Koreksi Awan Cirrus Citra Landsat 8 Menggunakan Cirrus Band

RATNA PRASTYANI, Abdul Basith, ST., M.Si., Ph.D.

2018 | Skripsi | S1 TEKNIK GEODESI

Teknologi penginderaan jauh tidak terlepas dari pengaruh interaksi antara energi elektromagnetik dengan atmosfer sebagai medium perambatan energi dan dinamika awan. Awan muncul dalam bentuk cumulus, stratus dan cirrus. Awan cirrus merupakan tipe awan dengan struktur yang paling tipis dan sulit dideteksi keberadaanya pada citra satelit. Indonesia sebagai negara tropis diliputi oleh awan cirrus hampir sepanjang tahun. Hal ini menyebabkan citra satelit optik di Indonesia terkontaminasi awan cirrus. Peluncuran satelit Landsat 8 tahun 2013 silam dilengkapi dengan cirrus band yang mampu mendeteksi awan cirrus secara efektif dan memberikan informasi kepada pengguna terkait keberadaan dan intensitas awan pada citra satelit yang diwujudkan dalam nilai digital number. Ketersediaan cirrus band tersebut dapat digunakan dalam mengestimasi dan mengoreksi kontaminasi awan cirrus. Namun penggunaan cirrus band belum begitu populer dalam tahap pre-processing citra satelit Landsat 8. Penelitian ini menjelaskan studi terkait metode koreksi awan cirrus menggunakan cirrus band berdasarkan pendekatan yang dilakukan oleh Xu dkk. (2014) dan dengan teknik sampling yang melibatkan sampel piksel dengan kontaminasi awan cirrus rendah, sedang dan tinggi. Koreksi awan cirrus dilakukan pada tiga area studi di Indonesia yang berada di Situbondo, Sumba dan Padang. Lokasi tersebut dipilih berdasarkan ketersediaan data Level 1 Precision Terrain (L1TP) Landsat 8 yang diakuisisi pada kala bebas dan terkontaminasi awan cirrus. Data L1TP Landsat 8 yang digunakan meliputi citra di masing-masing area studi yang diakuisisi pada kala bebas dan terkontaminasi awan cirrus dengan selisih waktu akuisisi selama 16 s.d.32 hari. Estimasi efek awan cirrus dihitung menggunakan metode regresi linier terhadap sampel piksel pada area homogen. Area homogen yang dipilih berada pada tubuh air karena kontaminasi awan cirrus yang umumnya meliputi tubuh air (laut) di ketiga area studi. Koreksi awan cirrus kemudian dilakukan menggunakan operasi aritmatik citra (band math) berdasarkan koefisien regresi slope yang memiliki koefisien determinasi tertinggi. Kualitas hasil citra terkoreksi awan cirrus kemudian ditentukan secara statistik berdasarkan nilai koefisien determinasi terhadap citra referensi yang dianggap benar. Penelitian ini telah berhasil mengoreksi citra dari kontaminasi awan cirrus. Koreksi awan cirrus dilakukan dengan estimasi efek awan cirrus menggunakan regresi linier berdasarkan sampel piksel di area homogen yang terkontaminasi awan cirrus dengan berbagai intensitas dan koreksi awan menggunakan operasi aritmatik saluran (band math). Dari hasil sampel yang diambil untuk estimasi efek awan cirrus, kualitas hitungan regresi yang paling baik cenderung pada sampel dengan kontaminasi awan cirrus rendah. Citra terkoreksi awan cirrus di ketiga area studi menghasilkan peningkatan koefisien determinasi. Area studi Sumba mengalami peningkatan koefisien determinasi sebesar 2,17%, 10,36% dan 7,44% serta area Situbondo sebesar 5,29%, 3,98% dan 2,24%. Sementara untuk area Padang, koefisien determinasi mengalami peningkatan pada band 3 dan 4 sebesar 4,12% dan 3,44% serta mengalami penurunan pada band 2 sebesar 0, 46%.

Atmospheric conditions as well as clouds are major factors that affect recorded data in remote sensing. Clouds appear in the form of cumulus, stratus, or cirrus. Cirrus are typically thin making it difficult to be detected on optical images. Indonesia as tropical country has cirrus coverage all year around. This causes cirrus contamination on optical images over the country. This problem has been addressed in Landsat 8 that was launched on 2013 by adding a new channel that can effectively detect cirrus cloud and is called cirrus band. Cirrus band is able to tell the users about cirrus contamination over the scene as well as its intensity which is represented in digital number. However, this channel has not yet been routinely used in image pre-processing to correct cirrus contamination in Landsat 8 images. This research describes a study on the utilisation of cirrus band in correction of cirrus clouds on Landsat 8 images based on the approach that has been developed by Xu et al. (2014) and with a modification on the sampling technique. The effectiveness of the proposed method in this research is examined in Landsat 8 images over tropical region especially Indonesia with variation of cirrus contamination. Correction of cirrus clouds is performed on three scenes with study area of Situbondo, Sumba and Padang. The study areas are chosen due to the availability of cirrus cloud-free images of Level 1 Precision Terrain (L1TP) Landsat 8 that is used as reference image. Data used in this research consists of two main L1TP datasets for each study area which are cirrus cloud-free image and contaminated image with temporal difference between two images vary from 16 up to 32 days and assumed that there were no land cover changes within temporal difference. Linier regression based on homogenous area pixel samples is performed to estimate the effect of cirrus contamination. Homogenous contaminated area over water bodies (sea) is defined as pixel sample due to cirrus contamination that covers vast area of water bodies over the scene. Correction of cirrus contamination is performed using band math based on slope coefficient which corresponds to the highest coefficient of determination. The effectiveness of proposed method is then statistically examined based on coefficient of determination between cirrus cloud-free image as reference and corrected image. This research has successfully corrected cirrus contaminantion over three study area images. Correction is performed by estimating cirrus effect using linear regession model and correcting the contaminated digital numbers using band math. Cirrus effect is estimated using linear regression based on pixel samples over homogen area with variety of cirrus intensity. Linier regression over homogen area with low cirrus intensity tends to have the best quality based on its determination coefficient. Corrected images of three study area shows satisfactory result with the increase of coefficient of determination. Sumba shows that the coefficient of determination increases by 2,17%, 10,36% and 7,44%. Situbondo coefficient of determination also increases by 5,29%, 3,98% and 2,24%. Meanwhile, Padang corrected images shows the increase of coefficient of determinantion only in green and red band by 4,12% and 3,44% yet decreased in blue band by 0,46%.

Kata Kunci : awan cirrus, tropical region, cirrus band, Landsat 8, regresi linier