Detecting forest areas and crops using vegetation indices

Detecting forest areas and crops using vegetation indices

Dr. Mohd. Ibrahim Seeni Mohd, Azhar Jj. Salleh
Centre for Remote Sensing
Faculty of Surveying
University Teknologi Malaysia
Locked Bag 791, 80990 Johor Bahru, Malaysia

Abstract
Vegetation Indices (VI) represent a linear transformation of spectral bands that express the spectral behavior of crops and natural vegetation. VI emphasizes spectral contrast between different surface, and therefore have been widely used for enhancing vegetation and classifying agricultural areas. This paper reports on studies carried out in the Bukit Kajang Forest Reserve and surrounding crops areas at Raub, Malaysia with the landsat-5 Thematic Mapper satellite data using various vegetation indices to evaluate their potential in differentiating forest area and major crops. The Perpendicular Vegetation Index ( PVI), Normalized Difference Vegetation Index (NDVI), and Ratio Vegetation Index (RVI) have been evaluated on the basis of their spectral band combinations and rationing for differentiating rubber, oil palm and forest areas. The combination of Mid-Infrared (MIR) band 5 and Visible (VIS) band 2 in the PVI gave the most difference among crops whereas the NDVI and RVI resulted in similar values. Significant differences were found between the forest areas and the other crops thus enhancing the forest areas. Better classification accuracies were achieved by using the PVI compared with other Vegetation Indices.

1.0 Introduction
Several empirical indices have been used as quantitative indicators of vegetation amount They reduce the multidimensional spectral space of vegetated scene to one dimension in order to sense variability in such properties as biomass, leaf areas index, fractional cover and types ( jasinski, 1990). During the past decade the wavelengths used have been restricted to VIS and Near-Infrared (NIR) spectral region. By combining differences and ratios of red ( VIS) and NIR, the vegetation indices respond to i) the relatively high radiation absorption of red light by leaves due to the presence of chlorophyll and ii) the high reflectance of NIR light due to scattering in the leaf internal structure ( Curran, 1980). Common ratio vegetation indices include the NIR/VIS index, and Normalized Difference Vegetation Index [NDVI= (NIR-VIS)/(NIR + VIS)] ( Richardson et al. ( `1991)). NDVI is commonly preferred because undesirable aspects on recorded radiance such as effect of variable illumination resulting from variation in topography can be reduced. Another VI that distinguishes the spectral response contributed by the soil background is the Perpendicular Vegetation Index which is described by Richardson et al. ( 1991) in the form.

PVI = √( RggNIR-RpNIR)2 = ( RggVIS-RpVIS)2 (1)

Where:
PVI is the perpendicular distance between the candidate vegetation point and the soil line,
R_p is the reflectance of a candidate vegetation point for NIR and VIS spectral region, and
R_gg is the reflectance of soil background corresponding to a candidate vegetation point.

Evert et al. ( 1989) described that MIR spectral region reflectance data obtained from TM5 ( 1.55-1.75mm) and TM7 ( 2.08-2.35mm) water absorption band of the Landsat were also useful for estimating vegetation parameters. He showed that the 1.65mmmm and 2.08mm wavelengths gave promise for use in discriminating bare soil from mature field crop. He studied VIS, NIR, and MIR reflectance data for winter wheat and corn, and found that the TM5 and TM7 bands in the MIR spectral region were more useful than the NIR/VIS index (TM4/TM3) for estimating agronomic variables.

This paper reports on the use of various VI calculated from satellite data at selected wavelengths in VIS and NIR spectral region for differentiating vegetation types at the study area. Since the MIR spectral region has been proven useful for agronomic variables by many researchers overseas, this paper will also examine the potential of using MIR wavelengths as input to the Via algorithm in a selected region in Malaysia.

2.0 Study area and data acquisition
The Bukit Kajang Forest Reserve and its surrounding agricultural areas which cover a ground area approximately 225 square km was selected as the study site. The areas is equivalent to 512x512 pixels of the TM image. The area is located approximately 25 km to the north-west of Raub, Pahang. This region is cultivated with major cover types such as oil palm, rubber, cocoa, banana and forest area. The location map of the study areas is shown in Figure 1.

Figure 1. Location map of study area
A Landsat Thematic Mapper ™ scene (WRS 127/57, A4) acquired on 15 June 1989 covering the study areas was used. This particular scene was chosen because it was the best quality data available and contains most of the major crops including forest areas. Other ancillary data such as Landuse maps, topographical maps, and related information were used in the study.

3.0 Data processing and calculation of VI
The image processing was carried out using the Intergraph IP225 system and the PCI EASI/PACE system available at the Centre for Remote Sensing, Universiti Teknologi, Malaysia. Contrast and linear stretching, band combination and image filtering were carried out to the satellite data to enhance vegetation area. The geometric rectification was performed using second order polynomial transformation with geometric rectification was performed using second order polynomial transformation with thirteen GCPs to sub pixel accuracy. The nearest neighbourhood resampling technique was then applied to the data since this technique will maintain the original DNs. Later the band combination of 5,4, and 3 was used because it has been found to be good for most forest and vegetation surveys where the absorptive and reflective properties of the vegetation are of importance.

The VIs were calculated by using the Complex Arithmetic Algorithm task on the Intergraph IP225 system. The VI algorithms used were

RVI = NIR (1) VIS NDVI = ( NIR-VIS) (2) ( NIR + VIS) PVI = √( RggNIR-RpNIR)2 = ( RggVIS-RpVIS)2 (3)

In the above equations , VI values are also obtained by replacing The NIR band with MIR band. For the PVI equation, the mean DNs for soil were found to be 47, 50, 72, 102, and 48 for TM bands 2,3,4,5, and 7 respectively. These values are assumed to be content throughout the areas since the soil in the areas is of the same type. All TM bands except TM1 and TM6 have been used to derive the VIs from equations (2) , (3) and (4).

Various band combinations were analysed and the results from some of the combinations are given in Tale 1. Figure 2 shows the PVI map of the study area from bands 5 and 2 of the Landsat TM.

Figure 2. Perpendicular Vegetation Index of study area from bands 5 and 2 of Landsat TM
Table 1: VI ranges for different types of vegetation and soil in study areas using selected band combinations

VI	FOREST	OIL PALM	RUBBER	COCOA	BARE SOIL
RVI ( TM4,3)	2.73 – 3.47	3.29 -3.68	2.96-3.36	2.59 -3.39	1.07-1.80
NDVI (TX4,3)	0.48-0.55	0.53-0.57	0.50.0.54	0.44-0.55	0-0.27
NDVI (TM5,2)	0.24-0.31	0.30-0.34	0.39-0.43	0.36 -0.43	0.33 – 0.47
PVI (TM4,3)	24-31	31-37	27-34	25-39	0-15
PVI (TM5,2)	48-52	42-48	25-33	18-25	0-24
PVI (TM5,3)	49-59	45-50	28-36	27-37	0 – 27

4.0 Results and Discussions
The RVI and NDVI using NIR and VIS spectral region showed similar VI values for most crop types. Thus, low classification accuracy were achieved using these indices. However, these indices gave promising results for use in discriminating between vegetation cover and bare soil. PVI in bands 4 and 3 gave distinct contrast between forest and crops. Analysis using MIR band 5 and VIS bands 2 and 3 produced better classification accuracy. By using the PVI equation (4) , highest classification accuracy of 73% was obtained using band 5 and 2 whereas , the combination of band 5 and 3 gave a classification accuracy of 71%.

5.0 Conclusions
It was found that crop classes separation based on RVI and NDVI was not good. These indices provide general values for vegetation types but identification of vegetation from non-vegetated area is feasible. The combination of MIR and VIS in the PVI resulted in the most difference among crops as well as significant difference between forest areas and crops. Therefore, it can be concluded that the combination of MIR and VIS spectral bands in the PVI equation is useful for the identification of crop classes and forest areas.

References

Curran, P. ( 1980), Multispectral Photographic Remote Sensing of Vegetation Amount and Productivity. In Proceedings of the Fourteenth International Symposium on Remote Sensing of the Environment, Ann Arbor, MI, pp. 623-637.
Evert, J.H., Escubar, D.E., and Richardson A.J. ( 1989), Estimating Grassland Phytomass Production With Near-Infrared and Mid Infrared Spectral Variables, Remote Sens. Environ. 30: 257-261.
Jasinski, M.F. ( 1990), Sensitivity of the Normalized Difference Vegetation Index to Sub pixel Canopy Cover Soil Albedo, and Pixel Scale, Remote Sens. Environ. 32: 169 – 187.
Richardson, A.J., Wiegand, C.L., Escobar, D.E., and Gerbermann, A.H. ( 1991), Vegetation Indices in Crop Assessment, Remote Sens. Environ 35:105-119.