The LP is located in the middle of the Yellow River basin (33°41′–41°16′N, 100°52′–114°31′E), with a total area of ​​approximately 6.4 × 10⁵ km², spanning the provinces of Gansu, Henan, Inner Mongolia, Ningxia, Qinghai, Shanxi, and Shaanxi in China (Figure 1). The overall topography of the LP presents a downward trend of high in the northwest and low in the southeast. It is located in the semi-humid and semi-arid regions of China, with typical temperate continental monsoon climate characteristics. The multi-year average temperature and precipitation are approximately 8°C and 400 mm, respectively. The heavy and concentrated rainstorms in summer, coupled with high evaporation, high sediment content in the water system, and loose loess soil, cause serious soil erosion. According to the national classification standard of China, we used the land use data set of 2000–2020 produced by Zhang et al. (2021), and divide the land use types of LP into grassland, cultivated, forest, and unused lands, construction land, and wetland and water areas. In recent decades, the implementation of restoration projects (such as the Grain for Green project), rapid urbanization, and global warming have significantly altered the vegetation cover and land structure of the LP (Song et al., 2014), impacting the regional ecological environment.

The fishing net (10 km × 10 km) was created using the processed NPP data, following which the NPP of each matrix unit was assigned using the partitioned table. Finally, the global spatial autocorrelation test was performed in ArcGIS on the NPP data of the LP from 2000 to 2020. If the results showed a significant global correlation, then a local spatial autocorrelation analysis was performed.

The annual mean value of vegetation NPP on the LP from 2000 to 2020 followed a fluctuating upward trend (p < 0.05), with a growth rate of 6.88  g C m − 2 y r − 1 . The increases prior to 2010 (3.61%) were more pronounced than those after (1.75%). The multi-year average NPP was 298.59  g C m − 2 y r − 1 . The total volume of NPP increased from 127.12 Tg (in 2000) to 212 Tg (in 2020), with an annual growth rate of 4.05  T g / y r . The mean NPP varied ​​in different years, with 2002 being the lowest at only 201.64  g C m − 2 y r − 1 , while the highest was in 2018. Before 2012, the average annual NPP of the LP was lower than the multi-year average; however, in 2012, it increased sharply to 338.13 g C m − 2 y r − 1 (Figure 2).

Three obvious mutation periods were more intense than other years, namely 2001 to 2002, 2011 to 2012, and 2017 to 2018. The average annual NPP of each province in the LP from 2000 to 2020 is listed in Table 2. Henan Province had the highest average annual NPP (391.57  g C m − 2 y r − 1 ) and the largest growth rate (+11.60  g C m − 2 y r − 1 ), followed by Shaanxi, Shanxi, Gansu, Ningxia, Inner Mongolia, and Qinghai (with a growth rate of 5.10  g C m − 2 y r − 1 ). The NPP of each province showed rapid growth from 2000 to 2020, and the time points of NPP mutation in the LP, aligned with the implementation of the Grain for Green project in 1999. All provinces actively responded to the Grain for Green project with rapid vegetation restoration on the LP, achieving remarkable results.

The inter-annual NPP of the LP presented a spatial distribution pattern of low in the northwest and high in the southeast. The areas with the highest NPP were situated primarily in the central and southeast LP (Figure 3). Approximately 16%–55% of the total area had an average annual NPP below 200  g C m − 2 y r − 1 , which decreased annually, but the regional annual average NPP did not vary and remained at 200–400  g C m − 2 y r − 1 . The proportion of NPP with an average of 400–600  g C m − 2 y r − 1 varied significantly between years, encompassing only 4% in 2000 compared with the 37% in 2020, demonstrating significant improvements over time (Figure 4).

Our results also indicated that before 2002, there were almost no areas with an average NPP above 600  g C m − 2 y r − 1 , which has gradually increased over the past 20 years. Therefore, the implementation of ecological construction projects (such as the project of grain for green project) has improved the vegetation on the LP and the ecological environment has achieved remarkable results.

The land cover variations impacted the terrestrial ecosystem structure. The variations in NPP of different vegetation types from 2000 to 2020 are illustrated in Figure 5. The mean NPP values ​​of different vegetation types (from high to low) were: forest (447.87 g C m − 2 y r − 1 ) > cropland (317.20 g C m − 2 y r − 1 ) > impervious surface (301.61 g C m − 2 y r − 1 ) > grassland (256.81 g C m − 2 y r − 1 ) > water (197.79 g C m − 2 y r − 1 ) > wetlands (162.56 g C m − 2 y r − 1 ) > unused land (194 g C m − 2 y r − 1 ). All types showed a significant increasing trend from 2000 to 2020. Among them, the annual average NPP of the forest was much higher than that of other land types, indicating that the forest ecosystem is a large carbon sink (Figure 5). Moreover, there were significant discrepancies in the total annual NPP among different vegetation types. The annual total NPP of cropland, forest, and grassland increased rapidly above 40 TgC, while the total NPP of unused land was much higher than that of the remaining land, which suggests the carbon sequestration potential of unused land is enormous.

To further explore the spatial-temporal variations and spatial anomalous clustering changes in vegetation NPP, we analyzed the responses of precipitation and temperature to vegetation NPP.

There was a positive correlation between NPP and precipitation in 94.09% of the LP, with only approximately 5.91% of the region revealing a negative correlation (Figure 8). A total of 58.92% of the regions had a significant positive correlation, mainly in Inner Mongolia, Ningxia and Shaanxi, southern Gansu and western Shanxi, and 35.17% of the regions showed significant irrelevance. Conversely, 5.90% of the regions showed no significant negative correlation, these areas were concentrated in Xi’an, southeastern Baoji in Shaanxi, and Sanmenxia and Luoyang in Henan, there were almost no significant negative correlations.

Compared with the variations in response to precipitation, the temperature influences on vegetation NPP in the LP were not obvious over many years. In terms of spatial distribution, 76.99% of the regions showed a positive correlation, while 23.01% showed a negative correlation. There were almost no significant positive or negative correlations in the LP, as demonstrated in Figure 9.

In the study area, 73.73% of the regions showed an insignificant positive correlation, mainly in the southeast of Gansu Province, Ningxia, and Shanxi Province. Conversely, 22.86% of the region showed an insignificant negative correlation, in Yan ‘an, Xianyang, and Xi ‘an in the Shaanxi Province (Figure 9). The results indicate the impact of temperature rise on vegetation NPP has not had significant effects.

To further quantitatively explore the contributing factors to the vegetation NPP on the LP, we used correlation analyses. We compared the p-values ​​of the correlation coefficients of precipitation and temperature; the least significant correlation coefficient was extracted as the highest value correlation coefficient (Figure 10). The two climatic factors (temperature and precipitation) on the LP and the highest significant NPP correlation revealed significant correlations in 48.96% of the regions, mainly in the northwest and northeastern Shanxi, with no significance in the southeast LP. A total of 48.96% of the NPP on LP was significantly influenced by temperature and precipitation, mainly in the northwest and northeast of Shanxi, while the influence of climatic factors was insignificant on NPP in the south and east of the LP (Figure 10A). On the LP, 86.31% of the area is heavily dependent on precipitation for vegetation NPP, whereas only 13.70% of the area relies mainly on temperature. Therefore, precipitation is the main climatic factor affecting variations in vegetation NPP on the LP (Figure 10B). Moreover, for provinces and cities, the areas mainly influenced by temperature were the Shaanxi Province (Xianyang, Tongchuan, Baoji, Xi’an), Guyuan City in Ningxia, Pingliang City in Gansu, Xining City in Qinghai, and Sanmenxia City in Henan, while other areas were mainly influenced by precipitation factors.

The interannual variation in NPP had “multi-peak” fluctuation increases since the implementation of the Grain for Green project on the LP. These peaks may be attributed to ecological engineering coupled with the effects of climate change and human activities. With global warming over the past 20 years, the climate of the LP has gradually increased in temperature and humidity; temperature and precipitation play an important role in vegetation growth (Sun et al., 2020). Additionally, many ecological construction projects have been implemented, including the natural forest protection project, the comprehensive management of small watersheds, and the Grain for Green Project. From artificial afforestation to natural restoration, the LP has an obvious “greening” trend (Li et al., 2017; Zhao et al., 2018; Deng and ShangGuan, 2021), thereby enhancing the carbon sequestration capacity of the vegetation. Our results are in line with previous findings (Feng et al., 2013; Jiang et al., 2019).

Interestingly, there are three distinct spikes in the NPP (2001–2002, 2011–2012, and 2017–2018). On the regional scale, the sudden increase in NPP from 2001 to 2002 corresponded with the pilot projects in Shaanxi and Gansu in 1999. After 2000, the pilot projects of returning farmland to forests were expanded to include 17 provinces in the central and western regions. The wave of forestry projects has continued, and vegetation growth improved as plants develop. The extreme NPP during 2011–2012 may be due to two factors, the implementation of a new round of Grain for Green Projects and sudden change in extreme temperatures which accelerated vegetation growth and abundance, improving NPP (Liu P. et al., 2022). In 2017, the State Council approved the conversion of Cropland to forests and grasslands in 17 provinces to expand the scale of agriculture and forestry in the region. The State Council also initiated rural revitalization and the mountains-rivers-forests-farmlands-lakes-grasslands program has favorably promoted the grand vision of “green waters and green mountains” (Li and Liu, 2022), which led to a sudden rise in NPP from 2017 to 2018. Moreover, the extreme precipitation levels correlated with NPP, especially the P-min with NPP, indicating that extreme precipitation may have strongly influenced the sudden increase in NPP (Figure 11). Similarly, at the provincial scale, the timing of sudden increases in vegetation NPP in each province also clustered with the three stages of distinct spikes. For example, the NPP in Shaanxi Province increased by 72.21  g C m − 2 y r − 1 from 2001 to 2002, indicating that increased vegetation effectively controlled soil erosion, reduced carbon loss from the ecosystem, and enhanced carbon fixation.

The annual average vegetation NPP on the LP had a significant gradient from southeast to northwest. The area with an NPP above 400  g C m − 2 y r − 1 gradually increased, and moved northward, with the main concentration in northern Shaanxi and southeastern Gansu and the Lüliang mountains in Shanxi. These results are in line with those of Zhang et al. (2016). The higher NPP locations are the key restoration areas of the early pilot sites of the Grain for Green project. The pilot areas focused on the sub-region with warm temperate deciduous oak forest, with relatively superior site conditions as identified through superb afforestation technology, to ensure vegetation restoration success. However, the poor natural conditions, lack of water resources, severe drought, and planting a vegetation monoculture (single species) have allowed unused land to occupy most of the LP, impacting afforestation attempts in Inner Mongolia, Gansu, and Ningxia, resulting in slow NPP growth. In general, a series of large-scale ecological construction projects implemented over the past 21 years has changed the ecological environment of the LP and promoted the sustainable development of the Yellow River Basin; the effect of the extensive vegetation restoration work was also outstanding.

Most previous studies have focused on the linear relationship and driving factors based on the “first-order effect,” ignoring the advantage of spatial autocorrelation which can remove the assumption of sample independence in classical statistics and performs well with potentially dependent samples. Presently, few studies have used the spatial autocorrelation method to analyze spatial variation in NPP (Ren et al., 2020). Therefore, we used the “second-order effect” via spatial grid to explore the spatial correlation and heterogeneity of NPP in the LP. Our results reveal that the NPP of the LP from 2000 to 2020 had a strong spatial positive correlation (p< 0.01), and an agglomeration effect, indicating that the ecosystems between the regions are not independent, yet interrelated with a tendency to gather. To avoid global autocorrelation masking local anomalies or instability, we explored the degree of local spatial aggregation on the LP and observed obvious polarization of cold and hot spots. The local spatial pattern of the LP is dominated by HH and LL clustering patterns, with a prominent hot spot area (p< 0.01), which is similar to previous research (Wang and Gong, 2022). The HH is dominated by forests and cultivated land with vast carbon storage and is located in the warm temperate zone, which allows for an effective carbon sequestration. Focused conversion of farmland to forests involved sloping farmland converted into forest and grassland; the net forest area increased to 4,726 km² in 2020 (Table 4). Forests promote the accumulation of forest carbon and areas with high forest cover have significantly higher NPP than other habitats. In the past 21 years, Ordos and Bayannur (cities in Inner Mongolia, northern Ningxia) and Baiyin City in the Gansu Province remain a LL district (cold spot), as previously indicated (Ren et al., 2020). The carbon sequestration services of ecosystems in the northwest of the LP remain low, with the fragile ecological environments impacting the success of vegetation restoration in this area.

The transfer matrix revealed the worst land categories (unused land and grassland) in Inner Mongolia, which accounted for 79.58% (Table 4) of this land category. In addition, the location is a temperate desert steppe zone with serious desertification of Mu Us sandy land. The ongoing coal mining in the Ordos City has also adversely impacted local vegetation and caused soil erosion.

Recently, the government has improved the area through ecological restoration, although vegetation rehabilitation remains slow (Wu et al., 2022).

To further explore the heterogeneity in the clustering patterns, we calculated the clustering patterns between counties and combined the annual average NPP of each county with LULC data to explore the abnormal changes between HH and LL and the significance levels of each stage.

The main reason for the disappearance of the LL in Ansai, Wuqi, Yanchang County, and Yan’an City in the Shaanxi Province and Shanxi (Xi and Yonghe County) from 2000 to 2005 was the transformation of sloping farmland into forest and grassland and of unused land into grassland, which increased the vegetation cover. The reason for the disappearance of the HH in Shanxi Province may be attributed to the conversion of grassland to cultivated land and the expansion of cultivated land to impervious surfaces, including the transfer of 76 km² in Yuanping County (from 2000 to 2005).

The main reason for the disappearance of HH from 2005 to 2010 was the transfer of cropland and grassland to impervious surfaces in some counties in Shanxi Province. For example, 11 km² of cropland and grassland in Gaoping County was changed to impervious surfaces. The expansion of urban construction has accelerated urbanization, thereby reducing the ability for vegetation to sequester carbon in many arable lands and grasslands. The emergence of LL is mainly the degradation of grassland. For example, the grassland was converted into unused land in Hangjin Banner in Inner Mongolia, creating an additional 90 km² of unused land from 2005 to 2010. The conversion of unused land to forest and grassland (in 2010–2015) was the main reason for the disappearance of the LL. Other clusters were less affected by land use changes. The disappearance of the LL in 2015–2020 and the expansion of the HH may be related to the transfer of grassland to cropland.

The local spatial anomaly of the NPP in the LP is driven by human-land relationship and natural factors. The effect of temperature on NPP was much smaller than that of precipitation, which is consistent with the results of Xie et al. (2014), however, Xie et al. (2014) did not provide the specific spatial distribution of NPP which is affected by climate indicators. This study revealed a positive correlation between NPP and precipitation in 94.09% on the LP, which indicates that precipitation has a strong synergy with plant photosynthesis and vegetation growth. The increased precipitation in the Qinling Mountains did not correspondingly increase the NPP or reduce vegetation resilience. Northwest China (especially Inner Mongolia, Gansu, and Shaanxi) is in arid and semi-arid areas. The lack of precipitation weakens the net primary production capacity of the vegetation, which might have caused consistent LL in the area. The quantitative results of the extreme value analysis revealed that the contribution of precipitation to NPP reached 86.31%; therefore, precipitation was the dominant climatic factor affecting variations in NPP in the LP. Combined with spatial autocorrelation analysis, the cold spot area was mainly influenced by precipitation, while the temperature had a stronger correlation with the hot spot area. This suggests that precipitation plays a decisive role in vegetation change in arid and semi-arid areas in the northwest of the LP, which is consistent with previous research on the impact of climate change on vegetation NPP in China (Ge et al., 2021). The southeast LP is in a warm temperate zone with abundant precipitation where vegetation growth is strongly affected by light conditions.

Our analysis of precipitation impacts identified a very close synergistic relationship between the interannual variation in NPP mean and minimum precipitation levels on the LP. There is both coordination and a tradeoff between the minimum and maximum precipitation.

The year of the tradeoff between the NPP and maximum precipitation almost coincides with the abnormal year in the local clustering model in the Shanxi and Henan provinces, which indicates that the heterogeneity of this spatial clustering model was affected by P-max (Figure 11). Hence, future studies should focus on the impact of extreme precipitation on vegetation restoration (Fischer et al., 2013).

The spatial autocorrelation model is based on the first law of geography and regularly applied in various industries. However, in ecology (especially in the application of remote sensing data) some mechanistic discussions have not been fully investigated. For example, Geoda and ArcGIS software is superior to previous ecological research which has not clearly defined data distance and the selection and rationality of thresholds, such as NPP. This study focused on a single variable in the spatial autocorrelations to reveal the spatial variation in vegetation NPP on the LP. Future studies should combine multivariate spatial correlation analysis with economic, natural, and other indicators to explore the high and low clustering relationship between the factors and vegetation NPP to advise effective ecological construction.

Global warming remains a major global challenge. As an ecologically fragile area, we focused on temperature and rainfall as the natural factors influencing LP and identified gaps in the land cover data in a specific year. There may be deficiencies in our selection of indicators. Previous research has suggested that altitude and species diversity can also impact vegetation NPP. In the future, we will increase the diversity of valuable indicators in our exploration of the driving forces in vegetation carbon sequestration capacity to achieve the dual-carbon goal in China and the world.

In general, the vegetation NPP in the LP possessed a strong spatial autocorrelation. This spatial heterogeneity and dependence are driven by both natural factors and human activities. The results reveal that extreme NPP are a response to natural factors and violent disturbances by human activities and clarify the driving mechanism underlying vegetation carbon sequestration on the LP, revealing the spatial dependence of vegetation NPP and relationships between regions to support ecological management plans and provide suggestions to formulate double carbon plans. Future studies and relevant government departments should focus on fragile areas for vegetation restoration and, using the results of this study, accurately identify these vulnerable locations. By exploring the driving factors in vegetation restoration in each county, the correlations between the vegetation restoration impacts on the adjacent areas in space can be comprehensively explored. The potential future risks are predicted so that measures can be taken to prevent them. Reasonable regulation of land resources should solve issues of optimal allocation of space resources, which is especially important in the northwest region which has a weak vegetation carbon sequestration capacity, due to the inferior local conditions for ecological construction. Our results support the realization of ecological protection and high-quality development in the LP.