Home Articles Manipulation of Normalized Difference Vegetation Index (NDVI) for Delineating Drought Vulnerable Areas

Manipulation of Normalized Difference Vegetation Index (NDVI) for Delineating Drought Vulnerable Areas

Uzma Rabab

Uzma Rabab*
Visitng Faculty Member
NUST Institute of Geographical Information System (IGIS)
Postal Address: 112-A, Street # 37, F-10/1, Islamabad, Pakistan
Tayyab I. Shah & M. Iqbal Tabbsum, GIS/RS Experts
E-mail: [email protected], [email protected], [email protected]

*The work was originally carried out as partial fulfilment of the requirement of Degree of Masters in Environmental Sciences (2002), Fatima Jinnah Women University, The Mall, Rawalpindi, Pakistan

Timely information about the onset of drought, extent, intensity, duration and impacts can limit drought-related losses of life, human suffering and decrease damage to economy and environment. The present research work has been carried out with the aim to integrate SRS and GIS for the identification of drought vulnerable areas in Sindh and major part of Balochistan. Arid and extremely arid conditions prevail in these areas and the amount of rainfall varies with time and space. This departure of rainfall results in the emergence of drought condition. The vegetation cover is directly linked with water availability and a decrease in vegetation cover can be alarming. . NOAA AVHRR derived NDVI can be used to obtain vegetation status on regular basis. Although spatial resolution of NOAA is coarse, yet, the onset of drought conditions for a large area in a given year can be predicted by comparative analysis of trend of derived NDVI of that year relative to the trend in a normal year. It is better to develop a multi date NDVI composite for the study area and consider it as a normal for comparison. Thus, for a developing country like Pakistan, regular monitoring of the vegetation status (application of NOAA derived NDVI data) along with the other layers including climate, soil type, hydrology and socioeconomic condition of people is needed to delineate the areas that are drought vulnerable. This multidisciplinary information can be effectively and accurately handled with GIS. Spatial analysis in GIS can lead to a decision support system for the concerned government departments, NGO’s and others to help drought vulnerable people and others living in potential drought areas.

1. Introduction
Drought is the single most important weather-related natural disaster. Its impacts on society result from the interplay between a natural event (less precipitation than expected resulting from natural climatic variability) and the demand people place on water supply. Recent droughts in both developing and developed countries and the resulting economic and environmental impacts and personal hardships have underscored the vulnerability of all societies to this “natural” hazard. A very key aspect to drought vulnerability is whether a population gets its water from a well or other reliable source, or if it relies on rainfall. Those that are using rain for the source of water are particularly vulnerable to food insecurity in times of drought, due to the lack of water for agriculture and domestic purposes. Drought has a differential impact by wealth status (i.e., access to labor, capital and improved input). In a study from Ethiopia, wealthier households achieved drought-year yields three times higher than poor households. The wealthier households did change their diet, but less than poor households changed their food intake (Webb, 1993).

Unlike earthquake, drought always has a slow onset, which is quite observable. It is not an event rather is a process, which can be understood and forecasted quite well before time (Bhatti, 2000). Many drought indices have been used over the globe to monitor and forecast drought. Drought indices assimilate thousands of bits of data on rainfall, snow pack, stream flow and other water supply indicators into a comprehensible big picture. For example, the Palmer Drought Severity Index has been widely used by the U.S. Department of Agriculture, but the Palmer is better when working with large areas of uniform topography. Western states, with mountainous terrain and the resulting complex regional microclimates, find it useful to supplement Palmer values with other indices such as the Surface Water Supply Index, which takes snow pack and other unique conditions into account. Australian Drought Authorities are using deciles (NDMC, 2000).

The new scientific technologies of remote sensing, satellite imaging, geographical information systems (GIS) and geographical positioning system (GPS) can be put to effective use in forecasting and monitoring drought. GIS/SRS if incorporated in drought mitigation and research process exhibits two principal advantages. First, the technology allows long-term time-series studies and storage of the information, which may prove invaluable in future situations. Secondly, GIS/RS improves information accessibility. Remote sensing platforms can provide large amounts of data quickly and inexpensively, relative to other means of collection, and GIS can bring together vast amounts of information from a wide variety of sources and make the information quickly visible and applicable in emergency situations (Verstappeen, 1995).

Satellite data processed into Normalized Difference Vegetation Indices (NDVI) can be used to indicate deficiencies in rainfall and portray meteorological and/or agricultural drought patterns both timely and spatially, thus serving as an indicator of regional drought patterns. NDVI is a measure or estimate of the amount of radiation being absorbed by plants. The amount of radiation absorbed is directly related to evapotranspiration, since the plant must cool primarily by evaporating water. The evapotranspiration is constrained by the amount of water in the soil. And for relatively low rainfall amounts, the amount of water in the soil is constrained by rainfall. Hence NDVI correlates with rainfall (Rowland et al. 1996). Drought will continue to occur, but the application of NDVI as a tool for decision making will allow better integration and more timely planning of methods to promote food security.

Drought in Balochistan in 1999 is said to be the worst drought in Pakistan’s history. For the development of the Action Plan for mitigation of drought and development of adaptation strategies systematic zoning of Balochistan province indicating the relative drought risk of various zones is required. This information will be helpful in selecting the interventions for the proposed Action plan and formulation of projects accordingly (Ahmad, 2000).

The aim of present research work was to integrate Satellite Remote Sensing and Geographic Information System for delineating drought vulnerable areas and the objectives were:

  • Climatic Classification of study area on the basis of moisture index.
  • Identification of the cases of drought emergence, moderate and severe drought
  • Application of NDVI to find out the area covered by vegetation.
  • Vegetation change detection at district level.

2. Methodology

2.1 Study Area
The study area for the present research work was Sindh and major part of Balochistan province (Figure 2.1). Sindh province is located in south east part of Pakistan. It lies between 23°-40′ and 28°-29′ north latitudes and 66°-40′ and 71°-05′ east longitudes. It is bordered by the provinces of Balochistan on the west and north, Punjab on the north east, the Indian states of Rajastan and Gujrat on the east, and the Arabian Sea in the south. The total area of the province is 140914 square kilometers. Indus is the main and only river that flows in the province. The climate of Sindh is reminiscent of the Sahara type and of that prevailing in the tropical region of low and dry lowlands. The scanty rainfall, the province gets is often due to cyclonic storms, caused by eastern and western disturbances, particularly the former. The annual rainfall is about 200mm in the Lower Sindh and less than 100 in the Upper Sindh. The average annual rainfall is hardly 125mm.

The province of Balochistan, with the geographical area of 347,000 sq.km (thirty five million hectares) is the largest province of Pakistan. However, only 4.9 percent of the national population lives there. It is located between 600-700 longitude and 250-300 latitude in the northern hemisphere. It has got a varied climate from the sea coast in the south to plateaus 8000 feet above the sea level in the north east. The province of Balochistan falls completely under arid /semiarid zone. The average annual precipitation ranges from 300-400 mm in the north western regions to 75-100 mm in the south western section. This province of Pakistan has faced many droughts in the history and today also it is home to many drought affected people.

2.2. The Datasets
The base data used in this study was 1.1 km NOAA AVHRR LAC Format datasets. The multi date images were acquired on February 9, 1997, February 6, 1998 and February 6, 2000. First two bands with spectral ranges 0.58- 0.68 μm and 0.72 -1.10μm were used for the calculation of Normalized Difference Vegetation Index. 1:1000 000 scale district level map of Pakistan being compatible to NOAA AVHRR coarse resolution was used to overlay administrative boundaries over satellite imageries and NDVI images and clip them according to the extents of study area.

The mean monthly rainfall data for years 1995 to 2000 was available for meteorological stations located in the study area. The rainfall and reference crop evapotranspiration measurements from 1961-1990 for all meteorological stations were used to calculate annual moisture index (percentage).

2.3. Techniques/Formulae

a. Normalized Difference Vegetation Index (NDVI)
The NDVI formula in particular was originally termed the VI (Vegetation Index) and devised by Rouse et al. in 1973 and applied to Landsat MSS data (Tucker 1979).
NDVI = (IR – R) / (IR + R)
The intensity or digital number (DN) value, of each visible red band is subtracted from the infrared band on a pixel-by-pixel basis. That value is then divided by the sum of the two. The result of NDVI is a theme consisting of continuous floating-point data that ranges from -1 to 1. Floating point data includes decimal and negative values.

In the case of AVHRR data, the Near Infrared Layer is Band 2 and the Visible Red Layer is Band 1. The near infrared band of the spectrum emphasizes the contrast between vegetation and water. In the Visible Red Layer, vegetation appears darker than man-made structures.
(AVHRR) NDVI = (Band 2 – Band 1) / (Band 2 + Band 1)
b. Classification
The NDVI image is a single band continuous data, only one value per pixel in each image. The differences in data can be evaluated and pixels having similar values can be assigned to a single class. This classification can be done by an unsupervised nonhierarchical approach Iterative Self-Organizing way of performing clustering. This type of clustering uses a process where new means are recalculated for every class after are assigned to existing cluster centers for each iteration. ISODATA has a default number of 12 iterations and a convergence threshold of 95%. These two parameters keep the tool from running indefinitely to create the classes.

c. Moisture Index
The Moisture Index is used to measure and compare moisture levels at different places. The equation used to compute moisture Index is:
MI= [ (R-Eto) /Eto]*100
MI = Annual /Seasonal Moisture Index (percent)
R = Mean Annual / Seasonal Rainfall in mm.
Eto = Mean Annual / Seasonal Reference Crop Evapotranspiration.
Where Eto is defined as evapotranspiration from an extended surface of 8-15 cm tall green grass cover of uniform height, actively growing, completely shading the ground and not short of water (Farooqi et al, 1995)(Table 2.1).

Manipulation of Normalized Difference Vegetation Index (NDVI) for Delineating Drought Vulnerable Areas

d. Percentage Departure of Rainfall
Considering the normal rainfall, as a standard of comparison the estimation of actual rainfall can be done by calculating the percentage departure of rainfall. It is determined by calculating the difference between actual rainfall and the normal rain fall. This difference (departure in mm) is then divided by the normal rainfall and multiplied with hundred (actual rainfall is the total amount of rainfall recorded and the normal rainfall is standard of comparison: average value of rainfall at a particular date, month or year over a specified 30-year period).

As a part of the Drought Monitoring Program, Pakistan Meteorological Department, Islamabad uses the percentage departure of rainfall to identify drought areas. The standard normal is the average rainfall recorded at the ground observatories from the period 1961-1990 (Table 2.2). The resultant values of percentage departure range from negative to positive. Positive values indicate the wet periods characterized by good rainfall whereas negative values show a below normal conditions.

Vegetation cover is a prime indicator of these three conditions. The survival of vegetation in the absence of available water is impossible so there is almost complete destruction of vegetation in rain fed areas in severe drought condition.

2.4. Methods
Arc View 3.1 with its extensions Image Analyst and Spatial Analyst was the main GIS software used. Each dataset was aligned to a map coordinate system for precise area location. Ground Control Points (GCPs) were extracted from a rectified NOAA AVHRR 1.1km resolution dataset for image to image registration. Normalized Difference Vegetation Index (NDVI) was calculated and image was classified into five categories using Isodata Clustering Technique. These categories were denser vegetation; dense vegetation; sparse vegetation; bare soil; and water bodies, clouds. A small number of pixels remained unclassified. The vegetation cover categorized as denser vegetation was assumed to be healthy, whether cultivated crops or natural vegetation. Dense vegetation corresponded to less healthy vegetation. Sparse vegetation was assumed to be environmentally stressed vegetation that was going to vanish due to unavailability of water.

Interpolation of moisture index (percentage) values of meteorological stations generated a continuous surface. This surface was reclassified to create climatic zones of Pakistan. The rainfall data recorded from time period 1995 to 2000 was organized at two levels. Firstly, the total amount of rainfall recorded during the year and secondly, the total amount of rainfall during the months December/January. The reason for considering December/January rainfall was that satellite imagery available was of early February and rainfall of the two preceding months was considered to be effective for vegetation condition in February. The percentage departure of rainfall was also determined at two levels. This was done to track the incidences of drought and wet conditions per year from 1995 to 2000 and during December/January from 1995-96 to 1999-00 at selected meteorological stations.

3. Results
The resultant NDVI images of study area calculated for the multidate images were the gray scale continuous datasets (Figure 3.1). The vegetation cover was highlighted in the form of bright parts. The shift towards darker shades of gray was due to the presence of other land features including bare soil, water bodies, and clouds. The range of maximum and minimum NDVI values was maximum for February 9, 1997 and minimum for February 6, 2000. This difference was reflected in the area covered by the bright and dark shades of gray in each scene.

The vectorization of categorized NDVI images resulted in land cover maps of the study area. The land features identified were vegetation, bare soil and water bodies (Figure 3.2). These features were common in all the multi date maps. However, the area covered by each feature varied.

The annual moisture index (percent) 1961 to 1990 resulted in the classification of study area in two zones only. These zones were extremely arid and arid zones. Extents of each of these zones are shown in Figure 3.3. The resultant values of percentage departure of total annual rainfall for the years 1995 to 2000 calculated for selected meteorological stations in study area are shown in Table 3.1. These values were used for the identification of drought situation at these meteorological stations. Below normal rainfall in 2000 resulted in the emergence of severe drought situations at all the meteorological stations. The values from 1995 to 2000 for Khuzdar and Dalbandin indicated a gradual decrease in the amount of rainfall as compared to the normal resulting in the shift from wet to severe drought conditions.

The drought situation during December/January 1995-96 to 1999-2000 estimated for selected meteorological stations in study area is shown in Table 3.2. This was based on the results of percentage departure of rainfall calculated for these meteorological stations. Except Karachi, Rohri and Badin, severe drought conditions occurred in all the stations in 2000. The values of percentage departure for Khuzdar indicated a gradual and continuous decrease in the amount of rainfall than normal resulting in severe drought.

4. Data Findings and Discussion
Aridity is a permanent feature of the climate of major parts of Balochistan and Sindh. The mean annual rainfall recorded in these arid areas fails to meet seventy five to ninety percent of mean annual ETo. The meteorological stations lying in the arid zone include Quetta, Kalat, Khuzdar, Chhor, Badin and Karachi. The situation is even more severe in south western Balochistan and south eastern Sindh where the moisture index (percentage) is less than minus ninety. The larger part of study area comes under this zone (Figure 3.3).

Water availability is one of the limiting factors for vegetation growth. If the moisture requirements of vegetation (cultivated and/or natural) are being fulfilled, then this healthy and dense vegetation will be evident from higher reflectance in near infra red region. However, decrease in water availability may limit healthy vegetation and result in decreased reflection in the near infrared region of EM spectrum. This can be supported by the two findings of present study. Firstly, denser vegetation is found mainly in the districts along Indus River. These districts have an irrigation network to support the crop cultivation. A decrease in denser vegetation is prominent towards lower Sindh, eastern Sindh and Balochistan where the irrigation is less or impossible. Secondly, in the rain fed areas the temporal change in vegetation cover occurs due to the change in rainfall pattern. The wet years are characterized by healthy vegetation where as in drought years vegetation growth gets limited and chances of occurrence of sparse vegetation also become less (Fig 4.1).

In order to study temporal change in the vegetation cover at district level two districts (Umerkot and Khuzdar) were selected. Umerkot district lies between 24°-54′ to 25°- 47′ north latitudes and 69°- 11′ to 70°- 18′ east longitudes. Topographically the district has two distinct portions the irrigated area in the west and north and the desert area in the east and south. The climatic conditions in both portions differ considerably. In the irrigated portion the climate is temperate, being neither exceedingly hot in the summer not very cold in winter as compared to the eastern desert area. The mean annual total precipitation from 1961 – 1990 is above 200mm Most of the rain falls in the monsoon months between June and September. The winter rains are insignificant. The mean December/January total precipitation from 1961- 1990 is just 1.5 mm. The distribution of denser, dense and sparse vegetation in the district as obtained from the categorization of NDVI images of February 1997 clearly distinguishes between the irrigated and desert area. The irrigated area has both the forest area as well as the crop cultivation. The exact delineation of the portion covered by forest and cultivated land cannot be done by NOAA satellite imagery only. A satellite image of higher resolution is required to support this result.

A decrease in the vegetation cover of both the irrigated area and the desert area in Umerkot as evident from Figure 4.2 can be due to decline in water availability. In irrigated area, shortage of water occurs as the district is situated at the tails of Nara canal and Mithrao canal. The data regarding the surface water level for the irrigated area was not available so the estimation of hydrological drought leading to agricultural drought was limited. However, the rainfall data available for the meteorological station in Chhor was evaluated for the identification of meteorological drought conditions. Below normal rainfall in December/January of years 1997-98 and 1999-00 resulted in emergence of severe drought conditions. The vegetation cover in this part became environmentally stressed and faced a severe decrease in area in February 2000.

The district Khuzdar of Balochistan lies between 25° 43′ and 28° 52′ north latitude and 65° 42′ and 69° 29′ east longitudes. According to the climatic classification of the study area on the basis of annual moisture index (percent), Khuzdar lies in arid zone with annual average rainfall 243mm. The percentage departure of total December/January rainfall from 1995-96 to 1999-00 shows a gradual increase in below normal rainfall resulting in meteorological drought. Severe conditions in 1997-98 and 1998-99 had an impact on the vegetation cover in the district. A shift in dense vegetation to sparse and environmentally stressed occurred as shown in Figure 4.3.

The example of each of these districts gave an idea that comparison of vegetation cover in later dates with that of previous ones of the same month/season can lead to monitor the vegetation status. This vegetation status determined is the direct indicator of the instances of rainfall occurrence. Further information of physical parameters including soil type, surface water hydrology, ground water potential etc. would be helpful in carrying out multilayer spatial analysis. Also, there is a great need of incorporating the socioeconomic data (population growth, education, health, drinking water availability, migration, employment etc.) to determine the vulnerability of people of theses areas to drought.

5. Summary and Conclusions
NOAA AVHRR derived NDVI can give the picture of vegetation status on regular basis. Although spatial resolution of NOAA is coarse, yet, the onset of drought conditions for a large area in a given year can be predicted by comparative analysis of trend of derived NDVI of that year relative to the trend in a normal year. It is better to develop a multi date NDVI composite for the study area and consider it as a normal for comparison. This warning capability can allow mitigating the effects of droughts through an appropriate redistribution of food supplies for humans and foddering for livestock.

When looking in Pakistan scenario, solving the problem of emergence of drought risk in Sindh and Balochistan areas needs a Permanent GIS Based Drought Forecasting System for the estimation of the complete area that is at risk along with calculation of other factors that are used for vulnerability analysis. This area estimation and analysis will certainly aid in taking such measures that will be useful for the people living in risk area. Moreover, it will provide the demarcation of those areas that may be at drought risk in near future.

The cost of reception and use of NOAA data is low. Thus, for a developing country like Pakistan, application of NOAA derived NDVI data may well represent a low-cost means of developing local expertise in the use of Earth observation systems for resource management and disaster mitigation.

The integration of satellite derived information with other layers including climate, soil type, hydrology and socioeconomic condition of people would be the next requirement. This multidisciplinary information can be effectively and accurately handled with GIS. Spatial analysis in GIS can lead to a decision support system for the concerned government departments, NGO’s and others to help drought vulnerable people and others living in potential drought areas.

Many thanks and gratitude to Dr. Najma Najam, Dr. Uzaira Rafique (FJWU, Rawalpindi), Dr. Abdul Raouf (SUPARCO, Islamabad), Mr. Anjum Bari Farooqi (PMD, Islamabad), Dr. Rakhshan Roohi (NARC, Islamabad) and the staff members of these departments for providing the data required for the present research.


  • Ahmed, S. (2000) Drought Mitigation Measures and Adaptation Strategies for Balochistan Province. Water Resources Research Institute. National Agricultural Research Center, Pakistan. p.3.
  • Bhatti, A. (2000) Investigating the Political Economy of Drought in Pakistan. Disaster Dispatch. May, June, July. pp. 15-19.
  • Farooqi, A. B., Rasual. G., and Chaudhry, Q. Z. (1995) Climatic Classification of Pakistan. Pakistan Metrological Department Agro Met Center, H-8 Islamabad.
  • Glantz, M H., and R. W. Katz. (1977) When is a Drought a Drought? Nature. Vol. 267, pp. 192- 194.
  • Kogan, F.N. (1995) Droughts of the Late 1980s in the United States as Derived from NOAA Polar-Orbiting Satellite Data; NOAA; American Meteorological Society; Camp Springs, MD.
  • Kogan, F. (1997) Global Drought Watch from Space. Bulletin of the American Meteorological Society. Vol. 78, pp. 621-636.
  • Lillesand, T.M., and Kiefer, R.W. (1987) Remote Sensing and Image Interpretation. John Wiley and Sons. New York.
  • Nicholson. and Sharon. (1989) Long-term Changes in African Rainfall. Weather. Vol.44, pp. 46-56.
  • Prince, S. 1991, Satellite Remote Sensing of Primary Production: Comparison of Results for Sahelian Grassland 1981-1988-Special Issue – Coarse Resolution Remote Sensing of Sahelian Environment. International Journal of Remote Sensing. Vol. 12, pp. 1301-1311.
  • Rowland, J., Nadeau, A., Brock, J., Klaver, R. and Moore, D. (1996) Vegetation Index for Characterizing Drought Patterns. Raster Imagery in Geographic Information Systems. Santa Fe, New Mexico, Onward Press, pp. 247-254.
  • Shepherd, A. (1998) Drought Contingency Planning: Evaluating the Effectiveness of Plans. Journal of Water Resources Planning and Management. Vol. 124, no. 5, pp. 246-251.
  • Timmerman, P. (1981) Vulnerability, Resilience and the Collapse of Society. Environmental Monograph No 1. Institute for Environmental Studies, University of Toronto, Canada.
  • Tucker, C.J., Dregne, H.E. and Newcomb, W.W., (1991) Expansion and Contraction of the Sahara Desert from 1980 to 1990. Science. Vol. 253, pp. 299-301.
  • Tucker, C.J., Townshend, J.R.G., and Goff, T.E. (1985) African Landcover Classification using Satellite Data. Science. Vol. 227, pp. 369-375.
  • Unganai, L. S., and Kogan, F. N. (1998) Drought Monitoring and Corn Yield Estimation in Southern Africa from AVHRR data. Remote Sensing of Environment. Vol. 63, No. 3, pp. 219-232.
  • Verstappen, H. (1995) Aerospace Technology and Natural Disaster Reduction. COSPAR Advanced Space Research. Vol. 15, No. 11, pp.14.
  • Warren, A. (1993) Desertification as a Global Environmental Issue. GeoJournal. Vol. 31, No. 1, pp. 11-14.
  • Wilhite, D.A. (1993) Drought Assessment, Management, and Planning; Theory and Case Studies. Kluwer Academic Publishers, Boston.
  • Wilhite, D. and Glantz, M. (1993) Understanding the Drought Phenomenon: the Role of Definitions in Planning for Drought. Westview Press, London. pp. 11-30.
  • WFP (1999) Food Security Zones of Pakistan. United Nations World Food Programme. Vulnerability Analysis and Mapping Unit, Regional Office, Islamabad.
  • Tannehill, I.R. (1947) Drought: Its Causes and Effects. Princeton University Press; Princeton, NJ.