Elsevier

Remote Sensing of Environment

Volume 89, Issue 4, 29 February 2004, Pages 497-509
Remote Sensing of Environment

Land surface phenology, climatic variation, and institutional change: Analyzing agricultural land cover change in Kazakhstan

https://doi.org/10.1016/j.rse.2003.11.006Get rights and content

Abstract

Kazakhstan is the second largest country to emerge from the collapse of the Soviet Union. Consequent to the abrupt institutional changes surrounding the disintegration of the Soviet Union in the early 1990s, Kazakhstan has reportedly undergone extensive land cover/land use change. Were the institutional changes sufficiently great to affect land surface phenology at spatial resolutions and extents relevant to mesoscale meteorological models? To explore this question, we used the NDVI time series (1985–1988 and 1995–1999) from the Pathfinder Advanced Very High Resolution Radiometer (AVHRR) Land (PAL) dataset, which consists of 10 days maximum NDVI composites at a spatial resolution of 8 km. Daily minimum and maximum temperatures were extracted from the NCEP Reanalysis Project and 10 days composites of accumulated growing degree-days (AGDD) were produced. We selected for intensive study seven agricultural areas ranging from regions with rain-fed spring wheat cultivation in the north to regions of irrigated cotton and rice in the south. We applied three distinct but complementary statistical analyses: (1) nonparametric testing of sample distributions; (2) simple time series analysis to evaluate trends and seasonality; and (3) simple regression models describing NDVI as a quadratic function of AGDD.

The irrigated areas displayed different temporal developments of NDVI between 1985–1988 and 1995–1999. As the temperature regime between the two periods was not significantly different, we conclude that observed differences in the temporal development of NDVI resulted from changes in agricultural practices.

In the north, the temperature regime was also comparable for both periods. Based on extant socioeconomic studies and our model analyses, we conclude that the changes in the observed land surface phenology in the northern regions are caused by large increases in fallow land dominated by weedy species and by grasslands under reduced grazing pressure. Using multiple lines of evidence allowed us to build a case of whether differences in land surface phenology were mostly the result of anthropogenic influences or interannual climatic fluctuations.

Introduction

Recent studies of land cover/land use change (LCLUC) have focused on data derived from spaceborne sensors with spatial resolutions <100 m acquired across several years (e.g., Brown et al., 2000, Peterson & Aunap, 1998). While much information can be gained by parsing the dynamics of decision-making in landscapes in this manner (Geoghegan et al., 1998), the observational scale occurs at too fine a resolution and too slow a tempo to expect significant linkages with the atmospheric boundary layer (Lambin, 1996). The boundary layer is the lower portion of the troposphere where the atmosphere can be directly influenced by the planetary surface. The atmospheric boundary layer plays an important role in numerical weather prediction models. Observations of land surface phenology at coarser spatial resolutions (1–16 km) have shown linkages with boundary layer dynamics Lim & Kafatos, 2002, Schwartz & Reed, 1999, White et al., 2002. However, the seasonality of surface vegetation in temperate climates and the interannual variation in onset, duration, and intensity of the growing season pose formidable challenges to LCLUC studies since it is necessary to distinguish between weather-induced variation and enduring changes. Given an image time series that has both the sufficient temporal density to characterize seasonality and the temporal depth to characterize interannual variability, how should we analyze changes in land surface phenology? LCLUC occurs on many different spatial and temporal scales and in multiple forms ranging from alterations in crop type to changes in land use category, e.g., from cultivated to residential. Here, we are interested in using land surface phenology as a means to detect changes in agricultural land cover and land management practices. Land surface phenology could change because of changing climate, leading to phenomena such as the earlier onset of spring Myneni et al., 1997, Zhou et al., 2001 or earlier senescence. However, land surface phenology could also change as a result of shifts in land cover proportions or alterations in land management practices.

Change analysis of image time series can be decomposed into four steps: (1) change detection to identify differences between images; (2) change quantification to determine the character, magnitude, and extent of the differences; (3) change assessment to decide whether the observed differences are significant; and (4) change attribution to identify possible causes associated with the observed changes.

Most change detection strategies commonly used in remote sensing studies were developed in an era of image scarcity and thus focus on comparing just a few scenes (Jensen, 1996). In an era of intensive earth observation, something more is required for change analysis. What sufficed for handfuls of data is inadequate when confronted with a “data tsunami”. Coarser spatial resolution satellites (e.g., AVHRR, MODIS, MERIS) are capable of observing broad regions in every overpass, resulting in a much higher temporal data record than for finer resolution satellites. Change analysis methods applicable to images with sparse temporal sampling may not provide efficient or effective analysis when applied to dense image time series where coherent, quasi-periodic spatio-temporal patterns may be observable. For example, when operational remote monitoring of the terrestrial environment is to contribute near-realtime data flows for assimilation into numerical weather prediction models Champeaux et al., 2000, Ehrlich et al., 1994, there is the need to determine whether “significant” change has occurred since the last data acquisition. Whether a detected change is “significant” depends on the research question. A similar question addresses whether there are significant trends in timing of the onset of boreal spring Myneni et al., 1997, Shabanov et al., 2002, Tucker et al., 2001, Zhou et al., 2001.

Kazakhstan has been the setting for several notable anthropogenic transformations of the planetary surface during the 20th century. Well known is the dramatic recession of the Aral Sea resulting from the upstream diversion of water to agriculture (Bos, 1995). Less familiar, perhaps, is the largest land cover change event in the 20th century-Khrushchev's “Virgin Lands” program. More than 13 million hectares of native steppe were plowed and sown to spring wheat during 1954–1956. To support this colossal effort, more than a million people immigrated to the region, which transformed Kazakhstan into the only Soviet state in which the native population was a numerical minority (McCauley, 1976). From a traditional economy, based largely on nomadic pastoralism, Kazakhstan was rapidly transformed into a principal provider of grain to the Soviet Union, supplying 27% of USSR's demand for wheat (Kaser, 1997). The total cultivated area for all crops increased from 11.4 million hectares in 1954 to 30.8 million hectares a decade later (McCauley, 1976). Besides grains, Kazakhstan also exported large quantities of wool and meat FAO, 2003, Suleimenov & Oram, 2000. The exceptionally strong emphasis on grain production had a large damaging effect on the environment in Kazakhstan. Excessive use of fertilizers and pesticides and large irrigation projects caused soil pollution, desertification, and deterioration of water quantity and water quality (Grote, 1998).

Following the collapse of the Soviet Union in 1991, Kazakhstan gained independence and became the second largest country cleaved from the USSR and the world's ninth largest country in land area. Independence caused myriad economic dislocations, including the end of the highly regulated Soviet trading bloc, centralized agricultural planning, and the political interest in agriculture in general. The political changes resulted in extremely high inflation, scarcity of food and other products, and a precipitous decline in production of exports such grain, wool and meat (Alaolmolki, 2001).

The question motivating our analysis is this: Given abrupt, sweeping changes in political, social, and economic institutions and the subsequent reallocation of land use decisions, are the consequences of change observable in land surface phenology at spatial resolutions relevant to interactions with the atmospheric boundary layer? To be relevant to boundary layer processes, any land surface transformation must be observable and significant at resolutions that are very coarse relative to conventional LCLUC studies. To quantify and assess change in the presence of high interannual variation in weather and NDVI response, we employ a suite of complementary statistical analyses and test hypotheses for significant differences in land surface phenology among different study areas and periods contained within a standard NDVI dataset.

We used the Pathfinder Advanced Very High Resolution Radiometer (AVHRR) Land (PAL) dataset to analyze the land cover dynamics of seven agricultural areas in Kazakhstan before and after institutional change and to place this episode in the larger context of climate variability and landscape dynamics. The Pathfinder AVHRR Land (PAL) data are frequently used in change detection studies Borak et al., 2000, Shabanov et al., 2002, Tucker et al., 2001, Young & Wang, 2001. To minimize clouds and atmospheric contaminants, maximum-value NDVI composites (Holben, 1986) have been generated for 10-day periods (dekads) by selecting the maximum NDVI value from the daily data during a dekad. Although there are numerous problems with the PAL data, such as satellite orbital drift and lack of correction for scattering and water vapor absorption, these data are still very attractive for change analysis because the data are global in extent, frequent in recurrence, long in duration, and freely available in a standard form. Filtering techniques have been developed to attenuate remaining cloud contamination (Lovell & Graetz, 2001), and other atmospheric conditions degrading the data (Shabanov et al., 2002). Kaufmann et al. (2000) concluded however that, despite the orbital drift and sensor changes, the data still could be used for research about interannual variability. Numerous studies have been performed to estimate crop yields from AVHRR satellite data. Labus et al. (2002) developed a model to estimate wheat yield with AVHRR data and concluded that NDVI data from AVHRR can provide good estimators of regional yields at the end of the growing season.

Retrospective analyses can be fraught with ambiguities that result from a lack of clear experimental manipulation. As in the case for accuracy assessment of small-scale land cover maps (Merchant et al., 1994), a multiple lines of evidence approach is preferred over reliance on a single type of analysis. Here, we focus on seven agricultural areas in Kazakhstan, ranging from grassland and dryland agricultural areas in the north to irrigated intensive agriculture in the south, because it is exactly in the agriculture sector—where centralized control and subsidies abounded—that repercussions of institutional change ought to be observable. To investigate changes in seasonality and interannual variation independent of variations in the bioclimatic regime, we apply to each study area three distinct but complementary statistical analyses: (1) nonparametric testing of sample distributions to investigate for difference in means for NDVI and growing degree-days (GDD); (2) simple time series analysis to evaluate trends and seasonality; and (3) simple regression models describe NDVI as a quadratic function of accumulated growing degree-days (AGDD). The methods proceed from simple to more involved, both in terms of implementation and interpretation. Method 1 is a basic comparison of the mean structure of the dataset performed to detect obvious average differences between the two time periods. Method 2 is a trend analysis performed to identify temporal trends. In method 3, the AGDD and NDVI are linked using simple quadratic models to identify any changes in NDVI that are not attributable to changes in AGDD. The remainder of the paper is organized into six sections: description of the study areas, description of data, methods, presentation of the results, discussion, and conclusions. Section 5 has been divided into three parts to allow separate discussions of the three analytical approaches. Section 6 is divided in four sections discussing similar regions in one section. The discussion is followed by a general conclusion.

Section snippets

Study areas

With an area of 2.72 million km2, Kazakhstan roughly equals one-third of the conterminous U.S. or one-quarter of China. It is sparsely populated with only 16.7 million people (Grote, 1998). As a landlocked country, Kazakhstan borders Turkmenistan, Uzbekistan, and Kyrgyzstan to the south, Russia to the north, China to the east, and the Caspian Sea to the west. Kazakhstan covers about 15 degrees of latitude from 40°N to 55°N and 35 degrees of longitude from 50°E to 85°E (Fig. 1).

The climate is

Satellite sensor data

The Pathfinder AVHRR Land (PAL) data were used to characterize the spatio-temporal dynamics of the land surface. The maximum value compositing method can create a relatively cloud-free dataset (Holben, 1986). The AVHRR scanner records near infrared and red radiance in two broad channels. The Normalized Difference Vegetation Index (NDVI) is calculated as the ratio of the difference between the near infrared and red divided by the sum of near infrared and red. Active green vegetation reflects

Exploratory data analysis

We want to test the null hypothesis that the mean NDVI from the NOAA-9 and NOAA-14 AVHRRs is equivalent. First, it is necessary to determine if the data follow a normal distribution. In case of normality, the data can then be submitted to the regular t-test with unequal variances to test for differences between the two groups. In case the data are not normally distributed, the data are submitted to the nonparametric Wilcoxon rank-sum test. This procedure was repeated for both the NDVI and GDD

Exploratory data analysis

Average NDVI and GDD values were compared for the periods before and after institutional change. Table 2 reports the mean NDVI and mean sum of GDD per dekad of each study area and each period together with the p-values for the difference between the periods. Notice that every region shows a higher mean NDVI after institutional change; however, the increase is significant only in study areas 5, 6, and 7. Five regions have a slightly lower sum of GDD after institutional change but nowhere are

Discussion

The seven study areas are distributed across an area greater than 1.6 million km2 and are located in eight different ecoregions as delineated by the World Wildlife Fund (Olson et al., 2001). They have distinct temperature and precipitation regimes, different native land cover types and land uses. However, they share the same historical events associated with institutional change in Kazakhstan. To assess significant differences between the periods on either side of the institutional change in

Conclusion

In this study we analyzed image time series with high temporal density for changes due to institutional change. We focused on the processes of change quantification and change assessment of the land surface phenology in seven areas across Kazakhstan. We presented three distinct but complementary statistical analyses to test for significant differences before and after institutional change. Testing of average NDVI values revealed significant differences between the two periods for three areas.

Acknowledgements

Research supported through the NASA LCLUC program. This manuscript benefited from careful reviews by Lei Ji, Andrés Viña, and two anonymous reviewers. Image data used here were produced through funding from the EOS Pathfinder Program of NASA's Mission to Planet Earth in cooperation with NOAA. The data were provided by EOSDIS DAAC at Goddard Space Flight Center, which archives, manages, and distributes this dataset. NCEP Reanalysis data were provided by the NOAA-CIRES Climate Diagnostics Center,

References (55)

  • J.C Brada

    The variability of crop production in private and socialize agriculture: Evidence from Eastern Europe

    The Journal of Political Economy

    (1986)
  • J.C Brada et al.

    Is private farming more efficient than socialized agriculture?

    Economica

    (1993)
  • J.C Brada et al.

    Differences in the technical and allocative efficiency of private and socialized agricultural units in pre-transformation Poland

    Economic Systems

    (1994)
  • J.F Brown et al.

    Advanced strategy for multi-source analysis and visualization in land cover characterization

  • J.L Champeaux et al.

    AVHRR-derived vegetation mapping over Western Europe for use in numerical weather prediction models

    International Journal of Remote Sensing

    (2000)
  • T.M Crawford et al.

    The insertion of a high-resolution land use data set into MM5/PLACE

    Journal of Hydrometeorology

    (2001)
  • K.M de Beurs et al.

    A statistical framework for the analysis of long image time series

    International Journal of Remote Sensing

    (2004)
  • E.J Dietz et al.

    A nonparametric multivariate test for monotone trend with pharmaceutical applications

    Journal of the American Statistical Association

    (1981)
  • P Doraiswamy et al.

    Evaluation of MODIS data for assessment of regional spring wheat yield in Kazakhstan

  • D Ehrlich et al.

    Applications of NOAA-AVHRR 1 km data for environmental monitoring

    International Journal of Remote Sensing

    (1994)
  • FAO

    FAO Statistical Database

  • K Gallo et al.

    Monthly fractional green vegetation cover associated with land cover classes of the conterminous USA

    Geophysical Research Letters

    (2001)
  • J Geoghegan et al.

    “Socializing the Pixel” and “Pixeling the Social” in land-use and land-cover change

  • U Grote

    Central Asian environments in transition

    (1998)
  • G Gutman et al.

    The derivation of the green vegetation fraction from NOAA/AVHRR data for use in numerical weather prediction models

    International Journal of Remote Sensing

    (1998)
  • G.M Henebry et al.

    Land surface dynamics in Kazakhstan: Dynamic baselines and change detection

  • R.M Hirsch et al.

    A nonparametric trend test for seasonal data with serial dependence

    Water Resources Research

    (1984)
  • Cited by (427)

    View all citing articles on Scopus
    View full text