Time-series satellite data analysis for assessment of vegetation cover of Mongolia

INTRODUCTION

Mongolia is a land-locked country located in Central Asia between Russia and China. The climate is continental with harsh winters and hot short summers. The total population of Mongolia is 2.3 million with territory of 1.565 million sq. km, making it one of the lowest population density areas (1.5 persons per sq. km) in the world. The main economy is nomadic animal husbandry, with the portion of 33.4% in GDP. Forty-eight point six percent of the work force is in the livestock- breeding sector. Until 1990, Mongolia had a central planned economy with state owned farms of cattle, sheep, goat, camels and horses. There were also state-supported monitoring systems for livestock breeding, water supply and pastureland quality. After 1990, when Mongolia shifted towards market -oriented economy, privatization took place for much of the state owned properties and former structures for monitoring failed to function effectively. Privatization has stimulated the livestock industry, which has reached its maximum and pastureland has approached its maximum carrying capacity (Erdenebaatar et al. 2001). Recent extreme climate variations and pastureland deterioration brought waves of problems in this sector: high number of livestock mortality; decrease in pastureland productivity; increase in livestock diseases and as a result, a dramatic decrease in quality of lifestyle of herders’ families. Currently, livestock managers need accurate and timely data in the vegetation conditions in their pasturelands (Tserendash 2000). Despite the statement in the constitution of the country that “livestock is under care of the state” the government is not able to provide essential information for the herders to assist them in a way of sustainable livestock breeding. It is very time consuming and expensive to use ground measurement for pastureland quality, which makes it unaffordable for the central and local governments as well as the individuals over vast areas.

Remote Sensing for Pasture Management

The time –series information about location and condition of the vegetation of pasture areas is one of the key elements for the effective management of a pastoral (extensive) livestock breeding. Remotely sensed data derived from satellites have successfully being utilized for decades in assessment of pastureland productivity, predicting biomass and monitoring the vegetation health status (Reeves, 2001) in temporal and spatial scales and proved to be economically feasible measurements (Tueller 1989). Many researchers have studied vegetation growth and its productivity in two different directions. One is to establish empirical relationships between spectral reflectance and biomass (Tucker et al. 1983; Wylie et al. 1995) the other is the use of spectral reflectance to estimate the amount of absorbed photosynthetically active radiation (Choudhury 1987; Franklin 2001) for ecosystem modeling. The first method is mainly used to estimate active growing biomass and the estimated biomass is well correlated with remotely sensed vegetation indices (Tucker et al. 1983; Kennedy 1989; Thoma 1998). However, this method does not take the existing dry mass into account (Reeves 2001). The second method is more successful for predicting of biomass, but it’s based on regression models and depends on local environment parameters (Kennedy 1989; Merill et al. 1993; Wylie et al. 1996). Many researchers use coarse- to- high- resolution satellite and aerial images with a ground scale of 0.2 to 60 meters (Wylie et al 1995; Yool et al. 1997; Weber 2001) for pastureland assessment. However, the barriers in the use of high- resolution data are the data availability and high cost of processing. High temporal resolution (for over 20 years), wide area coverage, and availability/affordability make NOAA/ AVHRR satellite images very attractive for application in pasture assessment in the context of Mongolia. A few studies have taken place in Mongolia to estimate pastureland and vegetation dynamics from the remote sensing viewpoint. Purevdorj has studied seasonal growth changes using NOAA AVHRR data, taking into account the soil background reflectance and percent vegetation cover (1995). As a result, monthly green vegetation cover images were produced based on global land 1 km AVHRR data in 1995. The information presented here is an intention to bring NDVI time-series measurement as first –aid on-hand tools for those in agricultural sector of the country who lacks essential data for nomadic livestock management

NOAA AVHRR pre-processing

Multi-temporal NOAA 9, 10, 11, 14 and 16 dataset from the NOAA Satellite Active Archive (SAA) were used for this study. The period from 1989 to 2001 were analyzed, with exception to the 1994 data that did not exist for the region of Mongolia in the database of SAA. Data were processed using image- processing Erdas –Imagine 8.5 (ERDAS Inc 2001) and ArcGIS (Esri 2001) systems software. Only the afternoon pass of NOAA series images were collected within the time frame of July 1 to August 15 in each year (45 images in a year), which coincide with maximum vegetation growth seasons in Mongolia (Tserendash 1996). Due to extensive cloud coverage and high off-nadir areas in the years of 1993 and 1996, the time frame was expanded from June 20th to August 20th to capture additional images. This extension could cause higher variability of vegetation spectral responses (Jensen, 2000), because Mongolia’s frost seasons starts in August (NAP 1991), which can contribute to early vegetation senescent and thus, decrease the vegetation spectral response compared to other years.

Image Registration

Raw images were registered through image- to- image registration procedures using 25 ground control points on average and that were taken from existing vector coverage of lake, river and current administrative boundary of Mongolia. Depending on the satellite nominal flight height, images were distorted differently according to the surface curvature and that influenced the use of 2nd and 3rd polynomial orders, where applicable. To keep original values, the nearest neighbor method was used for resampling.

AVHRR Data Correction

Estimation of the spectral properties of vegetation cover using remote sensing methods, have been successful for many applications. However, extensive processing efforts related to geometric and atmospheric corrections are required. The basic sun angle correction 1/cosine (sun angle) and afternoon pass selection procedure were applied for each image data to decrease the influence of different sun elevations and bi-directional reflectance distribution function (BDRD). The inter-satellite calibration factors, an aim of which to normalize the satellite at –sensor reflectance coefficient in each satellite are have been applied (CIT 1999) for each image data through the radiometric correction procedures

Methods

Normalized Difference Vegetation Index

AVHRR –derived vegetation indices for the last three decades the proved to be a useful tool in depicting the large scale distribution and phenological changes of vegetation cover over particular regions of the world (Jensen, 2000). The most common methods available for estimating vegetation spectral responses for AVHRR data are the simple difference of visible and near infrared reflectance (NIR) DVI = NIR- Visible (Gutman, 1991), simple ratio SR = NIR/Red (Birth, 1968) and NDVI (Rouse et al, 1974)
NDVI = NIR – Visible/NIR + Visible
The principle used for the vegetation index is based on discontinuity of reflectance curve of healthy green vegetation at the 0.7-mkm regions. Green vegetation absorbs and reflects more energy in the visible red and near infrared regions. Whereas, senescent vegetation absorbs and reflects less in that regions. Based on these properties, different vegetation indices were developed taking into consideration various factors of soil, vegetation density, atmospheric effects that influences the outputs (Rouse et all, 1974; Sellers et al, 1994; Hall. et al, 1995). Compared to the simple difference and ratio of NIR and Visible bands, NDVI has less influence from sun angle and illumination and thus, provides relatively reliable information about photosynthetic processes going on in green vegetation (Gutman, 1991).

Maximum NDVI Decision Rule

NDVI has been calculated for each of the image scenes in a year and then the maximum NDVI decision rule (Holben 1986; Spanner et al. 1990; Burgan 1993; Roberts 1994) was applied to reduce cloud -contaminated pixels in image scenes and eliminate the differences of vegetation spectral responses due to phenological processes captured in the long-compositing period. Maximum NDVI decision rule employs the selection of highest NDVI pixel values from a scene to make a composite consisting of maximum reflectance of the image area over the chosen period of time. The compositing period of images was 45 days and phenological variability of vegetation cover within this period was then apparent (Reed 1994) during the process of composition. This could be a potential for misinterpreting inter-annual variability of vegetation cover, therefore, application of maximum NDVI was necessary to avoid this type of errors.

Departure from Average Vegetation Greenness

Long-term time series image data provides an opportunity to assess quantitatively and qualitatively the vegetation cover status in the past and present, to determine trends, and for the predicting the ecosystem processes (Nemani et al. 1997). An average of twelve- year NDVI data was computed for each image pixel and departure from its average was then calculated for each participating year to evaluate a yearly vegetation growth rate or greenness visually and statistically. The algorithm to produce the departure from average (Burgan 1996) is:
Depi = NDVIcur /NDVIaveg * 100
Where,
Depi – ith pixel’s departure value
NDVIcur – current NDVI
NDVIaveg - 12 -year average NDVI
100 – a multiplier to scale the output to be 100 if there is no departure
Departure from average vegetation greenness can be applied for inter-annual assessment of vegetation cover status over the whole territory of Mongolia. This temporally and spatially distributed information provides support to decision makers, planners, and agricultural managers and give the opportunity to compare each year in terms of vegetation growth, stress and productivity, resource allocation and pastureland overgrazing.

RESULTS

Vegetation Greenness Distribution

NDVI values are distributed unevenly on the image scene covering the entire territory Mongolia.
As a parameter well correlated with ongoing photosynthetic process in vegetation, NDVI higher values ranging 0.5-0.7 were distributed over the north-central part of the country, where dominate forest cover areas. NDVI 0.3- 0.49 values are mainly distributed in forest fringe areas extending from north- central to east of the country. Lower values of 0.1-0.29 dominate in middle part and ranged from west to east. The lowest values are associated with the areas of Gobi desert and Great Lakes Hollow. With some variations from year to year, NDVI spatial distribution conforms the boundary and extent of dominant natural ecological zones of Mongolia (Ramsey et al 1995; Biodiversity report 1998). Despite efforts to minimize influence of undesirable noises, some areas of Huvsgul, Hangai and Hentii mountainous taiga and forest regions still have cloud coverage on some images. Clouds are regular in these regions of Mongolia and cause an obstacle for optical remote sensing devices. According to the yearly solar radiation map, cloudless days in these areas are sixty (NAP, 1991). Clustering (ISODATA) with 7 classes of NDVI values ranging from 0.001 to 0.7 and 6 iterations (convergence threshold 0.95) was carried out to assess spatial distribution of NDVI quantitatively, over entire area of Mongolia for each scene. Transitional matrix is calculated based on the clusters to assess yearly the change of NDVI classes.

Transition Matrix of NDVI classes across the 2000 to 2001

The spectral responses of vegetation cover in 2001 were lower in all classes, except class 4. Dominant decreases were in classes 5 (7.25%) and 6 (9.4%) compared to 2000, whereas the class with no NDVI response was increased by 4.4%. Overall, the cluster classes and transitional matrix show the decrease in vegetation spectral responses from 2000 to 2001. Standard deviation of pixel values for 12 years has been plotted on the map below to exhibit the variability. Variation of spectral responses were very high from north to south and west to east spatially and it were also true temporally.

Distribution of high standard deviations is shown in the northern half of the country, including high mountains, taiga forest, and steppe areas. All forested areas, except Hentii mountain region, have high spectral variations; the causes can be the changes in precipitation regimes, temperature fluctuations as well as human induced disturbance factors. It is interesting to observe the similar high variability in Eastern steppe areas as well in high elevation mountainous areas in western most Bayan-Ulgii Province, despite their distinctive ecosystem nature. The lowest values of standard deviation are distributed over the Gobi desert and Great Lakes Hollow areas. The sandy areas with sparse vegetation areas are more stable in terms of spectral response over time. It was obvious from scene to scene, the changes in vegetation dynamics, as to NDVI is a direct function of precipitation and temperature in a particular year (Ichii et al. 2002). Accurate climate data could give explanations for these variations.

Average Greenness and Departure from Average

Departure from Average Vegetation Greenness method was employed to assess the changes in spectral responses of vegetation growth, from year to year. Yearly assessment of grass growth and biomass in broad scale and its comparison to other years has practical meaning for correct allocation of pasture resources, pastureland conservation and justification of livestock numbers. After screening for quality, ten consecutive years AVHRR NDVI data, except 1989 and 1996 that had missed some coverage, were averaged to create the average greenness image.
Then the average image has been subtracted from each annual composite of images. The departure values on an image composite range between 0-255 and value 100 + 10 % represents no change from average. Values over 100 stand for positive change i.e. more growth than the average and have been assigned light and dark green colors. Values below 100 stand for negative change and have yellow (10-20% below average) and red (more than 20% below average). Pseudo color composite has been created for each year departure (See above images).

The image differencing technique (Ross et al. 1998) has been applied to assess the vegetation growth changes over ten years, averaging the first and last two -year periods. The result shows spectral responses of vegetation cover in eastern steppe and central regions have negative changes and the areas of taiga forest in the north and western areas of the Great Lakes’ Hollow have high spectral responses that correspondents to good vegetation growth condition.

Conclusion

Broad-scale time –series NDVI images can be applied in conjunction with traditional methods for monitoring grassland condition and to some extend assess grassland productivity, changes across the time and space. Especially, it is useful where is lack of primary measurements of vegetation and pastureland conditions in broad areas and where climate dependent nomadic livestock is primary source of living. With the launch of NOAA KLM series satellite with AVHRR/3 sensors make it available to use many formerly available for high- resolution sensors, vegetation indices like Moisture Stress Index, Infrared Index Mid Infrared Index described by Jensen (2000). Especially, 3A channel (1.58-1.64 mkm EMS wavelengths) that measures moisture content of vegetation cover will be beneficial for NOAA satellite data users particularly, in the developing countries.

Reference:

Biodiversity National Report. 1998. UB. Mongolia. pp56.
Birth, G.S. and G. McVey, 1968. Measuring the color growing turf with a reflectance spectrophotometer. Agronomy Journal, 60:640-643
Burgan, Robert E.; Hartford, Roberta A. 1993. Monitoring vegetation greenness with satellite data. Gen. Tech. Rep. INT-297. Ogden, UT:U.S. Department of Agriculture, Forest Service, Intermountain Research Station. 13 p.
Burgan, Robert E.; Hartford, Roberta A.; Eidenshink, Jeffery C. 1996. Using NDVI to assess departure from average greenness and its relation to fire business. Gen. Tech. Rep. INT-GTR-333.Ogden, UT:U.S. Department of Agriculture, Forest Service, Intermountain Research Station. 8 p.
Cohen Warren B., Maria Fiorella. 1998. Comparison of methods for detecting conifer forest change with thematic mapper imagery. Remote sensing change detection. Ann Arbor Press. pp89-102
Choudhury, B. J. 1987. Relationships between vegetation indices, radiation absorption, and net photosynthesis evaluated by a sensitivity analysis. Remote sensing of environment.22:209-233.
Erdas. Inc. 2001. Erdas-imagine v. 8.5 Image processing application manual.
Erdenebaatar B., Batjargal N. 2001. Disputes of pasture ownership. UB. Mongolia. In Mongolian.
Esri. 2001. ArcGIS software manual.
Franklin, Steven E. 2001. Remote sensing for sustainable forest management. Boca Raton, Florida: Lewis Publishers
Gutman G. Garik. 1991. Vegetation indices from AVHRR: An update and future prospects. Remote Sens. Environ. 35:121-136
Hall, F. G., Townshend, J.R., and Engmann E. T. 1995. Status of remote sensing algorithms for estimation of land surface parameters. Remote Sensing of Environment. 51: 138-156
Holben, Brent N. 1986. Characteristics of maximum-value composite images from temporal AVHRR data. International Journal of Remote Sensing, Vol.7,No. 11, 1417-1434
Jensen J.R., 2000 Remote Sensing of the Environment: An Earth Resource Respective, Prentice Hall, Upper Saddle River, New Jersey, pp. 353.
Ichii, Kazuhito, Akira Kawabata, Yasushi Yamaguchi. 2002. Global correlation analysis for NDVI and climatic variables and NDVI trends: 1982 to 1990. Accepted paper. International journal of remote sensing.
Kennedy, P.1989. Monitoring the vegetation of Tunisian grazing lands using the normalized difference vegetation index. Ambio 18: 119-123
Merrill, E.H. M.K. Bramble-Brodahl R.W. Marrs and M.S. Boyce.1993. Estimation of green herbaceous phytomass from Landsat MSS data in Yellowstone National Park. J. Range manage. 46: 151-157
NOAA/NASA AVHRR Pathfinder Oceans –GAC product Generation Appendix. California Institute of Technology, 1999
National Atlas of Mongolia. 1991. Ulaanbaatar Mongolia pp54
Nemani, Ramakrishna and Steve Running. 1997. Land cover characterization using multitemporal red, near –ir and thermal-ir data from NOAA/AVHRR. Ecological applications, 7(1), pp79-90
Reeves, Matthew C., Jerome C. Winslow, Steven W. Running. 2001. Mapping weekly rangeland vegetation productivity using MODIS algorithms. Journal of Range Management. 54:A90-A105 March 2001
Reed, Bradley C., Brown, Jesslyn F., VanderZee, Darrel, Loveland, Thomas R., Merchant, James W., Ohlen, Donald O. 1994. Measuring phenological variability from satellite imagery. Journal of Vegetation Sceince 5: 703-714
Roberts Miles, Thomas W. Doyle. 1994. Component Analysis for interpretation of time series NDVI imagery. ASPRS/ACSM
Rouse, J. W., R. H. Haas, J. A. Schell and D.W. Deering, 1974. Monitoring Vegetation Systems in the Great Plains with ERTS. Proceedings, Third Earth Resources Technology Satellite-1 Symposium, Greenbelt: NASA SP-351, 3010-317
Sellers, P.J., Los S.O., Tucker, C. J., et al. 1994. A global 10 by 10 NDVI data set for climate studies. Part. 2: The adjustment of the NDVI and generation of global fields of terrestrial biophysical parameters. Int. J. Remote Sens. 11: 95-111.
Spanner, Michael A., Lars L. Rierce, Steven W. Running, David L. Peterson. 1990. The seasonality of AVHRR data of temperate coniferous forests: Relationship with leaf area index. Remote sensing of environment. 33:97-112
Thoma, D.1998. Near real-time satellite and ground based radiometric estimation of vegetation biomass, and nitrogen content in Montana rangelands. MS Thesis. Montana State University, Bozeman Mont.
Tserendash S. 2000. Some policy related issues for pasture management. Newsletter No. 6 (38). Mongolian Academy of Sciences.
Tueller, P.T. 1989. Remote sensing technology for rangeland management. Journal of Range Management. 42:442-452
Weber, Keith T. 2001. A method to incorporate phenology into land cover change analysis. Journal of Range Management. 54:A1-A7 March 2001
Wylie, Bruce K., Issa Denda, Rex D. Pieper, Iohn A Harringston, Jr., Bradley C. Reed, and G. Morris Southward. 1995. Satellite-based herbaceous biomass estimates in the pastoral zone of Niger. Journal of Range Management. 48:159-164 March 1995
Yool S. R., M.J. Makaio, and J. M. Watts. 1997. Techniques for computer –assisted mapping of rangeland change. J. Range Manage. 50:307-314