Journal of Applied and Natural Science 12(1): 9 -12 (2020) ISSN : 0974-9411 (Print), 2231-5209 (Online) journals.ansfoundation.org A study on landuse and landcover classification using microwave data in Joida taluk of Uttara Kannada district, Karnataka Arjun. G. Koppad* Article Info Department of Natural Resource Management, College of Forestry, Sirsi (Karnataka), https://doi.org/ India 10.31018/jans.v12i1.2209 Syeda Sarfin Received: December 16, 2019 Department of Natural Resource Management, College of Forestry, Sirsi (Karnataka), Revised: January 24, 2020 India Accepted: February 2, 2020 Anup Kumar Das Space Application centre, Ahmedabad (Gujarat), India How to Cite *Corresponding author. E-mail: [email protected] Koppad, A.G. et al. (2020). A Study on Landuse and Abstract Landcover Classification The study has been conducted for land use and land cover classification by using SAR using Microwave data in data. The study included examining of ALOS 2 PALSAR L- band quad pol (HH, HV, VH Joida Taluk of Uttara and VV) SAR data for LULC classification. The SAR data was pre-processed first which Kannada district, Karna- included multilook, radiometric calibration, geometric correction, speckle filtering, SAR taka. Journal of Applied Polarimetry and decomposition. For land use land cover classification of ALOS-2- and Natural Science, 12 (1): 9 - 12 https://doi.org/ PALSAR data sets, the supervised Random forest classifier was used. Training samples were selected with the help of ground truth data. The area was classified under 7 different 10.31018/jans.v12i1.2209 classes such as dense forest, moderate dense forest, scrub/sparse forest, plantation, agriculture, water body, and settlements. Among them the highest area was covered by dense forest (108647ha) followed by horticulture plantation (57822 ha) and scrub/Sparse forest (49238 ha) and lowest area was covered by moderate dense forest (11589 ha). Accuracy assessment was performed after classification. The overall accuracy of SAR data was 80.36% and Kappa Coefficient was 0.76. Based on SAR backscatter reflec- tance such as single, double, and volumetric scattering mechanism different land use classes were identified. Keywords: Decomposition, Forest, LULC, Microwave, Polarization, Polarimetry INTRODUCTION monitoring (Sim, et al., 2011). Polarimetry SAR is an advanced imaging radar system; it plays an Remote sensing is an art of acquiring data/ important role in radar remote sensing. The suc- information about an object, process or a place cessful application of SAR technology to address without being in direct contact with the sensor a wide range of remote sensing problems helped (Campbell et al. 2011; James B et at.2011). Re- advancement of SAR systems to include polarisa- mote sensing provides real time data. It has capa- tion diversity and operation in interferometric bilities to address various global applications in- mode (Pampaloni et al., 2004; Paolo et al., 2004). cluding agriculture, forestry, meteorology, geology The fundamental data requirements to assess and and environmental studies. The remote sensing quantify changes in land use and land cover are provides a platform to analyze large area within images from earth observing satellite. The images short time period (Roy et al.2017; Behera et al. are analyzed and processed to deliver useful 2017). products. Image classification of remote sensing Microwave remote sensing uses microwave region data is widely used to extract information and of electromagnetic spectrum. The application of identify land cover types (Abdikan et al. 2016; microwaves in remote sensing is motivated be- Saygin, et al., 2016). The Accurate and appropri- cause of its cloud penetrating and all weather im- ate information of LULC is extremely important for aging properties (Pampaloni et al. 2004; Paolo et analyzing various socio-ecological concerns al. 2004). It provides unique capability towards (Sharma, et al., 2018). In an urban environment monitoring earth. RADAR or radio detection and natural and human-induced environmental chang- ranging is an active microwave sensor. It is the es are of concern today because of deterioration main tool used for natural resource mapping and of environment and human health (Jat et al., 2008, This work is licensed under Attribution-Non Commercial 4.0 International (CC BY-NC 4.0). © 2018: Author (s). Publishing rights @ ANSF. Koppad, A.G. et al. / J. Appl. & Nat. Sci. 12(1): 9 - 12 (2020) Garg, et al., 2008). Traditional methods of acquir- image under freeman-Durden decomposition in ing data of large area are not always possible and which red indicated double bounce; green indicat- it is time consuming. Remote sensing has become ed volumetric and blue indicated the surface scat- very popular subject in understanding and gather- tering which was used for classification. ing information about earth and processes. Keep- The supervised classification is performed by us- ing these points in view the experiment was ing random forest classifier. The first step towards planned to map the Land use Land cover classes classification is generation of vector data contain- using SAR data. er of different classes. The different classes were MATERIALS AND METHODS identified based on their scattering, colour and texture. supervised classification is a training The study area falls between 140 54' 13" N to 150 based methodology that classifies similar image 31'59" N latitude and 740 08' 11" E to 740 31' 10" E pixel values into training samples for a determined longitude. This is part of Joida Taluk in Uttar Kan- number of classes (Zakeri et al., 2017; Homa et nada (UK) district, Karnataka, India. It is tropical al., 2017). The supervised classification is con- evergreen forest which is covered mainly with ducted by using random forest classifier algorithm. dense forest. It has diverse land cover types such Random forest is a classification and regression as plantation, agriculture, forest, settlements and tree technique invented by Breiman. It randomly water body. The location map is shown in Fig1. collects the samples of data and variables to gen- The study was carried in Joida Taluk of Uattar erate a large group or forest or classification and Kannada district, Karnataka. The data used for regression trees. The area is classified into seven the study was ALOS 2 PALSAR, L band SAR da- different classes such dense forest, moderate ta. It had range resolution of 5.1m and azimuth dense forest, scrub/sparse forest, plantation, agri- resolution of 4.3 m, quad pol data with HH, HV, culture, water body, and settlements. The accura- VH and VV polarization. The ALOS 2 PALSAR cy assessment after the classification is shown in data is procured from NRSC Hyderabad. confusion matrix. There are many steps involved The SAR data contained I, q and intensity bands. in classification of ALOS 2 PALSAR. The general- These bands were used to create RGB image. ized flow chart of the methodology is given in Fig. The first step towards pre-processing of SAR data 2. involves multilooking to generate square pixel (Dini Das, 2018). The data was radiometrically RESULTS calibrated. The objective of calibration was to pro- The Freeman Durden decomposition allowes to vide an imagery in which the pixel values can be analyse and identify the different feature based on directly related to the radar backscatter. Calibrat- there scattering mechanism. The accuracy results ed SAR images are essential for quantitative use of Freeman Durden decomposition classification of SAR data. The quad pol data was used to gen- was 80.36% and kappa coefficent was 0.76. The erate Coherency matrix T3. This matrix was pre- confusion matrix of the results were produced to ferred because its elements had a physical inter- show the classified and referenced pixles. The pretation (Volume scattering, double bounce scat- total area is classified under 7 different classes. tering, and surface scattering) (Dos Santos, et al. The area classification is shown in Table 1. 2009). Further speckle filter was applied to gener- The present study observed that the total classi- ated matrix by using refined lee (5×5). This pro- fied area of dense forest was108647 ha, moderate cess removes speckle noises. The image may dense forest was 11589 ha, scrub/sparse forest contain topographical variations of scene and tilt was 49238 ha, horticulture plantation was 57822 of satellite sensor. This error can be removed by ha agriculture was 16061 ha, settlements was geometric correction. The data was geometrically 31251 ha and water body was 12393 ha. The corrected which included orientation angle correc- blue patch in supervised classified image was tion and terrain correction. Terrain corrections was water body (Supa reservoir). The river chanels done by using range doppler terrain correction were also identified correctly. The water acted as which is intended to remove the distortions so that plane surface which undergoes surface reflec- the geometric representation of the image will be tance. The red patches in the image indicated as close as possible to real world. The SAR Polar- settelemnts and small villages. The settelments imetry was carried out. Snap software was used mainly underwent double bounce scattering. The to pre-process SAR data. combination of dark green and sea greean colour The decomposition model used for this study was in the image indicated dense forest and moderate Freeman Durden decomposition. This decomposi- dense forest respectively. The plantation is repre- tion models the covariance matrix as the contribu- sented by green colour. The yellow colure in the tion of three scattering mechanism. They are vol- image indicated agriculture. SAR polarimetry is ume scattering, double bounce scattering and always said useful for forest mapping because of surface scattering. The components of these three its penetreting properties into dense canopy of mechanisms were employed to generate a RGB trees. The freeman durden decomposition, classi- 10 Koppad, A.G. et al. / J. Appl. & Nat. Sci. 12(1): 9 - 12 (2020) fied image is shown in Fig.