Version classiqueVersion mobile
OpenEdition Books

Le projet majeur africain de la Grande Muraille Verte

 | 
Abdoulaye Dia
, 
Robin Duponnois

Partie II. Mise en valeur et suivi de la Grande Muraille Verte

A Remote Sensing Technique for Monitoring Temporal Dynamics of a Forest Located at a Desert Fringe

A. Karnieli

Résumé

Drought years are a very frequent phenomenon in Israel. Between the years 1994/5 and 2001/2, Israel experienced four (non-consecutive) years of drought. Consequently the Yatir forest, a pine forest located in the desert fringe, suffered from a notable water shortage. The aim of this research is to detect and assess seasonal/phenological changes and inter-annual changes in the forest trees with respect to the drought effect. The use of a spectral vegetation index, namely the Normalized Difference Vegetation Index (NDVI) to detect stress conditions was implemented by using eight Landsat TM and ETM+ images. In addition, the change detection NDVI Image Differencing technique was applied for assessing seasonal and inter-annual variations in vegetation. Considerable NVDI decline was observed between 1995 and 2000 due to the drought events during these years, enabling assessment of the spatial and temporal effects of such a disaster. NDVI image differencing has proven to be a useful and accurate method for tracing physiological changes in the Yatir forest, which serves as a case study for a manmade forest in the desert fringe.

Texte intégral

Introduction

1The recurrence of severe droughts in the Sahel and other regions around the world has led to extensive discussions on the effect of droughts on the life of people and plants, particularly in the arid and semi-arid climatologic zones. Forest planting on the desert fringe represents an attempt to combat desertification and rehabilitate of drylands. In Israel, the Yatir forest is such a manmade pine forest situated in the transition between the semi-arid and sub-humid zones (275 mm mean annual rainfall). In order to explore changes and the effects of drought in the forest, the Normalized Difference Vegetation Index (NDVI) was used. The NDVI is the most widely used vegetation index, and it is based on the ratio between the maximum absorption of radiation in the red (R) spectral band vs. the maximum reflection of radiation in the near infrared (NIR) spectral band. Lacking the plants’ absorption/reflectance mechanisms, soil spectra typically do not show such a dramatic spectral difference. NDVI is formulated as:

2where R is the reflectance value in the indicated spectral bands. NDVI values range between -1.0 to +1.0 but are usually positive for soil and vegetation. Denser and/or healthier vegetation have higher values.

3NDVI values of vegetation usually offer an efficient and objective mean for evaluating phenological characteristics (e.g., Justice et al.,1985; Reed et al., 1994; Running et al., 1995) and have long been used to monitor vegetation conditions and changes in vegetation cover (e.g., Lyon et al., 1998; Mas, 1999; Woodcock et al., 2001).

4Change detection has become a major application of remotely sensed data because of repetitive coverage at short intervals and consistent image quality (Mas, 1999). Two categories are recognized for the change detection assessment (Yuan et al., 1998). The first is conversion from one land cover type (class) into another and the other is transformations within a given land cover type. The latter can be used for examining the effect of water shortage on the land use of a single renewable natural resource. Several methods for detecting land cover changes were reviewed by Yuan et al. (1998) and Mas (1999). The Vegetation Index Differencing method, and particularly the NDVI Image Differencing, was found to be suitable for the current investigation. NDVI Image Differencing (Δ NDVI) is a change detection technique that has been used for several applications such as studying the effect of extensive flooding on forest ecosystems (Michener and Houhoulis, 1997), monitoring the impact of urban development (Fung and Siu, 2000), and monitoring the regeneration of Mediterranean shrubland (Svoray et al., 2003). The following equation is applied:

5where the subscripts t1 and t2 are NDVI images from dates 1 and 2, respectively.

6The results of this operation correspond to an increase or decrease in vegetation state or cover.

7Nelson (1983) showed that using the Δ NDVI technique has stronger relationship to the phenomena of interest in the scene than any single spectral band alone.

8The objective of this research is to apply remote sensing and geographic information system (GIS) techniques to monitor changes in the forest on two temporal scales - seasonal and inter-annual changes. The hypothesis is that the NDVI provides a suitable tool to assess changes in the Yatir forest that are related to drought periods due to decrease in vegetation cover and consequent increase of soil background. Spatial data of the forest, coupled into a GIS, can provide a better understanding of the areal changes within the forest during the same hydrological year and patterns of change between the years.

Material and methods

Study area

9The forest area to be studied is located between the Mediterranean and Dead Seas, approximately 31° 20’ N 35° 00’ E and 650 m above mean sea level (Fig. 1). This southern part of the mountain chains in Israel is situated between two climate zones: the dry desert with less than 200 mm rainfall per annum and the semi-arid desert that receives between 200 mm and 300 mm rainfall per annum. In addition to its location on the desert fringe, its relative high elevation plays an important role in defining the climatic characteristics of the forest (Fig. 2). The mean annual temperature is about 17.6°C (ranging between 12.8 and 22.4°C).

10The mean winter temperature (November–March) ranges from 9.1 to 12.7°C, while in summer (June-September) the temperatures range from 23.2 to 24.3°C (Eshel, 2000). The long-term mean annual rainfall in this region (275 mm) is limited to the winter months

11(October – April) and characterized by high annual fluctuation, unequal distribution of the events within the rainy season, and above all its general scarcity. The current research covers a 7-year period (Fig. 3). The hydrological year 1994/5 is characterized by much more rainfall than the annual mean (360 mm). The next two years, 1995/6 and 1996/7 were drought years with 158 and 232 mm of rain, respectively. 1997/8 was an average year with 274 mm of rainfall. The following two years, 1998/9 and 1999/2000 were again drought years with 138 and 157 mm, respectively. Lastly, 2000/1 was a wet year with 297 mm. In summary, during five years (1994/5 – 1999/2000 the forest suffered four drought years and one average year. The trees in the forest were planted during four decades: 1960s (28% of the forest); 1970s (38%); 1980s (13.5%); and 1990s (20%).

Figure 1

Figure 1

A Landsat-TM image of central Israel.
Note the location of the Yatir forest on the desert fringe, visible as the sharp contrast between bright tones (semi-arid zone) and dark tones (sub-humid zone).

Figure 2

Figure 2

Ground photograph of the Yatir forest at the desert fringe.

Figure 3

Figure 3

Yearly distribution of rainfall amounts, cumulative rainfall, and dates of Landsat-TM and ETM+ images.

Dataset

12Assessment of drought effect was implemented by using eight Landsat TM and ETM+ images dating from: winter and spring 1994/5; fall 1995; winter and spring of 2000; and spring 2001. Fig. 3 shows the distribution of Landsat images with respect to monthly and cumulative yearly rainfall.

Image pre-processing

13The aim of the pre-processing operation was to bring all images to the same comparable format. Raw digital numbers of the images were converted to radiance values using the procedure published by EOSAT (1986). In order to perform atmospheric correction and to convert the radiance values to surface reflectance values, the Second Simulation of Satellite Signal in the Solar Spectrum (6S) (Vermote et al., 1997) was implemented. For this code aerosols and watervapor contents were acquired from a sunphotometer located at Sede-Boker Campus, about 50 km from the research site. All images were then registered to the New Israeli Grid using 20 ground control points with a root mean square error (RMSE) of less than one pixel. The area of interest (AOI), namely the Yatir forest, was extracted from the geometric corrected images. Finally, a masking of the areas with no trees was performed in order to provide NDVI images with a minimal effect of bare soil and/or annuals that grow in clear-cut plots. Landsat-derived NDVI, for selected dates in winter and spring 1995 and winter and spring 2000 is presented in Fig. 4.

Figure 4

Figure 4

NDVI images in similar seasons (winter and spring) but for a wet year (1995) and a drought year (2000).

Change detection

14The NDVI Change Detection method (Eq. 2) was selected for implementing the research goals. Changes within the same hydrological year were computed in order to assess the dynamics of NDVI during the phenological cycle of the trees. In addition, images from the same season in different years were computed to characterize the drought effect on vegetation cover due to differing rainfall regimes. Note that the 1994/5 hydrological year, which represents a wet year with above-mean annual rainfall, occurred after several wet years. Conversely, the 1999/2000 hydrological year represents a drought year, the fourth drought year in a five-year period. A common way to assess changes is based on determination of thresholds in terms of standard deviation levels from the mean Δ NDVI (NDVI Δ) (Fig. 5A). In this manner, one can distinguish between changed and unchanged pixels as well as between negative and positive changes (Jensen, 1986). In the current study, in the cases in which the entire forest changed in only one direction, the threshold is determined in the minimal NDVI value (≈0) and not in adjacency to the mean in order not to lose meaningful information (Fig. 5B). Steps of 0.5 standard deviation (STDV) from the NDVI Δ determined the magnitude of the change in both case studies described above.

Figure 5

Figure 5

Shematic illustration of two change detection approaches used in the study.

Results and discussion

Inter-annual dynamics of NDVI

15Fig. 6 represents the NDVI transects across the forest (8.5 km) during the rainy year in 1994/5 (360 mm) and the drought year 1999/2000 (130 mm of rain after three non-consecutive years of drought). The curve of January 1995 shows the highest values of NDVI that are related to the photosynthetic activity of a healthy forest. The curve of June 1995 shows lower NDVI values, as expected due to closing of the stomata during the late spring and summer. Looking at the same months five years later reveals the effect of water shortage on the NDVI due to four years of drought. The January 2000 curve has higher NDVI values than May of the same year, however compared to January 1995 these values are significantly lower. Note that the NDVI values of January 2000 almost perfectly match the ones of June 1995. Also note that the effect of rainfall amounts, after four years of drought, is less profound than the effect of rainfall on the forest during a rainy year. The amount of rain required to recover the trees into the growing mode needs to be greater than the amount of rain needed in 1994/5.

Δ NDVI images

16Numerous change detection products were computed in order to observe the magnitude of the change along the phenological cycle of different years and in the same season but between years. According to the previous discussion, a threshold value was derived from the image as one standard deviation (SD) from the Δ NDVI mean in cases where the mean was between -0.1 and +0.1. Otherwise, when the mean was either smaller -0.1 or greater than +0.1, Δ NDVI was set to 0 as the reference point. Table 1 summarizes the mean and SD of the nine case studies. Positive mean indicates recovery of the forest while negative mean indicates degradation.

Table 1.

Table 1.

Pairs of images used for the change detection analysis along with the change statistics – mean and standard deviation (STDV).
Position mean indicates recovery of the forest while negative mean indicates degradation.

Figure 6

Figure 6

NDVI values across the forest (west – east transect) for winter and spring of 1995 and 2000.

Figure 7

Figure 7

NDVI Image Differencing products for selected images of inter-annual analysis along the phonological cycle along with the respective frequency histograms of the change categories. Each category represents 1 step of standard deviation.
(A) January 24, 1995 – January 14, 2000;
(B) June 17, 1995 – May 21, 2000;
(C) May 21, 2000 – May 24, 2001.
Note the matching of colors between the images and the histograms.
VHD = Very High Decrease;
HD = High Decrease;
LD = Low Decrease;
VLD = Very Low Decrease;
NC = No Change;
VLI = Very Low Increase;
LI = Low Increase;
HI = High Increase;
VHI = Very High Increase.

17Inter-annual changes are illustrated in Fig. 7. The largest negative change is observed between January 1995 and January 2000 due to the droughts in-between the two years (Fig. 7A, Table 1). The change between June 1995 and May 2000 was less pronounced since both images represent the season with less photosynthetic activity (Fig. 7B, Table 1). The moderate recovery of the forest as a result of a new wet year expressed as positive change, is demonstrated in Table 1. Note however that most of the pixels remain unchanged.

Conclusions

18The aim of this study was to monitor temporal changes in NDVI values in the Yatir forest drought conditions. Despite limitations caused by the effect of differences in bare soil reflectance and a relatively low number of images it can be concluded that changes of NDVI values during the growing season show that changes in forest physiology could be detected due to changes in photosynthetic activity. During the winter, high photosynthetic activity values are detected due to the relatively low temperature (in comparison with summer temperatures) and the high water availability. Conversely, during the summer, stomata close and photosynthetic activity decreases as a result of high temperature and absence of water. All those variables affect the state of the forest greenness and are reflected in the NDVI fluctuations along the year. This result shows the changes in phenotype in Pines trees with their immigration from Europe to Israel: a change from high photosynthetic activity in the summer months in Europe to high photosynthetic activity in the Israeli winter. In summary, NDVI image differencing has proven to be a useful and accurate method for tracing physiological changes in the Yatir forest, which serves as a case study for a manmade forest in the desert fringe.

Bibliographie

Fung, T. Siu, W., 2000. Environmental quality and its changes, an analysis using NDVI. Int. J. Remote Sens.. 21 (5), 1011-1024.

Justice, C.O., Townshend, J.R.G., Holben, B.N., Tucker, C.J., 1985. Analysis of the phenology of global vegetation using meteorological satellite data. Int. J. Remote Sens. remote sensing, 6, 1271-1318

Lyon, G.J., Yuan, D., Lunetta, R.S., Elvidge, C.D., 1998. A change detection experiment using vegetation indices. Photogramm. Eng. Remote Sens. 64, 143-150.

Markham, B.L., Baker., J.L., 1986. Landsat-MSS and TM post calibration dynamic ranges, atmospheric reflectance and at-satellite temperature. EOSAT Landsat Notes 1, August 1986, Earth Observation Satellite Company, Lanham, Maryland, USA. pp. 3-8.

Maselli, F., Conse, C., Petkov, L., and Gilabert, M.A., 1993. Environmental monitoring and crop forecasting in the Sahel through the use of NOAA NDVI data. A case study: Niger 1986 -89. Int. J. Remote Sens.. 14, 3471-3487.

Mass, J.F., 1999. Monitoring land - cover changes: a comparison of change détection techniques. Int. J. Remote Sens.. 20, 139-152.

Michener, W.K. Houhoulis, P., 1997. Detection of vegetation changes associated with extensive flooding in a forested ecosystem. Photogramm. Eng. Remote Sens. 63, 1363-1374.

Nelson, R.F., 1983. Detecting forest canopy change due to insect activity using Landsat MSS. Photogramm. Eng. Remote Sens. 49, 1303-1314.

Nicholson, S.E., Davenport, M.L., Malo, A.R., 1990. A comparison of the vegetation response to rainfall in the Sahel and East Africa, using NDVI from NOAA AVHRR. Climatic Change. 50, 107-120.

Nicholson, S.E., Farrar, T.J., 1994. The influence of soil type on the relationship between NDVI, rainfall, and soil moisture in semi-arid Botswana. Remote Sens. Environ. 50, 107-120.

Raven, P.H., Evert, R.F., Eichhorn, S. E., 1999. Diversity. Biology of Plants, Worth Publishers. 875 p.

Reed, B.C., Brown, J.F., Vanderzee, D., Loveland, T.R., Merchant, J.W., Ohlen, D.O., 1994. Measuring phenological variability from satellite imagery. J. Veg. Sci. 5, 703-714.

Richard, Y., Poccard, I., 1998.

A statistical study of NDVI sensitivity to seasonal and interannual rainfall variations in Southern Africa. Int. J. Remote Sens.. 19, 2907-2920.

Running, S.W., Loveland, T.R., Pierce, L.L., Nemani, R. Hunt, E.R., 1995. A remote-sensing based vegetation classification logic for global land-cover analysis. Remote Sens. Environ. 51, 39-48.

Svoray, T., Shoshany, M., Perevolotsky, A., 2003. Monitoring the response of spatially complex vegetation formations to human intervention: A case study from Mediterranean rangelands. J. Mediterr. Ecol. 4, 3-12.

Vermote, E.F., Tanre, D., Deuze, J.L., Herman, M., Morcrette, J.J., 1997. Second Simulation of the Satellite Signal in the Solar Spectrum, 6S: An overview. IEEE T. Geosci. Remote Sens. 35, 675-686.

Woodcock, C.E., Macomber, S.A., Pax-Lenney, M., Cohen, W.B., 2001. Monitoring large areas for forest change using Landsat: Generalization across space, time and Landsat sensors. Remote Sens. Environ. 78, 194-203.

Yuan, D., Elvidge, C.D., and Lunetta, R.S., 1998. Survey of multispectral methods for land cover change analysis. In: Lunetta, R.S., Elvidge, C.D. (Eds.) Remote Change Detection. Environmental Monitoring Methods and Applications. Ann Arbor Press. Chelsea, Michigan, pp 21-39.

Table des illustrations

URL http://books.openedition.org/irdeditions/docannexe/image/2128/img-1.jpg
Fichier image/, 12k
URL http://books.openedition.org/irdeditions/docannexe/image/2128/img-2.jpg
Fichier image/, 12k
Titre Figure 1
Légende A Landsat-TM image of central Israel.Note the location of the Yatir forest on the desert fringe, visible as the sharp contrast between bright tones (semi-arid zone) and dark tones (sub-humid zone).
URL http://books.openedition.org/irdeditions/docannexe/image/2128/img-3.jpg
Fichier image/, 72k
Titre Figure 2
Légende Ground photograph of the Yatir forest at the desert fringe.
URL http://books.openedition.org/irdeditions/docannexe/image/2128/img-4.jpg
Fichier image/, 40k
Titre Figure 3
Légende Yearly distribution of rainfall amounts, cumulative rainfall, and dates of Landsat-TM and ETM+ images.
URL http://books.openedition.org/irdeditions/docannexe/image/2128/img-5.jpg
Fichier image/, 80k
Titre Figure 4
Légende NDVI images in similar seasons (winter and spring) but for a wet year (1995) and a drought year (2000).
URL http://books.openedition.org/irdeditions/docannexe/image/2128/img-6.jpg
Fichier image/, 76k
Titre Figure 5
Légende Shematic illustration of two change detection approaches used in the study.
URL http://books.openedition.org/irdeditions/docannexe/image/2128/img-7.jpg
Fichier image/, 60k
Titre Table 1.
Légende Pairs of images used for the change detection analysis along with the change statistics – mean and standard deviation (STDV).Position mean indicates recovery of the forest while negative mean indicates degradation.
URL http://books.openedition.org/irdeditions/docannexe/image/2128/img-8.jpg
Fichier image/, 180k
Titre Figure 6
Légende NDVI values across the forest (west – east transect) for winter and spring of 1995 and 2000.
URL http://books.openedition.org/irdeditions/docannexe/image/2128/img-9.jpg
Fichier image/, 76k
Titre Figure 7
Légende NDVI Image Differencing products for selected images of inter-annual analysis along the phonological cycle along with the respective frequency histograms of the change categories. Each category represents 1 step of standard deviation.(A) January 24, 1995 – January 14, 2000;(B) June 17, 1995 – May 21, 2000;(C) May 21, 2000 – May 24, 2001.Note the matching of colors between the images and the histograms.VHD = Very High Decrease;HD = High Decrease;LD = Low Decrease;VLD = Very Low Decrease;NC = No Change;VLI = Very Low Increase;LI = Low Increase;HI = High Increase;VHI = Very High Increase.
URL http://books.openedition.org/irdeditions/docannexe/image/2128/img-10.jpg
Fichier image/, 176k

Auteur

The Remote Sensing Laboratory, Jacob Blaustein Institutes for Desert Research, Ben Gurion
University of the Negev, Sede Boker Campus, 84990, Israel.
karnieli@bgu.ac.il

© IRD Éditions, 2010

Conditions d’utilisation : http://www.openedition.org/6540