34 Application of Remote Sensing for Vegetation Mapping

Dr Dinesh Kumar

 

CONTENTS

 

1.  Learning Objectives

 

2.  Introduction

 

3.  Vegetation Mapping

 

4.  Working with Classifier and Vegetation Indices

 

5.  Summary

 

6.  References

 

1. Learning Objective

  • To understand the concept of Image classification and band ratioing and its application in Vegetation Mapping

 

2. Introduction

 

Survival of human beings in the changed scenario lies upon many factors and vegetation is one of the most important factor. Coexistence of human with harmony with nature is the need of the hour. The growing concern of global environmental issues particularly deforestation, drags the attention to reevaluate the unethical human behavior towards nature and correcting things by adopting sustainable ways. Restoration and conservation of nature’s lung is more precisely left option to let the nature regain its equilibrium favorable to humans. In order to design plans, policies and strategies, status of vegetation as forest cover, agricultural practices, vegetation diversity and aquatic flora at present and in the past becomes reference line. Monitoring of the same also becomes equally important as an indicator of growth and progress. Preparing the status of vegetation is done by traditional method and applying advance sophisticated technology also. While traditional methods like field surveys, literature reviews, map interpretation and collateral and ancillary data analysis are time consuming, date lagged, too expensive and ineffective for large scale vegetation cover, advanced technology like remote sensing offers systematic observations at various scales, data archives from present to several decades back.

 

The present module emphasizes the potential of remote sensing and GIS technology in providing cost effective, at desired level and more precise way of solutions.

 

 

3.  Vegetation Mapping

 

Vegetation mapping or extraction is done from the remotely sensed images by means of

  1. Image classification methods
  2. Vegetation indices based on band ratioing

 

Preprocessing of satellite data using image processing techniques is pre-requisite before applying any of above mentioned methods. Satellite data quality depends upon season and sensor health. For example, data taken during bad weather or spot or line over sensor lens, reduced atmospheric visibility, need to remove or reduce these effects to enhance ground surface information which is met by applying atmospheric correction. Further, the images should also be radiometrically, geometrically corrected and ortho-rectified to avoid any spatial data errors.

 

Although most of aforementioned procedures are applied to raw data while processing at 1A, 1B, 2A, 2B, 3A, 3B level, it is always better to apply the prerequisite procedures for customize purposes. Raw data captured at sensor level is termed as zero “0” Level data which is sometimes acts as important inputs for many derived parameters. So, it is better to know the level of data (0,1A, 1B, 2A, 2B, 3A, 3B) before going for any satellite data procurement. Generally 3A level data is already gone through radiometrically, geometrically corrected and ortho-rectification processes.

 

 

4.  Working with Classifier and Vegetation Indices

 

4.1.  Vegetation mapping based on Indices:

 

The retrieval of information about the features of earth’s surface such as vegetation depends upon remote sensors; a key device of remote sensing systems. The information of surface features are captured at sensor as a unique pattern of spectral radiances at different spectral bands. The unique spectral pattern of surface features is termed as spectral signature (Fig-1). Like other earth’s surface features, spectral signature of vegetation shows its unique spectral reflectance peak at visible and NIR region and spectral information at these two peaks (Vis and NIR) is used to extract vegetation among the other features. The matter of the fact is that spectral signature curve for different vegetation species and diseased vegetation show average similar spectral curve pattern with varying peaks at visible and NIR region (Fig-2). The varying peaks provide the information about the type, stage and stress condition of vegetation (Fig-2).

 

Vegetation type, diversity, cover, species richness, agricultural pattern are among the different ways to represent vegetation details. To differentiate vegetation from other land cover type, classifying satellite imagery based on spectral radiance values can be used. However in most heterogeneous regions like metros and highly populated urban centers, more precise methods are also applied.

Fig-1: Spectral Signature of different features

Image Source: http://www.seos-project.eu/modules/classification/classification-c01-p05.html

 

When it comes to delineate vegetation species, species richness or abundance, forest cover types, more spectral information needs to be used in peculiar way. Here comes the role of band ratioing to reduce the spectral biasness where closer spectral information of vegetation needs to be segregated based on criteria such species, richness or abundance etc. usually band ratioing is performed by using spectral information in visible and NIR region in a mathematical way (Fig-3).

 

Using the band ratioing logic, several vegetation indices have been designed to enhance within species information or different stage of vegetation. Vegetation index were designed on the basis of potential use. They can be grouped into

 

a) broadband indices which uses broadband reflectance

b) narrowband indices which use narrowband reflectance (hyper-spectral)

c) leaf pigment indices which were designed to examine several pigments in the leaf

d) stress indices, developed for monitoring stress conditions in the canopy

e) water stress indices

Fig-2: Comparative Spectral Signature of different vegetation species

Image Source: http://www. researchgate.net

 

Fig-3: Example of spectral biasness due to topography

Image Source: http://www. researchgate.net

 

Spectral vegetation indices not only enhance the spectral reflectance information variability due to vegetation characteristics but also reduce the soil, atmospheric and sun-target-sensor geometry effects. Over the time, spectral vegetation indices have been advanced significantly and were optimized for particular application/sensors. Most of the vegetation indices works peculiarly and species specific and becomes little less effective when applied across different species, with different canopy architectures and leaf structures.

 

Tabel-1: List of VIs along with its equation

 

pNIR is the near infrared reflectance; pred is the red reflectance; pgreen is the green reflectance; pblue is the blue reflectance; px is the reflectance at a specific wavelength.

 

(a)  Difference Vegetation Index (DVI)

Probably the simplest vegetation index DVI = NIR – Red

Sensitive to the amount of vegetation Distinguishes between soil and vegetation

Does NOT deal with the difference between reflectance and radiance caused by the atmosphere or shadows

 

(b)   Simple Ratio (SR)

SR = NIR/Red

High for vegetation

Low for soil, ice, water, etc.

Indicates amount of vegetation

Reduces the effects of atmosphere and topography

 

(c)Normalized Difference Vegetation Index NDVI

NDVI = (NIR – Red)/(NIR + Red)

Ranges from -1 to 1

Indicates amount of vegetation, distinguishes vegetation from soil, minimizes topographic

effects, etc.

A good index

Does not eliminate atmospheric effects

 

4.2. Image classification:

 

Classification process takes into account of all pixel values and groups them into number of classes or “themes”. Such categorized data is then used for producing thematic maps of land cover such as vegetation of an image. These pixel values is considered from pan (single band) data or multispectral (more than one) or hyper-spectral data.

 

The classification process can be done using a number of methods listed below:

 

a)      Supervised

 

b)      Unsupervised

 

c)      Hybrid Approach

 

d)     Object based classifier

 

e)      Fuzzy classifier

 

f)       Image fusion classifier

 

Among the listed methods supervised and unsupervised is sometimes termed as traditional methods. While methods like hybrid approach, fuzzy classifier and ANN are considered as improved or advanced classifier. The accuracy of any classifier is assessed by its capacity to segregate feature class from mixed pixels particularly in heterogeneous surfaces. Vegetation extraction in highly populated urban regions becomes challenging task where patches or only few pixels of vegetation are scattered randomly in scene.

Fig-4: Comparative flow diagram of supervised and unsupervised classification method

 

4.2.1. The main difference between supervised and unsupervised classifier is as under:

  • The supervised method collects the information classes (i.e., vegetation) of interest in the image termed as “training sites”. Further signature analysis stage is followed in which reflectance of each class is statistically characterized by means of mean, variances and covariance over all bands. Finally the classes are assigned for whole image by validating each pixel with those of statistically characterized user specified samples achieved during signature analysis.
  • On the other hand unsupervised classifier creates natural groups or clusters using image pixels. This classification method lacks analyst-specified training data unlike supervised classification.
  • Initially Unknown spectral classes result from unsupervised classification based on natural groupings of the image values. Further, the unknown classes must be compared with reference data (such as larger scale imagery, maps, or site visits) in order to identity and assign spectral classes.
  • Thus, spectral saperability is examined after the traninig stage in supervised approach and class saperability is assigned on the basis of natural groups determined by computer in unsupervised approach.

Object based and fuzzy approaches fall under advanced or intelligent classifier used for vegetation delineation.

 

4.2.2. Fuzzy Approach

 

Fuzzy approach generally applied over suburban areas where mixed surface features are abundant. It is probability-based approach which considers the gradient and assigns the weightage to occur or not-occur at the same time. Fuzzy classifier assumes that transition from one to other happens gradually rather than abrupt. In order to apply its logic, decision tree (DT) is derived using regression approach to assign class probability within the pixel. Since this method uses probability-based approach, hence also called as soft classification technique.

 

4.2.3. Object-based classifier

 

Object-based classifier groups pixels based on similar texture, color, and tone into shapes or segments and classify the image into themes. This method can be used effectively on high-resolution imagery which has increased variability of spectral content of individual pixels related to same class. Since, geometry of objects mimics more closely with natural landscape, they become significant classification unit in place of pixels.

 

4.2.4. Image Fusion approach:

 

Another method for vegetation mapping is image fusion of satellite images. It can be better understood by an example i.e. high resolution pan data can be fused with multispectral coarser satellite data to enhance spatial information. The logic behind image fusion is that individual sensor information may have error in terms of consistency, preciseness for particular application. Image fusion enhances the spatial information to identify and extract the vegetation. The image fusion can be applied at different temporal, spectral and spatial resolutions

 

5. Conclusion:

 

Resource mapping is one of the important key factor which acts as base for decision and policy planning process. Forest cover, agricultural pattern and biodiversity status are pillars of natural equilibrium of the region. If not managed properly, the region may face severe environmental challenges. Remote sensing technique provides a powerful systematic tool to monitor, map and model the different vegetation cover and provides a precise and accurate road map for many aspects. Band ratioing extracts vegetation from heterogeneous surface features and reduces the spectral biasness also. A number of vegetation indices have been developed in order to create species diversity, canopy cover, agricultural growth cycle maps used for many decision making processes.

 

Image classification by means of traditional and intelligent classifier is also being applied for vegetation mapping. Fuzzy and object based classifier are used for automatic and heterogeneous surfaces while hard classification methods like supervised and unsupervised methods are used for less complex surface features.

 

On a whole it is pretty evident that remote sensing techniques is very powerful and potential technology ready to use for across a range of vegetation related mapping, analysis and modeling approaches which provides very intelligent and precise assessment.

 

References:

  • Gobron, N., Pinty, B., Verstraete, M. M., & Widlowski, J. L. (2000). Advanced vegetation indices optimized for up-coming sensors: Design, performance, and applications. IEEE Transactions on Geoscience and Remote Sensing, 38, 2489–2505.
  • Govaerts, Y. M., Verstraete, M. M., Pinty, B., & Gobron, N. (1999). Designing optimal spectral indices: A feasibility and proof of concept study. International Journal of Remote Sensing, 20, 1853– 1873.
  • http://www.sc.chula.ac.th/courseware/2309507/Lecture/remote18.html.
  • https://www.e-education.psu.edu/geog883/node/523.
  • Moulin, S., & Guerif, M. (1999). Impacts of model parameter uncertainties on crop reflectance estimates: A regional case study on wheat. International Journal of Remote Sensing, 20, 213–218.
  • Myneni, R. B., Hall, F. G., Sellers, P. J., & Marshak, A. L. (1995). The interpretation of spectral vegetation indexes. IEEE Transactions on Geoscience and Remote Sensing, 33, 481–486.
  • Yichun Xie, Zongyao Sha, and Mei Yu (2008). Remote sensing imagery in vegetation mapping: a review; Journal of Plant Ecology Vol-1, No-1, P 9–23.
  • http://gisgeography.com/image-classification-techniques-remote-sensing/.