INT. J. REMOTE SENSING, VOL. 25, NO. 20 OCTOBER, 2004, 20, 4177–4189 Assessment of grassland degradation near Lake Qinghai, West China, using Landsat TM and in situ reflectance spectra data Y. LIU Institute of Geographical Sciences and Natural Resources Research, Chinese Academy of Science, Beijing 100101, China; e-mail: [email protected] Y. ZHA College of Geographic Science, Nanjing Normal University, Nanjing 210097, China; e-mail: [email protected] J. GAO* School of Geography and Environmental Science, University of Auckland, Private Bag 92019, Auckland, New Zealand; e-mail: [email protected] and S. NI College of Geographic Science, Nanjing Normal University, Nanjing 210097, China (Received 25 February 2003; in final form 22 December 2003 ) Abstract. The severity of grassland degradation near Lake Qinghai, West China was assessed from a Landsat Thematic Mapper (TM) image in conjunction with in situ samples of per cent grass cover and proportion (by weight) of unpalatable grasses (PUG) collected over 1 m2 sampling plots. Spectral reflectance at each sampling plot was measured with a spectrometer and its location determined with a Global Positioning System (GPS) receiver. After radiometric calibration, the TM image was geometrically rectified. Ten vegetation indices were derived from TM bands 3 and 4, and from the spectral reflectance data at wavelengths corresponding most closely to those of TM3 and TM4. Regression analyses showed that NDVI and SAVI are the most reliable indicators of grass cover and PUG, respectively. Significant relationships between TM bands-derived indices and in situ sampled grass parameters were established only after the former had been calibrated with in situ reflectance spectra data. Through the established regression models the TM image was converted into maps of grass cover parameters. These maps were merged to form a degradation map at an accuracy of 91.7%. It was concluded that TM imagery, in conjunction with in situ grass samples and reflectance spectra data, enabled the efficient and accurate assessment of grassland degradation inside the study area. 1. Introduction Grassland in China is distributed mostly in Inner Mongolia, the Loess Plateau, and the Qinghai-Tibetan Plateau, totalling 313.33 Mha. These areas have a *Corresponding author. International Journal of Remote Sensing ISSN 0143-1161 print/ISSN 1366-5901 online # 2004 Taylor & Francis Ltd http://www.tandf.co.uk/journals DOI: 10.1080/01431160410001680419 4178 Y. Liu et al. semi-arid to arid climate that is highly vulnerable to degradation, desertification, and salinization. Such destructive processes have affected about one third of the grassland. Moreover, degraded grassland is expanding at an annual rate of 667 000 ha (Bao et al. 1998). Grassland degradation refers to the overall reduction in grassland productivity as a consequence of human activities and natural processes. It may manifest as a reduction in the extent of grass cover, density of grass cover, and output of forage, or as an increase in unpalatable grass species, and even as denudation of underlying soil. Grassland degradation is rather severe in Qinghai Province with moderately and severely degraded land (e.g. density of grass cover v31% and proportion of unpalatable grass (PUG) species w51%) totalling 7.266 million ha. This accounts for nearly 20% of the entire grassland. Grassland degradation has caused land carrying capacity to decrease from supporting 1.49 goats or their equivalent grass consumption per hectare in the 1950s and 1960s to 0.75 goats per hectare in 1998 (Bao et al. 1998). Grassland degradation is also responsible for causing a loss of 12 million tons of hay (Wang and Li 1999). Grassland degradation in this region is caused primarily by human factors, and secondly by environmental factors. In this predominantly grazing region, excessive reliance on animal husbandry under a growing population has exerted great pressure on the land (Wei et al. 1997). The degradation problem has also been exacerbated by attacks of locusts and rats, and by droughts (Shi 2000). In order to bring grassland degradation under control and make grazing sustainable, it is very important to study its severity and ascertain its spatial distribution. Grassland degradation is conventionally studied through field investigation during which its contributing factors are identified. Each of them is then graded, and the overall severity of degradation is derived by combining the effect of all contributors (Li 1997). This method of assessment is inefficient and costly because grassland usually covers a large spatial extent in remote and inaccessible areas (Asrar et al. 1986). In addition, the results obtained from the traditional method of assessment are unreliable due to the contributing factors being difficult to be mapped at a high level of accuracy. By comparison, remote sensing is much more efficient in assessing grassland degradation (Liu et al. 2001). Remotely sensing grassland degradation is achievable either indirectly or directly. In the indirect method a productivity model is constructed from climate and vegetation variables, some of which can be derived from satellite images. These variables include trends in average vegetation cover with distance from water at the end of every wet period, and trends in cover variance with distance from water (Pickup et al. 1994). The role of remote sensing in the assessment is to supply information on vegetative cover. However, grassland degradation cannot be assessed based solely on remotely sensed data. Thus, this indirect method falls outside the scope of this paper. Instead, this study focuses on the assessment from grass parameters such as per cent cover, biomass, and proportion of unpalatable grass species, all of which can be estimated directly from satellite data, usually through the use of a Vegetation Index (VI) (Williamson and Eldridge 1993, Hobbs 1995, Purevdorj et al. 1998). VI is an arithmetic disparity between pixel values in two or more spectral bands of the same imagery, aiming to maximize the vegetative signal while minimizing other effects (e.g. soil and other background covers) in the resultant image. Therefore, VIs are able to reveal potential and current status of grassland degradation. In particular, a new index called Moving Standard Deviation Index (MSDI) has been proposed to study degradation (Tanser and Grassland degradation in China 4179 Palmer 1999). This landscape diversity index is a powerful adjunct to Normalized Difference Vegetation Index (NDVI) in monitoring degradation patterns in a semiarid heterogeneous landscape. Several researchers have assessed severity of grassland degradation using satellite imagery (Shi et al. 1999, Liu et al. 2001). Shi et al. (1999) produced a map of alpine grassland degradation through visual interpretation of Landsat Thematic Mapper (TM) and SPOT images. Liu et al. (2001) mapped the severity of grassland degradation using both unsupervised and supervised classification of TM data. Wang et al. (2002) first classified NDVI images derived from Landsat Multispectral Scanner (MSS) and TM images using an unsupervised classification method, and then evaluated the severity of grassland degradation based on the classified results. Eve et al. (1999) took advantage of this computer-based automatic method to classify Advanced Very High Resolution Radiometer (AVHRR) data into several classes, one of which corresponded to irreversibly degraded grassland. These studies demonstrate that both visual interpretation and supervised classification have been applied to map grassland degradation. However, both methods are relatively slow. Although the unsupervised method is much faster, its results are not so useful. The aim of this study is to devise an efficient and simple method for quickly assessing grassland degradation by means of remote sensing. The specific objectives are: (a) to identify grassland degradation indicators that are critical to animal husbandry and that can be studied by means of remote sensing; (2) to determine the vegetation indices that are most effective in assessing grassland degradation; and (3) to develop an automatic method for efficiently and timely mapping severity of grassland degradation in West China. 2. Study area At approximately 37‡N and 99.5‡E, the study area is located in the northeast of Qinghai Province, western China (figure 1). It has a semi-arid climate with an annual temperature ranging from 21.5‡C to 3.4‡C. Terrain in this alpine province has an elevation between 3200 and 3800 m above sea level (asl). The amount of annual precipitation totals about 381.4 mm, most of which falls during the summer months between June and September. The frigid climate associated with the high altitude makes the growing season rather short. Consequently, grassland productivity is rather low at 144.9 g m22 during wet years, but only at 60.7 g m22 during dry years (Wei et al. 1997). Grass cover exhibits a clear distinction at around 3300 m asl, above which grassland is meadow or bushy meadow. Owing to the moist air, grass cover generally exceeds 90% without obvious traces of degradation. Below this elevation grassland is semi-arid steppe where grass cover varies considerably from below 20 to over 90%. This area was selected for study because it is one of the most important bases of animal husbandry in Qinghai Province and because grassland degradation has become a serious concern here. Overgrazing by yaks and sheep, together with locust- and rat-induced destruction, has resulted in grassland degradation (Chen et al. 1994). Grazing-induced degradation is especially severe during early winter and late spring (Ni et al. 1999). Monitoring of grassland degradation over such a vast region requires efficient mapping of grass cover parameters from satellite imagery. 4180 Y. Liu et al. Figure 1. Location of the study area and its vicinity. 3. Methodology 3.1. Remotely sensed data A half-scene Landsat TM image recorded on 17 July 2000 was acquired, from which a subscene of 2200 by 2200 pixels was delimited. This subscene image was radiometrically processed using the Pons and Solé-Sugrañes’ (1994) model, and geometrically rectified to the Gausse-Krüger coordinate system using six ground control points at a residual of 0.1454 pixels. All image processing was undertaken in ENVI1 (version 3.2). 3.2. In situ sampling Fieldwork was carried out on 21 and 22 July 2000 within five days of satellite overpass. In the field a total of 60 sampling plots of 1 m2 in size were randomly selected. Per cent grass cover within each plot was visually estimated by two to three experienced local rangers to the nearest 5%. Of these 60 sites, 55 estimates were useable. The remaining five sites either fell into the same pixel or were blocked by cloud on the satellite image. Grass within each plot was then clipped and immediately weighed. The process was repeated twice, first for unpalatable grass species and then for all remaining ones. Grass samples were collected at 22 of the 60 sampling plots. Their precise location was determined with a Garmin International 12XLC portable global positioning system (GPS) receiver. It has 12 channels with a horizontal accuracy of 10 m. At each site two consecutive readings were logged with their average as its final coordinates. In addition, the spectral reflectance of grass cover at the sampling plots was Grassland degradation in China 4181 measured at 17 wavelengths over the range of 400–1050 nm with a Japan-made PM12A spectrometer. The spectrometer was mounted on a tripod about 1 m above the ground. The intensity of the returned electromagnetic radiation from the target within its 10‡ field of view was converted into reflectance. Reflectance was measured twice at every sampling site, first for the grass cover and shortly afterwards for the reference board. 3.3. Data analysis Of the 17 measured spectral reflectance data, only three at 650, 675, and 850 nm were further analysed. The reflectance at 650 and 675 nm was averaged first and the averaged reflectance corresponds most closely to the approximate wavelength range of TM3 while the reflectance at 850 nm corresponds to that of TM4. These reflectance data were considered the equivalent spectral response of TM3 and TM4, from which more than 10 vegetation indices were derived. The in situ measured per cent grass cover and PUG were considered as the dependent variables. They were linearly regressed against these indices (independent variables) separately. Of the 10 established univariate regression models, the one with the largest R2 value was regarded as the most accurate, and the VI in the regression model was considered as the most effective estimator of the dependent grass variables, namely, per cent grass cover and PUG. The geometrically rectified and radiometrically calibrated TM bands 3 and 4 were used to derive VIs that had been found most effective in estimating grass cover parameters. On these resultant index images, pixels at positions corresponding to the ground sampling plots were determined. Regression analyses were undertaken to construct an empirical relationship between these raw index values and the grass cover parameters measured on the ground. The same regression analysis was repeated using calibrated index values, also. Calibration of the raw index derived from the TM bands for sampling plot j (denoted as Ij) to its reflectance counterpart Rj was achieved using equation (1) (Zha et al. 2003): Rj {Rmin Ij ~ |ðImax {Imin ÞzImin ð1Þ Rmax {Rmin where Imax and Imin stand for the maximum and minimum index values derived from TM bands 3 and 4, respectively; Rmax and Rmin are the maximum and minimum index values derived from the in situ measured reflectance data, respectively. The established regression models were used to transform the TM image data to maps of degradation distribution. After being assigned an equal weight, results from the two grass parameters were merged to form an overall degradation map that was visualized at four severity levels to show the spatial pattern of degradation risk. 4. Results 4.1. Effectiveness of VIs In order to find out which VI is the most suitable for studying grassland degradation in this semi-arid area, a number of VIs were derived and used in the univariate linear regression analyses. These indices included Difference VI (DVI), ratio VI (RVI), NDVI, Soil Adjusted VI (SAVI), Modified SAVI (MSAVI), Infrared Percentage VI (IPVI), and Modified Simple Ratio (MSR) (refer to Davidson and Csillag 2001 for their calculation formulas). In the analyses, two indicators of grassland degradation (per cent grass cover and PUG) were used as 4182 Y. Liu et al. the independent variables (table 1). The R2 value of these equations ranges from 0.33 to 0.74 for per cent grass cover. In general, the prediction from a single band is less accurate than from multiple bands. This outcome demonstrates that the use of more spectral bands will produce more reliable results for grassland. All R2 values are extremely similar to one another if the same number of spectral bands is used in deriving the index. Three indices (NDVI, IPVI and MSR) achieved the same largest R2 value of 0.74. We decided to use NDVI because it is relatively easy to derive, most commonly used, and hence most understood. The univariate regression models in which the dependent variable is PUG have an R2 value between 0.41 and 0.73. Four indices (DVI, SAVI, RDVI and MSAVI) achieved the same maximum R2 value, all of which are quite different from those for per cent grass cover. Therefore, the same VI may not be the best estimator for all grass parameters. Among these four indices, SAVI was selected because it is most commonly used. 4.2. Degradation from grass cover The in situ measured per cent grass cover varies widely from 35 to 75% with a mean of 58.1%. These values have a standard deviation of only 10.2. It seems that those covers outside this range are under represented in the 55 samples. This could have been caused by the limited spatial extent of sampling plots in the field. Grass cover percentages do not bear any definite relationship with the value of their corresponding pixels on the NDVI image derived from TM3 and TM4 as revealed by a scatter plot. However, grass cover at a given sampling plot is closely associated with its spectral reflectance-derived NDVI value. The R2 value (0.74) of Table 1. Accuracy of predicting per cent grass cover and proportion (by weight) of unpalatable grasses (PUG) from various vegetation indices judged by the R2 value of the regression equation. Regression model Grass parameter – dependent variable Per cent cover PUG Vegetation Index – independent variable Coefficient Interception R2 RED IR DVI RVI NDVI SAVIL~0.5 RDVI MSAVI IPVI MSR RED IR DVI RVI NDVI SAVIL~0.5 RDVI MSAVI IPVI MSR 2291.29 106.9 122.38 13.85 108.18 115.29 118.95 107.63 216.36 37.13 370.33 2252.26 2210.12 217.29 2137.26 2165.39 2174.14 2155.66 2274.52 246.59 104.41 9.63 22.10 17.56 6.36 12.57 13.53 16.52 2101.82 22.56 227.08 141.78 90.77 81.11 96.29 95.24 95.06 90.01 233.55 75.16 0.52 0.33 0.60 0.72 0.74 0.70 0.69 0.69 0.74 0.74 0.41 0.60 0.73 0.62 0.68 0.73 0.73 0.72 0.68 0.65 Grassland degradation in China 4183 the regression relationship suggests that grass cover can be mapped adequately from TM image data if the TM-derived NDVI values can be associated with their reflectance-derived counterparts. Nevertheless, these two sets of NDVI values do not appear to be correlated in any way. Upon further examination it was found that those NDVI values derived from the in situ measured reflectance have a broad range from 0.2 to 0.7 while those from the TM image have a narrow range of 0.4–0.6. Therefore, the TM-derived NDVI data were calibrated to the NDVI obtained from in situ measured reflectance. After calibration the relationship between grass cover visually estimated on the ground and TM image-derived NDVI values became much closer at an R2 value of 0.74 (figure 2). Such a close relationship (equation (2)) demonstrates that grass cover could be mapped from the TM data accurately. The NDVI image was later converted to a map of per cent grass cover based on equation (2). ð2Þ Grass Cover ð%Þ~210:57 NDVI{47:915 R2 ~0:74 4.3. Degradation from PUG The 22 PUG observations have a range from 13.0 to 60.0 with a mean of 29.9 and a standard deviation of 13.2. Again, there does not seem to be a huge variation among them. This could be explained by the fact that most of them were sampled within a limited spatial extent. The establishment of a regression model for PUG followed the same procedure for per cent grass cover except that NDVI was replaced by SAVI. The raw SAVI data derived from the TM bands, and those SAVI indices derived from in situ measured reflectance spectra data were plotted separately against PUG. A significant relationship was found to exist for the reflectance-derived SAVI only, but not for the TM image-derived SAVI. The SAVI data were calibrated to the former. After calibration, PUG had a linear and statistically significant correlation with the image-derived SAVI (figure 3). ð3Þ Unpalatable Grass ð%Þ~{318:11 SAVIz152:13 R2 ~0:73 The high R2 value of 0.73 (equation 3) suggests that PUG can be estimated from the Figure 2. Regression relationship between grass cover (%) estimated on the ground at sampling plots and its TM-derived NDVI that has been calibrated with 55 in situ measured reflectance values. GC~210.576NDVI247.915 (R2~0.74). 4184 Y. Liu et al. Figure 3. Regression relationship between proportion of unpalatable grass (%) and its TMderived SAVI that has been calibrated with 22 in situ measured reflectance samples. PUG~2318.116SAVIz152.13 (R2~0.73). TM image data at a reasonably high accuracy level. This equation was subsequently used to transform the TM image into another degradation map. 4.4. Assessment of overall degradation The two degradation maps independently produced from the two grass cover parameters were combined to form an overall map so that degradation could be assessed more efficiently. This was achieved through linearly adding the two variables up with a weight of 0.5 assigned to each (equation 4). Risk of degradation~0:5 ð1{grass coverÞz0:5 PUG ð4Þ This map contained continuously varying pixel values that had to be categorized into a few groups in order to effectively visualize the spatial distribution of degradation risk. In the literature several criteria have been proposed to define a specific level of degradation severity. In terms of per cent grass cover, less than 10% is considered severe, less than 30% moderate, and less than 50% slight (Shi et al. 1999). Similarly, Shi and Wang (1994) defined a reduction of 10% in per cent grass cover as slight, 10–35% as moderate, and over 35% as severe. For PUG, the thresholds are more than 70% as severe, 50–70% as moderate, 30–50% as slight. In 2002 the Chinese Ministry of Agriculture proposed the most authoritative criteria for the entire country of China. They are w50% as severe for both reduction in grass cover and PUG, 21–50% as moderate, 6–20% as light for grass cover change and 11–20 for PUG. Apparently, there are no universally accepted criteria. Based on these commonly used ones, we proposed and adopted a set of new criteria for the study area (table 2). Coincidentally, the threshold values were identical for both per cent grass cover and PUG. The overall risk map (figure 4) shows four levels of degradation severity: severe, intermediate, slight, and intact. Roughly three quarters of the study area has been affected by degradation to various levels of severity (table 2). Severely degraded areas make up the smallest class at 13.37%. Patches of denudated and compacted soil have emerged in these areas (figure 5(a)). Their overall productivity has dropped by 60% (Anonymous 1999). Intermediately degraded areas compose the largest class of the four at 35.32% (table 2). They have shown clear signs of change in grass species composition and appearance (figure 5(b)), and their overall productivity has Grassland degradation in China Table 2. Indicators of grassland degradation, criteria for degradation severity, and statistics of degraded grasslands. Severity level Severe Moderate Slight Intact 4185 Per cent cover Unpalatable grasses (%) Area (km2) % f30 31–50 51–70 w70 w70 51–70 31–50 f30 10.39 30.72 21.49 18.40 12.83 37.92 26.53 22.72 dropped by 30–60%. The exposed soil is compact with a dry surface. Slightly degraded areas (28.61%) make up the second most common severity class at an area of 23.17 km2. They have not shown profound changes in grass species composition and appearance (Figure 5(c)). The reduction in its overall productivity is less than 30% (Anonymous 1999). The areas that have not been affected by degradation yet (figure 5(d )) make up nearly a quarter of the study area. Consequently, they have not shown any void space between grasses. 5. Discussion 5.1. Criteria for selecting degradation indicators It is not easy to determine what criteria should be used for realistically assessing grassland degradation. Ludwig (1986) argued that estimates of net primary Figure 4. A map showing the distribution pattern of severity of grassland degradation. It has been median filtered with a widow size of 3 by 3 pixels. Ground area covered is 9 km by 9 km. 4186 Y. Liu et al. (a) (b) (c) (d ) Figure 5. Typical grassland areas that have been affected by degradation at various severity levels. (a) An example of severe degradation. Grass is sparsely distributed with a large proportion of unpalatable grasses. The denudated soil is compacted with a low fertility. (b) An example of moderate degradation. Bare and compacted soil is widespread with unpalatable grasses clearly visible. (c) An example of slight degradation. Patches of bare soil are clearly visible. (d ) An example of no degradation. Grass cover is mostly continuous. Grassland degradation in China 4187 production based on VI cannot be used reliably to indicate degradation status because degradation does not necessarily result in slowed plant growth. Instead, diversity should be used as the indicator. Unlike this view, we assessed grassland degradation from the productivity perspective because it is more relevant to animal husbandry. From this perspective, the identified criteria include above-ground biomass, ground cover, soil, structure of ecosystem, and recovery (Li 1997). Shi and Wang (1994) used only grass cover, productivity, and proportion of palatable grasses for the same purpose. In this study only two of these critical indicators (grass cover and PUG) were used, both of which can be estimated from remotely sensed data. Although the number of assessment criteria used is not comprehensive in this paper, the obtained results have practical values. 5.2. Importance of data calibration In this study we did not find a strong relationship between in situ sampled grass parameters and the value of their corresponding pixels on the TM bands-derived VI image. The absence of such a correlation is attributed to the atmospheric effect, even though the satellite image has been radiometrically rectified. It is speculated that the atmospheric effect was tackled during radiometric rectification, but was not eliminated completely. Residual atmospheric effect (e.g. atmospheric attenuation) suppressed the range of satellite data. Their correct range was restored after the satellite data-derived indices were calibrated with in situ measured reflectance spectra data which were closely correlated with in situ measured grass parameters. Data calibration based on equation (1) essentially stretches the range of satellite data-derived VIs. Data stretching enables the establishment of a strong correlation between in situ measured grass parameters and their corresponding pixel values on the satellite image. Thus, the combination of in situ measured reflectance spectra with a Landsat TM image appears to be a viable method in retrieving grassland parameters from satellite data (Zha et al. 2003), and hence in assessing grassland degradation. 5.3. Accuracy of assessment The accuracy of the produced degradation map (figure 4) was verified during a field visit in July 2002. In total, 36 evaluation points were selected for this purpose. Field-observed severity of degradation at 33 of these points was consistent with its counterpart shown in the map, leading to an accuracy of 91.7%. Admittedly, this high accuracy level is related to the fact that degradation was mapped at only four categories of severity. The accuracy would certainly be lower had the severity been visualized at more categories. The fact that a higher accuracy of assessment was not achieved and that the highest R2 value achieved in all regression analyses was only 74% is caused by a variety of factors, the most important being differential sampling sizes on the ground and from space. Recovery of physical features such as per cent grass cover and PUG from satellite data requires concurrent sampling. Logistic difficulty makes it impossible to carry out in situ sampling over an extensive area. Besides, PUG estimates become increasingly unreliable over a larger sampling plot. Therefore, it is rather common for the in situ sampling size to be smaller than that from space. As a matter of fact, the in situ sampling size of roughly 1 m2 adopted in this study is much smaller than the sampling size of 30 m by 30 m on the TM image. Friedl et al. (1994) speculated that such differential sampling intervals on the ground and from 4188 Y. Liu et al. space made up the largest source of error in the Kauth-Thomas greenness-based estimates of assigning point-based ground data to area-integrated measurements from satellite. This is certainly applicable to this study, even though the exact contribution by this factor is impossible to quantify. Nevertheless, the high accuracy level of 91.7% achieved demonstrates that differential sampling sizes did not invalidate the assessed distribution of grassland degradation. 6. Conclusions In this study two common degradation indicators (per cent grass cover and PUG) were used to assess severity of grassland degradation near Lake Qinghai in Qinghai Province, West China. The best VI for studying per cent grass cover in this semi-arid environment is NDVI whose performance is very similar to IPVI and MSR. The VI most suitable for estimating PUG from TM imagery is SAVI, even though three other indices achieved equally accurate results. It is impossible to map these two degradation indicators directly from the TM image data with the assistance of in situ grass samples collected over sampling plots of 1 m2 in size. However, statistically significant regression models were established between the two variables at an R2 of 0.74 after the TM bands-derived index values were calibrated with the in situ measured reflectance spectra data. The use of in situ measured spectral reflectance data successfully reduced the errors in the assessment caused by differential sampling sizes on the ground and from space. A simplified degradation map of four levels of risk was generated at an accuracy of 91.7%. It shows that overall, three quarters of the area under study has been affected by degradation at various levels of severity. This VI based method of assessment, in conjunction with in situ measured grass cover parameters and spectral reflectance data, enables grassland degradation to be assessed efficiently and accurately. Acknowledgments This research was partly funded by grants from the National Natural Science Foundation of China (grant no. 49 971 056 and 40 171 007). Additional financial support was received from the Key Science and Technology Project at the Key Land Use Laboratory, the Chinese Ministry of Land and Resources (grant no. 20 010 102). We appreciate the valuable comments from two anonymous reviewers on an earlier version of this manuscript. References ANONYMOUS, 1999, Local degradation criteria for natural grassland in Inner Mongolia Autonomous Region. Grassland of Inner Mongolia (in Chinese), 1999, 61–62. ASRAR, G., WEISER, R. L., JOHNSON, D. E., KANEMASU, E. T., and KILLEEN, J. M., 1986, Distinguishing among tallgrass prairie cover types from measurements of multispectral reflectance. Remote Sensing of Environment, 19, 159–169. BAO, W., SHAN, W., YANG, X., SUN, H., and LAN, Y., 1998, Ecological crises facing the grassland resources in northern China and their solutions. Grassland of China (in Chinese), 1998, 68–71. CHEN, G. C., PENG, M., ZHOU, L. H., and ZHAO, J., 1994, Preliminary study on the relationship between ecological evolution and human activities in the Qinghai Lake region. Chinese Journal of Ecology (in Chinese), 13, 44–49. DAVIDSON, A., and CSILLAG, F., 2001, The influence of vegetation index and spatial resolution on a two-date remote sensing-derived relation to C4 species coverage. Remote Sensing of Environment, 75, 135–151. EVE, M., WHITEFORD, W. G., and HAVSTADT, K. M., 1999, Applying satellite imagery to Grassland degradation in China 4189 triage assessment of ecosystem health. Environmental Monitoring and Assessment, 54, 205–227. FRIEDL, M. A., MICHAELSEN, J., DAVIS, F. W., WALKER, H., and SCHIMEL, D. S., 1994, Estimating grassland biomass and leaf area index using ground and satellite data. International Journal of Remote Sensing, 15, 1401–1420. HOBBS, T. J., 1995, The use of NOAA-AVHRR NDVI data to assess herbage production in the arid rangelands of central Australia. International Journal of Remote Sensing, 16, 1289–1302. LI, B., 1997, Rangeland degradation in northern China and strategies for its prevention. Scientia Agricultura Sinica (in Chinese), 30, 1–8. LIU, Z., YAN, M., WANG, G., MENG, H., ZHANG, W., and WANG, C., 2001, Analysis on grassland degeneration in West Jilin Province based on satellite remote sensing. Scientia Geographica Sinica (in Chinese), 21, 452–456. LUDWIG, J., 1986, Primary production variability in desert ecosystems. In Pattern and Process in Desert Ecosystems, edited by W. G. Whiteford (Albuquerque: University of New Mexico Press), pp. 5–17. NI, S. X., GONG, A. Q., JIANG, J. J., WANG, W. J., and WANG, J. C., 1999, Ecological and environmental problems and their rehabilitation in the Qinghai Lake region. Resources Science (in Chinese), 21, 43–46. PICKUP, G., BASTIN, G. N., and CHEWINGS, V. H., 1994, Remote-sensing-based condition assessment for non-equilibrium rangelands under large-scale commercial grazing. Ecological Application, 4, 497–517. PONS, X., and SOLÉ-SUGRAÑES, L., 1994, A simple radiometric correction model to improve automatic mapping of vegetation from multispectral satellite data. Remote Sensing of Environment, 48, 191–204. PUREVDORJ, T., TATEISHI, R., ISHIYAMA, T., and HONDA, Y., 1998, Relationship between percent vegetation cover and vegetation indices. International Journal of Remote Sensing, 19, 3519–3535. SHI, D., QIAO, A., SAI, W., HON, X., and HODGSON, N., 1999, Applied research on use of remote sensing to study alpine grassland resource and degradation. Grassland of Qinghai (in Chinese), 8, 1–6. SHI, S., and WANG, L., 1994, Status of grassland degradation in Qinghai and strategies for its prevention. Grassland of Qinghai (in Chinese), 3, 5–11. SHI, Y., 2000, Predictable huge environmental change in the Qinghai-Tibetan Plateau, Science Times (in Chinese), 8 May 2000. TANSER, F. C., and PALMER, A. R., 1999, The application of a remotely-sensed diversity index to monitor degradation patterns in a semi-arid, heterogeneous, South African landscape. Journal of Arid Environments, 43, 477–484. WANG, P.-X., CHEN, X.-L., and LI, F.-P., 2002, A study on dynamic monitoring rangeland degradation and its distribution in the Xilin Gol Plateau’s dry steppe. Agricultural Research in Arid Areas (in Chinese), 20, 92–94, 106. WANG, X., and LI, Y., 1999, Current status of grassland ecologic environment and strategies for its rehabilitation in Qinghai Province. Grassland of Qinghai (in Chinese), 8, 23–25. WEI, K., HU, R., MA, J., and WANG, X., 1997, Research on environmental issues of grassland ecology around Lake Qinghai and strategies for their prevention. Grassland of Sichuan (in Chinese), 1997, 1–5. WILLIAMSON, H. D., and ELDRIDGE, D. J., 1993, Pasture status in a semi-arid grassland. International Journal of Remote Sensing, 14, 2535–2546. ZHA, Y., GAO, J., NI, S., LIU, S., JIANG, J., and WEI, Y., 2003, A spectral reflectance-based approach to quantification of grassland cover from Landsat TM imagery. Remote Sensing of Environment, 87, 371–375.
© Copyright 2026 Paperzz