Mapping the potential distribution of high artemisinin-yielding Artemisia annua L. (Qinghao) in China with a geographic information system

The spatial distribution of A. annua L. was based on the following four sources: (1) the flora of China [30]; (2) scientific literature concerning the geographic distribution of A. annua L. in China [31]; (3) the Chinese Virtual Herbarium (CVH) [32], (4) germplasm accessions from the Sharing Information System for Chinese Medicinal Plant Germplasm Resources [33]; (5) field data of wild A. annua L. and interviews in Youyang County in 2008. Due to the excellent quality of A. annua L. from the habitats in Youyang County [31, 34‐36], a total of 180 accessions of A. annua L. germplasm were collected and used in the present study.

The potential distribution mapping program TCM-GIS and geo-referenced datasets were used to develop eco-adaptation models. The TCM-GIS package included three databases, namely (1) a basic geographic information database including digital line graphics and a digital elevation model (scale: 1:1,000,000), (2) a soil database (scale: 1:4,000,000), (3) and a climate database (mean values between 1971 and 2000). All three databases were used for spatial analysis and model calibration.

Raster and vector are two main data models in the TCM-GIS. Raster layers (1 × 1 km² resolution) were used for the eco-environmental analysis and cluster analysis. Vector layers were used to derive and identify the spatial extent and location of suitable habitats through overlay analysis. Moreover, global positioning system data on the locations of the 180 accessions were obtained for villages such as Banqiao, Zhongduo, Mawang and Nanmu and used in the TCM-GIS analysis (Figure 1).

In the present study, 14 eco-environmental variables were chosen for the predication of spatial distribution in Youyang County. These variables, namely (1) average temperature in January (ATJA), (2) average temperature in February (ATF), (3) average temperature in March (ATM), (4) average temperature in April (ATAP), (5) average temperature in May (ATMA), (6) average temperature in June (ATJ), (7) average temperature in July (ATJU), (8) average temperature in August (ATA), (9) average annual temperature (AAT), (10) annual sunshine time (AST), (11) total annual precipitation (TAP), (12) relative humidity (RH), (13) altitude (AL), (14) and soil properties (SP), were classified into three categories: topography, climate and edaphology (Table 1).

An optimal range was established by identifying minima and maxima for eco-environmental variables (e.g. elevation and temperature) at sample collection sites. The A. annua L. macro-habitats were characterized by examining the mean, minimal and maximal values, standard deviation (SD), standard error (SE), and coefficient of variation (CV) of these variables (Table 2). Prior to distance analysis, we normalized the raster grid data representing each variable. We derived the mean absolute deviation using the following equation:

where x_kf was the measured values of the variable f and m_f is the mean for the variable f. For the determination of similarity between grid data and eco-factor ranges, the statistical distance was calculated with the Minkowski distance equation [37]:

which is a generalization of the Euclidean distance and Manhattan distance; in general the shorter the distance, the greater the similarity. The comprehensive similarity index (SI) of each factor layer was calculated with an overlay analysis with various weighting values. Finally, maps with two ranks of predictive distributions were generated, followed by a grid-based spatial cluster analysis, vector-based overlaying, intersection analysis and an area calculation (Figures 2, 3, 4, Table 3).

The climatic, edaphic and topographic characteristics of known A. annua L. habitats are listed in Table 2. While low CV values for RH (CV: 0.33), TAP (1.28), AST (3.33), ATJU (4.60), AAT (4.69), ATA (6.23), ATJ (6.77) and ATMA (6.81) suggested that these could be the major limiting factors affecting the distribution of high quality A. annua L., high CV values for AL (29.79), ATJA (21.46) and ATF (21.43) suggested otherwise. According to the CV values, weighting value for each parameter was divided into levels I (0.15), II (0.08), III (0.06) and IV (0.03) and weighting values should add up to one. In addition, datasets of eco-factors from known habitats in Youyang County were as follows: ATJA = 1.2-5.6°C, ATF = 2.0-6.0°C, ATM = 4.0-10.0°C, ATAP = 10.0-16.0°C, ATMA = 14.0-20.0°C, ATJ = 18.0-24.0°C, ATJU = 21.6-27.3°C, ATA = 20.0-26.0°C, AAT = 15.9-21.0°C, AST = 1048-1200 h, TAP = 1169-1267 mm, RH = 79.2-80.6%, AL = 498-1010 mm. Soil types were mainly yellow soil, yellow sandy soil, limestone soil, paddy soil and brown soil with pH value at 6-7 and organic matter content ≥1.3%. Thus, we assumed that these conditions were optimal for the growth of high artemisinin-yielding A. annua L.

A. annua L. is a short-day plant. Non-juvenile plants are very responsive to short photoperiodic stimuli and flower about two weeks after induction. They require about 1000 hours of sunlight per year. Our results suggest that annual sunlight time is a critical factor for the growth of A. annua L., which is consistent with previous studies [5, 38]. Previous findings that A. annua L. requires a strict watering regime during the preliminary growth stages [5, 39] are also consistent with our results.

Figures 2 and 3 are the maps derived from the TCM-GIS analyses. The predicted areas were primarily located in the Wuling Mountain region in central China, covering Guizhou, Chongqing, Hunan, Hubei and Sichuan (25°14'-31°38' N to 104°31'-111°51'E). The predicted habitat density was high in northeastern Guizhou, southeastern Chongqing, northwestern Hunan, southwestern Hubei and parts of southern Sichuan.

The total favorable regions (SI 98%-99%) made up 1.60% of China's total land area covering 162 counties and cities (a total of 60,292 km²), among which Guizhou took the lead with 31,150 km² including 68 counties and cities. The most favorable region for A. annua L. (SI 99%-100%) was in the 58 counties and cities in Guizhou Province with a predicted area of 54,350 km². The second largest predicted area (14,330 km²) was in the 12 counties and cities in Chongqing, followed by Hunan, Hubei and Sichuan (Figure 4). The counties and cities with significant areas of potential habitat are listed in Table 3. The data indicated that Youyang County contained the largest favorable area with more than 4000 km². Unexpectedly, the total predicted areas in Wuchuan and Zunyi Counties in Guizhou exceeded 2000 km².

One of the world's largest artemisinin manufacturers and its affiliates operate A. annua L. farms in the Chongqing Wulingshan Mountain Range [40, 22]. Apart from this, Guizhou may be another important region for A. annua L. cultivation, particularly in the northeastern part of the province. Our model predicted that 13% of this area is potential A. annua L. habitats [41, 42]. Our model did not predict Guangxi Province, known for its habitats of A. annua L. of relatively low quality, as a region for A. annua L. cultivation possibly due to the subtropical climate, low altitude and red soil in Guangxi which are very different from those in other A. annua L. regions in China [9].

Interviews with the locals suggest that the Guizhou region and Youyang County have comparative advantages for A. annua L. growth with a high-yield variety and minimal pests. Furthermore, the northeastern Guizhou is home to wild populations of A. annua L. which may be an alternative source for artemisinin.

Using the TCM-GIS, we aimed to determine the optimal ecological factors from known habitats and the results showed that RH, TAP, AST, STJU, AAT and SP were important limiting factors. We also aimed to map the distribution of potential regions for the development of A. annua L. in China based on selected climatic, soil and topographical values. Using bioclimatic similarity theory and the TCM-GIS, we predicted the potential growing areas at the county level, particularly in northeastern Guizhou Province. The TCM-GIS is adequate for predicting and identifying potential areas for A. annua L. cultivation.

Using a higher resolution raster and vector spatial databases, we improved the resolution of species distribution considerably on the national surveys conducted in the 1960s, 1970s and 1980s. While most of the survey data were based largely on personal experiences and rough estimates, the model used in the present study is relatively objective.