Letters

Coherence-based land cover classification in forested areas of Chattisgarh, Central India, using environmental satellite—advanced synthetic aperture radar data

[+] Author Affiliations
Vyjayanthi Nizalapur

Forestry and Ecology Division, Land Resources Group, National Remote Sensing Centre, Indian Space Research Organisation, Balanagar, Hyderabad-500625, Andhra Pradesh, India

Rangaswamy Madugundu

Antrix Corporation Limited, Indian Space Research Organisation, Bangalore, Karnataka, India

Chandra Shekhar Jha

Forestry and Ecology Division, Land Resources Group, National Remote Sensing Centre, Indian Space Research Organisation, Balanagar, Hyderabad-500625, Andhra Pradesh, India

J. Appl. Remote Sens. 5(1), 059501 (March 14, 2011). doi:10.1117/1.3557816
History: Received November 24, 2008; Revised January 23, 2011; Accepted January 27, 2011; Published March 14, 2011; Online March 14, 2011
Text Size: A A A

Open Access Open Access

In the present work, the potential of synthetic aperture radar (SAR) interferometric coherence in land cover classification is studied over forested areas of Bilaspur, Chattisgarh, India using Environmental Satellite—Advanced Synthetic Aperture Radar (ENVISAT-ASAR) C-band data. Single look complex (SLC) interferometric pair ASAR data of 24th September 2006 (SLC-1) and 29th October 2006 (SLC-2) covering the study area were acquired and processed to generate backscatter and interferometric coherence images. A false colored composite of coherence, backscatter difference, and mean backscatter was generated and subjected to maximum likelihood classification to delineate major land cover classes of the study area viz., water, barren, agriculture, moist deciduous forest, and sal mixed forests. Accuracy assessment of the classified map is carried out using kappa statistics. Results of the study suggested potential use of ENVISAT-ASAR C-band data in land cover classification of the study area with an overall classification accuracy of 82.5%, average producer's accuracy of 83.69%, and average user's accuracy of 81%. The present study gives a unique scope of SAR data application in land cover classification over the tropical deciduous forest systems of India, which is still waiting for its indigenous SAR system.

Figures in this Article

The availability of reliable and up-to-date land cover information derived from remotely sensed data is necessary for regional land cover change studies, ecological monitoring, map updating, management, and planning activities.1 Besides the use of optical satellite sensor data in land cover classification reported in earlier studies, advancement in image classification techniques for all-weather capable synthetic operture radar (SAR) data has provided better methods for land cover classification such as segmentation methods2 and Bayesian classifiers for the classification of multifrequency polarimetric data.3

Among the different techniques for land cover classification using SAR data, the use of an interferometry technique has contributed in improving the accuracy of the land cover classification.4 Earlier studies5 have shown that the coherence component derived from an interferometric image pair provides additional information for land cover classification.6 The coherence is generally low over forested areas and high for open fields, which allows discrimination between forested and non-forested areas.7 Further, it was also observed that synergic use of SAR backscatter with InSAR coherence significantly enhances the sensitivity of the SAR system as a whole for forestry applications.8 Using coherence image, multi-temporal images and seasonal backscatter change images at appropriate seasons discriminate various classes within the forest with satisfactory accuracy.910 Based on the potential of interferometric SAR data, classification of different land cover classes in forested regions of Central India, using an interferometric pair of Environmental Satellite—Advanced Synthetic Aperture Radar (ENVISAT-ASAR) horizontal-horizontal (HH) polarization data is carried out in the present study. Though good amount of research is reported for the temperate forested regions,11 not much has been done on tropical regions to extract the potentials of SAR data in land cover classification. The study has its uniqueness and gains importance in the application potential of SAR interferometry over tropical regions like India, both in terms of an alternate/substitute to optical data sets due to persisting cloud cover and to the lack of availability of any earlier scientific work over the study region.

In the present study, ASAR data sets of 24 Sept. 2006 (SLC-1) and 29 Oct, 2006 (SLC-2) of IS3 beam position in HH polarization were acquired over parts of Mungeli and Pandaria villages of Bilaspur, Chattisgarh state, India. Satellite data for post-monsoon (September–October) was chosen to ensure a minimal effect of air temperature, wind speed, and rainfall on the SAR data backscatter.

The study area is characterized by major land cover classes, viz., agriculture, forests, water body, and barren area. Agriculture land use includes Kharif (September–November), Rabi (January – March), and Zaid (April–June) cropping patterns, with major crops such as paddy, wheat, cotton, and sugarcane apart from short-time rotation crops. The forests of the study area can be categorized into two major types, i.e., tropical moist deciduous and Sal gregarious forests.1213Shorea robusta and Tectona grandis are the two dominant tree species along with other notable species such as Madhuca indica, Diospyrus melanoxylon, Pterocarpus marsupium, Terminalia tomentosa, Anogessus latifolia, Emblica officinalis, Cleistanthus chollinus, and dendrocalamus strictus. The average annual rainfall is 1383 mm, with annual temperature ranging from a maximum of 25°C to 33°C and a minimum of 16°C to 17°C in the study area. The mean elevation in the study area is 660 m above mean sea level and varies from 500 to 800 m.

Single look complex (SLC) images of the acquired consecutive pairs was processed to generate backscatter coefficient images and then subjected to geo-coding using orbital parameters. The interferometric process of ENVISAT-ASAR data was carried out using the application software, “SARSCAPE,” from sarmap. The observed baseline in the interferometric data set is 203 m, which is well below the critical baseline. Co-registration of the acquired data set is done to use them in the same geometry, by taking SLC-1 as the master image and SLC-2 as the slave. An Interferometric synthetic aperture radar (InSAR) coherence image is obtained by using both amplitude and phase information from a complex SAR image pair. Coherence between the two images was calculated using the formula Display Formula

1γ=|s1(x)·2(x)*||s1(x)|2·|s2(x)|2,
where s1 and s2 are two complex co-registered images. The window size considered for the coherence image generation was 3×3. The obtained equivalent number of looks of the ASAR image for the study area was 1.47. Power images with range and azimuth in 7 multilooks were generated from the SLC images and calibrated to backscatter coefficient images. Radiometric calibration of the ASAR images was carried out following the radar equation principle, which involves corrections for the scattering area, antenna gain pattern, and the range spread loss.

A false color composite (FCC) [Fig. 1] is then generated using the derived “coherence image” (red), “mean backscatter” (green), “backscatter difference” (blue), and maximum likelihood supervised classification was carried out to delineate the land cover classes of the study area. The accuracy of the classification is assessed by computing Kappa statistics.

Graphic Jump LocationF1 :

(a) FCC of coherence, mean intensity, and intensity, difference. (b) Classified map in parts of Bilaspur, Chattisgarh.

Figure 1 shows an FCC of coherence, mean backscatter, and backscatter difference of the ASAR data and Fig. 1 shows the corresponding classified map of the study area. Visual analysis of the FCC showed clear discrimination of barren areas (in red color), because of high coherence values. Agriculture areas, which were characterized by crop-growth patterns during the study period, were observed to be in blue color due to differences in backscatter, while other areas under agricultural land use were observed to be in cyan color. Water appeared as black in the FCC due to its low coherence and low mean backscatter values. Forested regions of the study area were observed to be interspersed with agriculture land use [Figs. 1]. The discrimination of forest and agriculture classes is attributed to the large coherence values of agricultural areas compared to forests and marked backscatter difference between the two land cover types.

The coherence degree variation is related to the backscattering characteristics of the imaged targets and is an indicator for the phase stability between the two InSAR acquisitions. In the present study, forested areas have low coherence as canopy changes between acquisitions caused the decrease of the coherence degree. So, random changes of the position and the physical properties of the scatterers within a given resolution cell can be detected as low coherence. Thus, the coherence image, in combination with backscatter images, was used to separate different regions with different properties, which are difficult to be separated using SAR intensity alone.14

Coherence is a function of systemic spatial decorrelation, additive noise, and the scene decorrelation that takes place between the two acquisitions. It is observed from the coherence image that the average value of coherence in the study area was very small (<0.5) as the majority area is covered by vegetation. This can be attributed to the wind patterns in the study area that might alter the orientation of scattering objects (leaves, secondary branches) in the vegetation layer. The largest coherence values were observed in barren areas followed by agricultural areas, which are in accordance with the results reported in earlier studies,7 which suggested urban areas, agricultural areas, bushes, and forests have different correlation characteristics, with urban areas showing the largest correlation and forest the smallest. Medium coherence and large backscatter difference values were observed for the agricultural areas. As the temporal gap between the two images acquired is 35 days, agriculture areas showed some difference during the two acquisitions. It is observed that both the coherence and backscatter difference values of the forested areas were small compared to other land cover types. This is attributed to the temporal decorrelation effect that occurs on longer time scales such as the growth of the vegetation, human made changes, fires, etc.15

Figure 2 shows a scatter plot of interferometric coherence with SAR backscatter for the four land cover classes. It can be observed from Fig. 2 that land cover classes which had poor separability on the SAR backscatter image were clearly separable on the coherence image, which are in accordance to the observations of similar studies elsewhere.16

Graphic Jump LocationF2 :

Scatter plot showing the variation of interferometric coherence with SAR backscatter for four land cover classes.

Classification into different classes, i.e., water, barren, agriculture, moist deciduous forest, and Sal mixed forest was done using the bands of mean backscatter, backscatter difference, and coherence of the ASAR datasets [Fig. 1]. Confusion matrix for different land cover classes was derived and is given in Table 1. The overall classification accuracy and kappa statistics were 82.5% and 0.76, respectively. Statistical analysis based on kappa statistics was carried out to test the precision and significance of the results. Kappa statistics of given land cover classes are: Agriculture, ∼1; barren, ∼0.60; moist deciduous forest. 0.87; Sal gregarious forest, ∼0.61. The results are in accordance with earlier studies which showed that the mean backscatter, backscatter difference, and coherence band increased the classification accuracy.17

Table Grahic Jump Location
Confusion matrix (in percentage) of different land cover classes in parts of Bilaspur study area.

Though there is a marked difference between forest and non-forest, there was difficulty in differentiating mixed moist deciduous and Sal gregarious forests of the study area. In terms of backscatter and coherence, the effect of small changes in species composition within the forest is expected to be negligible. The classification accuracy was observed to be large within the forest in plain areas; however, there was some misclassification in hilly regions. This may be due to slopes, which is a major cause for misclassification.18 Forests in the sloping terrain areas were misclassified as the barren areas in sloping regions facing the sensors. The use of coherence for land cover classification is reasonable since low coherence is observed in forested areas, in comparison with bare soil and agriculture, which have high coherence. Interferometric coherence carries more land cover related information than the backscattered intensity.19

In the present study, an attempt was made to delineate four land cover classes, i.e., agriculture, moist deciduous forest, Sal gregarious forest, and barren using SAR interferometric coherence data over tropical forested regions of central India. It was inferred from the study that coherence from ASAR data can be used to differentiate major land cover classes with a marked difference between forest and non-forest types. More ground knowledge on sample points of forest types aided with value additions to the information given by ENVISAT-ASAR data can be very useful in discriminating different land cover classes in the Indian region.

The authors are thankful to Director and Deputy Director (RS&GIS–AA) and Group Director (LRG), and Head, Forestry and Ecology Division, NRSA for providing facilities. This work has been carried out as part of the RISAT-JEP project sponsored by ISRO.

Franklin  S. E., and Wulder  M. A, “ Remote sensing methods in medium spatial satellite data land cover classification of large areas. ,” Progr. Phys. Geogr.. 26, , 173–205  ((2002)).
Grover  K., , Quegan  S., , and Freitas  C. C., “ Quantitative estimation of tropical forest cover by SAR. ,” IEEE Trans. Geosci. Remote. Sens.. 37, , 479–490  ((1999)).
Saatchi  S., and Rignot  E., “ Classification of boreal forest cover types using SAR images. ,” Remote Sens. Environ.. 60, , 270–281  ((1997)).
Askne  J., “ C-band repeat-pass interferometric SAR observations of forest. ,” IEEE Trans. Geosci. Remote Sens.. 35, , 25–35  ((1997)).
Floury  N., , LeToan  T., , and Souyris  J. C., “ Relating forest parameters to interferometric SAR. ,” IEEE Int. Geosci. Rem. Sens. Symp.. 2, , 975–977  ((1996)).
Wegmuller  U., , Strozzi  T., , Farr  T., , and Werner  C. L., “ Arid land surface characterization with repeat-pass SAR interferometry. ,” IEEE Trans. Geosci. Remote Sens.. 38, , 776–781  ((2000)).
Wegmuller  U., and Werner  C. L., “ Retrieval of vegetation parameters with SAR interferometry. ,” IEEE Trans. Geosci. Remote Sens.. 35, , 18–24  ((1997)).
Srivastava  H. S., , Patel  P., , Sharma  Y., , and Navalgund  R. R., “ Detection and density mapping of forested areas using SAR interferometry technique. ,” Int. J. Geoinformatics, Forestry Special issue. , 3, (2 ), 1–10  ((2006)).
Martinez  J. M., , Beaudoin  A., , Wegmuller  U., , and Strozzi  T., “ Classification of land-cover and forest types using multi-date ERS tandem data acquired over hilly terrain. ,” IEEE Int. Symp. Geosci. Rem. Sens. , 1809–1811  ((1998)).
Askne  J., , Santoro  M., , Smith  G., , and Fransson  J. E. S., “ Multitemporal repeat-pass SAR interferometry of boreal forests. ,” IEEE Trans. Geosci. Remote Sens.. 41, , 1540–1550  ((2003)).
Williams  C. L., , McDonald  K. C., , and Chapman  B., “ Global boreal forest mapping with JERS-1: North America. ,” IEEE Int. Geosci. Rem. Sens. Symp.. 2, , 785–787  ((1999)).
Champion  H. G., and Seth  S. K.,  A Revised Survey of Forest Types of India. ,  Government of India ,  Delhi  ((1968)).
Joshi  P. K., , Roy  P. S., , Singh  S., , Agarwal  S., , and Yadav  D., “ Vegetation cover mapping in India using multi-temporal IRS WiFS data. ,” Remote Sens. Environ.. 103, , 190–202  ((2006)).
Abdelfattah  R., and Nicolas  J. M., “ Mixture model for the segmentation of the InSAR coherence map. ,” Int. J. Appl. Earth Observ. Geoinfo.. 12S, , 138–144  ((2010)).
Strozzi  T., , Dammert  P. B. G., , Wegmuller  U., , Martinez  J. M., , Askne  J. I. H., , Beaudoin  A., , and Hillakainen  M. T., “ Land use mapping with ERS SAR interferometry. ,” IEEE Trans. Geosci. Remote Sens.. 38, , 766–775  ((2000)).
Srivastava  H. S., , Patel  P., , and Navalgund  R. R., “ Application potentials of synthetic aperture radar interferometry for land cover mapping and crop height estimation. ,” Curr. Sci.. 91, (6 ), 783–788  ((2006)).
Araujo  L. S., , Santos  J. R., , Freitas  C. C., , and Xaud  H. A. M., “ The use of microwave and optical data for estimating aerial biomass of the savanna and forest formations at Roraima State, Brazil. ,” IEEE Int. Geosci. Rem. Sens. Symp.. 5, , 2762–2764  ((1999)).
Arne  E., “ SAR interferometry with ERS-1 in forested areas. ,” IEEE Int. Symp. Geosci. Remote Sens.. 1, , 202–204  ((1995)).
Engdahl  M. E., and Hyyppa  J. M., “ Land-cover classification using multitemporal ERS-1/2 InSAR data. ,” IEEE Trans. Geosci. Remote Sens.. 41, , 1620–1628  ((2003)).

Grahic Jump LocationImage not available.

Vyjayanthi Nizalapur did her master's degree in Physics from Osmania University, Hyderabad, India during 2002–04. She worked as a Senior Research Fellow during 2005–2010 in Forestry & Ecology Division, National Remote Sensing Center (NRSC), Indian Space Research Organization, India, during which she carried out her PhD thesis through department of Spatial Information Technology from Jawaharlal Nehru Technological University, Hyderabad, India. Her research interests are microwave remote sensing applications in forestry and agriculture. She is currently working as Research Scientist in Regional Remote Sensing Centre, NRSC, Jodhpur in the area of hydrological modeling.

Grahic Jump LocationImage not available.

Rangaswamy Madugundu has PhD degree in Environmental Science and Technology and working as a Project Scientist (RS&GIS) at Monitoring Evaluation Learning and Documentation Unit, Antrix Corporation Limited (ISRO), Bangalore, India. Presently, he is working on Prioritization of Watersheds, Impact Evaluation of Watershed Development. He has an experience of working on different available satellite and airborne optical as well as microwave sensors. His area of interest and expertise include Watershed Development, Environmental Impact Assessment, Land Use Land Cover Mapping, Plant Biodiversity, Biomass, NPP and Carbon Sequestration studies.

Grahic Jump LocationImage not available.

Chandra Shekhar Jha has PhD degree in remote sensing applications in forest ecology and is a senior scientist with National Remote Sensing Centre (ISRO), Hyderabad , India. His area is remote sensing and GIS applications for forest and environment. His area of interest and expertise includes mapping and monitoring of forest cover and type, plant biodiversity, biomass, timber volume, and carbon estimation. He has an experience of working on different available satellite and airborne optical and microwave sensors.

© 2011 Society of Photo-Optical Instrumentation Engineers (SPIE)

Citation

Vyjayanthi Nizalapur ; Rangaswamy Madugundu and Chandra Shekhar Jha
"Coherence-based land cover classification in forested areas of Chattisgarh, Central India, using environmental satellite—advanced synthetic aperture radar data", J. Appl. Remote Sens. 5(1), 059501 (March 14, 2011). ; http://dx.doi.org/10.1117/1.3557816


Figures

Graphic Jump LocationF1 :

(a) FCC of coherence, mean intensity, and intensity, difference. (b) Classified map in parts of Bilaspur, Chattisgarh.

Graphic Jump LocationF2 :

Scatter plot showing the variation of interferometric coherence with SAR backscatter for four land cover classes.

Tables

Table Grahic Jump Location
Confusion matrix (in percentage) of different land cover classes in parts of Bilaspur study area.

References

Franklin  S. E., and Wulder  M. A, “ Remote sensing methods in medium spatial satellite data land cover classification of large areas. ,” Progr. Phys. Geogr.. 26, , 173–205  ((2002)).
Grover  K., , Quegan  S., , and Freitas  C. C., “ Quantitative estimation of tropical forest cover by SAR. ,” IEEE Trans. Geosci. Remote. Sens.. 37, , 479–490  ((1999)).
Saatchi  S., and Rignot  E., “ Classification of boreal forest cover types using SAR images. ,” Remote Sens. Environ.. 60, , 270–281  ((1997)).
Askne  J., “ C-band repeat-pass interferometric SAR observations of forest. ,” IEEE Trans. Geosci. Remote Sens.. 35, , 25–35  ((1997)).
Floury  N., , LeToan  T., , and Souyris  J. C., “ Relating forest parameters to interferometric SAR. ,” IEEE Int. Geosci. Rem. Sens. Symp.. 2, , 975–977  ((1996)).
Wegmuller  U., , Strozzi  T., , Farr  T., , and Werner  C. L., “ Arid land surface characterization with repeat-pass SAR interferometry. ,” IEEE Trans. Geosci. Remote Sens.. 38, , 776–781  ((2000)).
Wegmuller  U., and Werner  C. L., “ Retrieval of vegetation parameters with SAR interferometry. ,” IEEE Trans. Geosci. Remote Sens.. 35, , 18–24  ((1997)).
Srivastava  H. S., , Patel  P., , Sharma  Y., , and Navalgund  R. R., “ Detection and density mapping of forested areas using SAR interferometry technique. ,” Int. J. Geoinformatics, Forestry Special issue. , 3, (2 ), 1–10  ((2006)).
Martinez  J. M., , Beaudoin  A., , Wegmuller  U., , and Strozzi  T., “ Classification of land-cover and forest types using multi-date ERS tandem data acquired over hilly terrain. ,” IEEE Int. Symp. Geosci. Rem. Sens. , 1809–1811  ((1998)).
Askne  J., , Santoro  M., , Smith  G., , and Fransson  J. E. S., “ Multitemporal repeat-pass SAR interferometry of boreal forests. ,” IEEE Trans. Geosci. Remote Sens.. 41, , 1540–1550  ((2003)).
Williams  C. L., , McDonald  K. C., , and Chapman  B., “ Global boreal forest mapping with JERS-1: North America. ,” IEEE Int. Geosci. Rem. Sens. Symp.. 2, , 785–787  ((1999)).
Champion  H. G., and Seth  S. K.,  A Revised Survey of Forest Types of India. ,  Government of India ,  Delhi  ((1968)).
Joshi  P. K., , Roy  P. S., , Singh  S., , Agarwal  S., , and Yadav  D., “ Vegetation cover mapping in India using multi-temporal IRS WiFS data. ,” Remote Sens. Environ.. 103, , 190–202  ((2006)).
Abdelfattah  R., and Nicolas  J. M., “ Mixture model for the segmentation of the InSAR coherence map. ,” Int. J. Appl. Earth Observ. Geoinfo.. 12S, , 138–144  ((2010)).
Strozzi  T., , Dammert  P. B. G., , Wegmuller  U., , Martinez  J. M., , Askne  J. I. H., , Beaudoin  A., , and Hillakainen  M. T., “ Land use mapping with ERS SAR interferometry. ,” IEEE Trans. Geosci. Remote Sens.. 38, , 766–775  ((2000)).
Srivastava  H. S., , Patel  P., , and Navalgund  R. R., “ Application potentials of synthetic aperture radar interferometry for land cover mapping and crop height estimation. ,” Curr. Sci.. 91, (6 ), 783–788  ((2006)).
Araujo  L. S., , Santos  J. R., , Freitas  C. C., , and Xaud  H. A. M., “ The use of microwave and optical data for estimating aerial biomass of the savanna and forest formations at Roraima State, Brazil. ,” IEEE Int. Geosci. Rem. Sens. Symp.. 5, , 2762–2764  ((1999)).
Arne  E., “ SAR interferometry with ERS-1 in forested areas. ,” IEEE Int. Symp. Geosci. Remote Sens.. 1, , 202–204  ((1995)).
Engdahl  M. E., and Hyyppa  J. M., “ Land-cover classification using multitemporal ERS-1/2 InSAR data. ,” IEEE Trans. Geosci. Remote Sens.. 41, , 1620–1628  ((2003)).

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging & repositioning the boxes below.

Related Book Chapters

Topic Collections

Advertisement
  • Don't have an account?
  • Subscribe to the SPIE Digital Library
  • Create a FREE account to sign up for Digital Library content alerts and gain access to institutional subscriptions remotely.
Access This Article
Sign in or Create a personal account to Buy this article ($20 for members, $25 for non-members).
Access This Proceeding
Sign in or Create a personal account to Buy this article ($15 for members, $18 for non-members).
Access This Chapter

Access to SPIE eBooks is limited to subscribing institutions and is not available as part of a personal subscription. Print or electronic versions of individual SPIE books may be purchased via SPIE.org.