Meteorological data are collected from 1915 stations nationwide, of which 335 stations for 1949-1996 are provided by the Institute of Remote Sensing and Digital Earth of the Chinese Academy of Sciences. The data from 1540 stations covering 1981-1990 have been obtained by inputting the national meteorological data compilation. The data for 10 stations in Taiwan are provided by the National Weather Service, which is the aggregated average data from 1949 to 1980. Data for some stations in Tibet and Kunlun Mountains (30 sites) are taken from the China Climate Atlas. All the above data are sorted and inspected to form an original database. The geographical coordinates of each meteorological site are obtained from the “District station Number List of National Meteorological Station” published by the National Meteorological Administration in September 1990.

(1)Thornthwaite method

The formula of the Thornthwaite method for calculating the potential evapotranspiration is as follows (Zhou et al., 2002; Fisher et al., 2011):

In these formulas, E represents the potential evapotranspiration (mm). t is the monthly average temperature (°C). α is a constant that varies from place to place. I is 12-month comprehensive heat index. PE is the corrected potential evapotranspiration. CF is the coefficient of the number of days of duration and the number of days per month depending on the latitude, which can be obtained by looking up the appropriate table.

(2) Wetting Index

In these formulas, IM represents the wetting index, P represents the annual precipitation (mm), and S and D are corrected values.

China Knowledge Infrastructure Engineering (CNKI) includes a full range of published resources including Chinese journals, masters and doctoral theses, newspapers, conference papers, yearbooks, encyclopedias, patents, standards, Frontier journals, economic information and political bulletins. As of July 13, 2019, the China Academic Journals full- text database (CAJ) included a total of 10454 journals, 2023587 issues and 66437162 articles.

This study establishes a standard toponym database according to the 2012 edition of the county-level administrative division map provided by the National Map Publishing House. The system of Chinese keywords used in this research contains 12 keywords (Fig. 1). Taking the historical evolution of administrative divisions into account, data for some toponyms have been revised and supplemented according to the 1:250000 basic geographic database provided by the State Bureau of Surveying and Mapping or the county-level administrative division database provided by the Earth System Science Data Sharing Platform of the Ministry of Science and Technology.

The Thornthwaite method is first used to calculate the Wetting Index (IM) to divide the study area into arid, semi-arid, and sub-humid arid regions (McCabe et al., 1990). Regarding the China Knowledge Network (http://www.cnki.net/) as data source, this study summarizes 12 keywords about desertification research, and automatically searches all the journal articles from the 1980-2018 based on the professional code of the web crawler. After the web page analysis, the titles, abstracts, keywords, author names and unit information of these articles are obtained, and this text information is downloaded and saved to the local SQLite database in real time. The area covered in the paper is spatially located after the Chinese word segmentation as well as toponym matching that relies on the abstracts of the 253, 055 desertification research papers retrieved. Then, the spatial statistics of all the localized articles are determined, and the research hotspots index is calculated for all the counties in the study area.

Finally, the spatial distribution thematic map of the desertification research hotspots is obtained. Chinese word segmentation (CWS) is an important task of Chinese natural language processing (Liu, Wu, et al., 2019). We use the model of Ansj to segment abstracts and use the model of Han LP to identify toponyms based on the results of word segmentation (Hu et al., 2017). The KMP algorithm (KMP: Knuth-Morris-Pratt) (Li et al., 2016) can realize character-level fuzzy matching based on the above-mentioned identified toponyms and standard toponyms in the spatial database, and it ultimately counts and records the number of occurrences in each district. After correcting the standardized comprehensive hot regions index, the Normalized Comprehensive Hotspots Index applicable to desertification research is obtained for describing the spatial distribution of the hotspots of desertification research of China from the 1980s-2018. At the same time, the trend analysis is carried out on the GEE platform to capture the dynamic changes of desertification research hotspots. The detailed workflow is shown in Fig. 2.

Firstly, we calculate the “Absolute Hotspots Index” and the “Relative Hotspots Index”. The Absolute Hotspots Index is the number of occurrences of a county in the 253055 papers retrieved after defining the theme as desertification (Hu et al., 2017). In fact, due to different standards of research, economy and different scales of scientific research personnel in different counties, some places have much higher research intensity than their natural intensity of desertification (Lu et al., 2014). Therefore, this study uses the Relative Hotspots Index to eliminate the statistical bias caused by this “information gap” to some extent. The formula is as follows:

in the formula, R represents the Relative Hotspots Index. N_gd is equal to the Absolute Hotspots Index, which is the number of occurrences of a county in the 253055 papers retrieved when the topic is desertification. N_all is the number of occurrences of a county in the 253055 papers retrieved when the topic is unlimited.

Previous studies have shown that the use of either the Absolute Hotspots Index or the Relative Hotspots Index alone to evaluate research hotspots cannot effectively describe their spatial distribution patterns and dynamic processes (Hu et al., 2017). In order to generalize the evaluation index, we normalized the values of both N_gd and R to be between 0 and 1. Then we use the field calculator in ArcGIS to perform overlay analysis based on the above normalized results. After the above calculation, we obtain the map of “Normalized Comprehensive Hotspots Index” (NCH Index). The formulas are as follows:

in these formulas, NCH represents the Normalized Comprehensive Hotspots Index whose value is between 0 and 1. O is the overlay result of the Absolute Hotspots Index and the Relative Hotspots Index. max_O is the maximum value of O, and max_O is the minimum value of O. ${{N}_{g{{d}_{S}}}}$ is the normalized result of the Absolute Hotspots Index whose value is between 0 and 1. R_S is the normalized result of the Relative Hotspots Index whose value is between 0 and 1. $\text{ma}{{\text{x}}_{{{N}_{gd}}}}$ is the maximum value of N_gd, and $\text{mi}{{\text{n}}_{{{N}_{gd}}}}$ is the minimum value of N_gd ^. max_R is the maximum value of R, and min_R is the minimum value of R.

In order to analyze the dynamic changes of research hotspots for desertification in China for the 1980-2018, we conducted the trend analysis on the Google Earth Engine platform based on the NCH Index. We use a one-dimensional linear regression method to simulate the trend of the NCH Index in the past forty years pixel by pixel, and take the linear regression slope of the NCH Index and time (i) as the characteristic trend of the index (Reynolds et al., 2007; Zhang et al., 2008). Compared with the method of simply comparing the differences between a field in two-time periods, the trend analysis method used here can eliminate the influence of individual cases in a specific year, and more objectively reflect the research progress and trends of desertification over a continuous period of time (Weller et al., 2019). The formula for the slope of the NCH Index is as follows (Hu et al., 2018):

in the formula, β_Slope is the slope of NCH Index. When β_Slope is above 0, it means the NCH Index rises. When β_Slope is below 0, it means that the NCH Index drops. When β_Slope is equal to 0, it means the NCH Index is unchanged. Y_i is the value of the NCH Index in year i. n represents the number of years for the monitoring period (n=4). In order to verify the confidence of the regression model, the P-value is calculated.

Null Hypothesis (H₀): Y₁=Y₂ (∀Y₁, Y₂)

If H₀ is rejected at 90% confidence, it would prove that the value of the NCH Index changes significantly year by year when P < 0.1. On the contrary, if H₀ is not rejected, then there are no significant changes. By means of ArcGIS, we performed overlay analysis on the layer of β_Slope and the layer of the P-values with the raster calculator. Then we obtained a change trend map divided into four categories: “increased significantly”, “increased”, “decreased significantly” and “decreased”.

China’s typical desert and desertification land is mainly distributed in arid, semiarid, and some semi-humid regions of Northern China, north of 35°N (Yao et al., 2019). The study area covers 18 provinces, including Inner Mongolia Autonomous Region, Northern Shaanxi Province, Northern Qinghai Province, Gansu Province, and Xinjiang Uygur Autonomous Region; and 781 counties, such as Linger County, Horqin District, Aru Kerqin Banner, Yuzhong County, Qilian County and Yanchi County (Fig. 3). We calculate the wetting index according to Eq. (1)-(7), and then map the climate type of desertification in China (Fig. 3), dividing the Chinese land into arid, semi-arid, sub-humid arid (dry semi-humid), wet sub-humid, wet, humid and over-wet. The sub-humid arid zone refers to the relatively arid part of the semi-humid zone, and it is also called the dry semi-humid zone (Zhou et al., 2002). We only discuss the potential areas of desertification in China here, namely the arid, semi-arid and sub-humid arid areas.

According to Fig. 3, the arid areas are concentrated in the south of the Tianshan Mountains, the foothills of the Kunlun Mountains, the north of the Qilian Mountains, and the vast areas in the northwestern part of the Qinghai-Tibet Plateau, and they are mainly composed of deserts and arid deserts. The eastern part of the semi-arid area is composed of typical grassland and desert grassland. After entering the Qinghai-Tibet Plateau, it becomes an alpine grassland and an alpine desert. Most of northern Xinjiang is a semi-arid area, mainly composed of desert and semi-desert. The Hulun Buir Sandy Land in the south of Greater Khingan Mountains in the sub-humid arid area is close to the boundary between the typical grassland and the meadow grassland. After crossing the northern part of the Loess Plateau, it continues west along the northern edge of the Qinghai-Tibet Plateau, bypasses the Qaidam Basin to the south, and then reaches the southwestern part of the Qinghai-Tibet Plateau. A small number of island-like areas belonging to the sub-humid arid regions are located in the northwestern, southwestern and Hainan provinces.

Analysis of the distribution of the absolute hotspots index of desertification for the 1980s-2018 (Fig. 4) reveals that the research hotspots for desertification are mainly distributed in the western, northern and northwestern regions of northeastern China. Nearly all districts and counties in Xinjiang Uygur Autonomous Region and Inner Mongolia Autonomous Region have reached extreme levels, and the research results of Gansu Province, Qinghai Province and Ningxia Hui Autonomous Region are also quite rich. There are some desertification research results in Southwest China, Beijing, Tianjin, Shanxi, Hebei and Northeast China. Obviously, this analysis reveals that the spatial distribution of the research is somewhat different from our traditional understanding of the spatial distribution of desertification in China.

Analysis of the relative hotspots index distribution for China's desertification research (Fig. 5) reveals that the research hotspots for desertification are mainly distributed in the western, northern and western parts of northeast China. Desertification research in the central and eastern parts of Inner Mongolia, Hulun Buir Sandy Land, Ningxia Region, Qaidam Basin and the southeastern foothills of the Kunlun Mountains to the Himalayas is the most abundant. In fact, the desertification land on the Qinghai-Tibet Plateau is mainly concentrated in the western and northern parts of the plateau (Li, Zhang, et al., 2018), and the research on desertification in Ningxia is mainly concentrated in the southeast. Therefore, considering the Relative Hotspots Index alone will cause some gaps in the research hotspots in some regions. However, the spatial distribution pattern of desertification research formed by the Relative Hotspots Index is generally consistent with our traditional understanding of the spatial distribution, and it can also reflect some differences among counties.

Analysis of the NCH Index distribution of China’s desertification research (Fig. 6) reveals that the hotspots of China’s desertification research for the 1980s-2018 are mainly distributed in five regions: the desert and desertification areas in Hulun Buir of Inner Mongolia (including Hulun Buir Sandy Land, Mu Us Sandy Land, Kubuqi Desert, Hunshandake Sandy Land, Horqin Sandy Land, etc.), the desertification zone of Qaidam Basin in the western alpine zone, the central Tarim Basin the area extending to the foothills of the Kunlun Mountains, and the south foothills of the Altai Mountains with the sporadic areas around the Tianshan Mountains. These areas are mainly distributed in the agro-pastoral ecotone or grassland zone of the sub-humid arid and semi-arid regions, the desertification area in the western alpine zone and the oasis-desert ecotone in the arid zone. Among them, the level of research hotspots on desertification in the agro-pastoral ecotone in central and eastern Inner Mongolia is the highest, so it is the region with the most serious desertification development in China and the most urgent need for governance. The reason why desertification land research hotspots are widely distributed in the sub-humid arid and semi-arid regions is that the environment here has natural factors for the potential development of desertification, and human activities are also frequent. This research shows that the NCH Index is superior to the Absolute Hotspots Index and Relative Hotspots Index in reflecting the differences in desertification research among counties, and at the same time it also ensures that the results are highly consistent with the spatial distribution pattern of our traditional understanding.

According to Fig. 7 and Fig. 8, since the 1980s, the number of counties with a level of severe or above in the study area has been reduced from 152 in the 1980s to only 37 in 2018, and the total area has been reduced from 1.226 million km² to only 307700 km². Different patterns also emerge if we divide the full time-span into different periods.

(1) Analysis of the data in the 1980-1990 shows that 56 counties reached the extreme level, mainly those in the agro-pastoral zone in central and eastern Inner Mongolia. Another 96 counties reached the severe level, involving Inner Mongolia Autonomous Region, Xinjiang Uygur Autonomous Region, Gansu, Ningxia and others. In Xinjiang, except for the sporadic distribution near the Tianshan Mountains, most of them were at the moderate level.

(2) Analysis of the data for 1990-2000 shows that 22 counties reached the extreme level, concentrated in the Horqin Sandy Land in Inner Mongolia Autonomous Region, and another 194 counties only reached the weak level. In the Xinjiang Uygur Autonomous Region, except for the values of NCH Index in the desertification zone of the Tianshan Mountains and the Tarim Basin, the values in most parts of Xinjiang were between 0.2 and 0.4.

(3) Analysis of the data for 2000-2010 shows that the number of desertification research studies in this period has generally increased compared to earlier periods. The main reason is that since 2002, the country has vigorously carried out the “Three North” shelter forest project, the “returning farmland to forest (grass)” project, the “Beijing-Tianjin sand source control” project, and others. During this period, 76 counties reached the extreme level, mainly distributed in the Qaidam Basin, the middle and eastern areas of Inner Mongolia, and the agro-pastoral area. Another 118 counties were at the moderate level, mainly concentrated in the Xinjiang Uygur Autonomous Region. The hotspots level of desertification research in Inner Mongolia remained at or above the level of 0.6-0.8.

(4) Analysis of the data for 2010-2018 shows that 481 counties are considered to be at the negligible level, and 196 counties are at the moderate level. Except for the NCH of Yanchi County, which is above 0.8, and the NCH values of Hulun Buir Sandy Land, Central and Eastern Inner Mongolia, and the Tianshan Mountains which reached 0.6, most areas had quite low values. One possible reason is that compared with 2000-2010, the 10-year “Beijing-Tianjin Sandstorm Source Control” project in Beijing, Tianjin, Hebei, Shanxi and Inner Mongolia lasted from 2010 to 2018, so this time period is close to the end of the project (Qi et al., 2006).

Analysis of the trends of NCH in desertification research for the 1980-2018 (Fig. 9) shows that since the 1980s, the hotspots of desertification research in 447 counties decreased, covering an area of 2524600 km² (51.80% of the study area). The hotspots of desertification research in 172 counties increased, covering an area of 1.627×10⁶ km² (37.39% of the study area). The NCH Index values decreased significantly for 219 counties located in Gansu, Shanxi, Shandong, Hebei, Inner Mongolia and other provinces, covering an area of 451200 km² (10.37% of the study area). The hotspots of desertification research in 8 counties located in Tibet and Ningxia increased significantly, covering an area of 19000 km² (0.44% of the study area). According to the statistics (Zhou, 2019), during 1975-2017, the desert and desertification areas in the northwest were mainly reduced. The desert and desertification areas of the Qinghai-Tibet Plateau and the North China-Loess Plateau mostly remained stable. The desert and desertification areas of Inner Mongolia-Greater Khingan Range increased in the eastern part and declined in the central part. During 2000-2017, the severe desertification areas of the Loess Plateau gradually decreased, and the mild and non-desert areas were extended to the northwest (Liu et al., 2019). This study found that from the 1980s to 2018, the hotspots of desertification research in the southern foothills of the Altai Mountains in Xinjiang, the scattered areas around the Tianshan Mountains, and the central part of the Tarim Basin to the extension of the Kunlun Mountains were all generally increasing. The research on desertification in the Qinghai-Tibet Plateau and the Qaidam Basin maintained an upward trend. The research intensity in central and eastern Inner Mongolia shows a downward trend. The reason for the mostly declining research intensity of desertification in the sub-humid arid and semi-arid areas in Inner Mongolia is that many projects have started since 2002. We can conclude that in recent years, the research on desert and desertification areas in the northwest has achieved remarkable effects, while the research on desertification areas in the Qinghai-Tibet Plateau and the Loess Plateau in North China has achieved significant effects. The central part of Inner Mongolia-Greater Khingan Range has also achieved significant governance effects, but the eastern region still needs to strengthen its governance.