IDENTIFIKASI MATRIK FENOLOGI DAN FREKUENSI PENANAMAN PADI TAHUN 2021 MENGGUNAKAN FUSI SPASIO-TEMPORAL LANDSAT-8 DAN MODIS DI LAHAN SAWAH DILINDUNGI KABUPATEN SRAGEN, PROVINSI JAWA TENGAH
PUTRI LAILA KARTIKA NINGRUM, Dr. Sc. Sanjiwana Arjasakusuma, S.Si., M.GIS.
2024 | Skripsi | KARTOGRAFI DAN PENGINDRAAN JAUH
Tracking the frequency of planting information through the repetition of phenology matrix parameters is crucial as an approach to intensify food production, especially in protected rice field areas with rice commodities. Multi-temporal analysis using remote sensing imagery is needed to record over a wide area and provide sufficient temporal data. The identification of planting frequency and multi-temporal phenology matrix requires an accurate and detailed identification process so that it can show the growth phase of plants using phenology matrix parameters. For this reason, remote sensing images with sufficient spatial and temporal resolution are needed. The unavailability of images with such quality for free access leads to the need for data fusion between free access images with medium spatial resolution but low temporal resolution with high temporal resolution but low spatial resolution. A comparison between the capabilities of fused image results with medium spatial resolution images as input is then needed to determine the ability of images in identifying planting frequency and phenology matrix related to the addition of data temporally through its accuracy aspects tested using RMSE and confusion matrix, and its spatial distribution.
This study compares the use of the Landsat-8 OLI/TIRS dataset and the fusion results with daily MODIS using the STARFM algorithm to obtain satellite images with higher spatial and temporal resolution. Phenology information, including SOS and EOS, is extracted using the EVI index, while planting frequency is extracted through the repetition of start of season (SOS) and end of season (EOS) using the LSWI index and classification using machine learning Decision Tree (DT) and Random Forest (RF). The results of phenology identification show that the fusion dataset provides higher accuracy with an error of about one month, but the opposite occurs in the identification of phenology-based planting frequency where the Landsat dataset provides higher accuracy, i.e., 65% compared to fusion at 42.5%. Furthermore, machine learning provides more accurate planting frequency identification results compared to phenology-based, where Landsat DT provides 95?curacy, fusion DT 85%, and Landsat & fusion RF 100%. This accuracy affects the spatial distribution of SOS, EOS, and planting frequency.
Kata Kunci : Lahan Sawah Dilindungi, Fusi Data, STARFM, Fenologi, Decision Tree, Random Forest/Protected Rice Fields, Data Fusion, STARFM, Phenology, Decision Tree, Random Forest.