Home Articles Application of Satellite Based Remote Sensing for Monitoring and Mapping of India’s...

Application of Satellite Based Remote Sensing for Monitoring and Mapping of India’s Forest and Tree Cover

Dr. J. K. Rawat
Director, Forest Survey of India

Dr. Alok Saxena
Joint Director, Forest Survey of India

Saibal Dasgupta
Joint Director, Forest Survey of India

Forests are ecological as well as socio-economic resource. These have to be managed judiciously not only for environmental protection and other services but also for various products and industrial raw materials. Considering the crucial role forests play in the country’s ecological stability and economic development, the current National Forest Policy (1988) in India aims at maintaining a minimum of 33 percent of country’s geographical area under forest and tree cover. This requires periodic monitoring of the forest cover of the country for effective planning and sustainable development. Forest Survey of India (FSI), an organization under the Ministry of Environment & Forests (Government of India) has been mandated to monitor and map country’s forest cover on a biennial basis. Consequently, FSI has been carrying out assessment of forest cover in the country using satellite based remote sensing data and has been publishing its findings in the State of Forest Report (SFR) every two years. Its first assessment of forest cover made in 1987 was published as SFR 1987 and the latest i.e., the eighth one as SFR 2001. With the improvement in satellite data resolution and adoption of digital image processing by FSI, it has been possible to assess forest cover patches as small as 1ha. However, there exists a significant tree cover wealth outside conventional forest areas, most of which is less than 1 ha in extent. These include small patches of trees in plantations and woodlands, or scattered trees on farms, homesteads and urban areas, or trees along linear features such as roads, canals, bunds etc. and constitute significant area. In a study done in Haryana State by FSI in 1997, it was found that growing stock of trees outside forest was approximately seven times than that from natural forests. Trees outside forests (TOF) are therefore considered an alternative but significant source of fuel, fodder, timber and environmental services to the local people. In 2001 assessment, FSI assessed tree cover (less than 1ha in extent) in the country using a stratified sampling and field inventory, and estimated it to be 2.48% of country’s geographical area. Thereafter, FSI has developed a methodology based on high-resolution satellite data for mapping and stratification of TOF leading to an improved sampling design for field inventory.

In the present paper, methodologies of forest cover and tree cover assessments as used by the FSI are discussed.

Forest Cover Assessment:
Till recently, FSI was using mostly visual interpretation of satellite data on 1:250,000 scale for assessment of forest cover. However, in its latest assessment i.e., 2001 assessment, it used digital interpretation of satellite data on 1:50,000 scale for mapping and monitoring forest cover. The present methodology uses Digital Image Processing software and involves the following steps:

Acquisition of satellite data: The digital data of IRS-1C and 1D LISS III is acquired from NRSA in CD.. India is covered in about 340 scenes, of IRS 1C and 1D. One scene covers an area of about 20000 km2, having an overlap of about 10% with adjoining scenes. While procuring the data, care is taken to ensure that it is cloud free (with not more than 10% cloud cover) and therefore data pertaining to the period from October-December is preferred.

Geometric Rectification of raw data: After downloading the data into computer, rectification is carried out in each image to provide Latitude and Longitude information into raw satellite scene using raster based geometric corrections. Rectification carried out in geographic projection is re-projected in shape of polygonal projection and the scene is geo-coded with using SOI toposheets.

Mosaicing of rectified scenes: Different scenes, which are already rectified, may have to be merged together to get one combined FCC (False Colour Composite). FCC of sheet is extracted from mosaiced scene in a chosen area of interest. Image is displayed in three bands 3, 2, 1. Masking of non-forest areas is done separately to extract forest areas on the basis of ground knowledge, cover map of previous cycles and on the basis of information available through SOI toposheets in the area of interest.

Classification of forest cover using NDVI: Interactive method of display is used for assigning threshold values for each class (open, dense and scrub) on the basis of the ground knowledge to highlight forest/vegetated areas. Density class of forest cover and colour is accordingly allocated. Survey of India toposheets is used for delineating boundaries of each district and classified map of forest cover is generated.


Flow chart of methodology of dynamic forest cover mapping using remote sensing is shown in figure-1

Figure1- Flow chart of Forest cover mapping using remote sensing
The output includes forest cover maps on 1:50,000 scale. These maps show forest cover in three classes- (i) Dense forest, having canopy density of more than 40%, (ii) Open Forests with canopy density between 10-40% and (iii) Scrub which are forest areas having less than 10% canopy density. These maps are also generated for district and States/Union Territories by overlaying the respective District/State/UT administrative boundary. Area under forest cover at District/State/country level is then assessed. Change maps are also prepared to depict changes taking place under different land cover classes.

In its latest assessment of 2001, taking advantage of advancements in remote sensing and improvement in digital interpretation qualities, FSI has provided a much more comprehensive status of forest cover in the country than in the previous assessments. Some of the new features incorporated in this assessment are:

  • For the first time FSI has interpreted the satellite data of the entire country digitally. In earlier estimates, interpretation has been largely visual. Digital interpretation has the advantage of overcoming subjectivity prevalent in visual method.
  • Due to absorption of digital image processing technique, it has been possible for FSI to interpret the data on 1:50,000 scale. This has resulted in providing more realistic information on forest cover as areas having forest cover down to 1 ha could be delineated while in earlier assessments, forest cover down to 25 ha could only be delineated. Similarly blanks down to 1 ha within forested areas can be separated. The entire exercise has resulted in new base-line information on forest cover.
  • As perennial woody vegetation (including bamboos, palms, coconut, apple, mango etc.) has been treated as tree and thus all lands with tree crops, such as agro-forestry plantations, fruit orchards, tea and coffee estates with trees etc., have been included in forest cover.
  • Mangrove cover has been classified into dense and open mangrove cover. The area of mangrove cover so assessed has been merged in the respective area figures of dense and open forest cover.
  • A classification is not complete unless its accuracy is assessed. For the first time an independent and systematic assessment of accuracy of satellite data interpretation was made. An error matrix was generated by comparing classified forest cover with the actual forest cover on the ground at 3,608 locations spread throughout the country. High resolution PAN data was used as proxy for ground verification. The overall accuracy of forest cover classification was found to be 95.9%.
  • Though forest cover in areas as less as 1 ha in extent could be assessed using satellite data, significant tree cover exists in patches of less than 1 ha and in linear shapes along roads, canals, etc. and scattered trees that can not be assessed using remote sensing. An attempt is made for the first time to assess such tree cover using ground inventory method.

The abstract of forest cover assessment 2001 is given in Table 1.
Table 1: Forest Cover as per 2001 assessment

Class Area (km2) Percent of Geographic Area
Forest Cover
a) Dense 416,809 12.68
b) Open 258,729 7.87
Total Forest Cover* 675,538 20.55
Scrub 47,318 1.44
Total Non-forest** 2,611,725 79.45
Total Geographic Area 3,287,263 100.00

*includes 4,482 km2 under mangroves (0.14 percent of country’s geographic area)
**includes scrub

Forest Cover Assessment 2001

Figure 2- Forest cover in India
Tree Cover Assessment:
TOF is assessed in rural as well as urban areas, although the greater part exists in rural areas. Initially, conventional field method was used for TOF assessment in rural areas. The state or a group of districts is considered as the study area. Since this area is fairly large there is every possibility of heterogeneity of the study variable i.e. growing stock. TOF being planted along with agricultural crops is likely to be influenced by the Agro-ecological variables. Therefore, study area is stratified according to agro-ecological zones (AEZ), which has already been demarcated by other agencies. Districts, in India, are the basic planning and administrative units, which influence the TOF and therefore, is considered for further stratification of AEZs. Villages are treated as sampling units. Optimum number of sample villages is selected randomly from different districts proportionate to the TOF area of the same. Complete enumeration of all the trees with diameter of 10 cm and above at breast height in the randomly selected villages in each district is carried out. Data is collected on pre-designed formats following prepared instructions for fieldwork and collected data is processed following appropriate formula. The above-mentioned methodology was providing accurate estimates but was very time consuming. It was not able to provide precise information at district level, which is the basic unit for economic planning.

Methodology Using Remote Sensing Data
To do away with these constraints many alternatives were tried and finally a methodology based on digital image processing and GIS analysis using multi spectral and panchromatic data for mapping of trees outside forests (TOF) was devised. The remote sensing data is used to provide stratification of the TOF resources, which is utilized to increase the precision and is time effective. In addition, sometimes the objectives of TOF resource assessment may require spatial distribution of resources on maps along with several other features. This objective can also be appropriately tackled by this methodology.

High-resolution satellite imageries provide information even up to identification of a single tree but these are cost prohibitive. The IRS LISS III data, which is multi spectral, and has a resolution of 23.5 m ×23.5 m, provide information on vegetation cover. There are techniques available through which tree vegetated land can be segregated from agriculture land if the tree vegetated patch is about one ha and more. However, LISS data cannot be used for smaller patches or scattered trees. The IRS PAN data, which is monochromatic, having resolution of 5.8 m × 5.8 m can identify a tree vegetated land even less than 0.1 ha. Therefore, both LISS III and PAN imageries are used for stratification of TOF resources on the basis of geometrical formation of trees i.e. block plantation (group of trees), linear plantation and scattered trees.

Raw images of IRS IC/D PAN and LISS III data for the period between Oct.-Dec. 2002 are acquired from National Remote Sensing Agency, Hyderabad. Thereafter, the PAN image is geometrically rectified with the help of Survey of India toposheets on 1:50,000 Scale. The LISS III image is then co registered with the rectified PAN images. PAN and LISS III images are fused using appropriate algorithm. Since mapping of TOF areas is the objective, the boundary of forest area is digitized and masked out. The remaining fused image are classified into settlement, water bodies, burnt areas, tree cover and agriculture area using appropriate classifier viz. Maximum likelihood. This classification enables the interpreter to distinguish between tree cover and other classes on fused image. This classified image is visually analyzed with respect to fused images for editing and refinement for inclusion and omissions. Since a cluster of trees having 0.1 ha area or more is defined as block plantation, pixels are clumped and cluster of pixels having area less than 0.1 ha are eliminated. After editing of the classified image the final classified map is generated which is done by taking the PAN, LISS-III and the fused images. Incorporating these corrections final classified image is prepared having three classes in TOF areas, namely, Block, Linear and Scattered. From the classified TOF map information pertaining to area under Block, Linear, Scattered and water bodies can be calculated. In addition, such areas, which do not support tree vegetation, like rivers and water bodies, snow covered mountains, marshes,etc. which is termed as Culturable Non Forest Area (CNFA)can also be calculated. Such information is very helpful for district level planning.

Flow chart of methodology of Tree Cover mapping using remote sensing is shown in Figure-3

Figure 3- Flow chart of methodology of Tree Cover mapping

Sampling Method
Besides generation of TOF maps, the information on block, linear and scattered patches can be used to estimate the number of trees and the corresponding volume (species wise) using appropriate sampling design by laying out optimum number of plots randomly selected in every stratum. Since the variability in each stratum is expected to be different demanding different sample and plot sizes, pilot studies were conducted to ascertain these so that the variability of the stratum can be properly addressed. In this pilot study, 0.1 ha, 0.2 ha and 0.3 ha plots were considered for Block Stratum. Similarly, strip of size 10 m × 75 m, 10 m × 100 m, 10 m × 125 m, 10 m × 150 m, 10 m ×175 m & 10 m × 200 m were considered for Linear Stratum. For scattered stratum plot of size 0.5 ha, 1.0 ha, 1.5 ha, 2.0 ha, 2.5 ha and 3.0 ha were considered for non-hilly districts and 0.25 ha, 0.50 ha, 0.75 ha and 1.00 ha were considered for the hilly districts. Twenty concentric plots in each stratum were randomly selected and data were recorded. After analysis it was concluded that optimum plot size for Block, Linear and Scattered stratum are 0.1 ha, 10 × 125 m strip and 3.0 ha respectively for non-hilly districts and 0.1 ha, 10 × 125 m strip and 0.5 ha for hilly district. It was also concluded through pilot study that the sample sizes for Block, Linear and Scattered stratum are 35, 50 and 50 respectively for non-hilly districts and 35, 50 and 95 for hilly district.

Desired number of sample points was randomly generated in each stratum separately and the data on pre decided variables were collected on designed formats, following Manual for Assessment for Trees Outside Forests (FSI, 2003). Thereafter, data processing is carried out following appropriate formulae corresponding to sampling design. The following table indicates the results obtained with regard to stems/ha, total number of stems, volume/ha and total volume of trees outside forests in rural areas of Gurdaspur district of Punjab, India. Likewise, similar results obtained from different districts spread across the country are aggregated to generate national level figures (Table 2).
Table 2: District level estimates (Gurdaspur, Punjab, India)

Geographical Area 3,551 sq.km.
Urban Area 76.42 sq.km.
Forest Area 368 sq.km.
Water bodies 94.58 sq.km.
CNFA (Rural) 3,013 sq.km.
Stems / ha 18.5
Total Stems 5,563,798 (5.56 M)
Volume / ha 3.5 cu.m
Total Volume 1,054,577 cu.m(1.05 M cu.m)

Accuracy of Classification
Any classification is not complete unless and until its accuracy is assessed. For the present study the accuracy of classification was assessed by taking 53 points in block, 65 in linear and 65 in scattered stratum for a particular district. It is recommended that 50 or more points should be located for ground verification in each class. The accuracy of this classification was high as evident from the following confusion matrix of Kapurthala district of Punjab state.
Table 3: Confusion Matrix

  Block Linear Scattered Row Total User’s Accuracy (%)
Block 41 0 0 41 100
Linear 0 63 0 63 100
Scattered 12 2 65 79 82
Column Total 53 65 65 183  
Producer’s Accuracy (%) 77 97 100    
Overall Accuracy = 92 %

The main objective of Forest survey of India in mapping and monitoring forest and tree cover of the country on a two-year cycle is to know the dynamic changes of forest resources in terms of quantity and quality over a period of time so that appropriate planning and management interventions can be developed for their conservation and sustainable utilization. Remote Sensing based forest cover mapping and monitoring adopted by FSI has proved to be cost and time effective over traditional forest resource monitoring. The methodology using digital image processing and geographical information system, as explained above can be effectively employed using multi spectral and high-resolution satellite imageries to stratify the TOF resources in such a way that the classification system of TOF resource remains valid. In addition, spatial distribution of TOF resources on maps along with other features will provide information for planning and implementation and utilization of these resources in a sustainable manner. Since, this methodology enables resource-based stratification, it is expected to provide better estimates of TOF resources than the one generated through field survey alone.


  • Belouard, T. 2002. Trees outside forests: France. FAO Conservation Guide, 35
  • Forest Survey of India, Dehradun, 2003: Manual on Assessment of Trees Outside Forests.
  • Forest Survey of India, Dehradun 2003: State of Forest Report 2001.
  • Glen, W.M. 2002, Trees Outside Forests: Sudan. FAO Conservation Guide, 35.
  • Hidalgo, D.M. and Kleinn, C. 2002; Trees Outside Forests: Costa Rica. FAO Conservation Guide, 35.
  • Holmgren, P., masakha,E.J. and Sjoholm,H.1994. Not all African land is degraded:A recent survey of trees on farms in Kenya reveals rapidly increasing forest resources.Ambio 23(7): 390-395
  • Kleinn, C. 2000. On large area inventory and assessment of trees outside forests. Unasylva, 200(51), 3 – 1-.
  • Legilisho-Kiyiapi, J. 2002. Trees outside forests:Kenya. FAO Conservation Guide, 35
  • Proceedings of the International Training Workshop on Assessment of Trees outside Forests conducted by FSI in April,2002 in collaboration with FAO of the United Nations.
  • Rawat,J.K.,Dasgupta,S.,Kumar,Rajesh.,Kumar,Anoop.,Chauhan,K.V.S.2003: Training Manual on Inventory of Trees Outside Forests published by FAO under the EC-FAO Partnership Programme.