Precise digital classification for Landsat 8 data of remote sensing images require pre-processing steps. The preprocessing consist of conversion from digital numbers (DN) to top of atmosphere (TOA) reflectance, cloud and cloud shadow masking, topographic correction and image normalization. In general, pre-processing steps were implemented to National scale (Indonesia) excluding topographic correction. The topographic correction algorithm is required to avoid reflectance bias from terrain effects due to shading. The highest mountains in Indonesia were selected as window areas, considering the reflectance bias is produced due to terrain effects. The results showed that algorithm is able to solve overcorrection problems and will be implemented into LAPAN’s system of image pre-processing for National scale. This research is a collaboration between Bogor Agricultural University (IPB) with National Institute of Aeronautics and Space (LAPAN) under Forests2020 Programme, in order to produce Landsat 8 data with the minimal cloud over Indonesia annually and then to automatically digital classification for forest monitoring. The automated system of preprocessing was developed with Perl and Python programming languages.
Agroforestry is a land use management-system represents unique vegetation characteristics among tree vegetation types. Tree height is a vegetation variable used to characterize vertical structure, including mixed vegetation structure in agroforestry. Estimation of tree heights with multispectral imagery is a relatively new application and is dependent on integrating synoptic coverage optical data with samples of height data, often from LiDAR-derived reference data. In this study, multispectral Landsat 8 data, Unmanned Aerial Vehicle (UAV)-based LiDAR height data and a log-linear regression model were used to estimate tree height for agroforestry land use in western part of Java Island, Indonesia. We generated a Canopy Height Model (CHM) directly from height-normalized LiDAR points and used as reference data in modeling the key height variable in the multispectral bands of Landsat 8. The analysis showed that red band was the best band to estimate tree height in agroforestry land use, followed by swir band. The log-linear regression algorithm of red band accurately reproduced the LiDAR-derived height training data using Landsat 8 data with overestimate 1.46 m in estimating tree height < 5 m and underestimate 7.79 m for tree height > 20 m.