of poverty, were concentrated in Beijing, Shanghai, Guangzhou, Shenzhen and the surrounding cities. The analysis on the counties with multidimensional poverty identified by CMDI (Figure 5b) showed that 505 counties with LL type, which represented poor multidimensional development and high levels of poverty, were concentrated in Xinjiang, Tibet, Qinghai, Gansu, Sichuan, Guizhou, and other regions. Furthermore, 226 counties with HH type, which represented better multidimensional development, were concentrated in coastal areas. Besides, both of the local autocorrelation analysis on CMDI and ANLI showed that most of the identified counties were with the LL and HH types, and the counties with LL type were mainly concentrated in the 14 contiguous destitute regions of the country.

3.2.1 Analysis of coupling coordination types

Although China has eliminated absolute income poverty, it still faces the problem of relative poverty and multidimensional poverty. And at present there are four main types of poverty alleviation pathways at county level in China including industry poverty alleviation, education poverty alleviation, tourism poverty alleviation and agriculture poverty alleviation. In order to select a suitable poverty alleviation pathway for each poverty county, this study took the counties with multidimensional poverty identified by CMDI and counties with severe relative poverty as the research objects, namely poverty-stricken counties, and then used the coupling coordination model to divide the coupling coordination types of poverty-stricken counties based on the county development level as well as the development level of industry, education, tourism, and agriculture in counties (Table 4 and Figure 6).

On the whole, the poverty-stricken counties with different coupling coordination types presented a spatial distribution state with large scale dispersion and small scale aggregation in space. Among them, approximately 33% of the counties had the type of coordinated development between county development and education development, whereas 15%, 16%, and 23% of the counties had the type of coordinated development between county development and industry development, tourism development, and agriculture development, respectively. Compared with others, the degree of coupling coordination between education development and county development were relatively high, which further showed the effectiveness of the implementation of compulsory education. As seen from the perspective of industry development, 85% of the poverty-stricken counties had the imbalance type between industry development and county development, and even 68% of the counties corresponded to the recession type, indicating that most poverty-stricken counties in China had unbalanced industry development and required further industrial adjustments. As seen from the perspective of education development, 67% of the poverty-stricken counties corresponded to imbalance type or recession type, mainly distributed in Guizhou and Gansu, which meant that it was important to pay more attention to the development of education in these areas. As seen from the perspective of tourism development, most poverty-stricken counties corresponded to the recession type, accounting for approximately 73%, scattered across the country, indicating that most of the poverty-stricken counties in China cannot properly balance the relationship between tourism development and county development. Therefore, it is necessary to explore the internal mechanism between tourism development and county development to determine whether the development of tourism can alleviate the poverty level of the county. As seen from the perspective of agriculture development, the poverty-stricken counties corresponded to the imbalance type accounted for approximately 55%, and had agglomeration in Lvliang Mountains, Liupan Mountains, and the mountainous areas on the western border of Yunnan. In general, China is still facing a problem that the overall development of a county is synchronized with the development of industry, education, tourism and agriculture.

3.2.2 Analysis of coupling coordinated difference types

Based on the coupling coordination types, the poverty-stricken counties in China were divided into four types according to the dividing principle of coupling coordinated difference types, including common development type, common lagging type, county development lagging type, and industry/education/tourism/agriculture development lagging type (Figure 7).

On the whole, the spatial distribution of the coupling coordinated difference types of poverty-stricken counties were scattered across the country. As seen from the perspective of coupling coordinated difference types of industry, only about 4% of the poverty-stricken counties corresponded to common development type, which meant that both the overall development and industry development of the counties were relatively good, and could promote each other. The poverty-stricken counties corresponded to common lagging type accounted for the highest proportion at approximately 37%, which meant that low levels of industry development restricted the county development in these counties. Moreover, ap proximately 32% of the poverty-stricken counties corresponded to county development lagging type, which meant that the low levels of county development restricted industry development to a certain extent. As seen from the perspective of coupling coordinated difference types of education, approximately 62% of the poverty-stricken counties corresponded to county development lagging type, mainly distributed in Tibet, Qinghai, and Sichuan, where had better education development compared with its overall county development. Approximately 15% of the poverty-stricken counties corresponded to common development type, mainly concentrated in Gansu, Hebei, and Heilongjiang, while counties in education development lagging type were mainly concentrated in Guangxi, Anhui, and Guizhou. As seen from the perspective of coupling coordinated difference types of tourism, the number of poverty-stricken counties in tourism development lagging type was the largest, accounting for approximately 45% and scattered throughout the country. The tourism development of these counties lagged its county development. As seen from the perspective of coupling coordinated difference types of agriculture, the poverty-stricken counties with common development type accounted for the smallest proportion, followed by agriculture development lagging type, indicating that the overall level of agriculture development in counties with poverty was relatively high. In general, balancing synchronous development within the county and preventing poverty caused by lagging development is an important issue need to be solved in poverty alleviation.

3.2.3 Selection for the poverty alleviation pathways

Combining the coupling coordination types and coupling coordinated difference types of poverty-stricken counties in China, the poverty alleviation pathway suitable for each poverty-stricken county was analyzed from the perspectives of industry poverty alleviation, education poverty alleviation, tourism poverty alleviation and agriculture poverty alleviation (Table 5).

Overall, 414, 172, 442, and 298 poverty-stricken counties were found to be suitable for industry poverty alleviation, education poverty alleviation, tourism poverty alleviation, and agriculture poverty alleviation in China, respectively, and their spatial distribution is shown in Figure 8. Counties suitable for industry poverty alleviation were distributed in a state of largely scattered and small clusters, scattered throughout the country but concentrated in contiguous poverty areas. Only 36 counties, including Baishui county in Shanxi, Leye county in Guangxi and Wushan county in Gansu, were in a coordinated state between industry development and county development, which were responded well to poverty alleviation measures. However, although the industries in these counties can promote county development, the development of industries was still in a state of lagging. Therefore, it is necessary to optimize the industrial layout through industrial integration and the creation of industrial parks. At the same time, it is also possible to increase the economic effect and accelerate industry development through actively develop characteristic industries. For example, Wushan county in Gansu Province, where all the mountains have hidden jade, can develop a characteristic jade industry chain by introducing new technologies, improving process flow, and corresponding to the market demand to achieve the purpose of industry poverty alleviation. Approximately 91% of the poverty-stricken counties were in a state of imbalance between county development and industry development. Their low level of industry development restricted county development. In order to alleviate the inhibitory effect of industry development on county development in these counties, it is necessary to adopt poverty alleviation methods such as optimizing industrial structure and developing characteristic industries to accelerate industry development, and select leading enterprises or cooperatives to accelerate economic development and solve the employment problem of poverty by building a corporate brand and enhancing corporate strength according to the characteristics of the county.

Counties suitable for education poverty alleviation were mainly concentrated in Tibet, Gansu, and Guizhou. Among them, 43 counties were in coordinated development type, while 129 counties were in imbalance type. The measures of education poverty alleviation mainly took poverty population as the core and tried to achieve human-oriented development and refined education, which was the fundamental approach to eliminating the intergenerational transmission of poverty. However, counties suitable for education poverty alleviation in China generally had problems such as conservative thinking and lacking of the emphasis on education. Therefore, it is necessary to increase awareness on the significance of education among the counties with poverty as well as reduce or exempt the cost of high school education or vocational education for the poor to improve their own development capabilities. In addition, the education should be fully integrated with the characteristics of the region in order to achieve targeted poverty reduction of education. For example, there was a shortage of talent in Tibet because of the harsh natural conditions, which made Tibet difficult to form a self-development education system even with the financial support of the state. Therefore, more attention should be paid to increasing the talents to support Tibet, so as to alleviate the low level of education. The primary education issue in Gansu and Xinjiang was that they should take their multi-ethnic characteristics into account, which required that their poverty alleviation of education should be appropriately tilted towards ethnic minorities, and promote bilingual education and national unity education. As for the education in Guizhou, the main problems were uneven development of education and numerous left-behind children, which made it necessary to increase educational assistance for left-behind children and address the issue of equity in education, as well as make full use of location advantages and absorb educational resources in the surrounding areas.

Counties suitable for tourism poverty alleviation had a large number and were widely distributed in China. Although the coupling coordination degree between tourism development and county development was poor compared with industry development, education development and agriculture development, the analysis on the coupling coordinated difference types showed that the tourism development had not been fully exploited in most poverty-stricken counties. As a result, the low level of tourism development restricted the development of the county, which made it necessary to promote the development of county by supporting the development of tourism. Among them, only about 11% of the counties had tourism development in a coordinated state with county development, mainly in the northwest and southwest regions, such as Wushi county in Xinjiang, Jinchuan county in Sichuan, and Longde county in Ningxia. The tourism development of these counties had a certain basis, which made it possible to facilitate tourism development by the assistance measures of exchange type, such as the training of tourism talents, scale expansion of the tourism, and the innovation of the tourism industry. However, there were 394 poverty-stricken counties, including Wuxi county in Chongqing, Wangqing county in Jilin, and Nanpi county in Hebei, were in a state of imbalance between tourism development and county development, which meant that the county development lagged behind tourism development due to insufficient development of tourism, thereby inhibiting county development. Therefore, the geographical location, natural resources, and human environment should be fully considered to develop tourism in these counties, besides, it is important to select appropriate tourism poverty alleviation pathways to maximize the effect of poverty alleviation. For example, counties with good traffic locations or accessible tourist attraction sites can rely on tourist routes to receive passing passengers to develop a “passing economy” and share the customer source. Counties with abundant natural resources can increase the stock of scenic spots, build tourism brands, and promote the tourism development of surrounding counties with the government financial assistance and leading enterprises. They could also set up tourist spots in surrounding areas, thus forming a fast-developing tourism industrial chain.

The spatial distribution of the counties suitable for agriculture poverty alleviation was relatively scattered with the obscure agglomeration effect. Only about 11% of the counties had agriculture development in a coordinated state with county development, and others were in an imbalance state. As the key to realizing rural revitalization, it is necessary to develop county-level superior agriculture and characteristic agriculture, and clarify whether the cause of agricultural backwardness is insufficient output or unsalable products to take corresponding measures to accelerate agriculture development. As for the counties in a coordinated state, although agriculture development was lagging, it still promoted the county development, which meant that the counties should focus on the use of innovative technologies and the construction of industrial chains, and realize the transition from “blood transfusion” poverty alleviation to “hematopoietic” poverty alleviation by developing agricultural mechanization and large-scale agriculture. As for the counties in a state of imbalance, their agriculture development restricted the county development, which meant that it was possible to rapidly alleviate the restriction by increasing government financial support and urging enterprises, cooperatives, and other institutions to play an active role. At the same time diversified agriculture poverty alleviation should be carried out by using multiple methods such as relocation, e-commerce, and skills training. In addition, high-end characteristic agriculture can be developed in some counties according to the local climate conditions to realize high-quality agriculture.

The targeted poverty alleviation in China is an in-depth and multidimensional comprehensive poverty alleviation development strategy. Accordingly, most poverty-stricken counties in China are suitable to adopt a variety of poverty alleviation method for poverty alleviation. This study extracted poverty-stricken counties suitable for one, two, three, and four poverty alleviation pathways, respectively. The results were shown in Figure 9. Except for 49 poverty-stricken counties requiring other alternate poverty alleviation pathways, 61% of the counties were suitable for the implementation of multi-pathways for poverty alleviation, reflecting the importance of multidimensional poverty alleviation. These counties should not only pay attention to the status of county development in poverty alleviation, but also analyze the links between multiple pathways of poverty alleviation, and formulate appropriate comprehensive poverty alleviation strategies. The results also showed that most of the counties were suitable for three kinds of poverty alleviation pathways. Among them, approximately 59% of the counties were suitable for industry poverty alleviation, tourism poverty alleviation, and agriculture poverty alleviation, which were mainly distributed in Tibet, Yunnan, and Guizhou. Such counties can promote agriculture development by developing rural tourism projects to promote the sales of characteristic and high-end agricultural products, as well as building beautiful villages, in order to accelerate the poverty alleviation rate of the county. The number of counties suitable for two poverty alleviation methods was relatively large. Among them, approximately 65% of the counties were suitable for the pathways of industry poverty alleviation and tourism poverty alleviation. Such counties can achieve the goal of poverty alleviation by building tourism brands to attract tourists, as well as increasing product promotion efforts to achieve the coordinated development of industries.

This study aimed to accurately explore the current situation of poverty at the county level in China and put forward corresponding poverty alleviation pathways. To this end, the average nighttime light index, county multidimensional development index, and relative poverty identification method were used to identify the poverty level of counties in China from the aspects of multidimensional poverty and relative poverty. Then, poverty alleviation pathways were explored by analyzing the coupling coordination types and coupling coordinated difference types from the four aspects of industry, education, tourism, and agriculture. The main conclusions are as follows:

(1) As seen from the results of multidimensional poverty identification, 602 and 611 counties with multidimensional poverty were identified by ANLI and CMDI, respectively. These counties accounted for 63% and 79% of the national poverty-stricken counties, indicating that CMDI can identify multidimensional poverty counties with higher accuracy. ANLI showed worse performance because it only took the brightness of nighttime light into account, whereas CMDI also considered the influences of multidimensional factors such as economy, materialization, and nature, which can reflect the sustainability and resistance of counties. The spatial distributions of counties with multidimensional poverty identified by the two methods were roughly the same, which showed that counties with multidimensional poverty in China were mainly concentrated in Xinjiang, Tibet, Qinghai, Sichuan, Yunnan, Gansu, Shanxi, Shaanxi, Jilin, Heilongjiang and Liaoning. Among them, the counties in Xinjiang, Qinghai, Yunnan, and Sichuan were generally in a higher poverty level, which made most of the counties remain in relative poverty and had a high risk of returning to poverty after successfully lifting out of poverty in 2020. Therefore, it is necessary for the government to pay more attention to these counties when formulating the next step of poverty alleviation policies. At the same time, it is important to increase support for the development of the northeast region in order to achieve the goal of overall national development. In addition, spatial correlation analysis showed that counties with multidimensional poverty were mainly concentrated in 14 contiguous destitute areas in China, which increased the difficulty of achieving rapid county development through state assistance.

(2) As seen from the results of relative poverty identification, 60% of the counties in China were under relative poverty. Among them, most of the counties were identified as light relative poverty and were scattered throughout the country. Counties with severe relative poverty were mainly concentrated in Xinjiang, Qinghai, Tibet, and Gansu. Counties with non-relative poverty were mainly distributed in southeastern China. Compared with the national poverty-stricken counties and counties with multidimensional poverty identified by CMDI, counties with relative poverty were found to basically consistent with the identified multidimensional poverty counties. And the national poverty-stricken counties, which were also identified as counties with light relative poverty, were found to have greater potential for overcoming poverty than those identified as counties with severe relative poverty. Therefore, it is also important for the government to place more emphasis on relative poverty and multidimensional poverty, and look for a scientific method of identifying relative poverty multidimensional poverty to implement corresponding poverty alleviation strategies based on the dominant core elements.

(3) Among counties with multidimensional poverty identified by CMDI and counties with severe relative poverty identified by the relative poverty identification method, 414, 172, 442, and 298 counties were found to be suitable for industry poverty alleviation, education poverty alleviation, tourism poverty alleviation, and agriculture poverty alleviation, respectively. The number of counties suitable for tourism poverty alleviation was relatively large. Compared with the development of industry, education, and agriculture, the degree of coupling coordination between tourism development and county development was relatively poor, but the analysis on coupling coordinated difference types showed that the tourism development restricted county development in most of the counties, which means that it is necessary to vigorously support tourism to promote county development. Tourism poverty alleviation is a highly feasible pathway for poverty alleviation in China at this stage, thus the government should pay more attention to tourism poverty alleviation, and increase the tourism competitiveness of a region by appropriately strengthening the degree of opening up of the region, the ability to attract foreign investment, and the support to the tourism industry. In addition, among the four poverty alleviation pathways, 61% of the counties were suitable for the implementation of multiple poverty alleviation pathways, which highlighted the importance of multidimensional poverty alleviation. Therefore, it is necessary to pay more attention to the comprehensive integration of multiple modes and form a higher-level comprehensive poverty alleviation mode through the integration of multiple poverty alleviation pathways in practice in order to accelerate regional development. Although poverty-stricken counties were clustered spatially, poverty alleviation pathways did not exhibit any significant spatial characteristics, which means that poverty alleviation strategies need to be implemented according to the specific circumstances of the county. Moreover, traditional poverty alleviation pathways have certain limitations, which means that poverty alleviation policies should be innovative under the premise of scientific deployment to accelerate the realization of the common prosperity goal.

The identification results showed that counties with severe relative poverty were mainly concentrated in Xinjiang, Qinghai, Tibet, and Gansu, which were basically consistent with the identification results of Fan et al. (2020). Counties with multidimensional poverty identified by CMDI in this study were mainly concentrated in Gansu, Yunnan, and Shanxi, which were basically consistent with the recognition results of Jin et al. (2020). The levels of multidimensional poverty at county level in China presented a stepped spatial distribution pattern which gradually increased from east to west with non-uniformity. In addition, this study also quantitatively analyzed poverty alleviation pathways suitable for different counties, and provided a reference for the sustainable development of poverty alleviation in China. However, as for the poverty alleviation pathways of counties in China, only the aspects of industry, education, tourism, and agriculture were considered in this study. Actually the suitable poverty alleviation modes for different counties were various. For example, counties suitable for agriculture poverty alleviation have significant differences in the types of favorable agricultural products. As for the counties suitable for tourism poverty alleviation, it is necessary to determine whether it is suitable for the farmhouse tourism or characteristic tourist attractions, or a mixture of multiple strategies is more suitable. As for the counties suitable for industry poverty alleviation, the industrial structure and type are different. As for the counties suitable for education poverty alleviation, the poverty alleviation efforts are also different. As the data for different industries and different agricultural products at the county level in China were unavailable, more specific poverty alleviation pathways were not considered in this study. Therefore, further studies should be carried out to determine specific poverty alleviation strategies according to the local regional characteristics in the future. In addition, the basic unit of poverty alleviation should be downscaled from the county level to the village level in order to implement poverty alleviation measures more efficiently and effectively when the relative poverty problem was to be resolved in local areas.