Poster Session 1
“Global extent of tropical sandy soils and
their pedogenesis”

Spatial heterogeneity in sandy soils of the Sahel region in West Africa:
implications for desertification processes

Shinjo, H.¹; K. Ikazaki¹; U. Tanaka² and T. Kosaki²

Keywords: desertification, geostatistics, NDVI, satellite image, spatial heterogeneity

Introduction

Desertification, the land degradation in arid, semi-arid, and dry subhumid areas (UNEP 1997), is one of the challenging issues for mankind due to its serious impact on human welfare and environment. Desertification risk assessment would help control desertification and achieve sustainable land use. The remotely sensed data such as aerial photos and satellite images have been employed for this purpose since archival data enable us to obtain land information in the past and assess temporal change in land characteristics (UNEP 1997; Mouat et al. 1997). However, the large inter-annual fluctuation in precipitation and the spatial heterogeneity in soil inherent fertility often make it difficult to assess desertification temporally over a given period. One of the measures proposed to overcome this difficulty is indexing spatial heterogeneity. Schlesinger et al. (1990, 1996) found that soil moisture and nutrient status and subsequently vegetative conditions had become spatially heterogeneous as desertification proceeded and proposed “the island of fertility” theory. On the other hand, our field observation in the Sahel region of Northern Burkina Faso suggested that spatial heterogeneity of vegetation decreased as land surface become bare. Thus, we integrated Schlesinger’s theory with the observation to hypothesize that spatial heterogeneity of soil and vegetation will reach a maximum at the middle of desertification processes, while soil and vegetation would in general have a homogeneous character at its beginning and end (Figure 1). Following above-mentioned finding of Schlesinger’s, Tucker et al. (1991), Milich and Weiss (2000), and Seixas (2000) employed coefficients of variance (CV) of normalized difference vegetation index (NDVI) of satellite images for desertification assessment. However, since spatial resolution in the first two studies was as coarse as 7.6 km and the last study did not relate NDVI with ground truth data, the assessment at the finer scale with the validation of NDVI by ground truth data would be necessary for the purpose of sustainable land use at a village scale. Thus, the objectives of our study are to understand the actual status of spatial heterogeneity in sandy soils of the Sahel region in West Africa and examine the possibility of spatial heterogeneity as the index of desertification processes.

Figure 1. Hypothetical diagram of desertification process

Materials and methods

The study area is located in Takabangou Village, Oudalan Province, Burkina Faso receiving about 400 mm of annual rainfall (Figure 2). In this area, millet fields and fallow land occupy strips of ancient sand dunes extending from east to west, while the area between the ancient sand dunes is less covered with sand and used as rangeland. Dominant native plant species were Aristida mutabilis, Cenchrus biflorus, Schoenefeldia gracilis and Eragrostis tremula.

Figure 2. Location of the study site. Red crosses represent the 50 m transect survey sites and the black cross the 450 m transect survey site. Background image is a false color composite of Landsat TM captured on Oct. 14, 2002 by assigning digital number to band 4, 3 and 2 to R, G and B respectively

Figure 3. Design for 450 m transect survey. Three 450 m rows were set 30 m apart from each other. Dots show soil sampling points. Squares show units for calculation of bare rate. Bare rates in nine units shown in the figure are complied to calculate their mean value and coefficient of variance

Our field survey consisted of 50 m transect surveys at 9 sites in October 2002 and 450 m transect survey at one site in October 2003 (Figure 2). At each site for the 50 m transect survey, plant species were identified at intervals of 50 cm in the 50 m transect and aboveground plant biomass in 1 m² was measured at 3 points in the transect. Shannon-Weaver’s diversity index, H’, was calculated at each site by

where P_i is the frequency of plant species i in the 50 m transect.

At the site of the 450 m transect survey, three 450 m rows were set 30 m apart from each other (Figure 3). After plant species were identified at intervals of 1 m in the rows, bare rate was calculated every 30 m along the rows as the percentage of the frequency of plant absence. Soil samples from the depth of 0-5 cm were also taken every 30 m, air dried and analysed for soil texture.

Six satellite scenes of Landsat-TM in 1986, 1992, 1999, 2000, 2001 and 2002 and one scene of ASTER in 2003 were used. All scenes were captured at the end of rainy seasons. After geometrical correction with ground control points and conversion to absolute radiance, temporal variation caused by path radiance was corrected by subtracting a minimum pixel value in each scene from each pixel value. With the corrected pixel values, NDVI was calculated for each scene by

Since inter-annual variation in NDVI values of each pixel due to climatic variation such as variation in amount and distribution of rainfall make it difficult to estimate the long term trend, the NDVI values were standardized using the following:

where NDVI_mode is mode value of NDVI in each scene, and NDVI_std is standard deviation of NDVI in each scene (Koizumi et al. 2003).

Results and discussion

Since the 50 m transect survey revealed that maximum diversity of plant species was found at the medium level of plant biomass (Figure 4), it is implied that land characteristics were heterogeneous to realize many plant species at the medium level of land productivity. Thus, the hypothesis was validated at the plant species level.

Figure 4. Relationship between diversity index and plant biomass at the 50 m transect survey sites

The 450 m transect survey resulted in a negative correlation between the bare rate and sand content, implying that plant growth would be reduced once the sandy surface soil was eroded and hard loamy subsurface soil with low water permeability was exposed on the surface (Figure 5). Thus, in this region, soils with a sandy surface layer could be regarded as productive. The bare rate calculated every 30 m in the 450 m transect survey was compiled for 9 neighbouring units to obtain mean values and coefficients of variance (CV) of bare rates (Figure 3). As shown in Figure 6, a negative correlation between CV and mean value of the bare rate indicated that plant coverage at its medium level of bare rates was spatially rather heterogeneous as expected by the hypothesis. However, since no high plant coverage with the less than 40% of mean bare rates was found at the site, the hypothesis could be only validated at the plant coverage level. In addition, it is of note that the desertification process at the site had already passed the one Schlesinger described.

Figure 5. Relationship between bare rate and sand content at the 450 m transect survey site

Figure 6. Relationship between mean values and coefficients of variance (CV) of bare rate in 3 × 3 unit at the 450 m transect survey site

Before Landsat-TM data was employed to compensate the insufficient field data set for validating the hypothesis, two conditions were examined. The first was the correlation of NDVI derived from satellite images with ground truth data and the other was whether the spatial resolution of satellite data was fine enough to evaluate the spatial heterogeneity of ground truth data. Figure 7 proves that ground truth data of bare rates can be estimated by NDVI. To examine the second condition, a geostatistical analysis was performed. Assigning 1 to the points without plants and 0 to the points with plants in the rows of the 450 transect survey, the semivariance was calculated as follows:

where γ(h) is semivariance at lag h, N(h) is number of the observation pairs which were apart from each other by the lag h, Z(x) is observation at point x, and Z(x + h) is observation at point (x + h). Figure 8 shows that the semivariance reached its maximum at the lag of 67 m to which the spatial dependence was limited (Webster and Oliver 2001). Thus, it is suggested that average size of vegetative patches was 67 m and the spatial resolution of 30 m for Landsat-TM data was adequate enough to distinguish vegetative patches.

Figure 7. Relationship between bare rate and normalized difference vegetation index (NDVI) in October 2003 at the 450 m transect survey site

Figure 8. Semivariogram of plant absence at the 450 m transect survey site

Now that the two conditions were met by NDVI of Landsat-TM data, the hypothesis was examined by the time series of standardized NDVI. For this purpose, 11 points in the study area were selected that were observed to have the largest temporal variation in NDVI, covered the whole range of NDVI found in the study area, and scattered over the whole study area.

Figure 9 shows that the maximum CV of NDVI was found at the moderate level of mean NDVI, suggesting that the hypothesis could be correct spatially. However, since no point had experienced the whole process of the hypothesis from 1986 to 2002, it seemed that the drastic desertification had not taken place in the study area during the period examined and that vegetative coverage were rather controlled by inherent land productivity. Further validation in extended space and time would be necessary.

Figure 9. Relationship between mean values and coefficient of variance of standardized NDVI at the selected 11 points in the study area. Symbols with the same color and shape represent values at one point from 1986 to 2002

References

Koizumi, T., Hagiwara, K., Yamashita, M., and Kokubun, S. 2003. A study on environmental evaluation method using the satellite image – Radiation quantity adjust­ment of multi-temporal data. Photogrammetry and Remote Sensing, 42(4), 6-17. (in Japanese with English summary)

Milich, L., and Weiss, E. 2000. GAC NDVI interannual coefficient of variance (CoV) image: ground truth sampling of the Sahel along north-south transects, International Journal of Remote Sensing, 21, 235-260.

Mouat, D., Lancaster, J., Wade, T., Wickham, J., Fox, C., Kepner, W., and Ball, T. 1997. Desertification evaluated using and integrated environmental assessment model, Environmental Monitoring and Assessment, 48, 139-156.

Schlesinger, W.H., Reynolds, J.F., Cunningham, G.L., Huenneke, L.F., Jarrell, W.M., Virginia, R.A., and Whitford, W.G. 1990. Biological feedbacks in global desertification, Science, 247, 1043-1048.

Schlesinger, W.H., Raikes, J.A., Hartley, A.E., and Cross, A.F. 1996. On the spatial pattern of soil nutrients in desert ecosystems, Ecology, 77, 364-374.

Seixas, J. 2000. Assessing heterogeneity from remote sensing image: the case of desertification in Southern Portugal, International Journal of Remote Sensing, 21, 2645-2663.

Tucker, C.J., Newcomb, W.W., Los, S.O., and Prince, S.D. 1991. Mean and inter-year variation of growing season normalized difference vegetation index for the Sahel 1981-1989, International Journal of Remote Sensing, 12, 1133-1135.

UNEP 1997. World Atlas of Desertification.2nd ed. London, Arnold, 69 p.

Webster R., and Oliver, M. 2001. Geostatistics for environmental scientists. Chichester, John Wiley & Sons, Ltd., 271 p.

¹ Graduate School of Agriculture, Kyoto University, Kyoto 606-8502, Japan, [email protected]
² Graduate School of Global Environmental Studies, Kyoto University, Kyoto 606-8501, Japan

The potential productivity of sandy soil for sugarcane in some regions of
Thailand associated with micromorphological observation

Sindhusen, P.¹; H. Pichainarong and K. Marlairodsiri

¹ Office of Science for Land Development, Land Development Department, Phaholyothin Rd., Chatuchak, Bangkok 10900, Thailand