Spatial and temporal patterns of China s cropland during : An analysis based on Landsat TM data

Transcription

1 Remote Sensing of Environment 98 (2005) Spatial and temporal patterns of China s cropland during : An analysis based on Landsat TM data Jiyuan Liu a, *, Mingliang Liu a,b, Hanqin Tian a,b, Dafang Zhuang a, Zengxiang Zhang c, Wen Zhang d, Xianming Tang a, Xiangzheng Deng a a Institute of Geographical Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing , China b School of Forestry and Wildlife Sciences, Auburn University, AL 36849, USA c Institute of Remote Sensing Applications, Chinese Academy of Sciences, Beijing , China d Institute of Atmospheric Physics (IAP), the Chinese Academy of Sciences, Beijing , China Received 24 January 2005; received in revised form 20 August 2005; accepted 25 August 2005 Abstract There are large discrepancies among estimates of the cropland area in China due to the lack of reliable data. In this study, we used Landsat TM/ ETM data at a spatial resolution of 30 m to reconstruct spatial and temporal patterns of cropland across China for the time period of Our estimate has indicated that total cropland area in China in 2000 was million hectares (ha), including 35.6 million ha paddy land and million ha dry farming land. The distribution of cropland is uneven across the regions of China. The North-East region of China shows more cropland area per capita than the South-East and North regions of China. During , cropland increased by 2.79 million ha, including 0.25 million ha of paddy land and 2.53 million ha of dry farming land. The North-East and North-West regions of China gained cropland area, while the North and South-East regions showed a loss of cropland area. Urbanization accounted for more than half of the transformation from cropland to other land uses, and the increase in cropland was primarily due to reclamation of grassland and deforestation. Most of the lost cropland had good quality with high productivity, but most gained cropland was poor quality land with less suitability for crop production. The globalization as well as changing environment in China is affecting land-use change. Coordinating the conflict between environmental conservation and land demands for food will continue to be a primary challenge for China in the future. D 2005 Elsevier Inc. All rights reserved. Keywords: Cropland; China; Land use; Remote sensing 1. Introduction China is the world s third largest country, the most rapidly developing nation and home to 1.3 billion people. Since the early 1980s, the unprecedented combination of economic and population growth has led to a dramatic land transformation across the nation (Chen, 1999; Houghton & Hackler, 2003; Liu et al., 2005; Tian et al., 2003). Accurate information on cropland area in China is of critical importance for assessing China s food security (Fischer et al., 1998) and greenhouse gas * Corresponding author. Institute of Geographical Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing , China. Tel.: ; fax: address: liujy@igsnrr.ac.cn (J. Liu). URL: (J. Liu). emission such as N 2 O, CO 2 and CH 4 (Houghton & Hackler, 2003; Li et al., 2003; Tian et al., 2003; Verburg & Denier van der Gon, 2001; Wang, 2001). It was reported that cropland area is decreasing due to urban expansion and increasing population, which has caused many concerns regarding China s food security in the near future (Brown, 1995). Data quality and reliability, however, is one of the greatest problems in generating a clear picture of China s food system (Brown, 1995; Fischer et al., 1998; Huang & Rozelle, 1995), the carbon cycle (Houghton & Hackler, 2003; Tian et al., 2003), nutrient cycles (Galloway et al., 2002; Verburg & Veldkamp, 2001; Zhang et al., 2005) and ecosystem sustainability (Smil, 2000). There are large discrepancies among estimates on the state and change of land use in China based on several data sources (Tian et al., 2003). The first data source is the official statistical data by State Statistical Bureau (SSB, hereafter) (Li, 1999), /$ - see front matter D 2005 Elsevier Inc. All rights reserved. doi: /j.rse

2 J. Liu et al. / Remote Sensing of Environment 98 (2005) which is generated based on agricultural census data at the county level. It was reported this data source could possibly underestimate the actual cultivated area (Crook, 1993; Frolking et al., 2002; SSB, 1994) by 27.0% (Liu, 2000). The second data source is the national land resources inventory data sponsored by the Ministry of Land and Resources (MLR, hereafter). This program was completed in 1996 after a 14-year survey and mapping based on aerial photos and census data with different scales (Fischer et al., 1998; Li, 1999; Liu, 2000; Ma, 2000; SLA, 1996). Since it covers a long time period, this data source is not suitable for retrieving land-use change information over a short period of time with fine spatial resolution at national scale. The third data source is 1:1,000,000 scale maps of China s land use by Wu (1990), which was generated from extensive field surveys, interpretation of aerial photos, and Landsat images during the late 1970s and the early 1980s. This is the first national land-use survey and mapping program which used standard method and routines (Wu, 1990; Wu & Guo, 1994). The fourth data source is land resources maps at the scale of 1:1,000,000, which are based on Landsat MSS images and field land survey. In this data set, the land was classified as different types according to land-use/land-cover classes and its suitability for cultivation, forestry and grazing, respectively (CISNR, 1993; Ge et al., 2000). The fifth data source is an IGBP DIScover data set which was produced from 1-km resolution National Oceanic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR) data (Loveland & Belward, 1997; Loveland et al., 2000). Although the IGBP DIScover data set has been used widely in global scale ecological and geographical studies (Ramankutty & Foley, 1998), it is inadequate for estimating cropland area and its change in China due to the coarse resolution (Xiao et al., 2003). Frolking et al. (1999) reported that cropland area estimate in China based on the AVHRR data set is about 50% higher than that using agricultural census data. According to the same classification system with assistance from other geographical databases, Chinese scientists developed a new version of a 1 km land-cover map, which was validated against field survey and more accurate than the IGBP DISCover map (Liu et al., 2003b). The most recently available data with 1 km resolution is land-cover production from MODIS (Moderate Resolution Imaging Spectroradiometer) data ( As described above, all of the data and maps above do not provide information on the detailed spatial patterns of cropland across China. Moreover, the spatial-temporal dynamics of cropland in the recent decade remain unknown, which is important to the assessment of how land transformation affects biogeochemical and hydrological cycles at regional and global scales. To characterize spatial and temporal patterns of land use/land cover in China, we used Landsat Thematic Mapper (TM) to develop the National Land-Use/Land-Cover data sets (NLCD, hereafter) for the three time periods: the end of the 1980s, 1995/96, and the end of the 1990s, with a mapping scale of 1:100,000 (Liu et al., 2002, 2005). Based on this data set, we analyzed the spatial and temporal patterns of China s cropland during We also compared our analysis with other studies to identify the uncertainties in the estimation of cropland area and its spatial distribution. The results drawn from our study are important for scientists as well as policy makers for assessing a number of cutting-edge issues associated with global change and sustainability. 2. Data sources and method In the late 1990s the Chinese Academy of Sciences organized eight research institutions and about 100 scientists to conduct its second nationwide land cover and land use classification project. In 1997, using 520 Landsat TM images primarily from 1995/ 1996, we developed the national land use databases at a spatial scale of 1: 100,000 by visual interpretation and digitalization with technical support from Intergraph MGE (Modular GIS Environment) software (Liu et al., 2002; Zhuang et al., 1999). Before the interpretation began, remote sensing images were geo-referenced by using 1:50,000 relief maps. For each TM/ ETM scene, there are at least 20 evenly distributed sites served as Ground Control Points (GCPs). The Root Mean Squared Error (RMS error) of geometric rectification was less than 1.5 pixels (or 45 m). Interpreters used MGE software to identify the landuse types on the computer screen, based on his/her understanding on the object s spectral reflectance, structure and other information. Then they drew the boundaries of the objects and added the attributes (labels) of the polygons to produce the digital map. Finally we edited and compiled the vector digital maps, which are the core of the national spatial database. A hierarchical classification system of 25 land-cover classes was applied to the Landsat TM/ETM (Enhanced Thematic Mapper) data (Table 1). The 25 classes of land cover were further grouped into 6 aggregated classes of land cover: croplands, woodlands, grasslands, water bodies, unused land and built-up areas including urban areas. Croplands include paddy and dry farming land. Woodlands include forest, shrub and others (e.g., orchards). Grasslands include three densitydependent types: dense, moderate and sparse grass. Water bodies include stream and rivers, lakes, reservoir and ponds, permanent ice and snow, beach and shore, and bottomland. Unused land includes sandy land, Gobi, Salina, wetland, bare soil, bare rock and others such as alpine desert and tundra. Built-up land includes urban area, rural settlements and others such as roads and airports. The definition of each land-cover class is given in Table 1. The cropland in our national database is defined as all agricultural lands. It includes permanently cultivated land, new cultivated land, fallow, and grasslandfarming rotated land. It also includes intercropping land such as crop-fruiter, crop-mulberry, and crop-forest land in which a crop is a dominant species. At the second classification level, the cropland is further divided into paddy land and dry farming land according to hydrological condition and crop types. The paddy land has enough water supply and irrigation facilities for planting paddy rice, lotus, etc. including the rotation between paddy rice and dry farming crops. To support image interpretation and the validation of land cover classification, we have used a variety of data including soil type, DEM, roads and rivers and climate. In 1999, we

3 444 J. Liu et al. / Remote Sensing of Environment 98 (2005) Table 1 The land use classification system 1st level classes 2nd level classes Descriptions Cropland Cultivated lands for crops. Including: mature cultivated land, newly cultivated land, fallow, shifting cultivated land; intercropping land such as crop-fruiter, crop-mulberry, and crop-forest land in which a crop is a dominant species; bottomland and beach that cultivated for at least 3 years. Paddy land Cropland that has enough water supply and irrigation facilities for planting paddy rice, lotus etc., including rotation land for paddy rice and dry farming crops. Dry land Cropland for cultivation without water supply and irrigating facilities; cropland that has water supply and irrigation facilities and planting dry farming crops; cropland planting vegetables; fallow land. Woodland Lands growing trees including arbor, shrub, bamboo and for forestry use. Forest Natural or planted forests with canopy cover greater than 30%. Shrub Lands covered by trees less than 2 m high, the canopy cover >40%. Woods Lands covered by trees with canopy cover between 10 30%. Others Lands such as tea-garden, orchards, groves and nurseries. Grassland Lands covered by herbaceous plants with coverage greater than 5%, including shrub rangeland and mixed rangeland with the coverage of shrub canopies less than 10%. Dense grass Grassland with canopy coverage greater than 50%. Moderate grass Grassland with canopy coverage between 20% and 50%. Sparse grass Grassland with canopy cover between 5% and 20%. Water body Lands covered by natural water bodies or lands with facilities for irrigation and water reservation. Stream and rivers Lands covered by rivers including canals. Lakes Lands covered by lakes. Reservoir and ponds Man-made facilities for water reservation. Permanent ice and snow Lands covered by perennial snowfields and glaciers. Beach and shore Lands between high tide level and low tide level. Bottomland Lands between normal water level and flood level. Built-up land Lands used for urban and rural settlements, factories and transportation facilities. Urban built-up Lands used for urban. Rural settlements Lands used for settlements in villages. Others Lands used for factories, quarries, mining, oil-field slattern outside cities and lands for special uses such as transportation and airport. Table 1 (continued) 1st level classes 2nd level classes Descriptions Unused land Lands that are not put into practical use or difficult to use. Sandy land Sandy land covered with less than 5% vegetation cover. Gobi Gravel covered land with less than 5% vegetation cover. Salina Lands with salina accumulation and sparse vegetation. Swampland Lands with a permanent mixture of water and herbaceous or woody vegetation that cover extensive areas. Bare soil Bare exposed soil with less than 5% vegetation cover. Bare rock Bare exposed rock with less than 5% vegetation cover. Others Other lands such as alpine desert and tundra. conducted a field survey to evaluate the classification accuracy, including an accumulated survey length of 75,271 km across China (average 2509 km for each province), a total of 13,300 patches and 8000 photos located with GPS facilities. Our results showed that the overall accuracy of the land-cover classification for the 25 sub-classes is 92.9%. For cropland, the accuracy is 94.9% based on an evaluation of 5058 patches. The identification of built-up area with 1714 patches selected has the highest accuracy of 96.3%. For the forest and grassland, the accuracy is 90.1% and 88.1%, respectively, which is based on the evaluation of 4104 and 1512 patches, respectively. We collected 39,663 slices to evaluate the accuracy of location during mapping process. The result indicates that 99.85% of the polygon boundaries show less than 1.5 pixels (45 m) shift from the real boundary. We updated the national land-use map of 1995/1996 to the end of 1990s by using the remotely sensed data in 1999/2000, which include a total of 512 scenes of Landsat ETM data imaged in 1999/2000 along with some images from 17 scenes of China Brazil Earth Resources Satellite (CBERS-1) ( (Liu et al., 2003a). The CBERS images have 20 m ground resolution and almost the same spectral bands as Landsat ETM. The most important principle for us to choose the remotely sensed image for updating the land-use map is that the imaging date (season) should be consistent. To reconstruct the land-use change history during , we collected Landsat TM images at the end of the 1980s and/or the beginning of the 1990s (thereafter X1990). Using MGE software, the interpreters drew the land-use change patches by comparing the TM images (showed as a combination of band 432 RGB) from X1990 with those from 1999/2000 (e.g., Fig. 1I and II). Also, the land-use maps of 1999/2000 have been used as the supporting information to identify the type of land-use change for each patch. The smallest patch of land-use change we selected is not less than 36 pixels (3.24 ha) and the shortest edge must be longer than 4 pixels (120 m). By comparing images directly, we can decrease the possible errors

4 J. Liu et al. / Remote Sensing of Environment 98 (2005) from classification by different peoples who used classified maps to make change detections. The land-use change patches can be considered as real change instead of classification errors and the small location shift. In total, the vector data of land-use change include 364,379 polygons containing land conversion information, i.e., the original and present land use types. All of 445 these were identified at the second level of classification systems (Table 1). We chose land-use change data from 249 counties to make the evaluation. In total, 44,381 patches (12.18% of all) were evaluated by the specialists and the result showed that the overall identification accuracy is 97.2%. Because cropland is relatively easy to be identified, the Fig. 1. Landsat TM/ETM imagery showing land-use change in Nenjiang County, Heilongjiang Province in 1988 and (a) LANDSAT-5 Thematic Mapper (TM) image. Date: Mar. 8, (b) LANDSAT-7 Enhanced Thematic Mapper Plus (ETM+) image. Date: Jun. 13, Path Row: In this area, a large area of forest was converted to cropland during this period. II. Landsat TM/ETM imagery showing land-use change in Wuxi City, Jiangsu Province in 1988 and (a) LANDSAT-5 Thematic Mapper (TM) image. Date: Nov. 13, (b) LANDSAT-7 Enhanced Thematic Mapper Plus (ETM+) image. Date: May. 4, Path Row: In this area, a large area of cropland was transformed into urban, transportation and living places. In all images, yellow polygons are the interpreted land-use change. The images are a composite of Band 4 (Red), 3 (Green) ans 2 (Blue). Spectral Range of Band 4 is Am, Band 3 is Am and Band 2 is Am. Using this composition, white and cyan color indicates urban and built-up objects, blue color is water body, and red color is vegetationcovered surface.

5 446 J. Liu et al. / Remote Sensing of Environment 98 (2005) Fig. 1 (continued). interpretation on cropland change has higher accuracy. The accuracies of cropland change interpretation in 26 provinces are above 95%, only in 3 provinces they are below 90%. We developed a new technique for data fusion that converts vector data into a series of grid data with 1 km resolution without destroying the acreage information, i.e., the area of each type in the 1 km2 grid is consistent with that in the original high resolution vector format data (Liu et al., 2002, 2003a, 2001, 2003c; Tang, 2000). The method includes three steps. Firstly, we build up a standard grid frame with vector format and each grid cell has 1 km width and 1 km length and each cell is identified with a unique ID. Secondly, we use the frame to intersect with the input vector data to group the input information into each cell. Lastly, we provide a summary of area, length, etc. for each cell group by class or level. In this study, we used the same method to process the land-use change data. Each cell with 1 km2 area contains the land transformation information, e.g., cropland area in 2000 and the conversion between cropland and other land cover types during the 1990s (Liu et al., 2003c). When surveying cropland area by using remote sensing techniques, we need to take into account the influence of spatial scale on the accuracy in calculating the area in the GIS software environment (Frolking et al., 1999; Liu et al., 2001; Wu & Guo, 1994). In our research project, a stratified, multi-layer sampling framework was established to account for the fraction of noncultivated land (e.g., narrow roads and footpaths, small rice paddy levees, irrigation channels, etc.) within the polygons of cropland (Zhang et al., 2000a,b) to rectify the cropland area estimation. Firstly, we divided China mainland into 10 zones and 814 sub-zones according to land-use characters and environmental condition. Secondly, we used 1 : 100,000 line feature maps including railway, national road, highway, river, stream to estimate area covered by the line features within each identified cropland polygon. Thirdly, we designed sampling frames for estimating errors from TM data interpretation. For each 7 km, we built sampling areas with 200 m width. We estimated which areas of other land-cover types are identified into cropland type due to mapping criterion (land features

6 J. Liu et al. / Remote Sensing of Environment 98 (2005) smaller than 33 pixels cannot be identified). We use the same average value in each sub-zone for these three periods. Finally, we collected about 10,000 aerial photographs which were mainly taken by national survey and mapping organizations in the 1980s to estimate the non-cropland parts (e.g., crop field ridge, woods beside crop field, etc.) within cropland polygons in each sub-zones. These non-cropland parts were not identified in image interpretation because of the mixed pixel problems in remote sensing technology. In estimating the national land-use status from the maps, we used the same average non-cropland fraction to all periods. After the above three correction processes, we estimated the national and regional real cropland area, i.e., excluding non-cropland area in each patch. For cropland area in 2000, the gross cropland area (without rectification) is 27.5% more than the rectified cropland area. The gross area of paddy land and dry farming land was overestimated by 32.7% and 25.7%, respectively. Generally, the non-cropland types in highly urbanized regions such as Beijing, Tianjin, Jiangsu, Guangdong, and mountainous regions such as Yunnan and Sichuan were higher than other regions. After the adjustment of cropland area, the net areas of build-up, water bodies and woodland at a national scale have increased by 267%, 121%, and 102%, respectively. Unless otherwise noted, the cropland area in this paper is the real area. We did not make the rectification for Tibet and Taiwan because of data limitation. In this paper, the land-use change area was the gross estimation directly from map without the rectification processes. To investigate difference in spatial and temporal patterns among subregions across China during the 1990s, we divided the land of China into 6 sub-regions: North-East region including Inner Mongolia, Liaoning, Jilin and Heilongjiang; North region including Beijing, Tianjin, Hebei, Shanxi, Shandong and Henan; South-East region including Shanghai, Jiangsu, Zhejiang, Fujian, Guangdong (data for Hong Kong and Macao are included into Guangdong province), Hainan and Taiwan; Central region including Anhui, Jiangxi, Hubei and Hunan; South-West region including Guangxi, Chongqing, Sichuan, Guizhou and Yunnan, and North-West region including Tibet, Shannxi, Gansu, Qinghai, Ningxia and Xinjiang. (Fig. 2, Table 2). For each sub-region, we analyzed the current pattern of cropland and its changes during the 1990s. 3. Result and analysis 3.1. Cropland area at national and provincial levels in 2000 Our analysis estimated that the total area of cropland in China was million (10 6 ) hectares (Mha) in 2000, including 35.6 Mha of paddy land and Mha of dry farming land. In general, provinces with high populations showed larger cropland areas, and the average cropland per capita in China in 2000 was about 0.11 ha, which was much lower than the world average of 0.35 ha per capita (Raman- Fig. 2. China s geographical regions. North-East region includes: 15 Inner Mongolia, 21 Liaoning, 22 Jilin and 23 Heilongjiang; North: 11 Beijing, 12 Tianjin, 13 Hebei, 14 Shanxi, 37 Shandong and 41 Henan; South-East: 31 Shanghai, 32 Jiangsu, 33 Zhejiang, 35 Fujian, 44 Guangdong (data for Hong Kong and Macao are included into Guangdong province), 46 Hainan and 71 Taiwan; Central: 34 Anhui, 36 Jiangxi, 42 Hubei and 43 Hunan; South-West: 45 Guangxi, 50 Chongqing, 51 Sichuan, 52 Guizhou and 53 Yunnan; North-West: 54 Tibet, 61 Shannxi, 62 Gansu, 63 Qinghai, 64 Ningxia and 65 Xinjiang.

7 448 J. Liu et al. / Remote Sensing of Environment 98 (2005) Table 2 China cropland in 2000 and its change during Region Province P D C_All C_Per P_CH D_CH C-O O-C C-CH North-East Inner Mongolia Liaoning Jilin Heilongjiang ,526 14, Total ,627 35, (11.10%)* (29.98%)* (25.21%)* North Beijing Tianjin Hebei Shanxi Shandong Henan Total ,474 29, (3.20%)* (26.99%)* (20.98%)* South-East Shanghai Jiangsu Zhejiang Fujian Guangdong Hainan Taiwan Total , (26.29%)* (4.27%)* (9.83%)* Central Anhui Jiangxi Hubei Hunan Total 11, , (33.20%)* (6.96%)* (13.59%)* South-West Guangxi Chongqing Sichuan Guizhou Yunnan Total ,141 24, (23.40%)* (15.30%)* (17.35%)* North-West Tibet Shannxi Gansu Qinghai Ningxia Xinjiang Total ,414 18, (2.80%)* (16.51%)* (13.05%)* Whole China 35, , , (Unit: area: 1000 ha; cropland per capita: ha/person) Note: P: paddy land; D: dry farming land; C_ALL: all cropland; C_Per: cropland area per capita (ha/person) P-CH: paddy land change; D-CH: dry farming land change; C-CH: all cropland change; C-O: conversion area from cropland to others; O-C: conversion area from others to cropland; * Percentage of that item in China. The estimated total areas (P, D and C_ALL) indicate the net areas with rectification corrections on non-cultivated land. The changed areas are the gross estimation without the rectification processes. kutty et al., 2002). The North-East region and North region of China, such as Inner Mongolia, Heilongjiang, Jilin, Gansu and Xinjiang provinces had higher average cropland area per capita than the national average. However, most provinces in the South-East and South regions of China, such as Guangdong, Zhejiang, Fujian, Jiangsu, and Hunan provinces had less average cropland area per capita than the national average (Table 2). The cropland in the South of China, in general, shows high productivity (SSB, 2003). For example, Southern China (including South-East, Central and South-West regions) has 40.8% of total cropland area, but it produced 51.3% of the food of whole China in the year 2002 according to SSB (SSB, 2003). Nevertheless, Southern China still need to import food due to its large population (more than 57.7% of the total population in China (not including population of Taiwan)) and decreasing cropland area.

8 J. Liu et al. / Remote Sensing of Environment 98 (2005) Most of the paddy land in China was distributed in South China, e.g., 59.5% in Central and South-East regions and 23.4% in the South-West region. In provincial level, Hunan, Jiangsu, Anhui and Sichuan provinces had more than 12.9 Mha of paddy land, which accounted for 36.2% of total paddy area. While Hubei and Jiangxi provinces are included, the six provinces accounted for 51.2% of total paddy land. The spatial distribution of dry farming land was quite different from that of paddy land. About 57.0% of total dry farming land was located in the North-East (30.0% of total dry farmland) and North regions (27.0% of total dry farmland). At the provincial level, Heilongjiang province had the largest area of dry farming land with 12.5 Mha, followed by Inner Mongolia of 9.4 Mha Spatial distribution of cropland in 2000 Fig. 3 shows the spatial distribution of China s cropland. It is clear that most of the grids with a high percentage of cropland in China were distributed on the big plains of China, for example, the Song-liao plain and the San-jiang plain in the north-east, the North-China and East-China plains in the north and east, the Sichuan basin in the south-west, the plain of Middle Delta of the Yangtse River in central China, and the Loess Plateau in the north-west (Fig. 3). In the North-West region of China, the large area of cropland was located in the Loess Plateau, the middle and lower parts of the pro-mountain region, alluvial plains in some large river s delta, and in oasis in deserts. The spatial distribution of cropland on a large scale was determined by environmental factors such as climate, soil moisture, soil properties, and economic factors such as agricultural technology and historical culture (Cramer & Solomon, 1993; Leemans & Solomon, 1993; Ramankutty et al., 2002). Paddy land and dry farming land in China have significant differences in suitability to the geographical background, which can be characterized by their spatial distribution along environmental zone (Fig. 4). We found that 67.4% of the cropland was located on lowland with the elevation of less than 500 m (Fig. 4a). Compared to dry farmland, the paddy land tended to be aggregated in lowland. For example, 83.2% of paddy land compared to 61.6% of dry farmland was distributed in the areas with the elevation of lower than 500 m. Small amount of cropland was distributed in Fig. 3. Cropland distributions across China in The value indicates the percentage of cropland within a 11 km grid cell. Darker red color means more cropland area in the grid cell and darker blue color means less cropland area in the grid cell.

9 450 J. Liu et al. / Remote Sensing of Environment 98 (2005) Fig. 4. Cropland distribution under different geophysical and climate conditions. ( Y axis units: percentage). Note: y axis represents the percentage of each type of cropland area located in each zone. (a) Cropland distribution along elevation (X axis units: meter); (b) cropland distribution along annual mean precipitation (X axis units: mm/year); (c) cropland distribution along annual mean temperature (X axis units: degrees Celsius). widespread mountainous region and plateau in China. Precipitation showed the largest control on the cropland distribution (Fig. 4b). About 71.0% of paddy land in China was distributed in wet areas with annual precipitation between 900 and 1650 mm. In contrast, 60% of the dry farmland was located in dry areas with annual precipitation between mm (Fig. 4b). We also found that large area in dry region of China is not suitable for crop cultivation. Cropland distribution was also controlled by temperature. Compared with the wide-distribution of dry farmland along temperate zones, paddy land occupied a narrower range, e.g., 66.4% of paddy land in China was distributed within annual mean temperature range of C, while most of the dry farmland occupied a wider range of C (Fig. 4c).

10 J. Liu et al. / Remote Sensing of Environment 98 (2005) Changes in cropland area in the 1990s During , China s cropland increased by 2.79 Mha including 0.25 Mha of paddy land and 2.53 Mha of dry farming land. However, more than 70% of the increased cropland area occurred in Heilongjiang and Inner Mongolia two provinces. The former increased 1.66 Mha and the later increased 0.97 Mha. In contrast, 19 provinces representing 60% of total land showed a decrease in cropland area (Table 2). For the decreased cropland, more than 50% took place in Jiangsu, Guangdong, Hebei and Shandong provinces with a decrease of 0.18, 0.13, 0.12, and 0.10 Mha, respectively. Most Fig. 5. Cropland transformation under different geophysical and climate conditions ( Y axis units: percentage). Note: y axis represents the percentage of each cropland transformation distribution in each zone. (a) Along elevation (X axis units: meter); (b) along annual mean precipitation (X axis units: mm/year); (c) along annual mean temperature (X axis units: degrees Celsius).

11 452 J. Liu et al. / Remote Sensing of Environment 98 (2005) of the lost cropland had good soil fertility, while most of the new cultivated cropland is located either in places with poor soil fertility and limited water resources (Fig. 5b) or in the places that has not enough agricultural facility. Spatially, cropland expanded in the North-East and North- West regions and shrank in the North and South-East regions (Table 2, Fig. 6). Cropland area decreased significantly in many traditional agricultural regions such as the Yangtze delta, the Zhujiang delta, the Huang-Huai-Hai plain, and the Sichuan basin (Fig. 6). Changes in cropland area on regional and national scales are the net result of reclamation, abandonment, and land-use transformation. Table 3 indicates that humans had intensive effects on cropland during this short time period, e.g., more than 5% of the total cropland was involved in transformation. The transformation between paddy land and dry farming land was significant in the North-East region of China (Liu et al., 2003a). Among the land transformations from cropland to others, urbanization accounted for 46.1%, while transformation from cropland to grassland and forest land accounted for 21.6% and Table 3 Conversion between cropland and others (units: 1000 ha) Region Type 1 Fr C Gr C Wa C Ur C Un C 16.7%, respectively (Table 3). It should be noted that great regional difference existed in land transformation in China. For example, most of the decreased cropland area in the North, South-East, and Central regions of China was due to urbaniza- C Fr C Gr C Wa C Ur C Un North-East North South-East Central South-West North-West Total Total increases: 5219 Total decreases: A B: represents that land-use A transformed into land-use B. C: Cropland; Fr: Forest land; Gr: Grassland; Wa: Water body; Ur: Urban area and other built-up; Un: Unused land. All the areas are the gross estimates without rectification correction on noncultivated land. Fig. 6. The spatial distribution of cropland change during (units: percentage). The value shows the fraction of cropland change in each grid during this period, e.g., 50% means 50% of grid area (i.e km 2 =50 ha of cropland) has changed in each grid. The cropland increase is represented by blue color and the cropland decrease is represented by red color.

12 J. Liu et al. / Remote Sensing of Environment 98 (2005) tion. However, for the North-East, North-West, and South-West regions, this cropland declination was attributed to cropland abandonment, reforestation, and afforestation (Table 3). In the new cultivated cropland, conversion from grassland and forest accounted for 55% and 30%, respectively. During the last ten years, the largest deforestation occurred in the North-East region in which 1.26 Mha of forest land was converted into cropland (Table 3). In Northern China, including North-East, North and North-West regions, 57.6% of the new cropland was converted from grassland, and 26.9% from forest land. In Southern China, including South-West, South-East and Central regions, the percentage of new cropland converted from grassland and forest land was 18.0% and 65.7%, respectively (Table 3) Cropland transformation under different geophysical environment Cropland transformation took place in different geophysical zones with different intensities. The annual mean precipitation and temperature data were extrapolated from the observation data of 1915 stations with 500 m resolution, which were developed by the Institute of Natural Resources and Regional Planning, the Chinese Academy of Agricultural Sciences. Over 80% of cropland transformation occurred in plains and low hilly areas (Fig. 5a). Most of the increase of cropland, especially the new cultivated paddy land, took place in dry area with low precipitation (Fig. 5b). Large area of paddy land in wet region was found to be transformed into other usage. Fig. 5c indicates that most of the new cultivated paddy land was distributed in temperate regions and that paddy land decreased in subtropical and tropical regions. 4. Discussion 4.1. Comparison with census data and other studies Large discrepancies among the estimates of the cropland area in China exist (Heilig, 1999; Houghton & Hackler, 2003; Seto et al., 2000; Tian et al., 2003). Our estimates of cropland area over the 1990s based on Landsat TM/ETM differ from those reported by other Chinese Government sources based on surveys using other techniques. Our results for the years of 1990, 1995 and 2000 are almost the highest among these estimates of cropland area (Fig. 7). For example, our estimate of cropland area for 1995 (about 140 Mha) is almost 50% larger than the State Statistical Bureau s estimate (SSB, 1996). It has been widely recognized among experts that the SSB s statistics of cropland areas were clearly underestimated and extremely unreliable in the past years (Heilig, 1999). Estimates from the late 1980s and early 1990s range from 95 to 150 million ha, which has been used by the Food and Agricultural Organization (FAO). Thus environmental studies based on the FAO data of China s cropland areas need to be reexamined (Houghton & Hackler, 2003). Based on land surveys, China s Ministry of Land and Resources (MLR) estimated that the area of farmland was in the order of 131 million ha in 1995 (MLR, 2000). Heilig (1999) reconstructed the cropland areas for the time period from 1990 to 1995, by using the MLR s estimate for the year 1995 and other information about the increases and decreases of various types of cultivated land in China, and concluded that SSB s estimates are about 28% lower than their estimates. Experts have recognized that the MLR s estimates of cropland areas are more accurate than SSB s estimates. Since 2000, therefore, China s government began using the MLR s Fig. 7. China s cropland area in the 1990s as estimated by different investigators (unit: million hectares). FAO: FAOSTAT data, 2004; this paper (for 1990 and 2000): this study result for the year 1990 and 2000; this paper (for 1995): cropland acreage data is based on this paper result for 2000 and cropland change data during (Liu et al., 2003a); MLR: from State Land Administration (now The Ministry of Land and Resources of China) (China Land and Resources Almanac, 2003, does not include Taiwan data); SSB: State Statistical Bureau ( ) (does not include Taiwan data); IGBP-DIS (pure): include only pixels classified as pure cropland in IGBP DISCover land cover data set derived from AVHRR satellite data for the period of (Frolking et al., 1999; Loveland & Belward, 1997); Heilig, 1999: annual mean cropland area reconstructed by Heilig (1999).

13 454 J. Liu et al. / Remote Sensing of Environment 98 (2005) estimates as a standard. However, our estimates of cropland areas were about 8.1 Mha larger for 1995 and 12 Mha larger for 2000 than the MLR s estimates (Fig. 7). This difference between MLR and our estimates may be caused by the differences in methods and data we used. Although MLR used different sources of remote sensing data, large-scale relief maps and field surveys to estimate cropland areas, the accuracy of MLR estimates for annual cropland area largely relies on the size of image samples. As we described above, however, we used more than 500 TM/ETM scenes covering the entire China for the time period we examined. From 1990 to 2000, our result shows a 2% increase in cropland area, but the MLR s estimate indicates a 2% decrease in cropland area. This difference is primarily because our Landsat-based analyses have counted for the new croplands converting from forests or grasslands, particularly in North China. Uncertainty exists in our analysis of land transformations across China. Although the average location errors of TM images are less than 45 m, the small errors could bias our detection of land transformations. Uncertainty also arises from the lack of direct ground truth in the late 1980s and the middle 1990s. However, we are confident that our estimates of cropland area and its change are a significant improvement over other earlier estimates. These new estimates should improve the ability of policy makers to manage China s environment as the nation transitions to sustainability. Improved estimates of cropland areas and its changes will lead a right way for decision-makers to manage land resources in a sustainable way in order to meet the increasing food needs of the great population Implications of cropland change for China s food security Our Landsat-based analysis shows that cropland area in China increased by 2.79 million ha over the 1990s. This increase in cropland area was primarily due to the conversion from forest and grassland to cropland. Although the cropland expansion in China could offset the pressure of increasingly need from its population increase in recent years (increased by about 132 million people during ), the soil fertility of the gained cropland is in general low (Li et al., 2004). The loss of croplands was largely because of urban development. These croplands converted to urban were very productive. With increasing population and growing demand for food, cropland protection for agricultural production is of utmost importance to ensure food security in China (Chen, 1999; Fischer et al., 1998; Heilig, 1999). Even though there are some regulations such as Land Management Law of China for protecting cropland, the conflict between cropland protection and urban expansion still exists. New agricultural techniques stimulate cropland expansion by converting woodland and grassland to cropland. However, the policy of returning cropland into woodland or grassland resulted in the decrease of cropland (Li, 1999). During the late 1990s, the National Project on Wild Wood Protection was issued along with the implementation of returning cropland to woodland or grassland policy. In addition, the new forest policy the Natural Forest Conservation Program (NFCP) (Zhang et al., 2000) will decrease the transformations from forest into cropland that took place during the last ten years. Afforestation has occurred in some areas, such as the Southeastern coasts and central part of China. The implementation of the Grain-for-Green Project in West China during the past 6 years resulted in about 2.93% decrease in cropland area in West China (Li et al., 2004). Chinese government has promoted to use biotechnology for improving the productivity of existing crops in order to reduce the pressure of food demand for cropland (Huang et al., 2002). However, protecting current arable agricultural lands of China will still be the top priority for Chinese government. The challenge encountered the Chinese government and scientists is how to ensure that good cropland is not lost, that agricultural production is increased while without destroying the environment, and that enough land resources are available for overall economic development. A primary challenge for China s future is to solve the conflict between environmental conservation and land demands for food. 5. Conclusion Based on high resolution Landsat TM/ETM data (30 m ground resolution), we have estimated that China s cropland area was Mha in 2000, including 35.6 Mha paddy land and Mha dry farmland. From 1990 to 2000, total cropland increased by 2.79 Mha, including 0.25 Mha in paddy land and 2.53 Mha in dry farmland. Recent changes in cropland showed substantial spatial variations across the nation. A significant increase in cropland areas occurred in Northeast China, Northern China and the Xinjiang oases. However, a remarkable decrease in cropland existed in the Huang-Huai-Hai Plain, the Yangtze River Delta, the Huanghe River band in the vicinity of Baotou and Datong sections, and the Sichuan Basin. In terms of cropland area per capita, the North-East region was higher than the South-East and North regions. Annual mean temperature, precipitation and elevation are the primary controls of the cropland spatial distribution across China. The North-East and North-West regions of China showed large expansions in cropland, while the North and South-East regions appeared to lose cropland. Urbanization accounted for more than half of the transformation from cropland to other land uses, and most new cropland was converted from grassland and deforestation. Our results have shown the important trend of cropland change during in a spatially explicit way. China s economic reform began at the end of the 1970s. Large-scale land transformation has occurred since the early 1980s. It will be valuable to characterize spatial and temporal patterns of cropland since 1980, which could further reflect impacts of social-economical forces on China s cropland. To improve our ability in predicting cropland change in the near future and its impacts on China s food security and environment, we need to investigate mechanisms that drive land-use change at multiple scales from landscape to the nation.

14 J. Liu et al. / Remote Sensing of Environment 98 (2005) Acknowledgements This study was supported by Ministry of Science and Technology (MOST) 973 Program (2002CB412507), NASA Interdisciplinary Science Program (NNG04GM39C), the Project of National Science Foundation of China ( ), Chinese Academy of Sciences ODS Program, and National Science Foundation of China ( ). We would like to thank Hua Chen, Yaoqi Zhang, Chi Zhang, Guangsheng Chen and four anonymous reviewers for their comments and suggestions. References Brown, L. R. (1995). Who will feed China? Wake-up call for a small planet. New York WW Norton & Company. 160 pp. Chen, B. (1999). The existing state, future change trends in land-use and food production capacities in China. Ambio, 28(8), CISNR, (Ed.). (1993). The database on territorial resources of China. Beijing China Renmin University Press. Cramer, W. P., & Solomon, A. M. (1993). Climate classification and future global redistribution of agricultural land. Climate Research, 3, Crook, F. W. (1993). Underreporting of China s cultivated land area: Implications for world agricultural trade. USDA, international agriculture and trade report, China. Document RS Washington, D.C. ERS, USDA. FAOSTAT data. (2004). Statistical database of the food and agricultural organization of the United Nations. Available on the World Wide Web at Fischer, G., Chen, Y., & Sun, L. (1998). The balance of cultivated land in China during IR Laxenburg International Institute for Applied Systems Analysis. Frolking, S., Qiu, J., Boles, S., Xiao, X., Liu, J., Zhuang, Y., et al. (2002). Combining remote sensing and ground census data to develop new maps of the distribution of rice agriculture in China. Global Biogeochemical Cycles, 16(4), /2001GB Frolking, S., Xiao, X., Zhuang, Y., Salas, W., & Li, C. (1999). Agricultural land-use in China: A comparison of area estimates from ground-based census and satellite-borne remote sensing. Global Ecology and Biogeography, 5, Galloway, J. N., Cowling, E. B., & Kessler, E. (Eds.). (2002). Optimizing nitrogen management in food and energy productions, and environmental change. Ambio, vol. 31(2). Ge, Q., Zhao, M., & Zheng, J. (2000). Land use change of China during the 20th century. Acta Geographica Sinica, 55, (in Chinese). Heilig, G. K. (1999). Can China feed it self? A system for evaluation of policy options. Laxenburg IIASA ChinaFood/index_m.htm Houghton, R. A., & Hackler, J. L. (2003). Sources and sinks of carbon from land-use change in China. Global Biogeochemical Cycles, 17, 1034, doi: /2002gb Huang, J., Carl, P., & Scott, R. (2002). Enhancing the crops to feed the poor. Nature, 418, Huang, J., & Rozelle, S. (1995). Environmental stress and grain yields in China. American Journal of Agricultural Economics, 77, Leemans, R., & Solomon, A. M. (1993). Modeling the potential change in yield and distribution of the Earth s crops under a warmed climate. Climate Research, 3, Li, C., Zhuang, Y., Frolking, S., Galloway, J., Harriss, R., Moore III, B., et al. (2003). Modeling soil organic carbon change in croplands of China. Ecological Applications, 13, Li, X. (1999). Change of arable land area in China during the past 20 years and its policy implications. Journal of Natural Resources, 4, (in Chinese). Li, X., Ming, T., & Hua, L. (2004). Impact of land-use change on grainland resources and food security in China. Abstract of the 4th international colloquium on land use/land cover change and environment issues in Asia, Beijing. Liu, J., Liu, M., Deng, X., Zhuang, D., Zhang, Z., & Luo, D. (2002). The land use and land cover change database and its relative studies in China. Journal of Geographical Sciences, 12, Liu, J., Liu, M., Zhuang, D., Zhang, Z., & Deng, X. (2003a). Study on spatial pattern of land-use change in China during Science in China. Series D, 46, Liu, J., Tian, H., Liu, M., Zhuang, D., Melillo, J. M., & Zhang, Z. (2005). China s changing landscape during the 1990s: Large-scale land transformations estimated with satellite data. Geophysical Research Letters, 32, L02405, doi: /2004gl Liu, J., Zhuang, D., Luo, D., & Xiao, X. (2003b). Land-cover classification of China: Integrated analysis of AVHRR imagery and geo-physical data. International Journal of Remote Sensing, 24, Liu, M., Tang, X., Liu, J., & Zhuang, D. (2001). Research on scaling effect based on 1 km grid cell data. Journal of Remote Sensing, 5, Liu, M., Tang, X., Zhuang, D., & Liu, J. (2003c). A new technique on spatial temporal data fusion and the construction of structural grid data platform. Geo-Information Science, 5(4), (in Chinese). Liu, Y., (Ed.) (2000). Data volume of land resources investigation in China. Beijing State Land Resources Investigation Office (in Chinese). Loveland, T. R., & Belward, A. S. (1997). The IGBP-DIS global 1km land cover dataset, DISCover: First result. International Journal of Remote Sensing, 18, Loveland, T. R., Reed, B. C., Brown, J. F., Ohlen, D. O., Zhu, J., Yang, L., et al. (2000). Development of a global land cover characteristics database and IGBP DISCover from 1-km AVHRR Data. International Journal of Remote Sensing, 21, Ma, K., (Ed.) (2000). Technology of land resources investigation in China. Beijing Chinese Land Publisher. Ministry of Land and Resources of China (MLR). (2000, 2001, 2002, 2003). Annual report on chinese land resources. Beijing Geological Publishing House (in Chinese). Ramankutty, N., & Foley, J. A. (1998). Characterizing patterns of global land use: an analysis of global croplands data. Global Biogeochemical Cycles, 12, Ramankutty, N., Foley, J. A., Norman, J., & Mcsweeney, K. (2002). The global distribution of cultivable lands: Current patterns and sensitivity to possible climate change. Global Ecology and Biogeography, 11, Seto, K. C., Kaufman, R. K., & Woodcock, C. E. (2000). Landsat reveals China s farmland reserves, but they re vanishing fast. Nature, 406, 121. SLA. (1996, March). Department for planning and finance of the State Land Administration. Summary of national land use during the eighth five-year plan. Smil, V. (2000). China s energy and resource uses: Continuity and change. In R. L. Edmonds (Ed.), Managing the Chinese environment (pp ). Oxford Oxford University Press. State Statistical Bureau (SSB). (1991, 1992, 1992, 1994, 1995, 1996). Statistical yearbook of China. Beijing China Statistical Publisher House. State Statistical Bureau (SSB). (2003). Statistical yearbook of China. Beijing China Statistical Publisher House. Tang, X. (2000). Studies on geo-spatial data fusion and its applications, Ph.D. thesis, Beijing: Institute of Remote Sensing Applications, Chinese Academy of Sciences (in Chinese). Tian, H., Melillo, J. M., Kicklighter, D. W., Pan, S., Liu, J., McGuire, A. D., et al. (2003). Regional carbon dynamics in monsoon Asia and its implications for the global carbon cycle. Global and Planetary Change, 3, Verburg, P. H., & Denier van der Gon, H. A. C. (2001). Spatial and temporal dynamics of methane emissions from agricultural sources in China. Global Change Biology, 7(1), 31 47, doi: /j x. Verburg, P. H., & Veldkamp, A. (2001). The role of spatially models in land-use change research: A case study for cropping patterns in China. Agricultural Ecosystems and Environment (1 3), Wang, M. (2001). Methane emission from paddy land in China. Beijing Science Press (in Chinese).