Please use this identifier to cite or link to this item:
|Title:||Estimation of forest biomass using remote sensing|
|Authors:||Sarker, Md. Latifur Rahman|
|Keywords:||Hong Kong Polytechnic University -- Dissertations|
Forest biomass -- Measurement.
Forest biomass -- Environmental aspects.
Remote sensing -- Environmental aspects.
|Publisher:||The Hong Kong Polytechnic University|
|Abstract:||Forest biomass estimation is essential for greenhouse gas inventories, terrestrial carbon accounting and climate change modelling studies. Although a lot of effort has been made in estimating biomass using both field-based and remote sensing techniques, no universal and transferable technique has been developed so far to quantify biomass carbon sources and sinks due to the complexity of the environmental, topographic and biophysical characteristics of forest ecosystems. So far traditional field-based methods are considered to be the most accurate, but the procedures are time consuming, expensive, destructive in nature and most importantly, their implementation is only possible over small and accessible areas. Over the last few decades, remote sensing techniques have been explored for forest biomass estimation and considerable improvement has been achieved regarding time, costs and scale (local, regional and global), but the accuracy and transferability of the techniques developed remain low. Remote sensing-based biomass estimations have mainly been carried out using airborne SAR, space borne single polarization SAR (JERS-1, ERS-1/2 & RADARSAT-1) and optical sensors with low spatial resolution, including Landsat TM, MODIS, and SPOT-4. However, the potential of remote sensing for biomass estimation could not be realized due to the limitations of the sensors. For SAR sensors, these included i) the unavailability and high cost of airborne SAR data, ii) the lack of dual polarization SAR data in the previous space borne sensors JERS-1, ERS-1/2, and RADARSAT-1 and iii) the low resolution of space borne SAR (JERS-1, ERS-1/2). For optical sensors, the main limitation was low spatial resolution. The availability of new SAR (C-band RADARSAT-2 and L-band PALSAR) and optical sensors (SPOT-5 and AVNIR-2) has opened new possibilities for biomass estimation. The new SAR sensors (RADARSAT-2 and PALSAR) can provide data with varying polarizations, incidence angles and fine spatial resolutions. On the other hand, the new optical sensors (SPOT-5 and AVNIR-2) have the capability of providing data with significantly higher spatial, spectral and temporal resolutions. Thus, the potential for biomass estimation from remote sensing has greatly improved, although some sensor limitations and unknown criteria still need to be addressed. These include i) the speckle noise of SAR sensors, ii) the canopy penetration depth of C-band and L-band SAR, iii) the resolution of RADARSAT-2 and PALSAR sensors, iv) the saturation level of C-band and L-band SAR, v) the extent of canopy penetration of optical and SAR sensors, vi) the saturation problem of optical data, vii) the incidence angles of different sensors, and viii) the spectral bands of AVNIR-2 and SPOT-5 sensors. Although the differences between sensor types may make one sensor more useful than the others, their individual usage may not give the best results compared to the potential synergistic usage of multiple sensors.|
Therefore, this study investigated the potential of two dual polarization SAR sensors (RADARSAT-2 with C-band and PALSAR with L-band) and two optical sensors (SPOT-5 and AVNIR-2) for the estimation of biomass in a challenging study area with mountainous subtropical terrain where biomass levels are far beyond the previously recognized saturation levels for all sensors, and where the forest is a mixture of native and non-native species and plantations. The biomass estimation used five types of sensor combinations i.e. i) dual polarization C-band SAR data (chapter 6), ii) two-date dual polarization PALSAR data (chapter 7), iii) AVNIR-2 data (chapter 8), iv) the data from two optical sensors AVNIR-2 and SPOT-5 (chapter 9), and v) the fusion of SAR (RADARSAT-2) and optical (AVNIR-2) data using wavelet transform (chapter 10). Although slightly different methodologies were applied for the processing of data from optical and SAR sensors, three common major processing steps were used, namely i) spectral reflectance/intensity, ii) texture measurements and iii) polarization or band ratios of texture parameters. Simple linear and stepwise multiple regression models were developed in order to establish a relationship between the image parameters extracted from the different stages of image processing and the biomass of field plots, which was estimated using a newly developed allometric model for the study region. The results clearly demonstrate the ineffectiveness of raw data (whether L-band or C-band SAR, or optical (SPOT-5 or AVNIR-2), or raw data in multisensor combinations) which was probably due to the high biomass level in this study area. Significant improvements in performance (r2) (RADARSAT-2=0.78; PALSAR=0.679; AVNIR-2=0.786; SPOT-5=0.854; AVNIR-2 + SPOT-5=0.911) were achieved using texture parameters, although the performance was variable between sensors and sensor combinations. The accuracies of biomass estimation were further improved and very promising accuracies (r2) were obtained using the ratio of texture parameters (RADARSAT-2=0.91; PALSAR=0.823; PALSAR two-date=0.921; AVNIR-2=0.899; SPOT-5=0.916; AVNIR-2 + SPOT-5=0.939). These accuracies suggest four main contributions arising from this research, namely i) biomass estimation can be significantly improved by using texture parameters, ii) further improvements can be obtained using the ratio of texture parameters, iii) multisensor texture parameters and their ratios have more potential than texture from a single sensor, and iv) biomass can be accurately estimated far beyond the previously perceived saturation levels of SAR and optical data using texture parameters or the ratios of texture parameters. A further important contribution resulting from the fusion of SAR & optical images produced accuracies (r2) of 0.706 and 0.77 from the simple fusion, and the texture processing of the fused image, respectively. Although these accuracies were not as attractive as the accuracies obtained from the other four processing steps, the wavelet fusion procedure improved the saturation level of the optical (AVNIR-2) image very significantly since, the performance (r2) of AVNIR-2 (NIR) alone (r2=0.49) was increased to 0.77 after fusion with SAR in a high biomass situation using only one input image (the fused image) in the model.
|Description:||xvi, 17-221 leaves : ill. (chiefly col.) ; 30 cm.|
PolyU Library Call No.: [THS] LG51 .H577P LSGI 2010 Sarker
|Rights:||All rights reserved.|
|Appears in Collections:||Thesis|
Show full item record
Files in This Item:
|b24250211_link.htm||For PolyU Users||162 B||HTML||View/Open|
|b24250211_ir.pdf||For All Users (Non-printable)||7.81 MB||Adobe PDF||View/Open|
Checked on Feb 26, 2017
Checked on Feb 26, 2017
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.