Forest Above-Ground Biomass Inversion Using Optical and SAR Images Based on a Multi-Step Feature Optimized Inversion Model

Potential of texture from SAR tomographic images for forest aboveground biomass estimation

Matter of discussion in this Ph. D. thesis is SAR (Synthetic Aperture Radar) image denoising. Main elements of innovation are the introduction of SAR-BM3D, a denoising algorithm optimized for SAR data, and the introduction of a benchmark which enables the objective performance comparison of SAR denoising algorithms on simulated canonical SAR images. In the first part of the thesis, basic concepts on SAR images are introduced, with special emphasis on its peculiar multiplicative noise, called speckle. A description of key ideas and tools of denoising techniques known in literature then follows. After introducing the basic elements of SAR data processing, the main statistical features of SAR images are described, and it is clarified in which context the denoising techniques operate. Techniques are then classified as those that follow the homomorphic approach, where the multiplicative noise is turned into additive noise through a logarithmic transform, and those that take explicitly into account the multiplicative nature of noise. Afterwards, it is described how the introduction of the wavelet transform has brought new ideas into SAR image denoising and how the non-local filtering strategy, originally proposed in the AWGN field, has provided relevant results also in the application to SAR. In this context, the novel SAR-BM3D algorithm is introduced which, starting from key elements of wavelet-based and non-local filtering implemented in BM3D, optimizes the elaboration for SAR data, following a non-homomorphic approach. A very detailed experimental analysis on simulated SAR images, obtained as optical images corrupted by artificial speckle, has been performed: results proved the SAR-BM3D algorithm to outperform traditional approaches, both in terms of PSNR and visual inspection. Due to the well-known difficulties of evaluating the performance of denoising techniques on real SAR images, a workaround has been proposed. Rather than resorting to images corrupted by artificial speckle, a physical SAR simulator, SARAS (developed by the remote-sensing group of the Federico II University of Naples) has been used to generate a set of canonical benchmark SAR scenes. The main advantage of SARAS images is the availability of both the noisy and clean versions of the images, the latter acting as a reference to objectively evaluate the performances of different algorithms. We have shown in detail the procedure which leads to a definition of an objective criterion to compare results provided by different algorithms when working on real SAR images. For this purpose, different test cases have been selected and specific measures, suitable for the various scenes have been proposed for the characterization. At the end of the thesis, open issues are pointed out and future research is outlined.

Recent synthetic aperture radar (SAR) sensors with a capability of providing data with varying spatial resolutions, polarizations, and incidence angles have attracted greater interest for forest biomass and carbon storage estimation. This study investigates the capability of RADARSAT-2 fine-beam dual-polarization (C-HV and C-HH) data for forest biomass estimation in complex subtropical forest, with different types of processing: 1) raw intensity data (both polarizations separately and as polarization ratio) and 2) texture parameters of both polarizations (separately, jointly, and as polarization ratio). Field data (diameter at breast height and height) were collected from 53 field plots and converted to biomass (dry weight) using a newly developed allometric model. Finally, biomass estimation models were developed between SAR signatures from different processing steps and field plot biomass using stepwise multiple regression. All biomass estimation models using radar intensity data (C-HV, C-HH, and ratio of C-HV and C-HH) proved ineffective, but texture parameters derived from intensity data showed potential. We were able to estimate forest biomass amounts up to 360 t/ha with a goodness of fit of 0.78 (adjusted r2) and an rmse of 28.68 t/ha using the combination of texture parameters of both polarizations (C-HV and C-HH). However, goodness of fit could be improved to 0.91 (adjusted r2) and an rmse of 26.95 t/ha for biomass levels up to 532 t/ha using the ratio of texture parameters of C-HV/C-HH. The result is very encouraging and indicates that the dual-polarization C-band SAR sensor has a potential for the estimation of forest biomass, particularly using the polarization ratio of texture measurements, and biomass estimation can be improved substantially beyond the previously stated saturation level for C-band SAR.

In this paper, we develop a novel method for forest biomass estimation. The intensity values of Advanced Land Observation Satellite-Advanced Visible and Near Infrared Radiometer type 2 and PRISM images and the texture features of the Japanese Earth Resources Satellite 1 image are used in a multilayer perceptron neural network (MLPNN) that relates them to the forest variable measurements on the ground. A proposed speckle noise model is also applied for modeling and reducing the speckle noise in the synthetic aperture radar (SAR) image. Reducing the speckle would improve the discrimination among different land use types and would make the textual classifiers more efficient in SAR images. Ideally, filters will reduce the speckle without loss of information. In the process of the forest biomass estimation, the filters should preserve the backscattering coefficient values and edges between different areas. We investigate both quantitative and qualitative criteria in speckle reduction and texture preservation to evaluate the performance of the proposed filter in the forest biomass estimation. We will also show that the biomass estimation accuracy is significantly improved in an MLPNN when the radar and the optical data are used in combination compared to estimating the biomass by using a single datum only. The root-mean-square error (rmse) value is decreased when the proposed method is used (rmse = 2.175 ton) compared with that of the classic method (rmse = 5.34 ton).

The forest aboveground biomass and its dynamics are essential for researches of carbon cycling. The estimation of forest aboveground biomass is an important topic in the application of remote sensing. However, the estimation accuracy is limited by the lack of forest structures. Recent progresses over the past decade in the estimation of forest biomass from remote sensing data was mainly due to the success in the extraction of related forest structure parameters[1]. For example, LiDAR data is widely used in the estimation of forest biomass at local scales because it can directly measure the vertical structure of forests [2, 3]. Polarimetric Interferometric SAR (PolInSAR) is a another promising technique for the estimation of forest height and biomass which employs the dependence of penetration depth of SAR on polarization [4]. However, spaceborne LiDAR system cannot acquire wall-to-wall data because it mainly worked by point sampling while airborne LiDAR system only worked on at local scaled due to its cost. The P-band PolInSAR system onboard European BIOMASS satellite which is suitable for retrieval of forest biomass is still under construction and scheduled to be launched around 2020 by European space agency (ESA)

Accurate estimates of aboveground biomass in tropical forests are important in carbon sequestration and global change studies. Tropical forest biomass estimation with microwave remote sensing is limited because of the strong scattering and attenuation properties of the green canopy. In this study a microwave/optical synergistic model was developed to quantify these effects to Synthetic Aperture Radar (SAR) signals and to better estimate woody structures, which are closely related to aboveground biomass. With a Leaf Area Index (LAI) retrieved from Japan Earth Resources Satellite (JERS)‐1 Very Near Infrared Radiometer (VNIR) imagery, leaf scattering and attenuation to woody scattering were quantified and removed from the total backscatter in a modified canopy scattering model. Woody scattering showed high sensitivity to biomass >100 tonnes/ha in tropical forests. Tree height and stand density were derived from the JERS‐1 SAR image with a root mean square error (RMSE) of 4 m and 161 trees/ha, respectively. Aboveground biomass was calculated using a general allometric equation. Biomass in secondary dry dipterocarps (Dipterocarpaceae family of tropical lowland deciduous trees) was overestimated. The modelled biomass in mixed deciduous and dry evergreen forests fit better with ground measurements. In mountainous areas with steep slopes, the topographic effects in the SAR image could not be properly corrected and therefore the results are unreliable.

Estimation of forest biomass with synthetic aperture radar (SAR) and interferometric SAR (InSAR) observables has been surveyed in 186 peer-reviewed papers to identify major research pathways in terms of data used and retrieval models. Research evaluated primarily (i) L-band observations of SAR backscatter; and, (ii) single-image or multi-polarized retrieval schemes. The use of multi-temporal or multi-frequency data improved the biomass estimates when compared to single-image retrieval. Low frequency SAR backscatter contributed the most to the biomass estimates. Single-pass InSAR height was reported to be a more reliable predictor of biomass, overcoming the loss of sensitivity of SAR backscatter and coherence in high biomass forest. A variety of empirical and semi-empirical regression models relating biomass to the SAR observables were proposed. Semi-empirical models were mostly used for large-scale mapping because of the simple formulation and the robustness of the model parameters estimates to forest structure and environmental conditions. Non-parametric models were appraised for their capability to ingest multiple observations and perform accurate retrievals having a large number of training samples available. Some studies argued that estimating compartment biomass (in stems, branches, foliage) with different types of SAR observations would lead to an improved estimate of total biomass. Although promising, scientific evidence for such an assumption is still weak. The increased availability of free and open SAR observations from currently orbiting and forthcoming spaceborne SAR missions will foster studies on forest biomass retrieval. Approaches attempting to maximize the information content on biomass of individual data streams shall be pursued.

Forest plays an important role in carbon sequestration and biosphere-atmosphere interaction. Knowledge of forest biomass content helps in assessing its sustainability and thus mitigating climate change. The advancement in remote sensing technology provides the capability of estimating biomass at a large scale. Hybrid polarimetry has gained significant attention among other radar missions due to its fundamental advantages. In this article, the potential of hybrid polarimetric SAR is evaluated for the efficient forest aboveground biomass (AGB) estimation for Barkot Forest, Uttarakhand, India. Forest biomass is calculated by means of the extended water cloud model. Scattering parameters are derived using two widely used hybrid polarimetric decomposition techniques, m-δ and m-χ decompositions. Potential insight into the efficacy of these decomposition techniques toward biomass estimation is brought forth. The modeled AGB estimates were compared with fully polarimetric data-based estimated AGB. The estimation based on the m-χ and m-δ decomposition resulted in biomass estimation with an accuracy of 75.8% and 73.4%, respectively.

Estimation of forest biomass is still a challenging task over large areas because of the saturation problem of remote sensing data as well as the environmental, topographic, and biophysical complexity of forest ecosystems. However, Synthetic Aperture Radar (SAR) is still one of the most attractive choices for the estimation of biomass or carbon storage capacity of vegetation due to its sensitivity to plant canopy structure, and new SAR sensors have attracted greater interest as they are able to provide data with varying spatial resolutions, polarizations, and incidence angles. This research investigates the potential of C-band dual polarization (HH and HV) SAR (Radarsat-2) imagery for forest biomass estimation using different combinations of raw backscattering (intensity), polarization texture parameters and texture polarization ratios. Several models have been developed between field biomass and SAR signatures using stepwise multiple regression. Results indicate that SAR intensity images (C-HV and C-HH) and the ratio of intensity data (C-HV/C-HH) have relatively low potential (r2 = 0.20) for biomass estimation. However, the SAR polarization (C-HV and C-HH) texture parameters were found to be effective and about 82 % (r2 = 0.82 and RMSE = 28.68 t/ha) of the variability in the field data (forest biomass up to 360 t/ha) was explained by the model. Further improvement of the estimation was achieved (r2 = 0.90 and RMSE = 21.55 t/ha) using the texture polarization ratio (C-HV/C-HH). The outcomes suggest that a clear improvement in forest biomass estimation can be obtained using the texture parameters of dual polarization C-band SAR and more improvement can be achieved using the ratio of texture parameters, as this combines the advantages of both texture and ratio.

Sufficient remote sensing observation data during crop main growing season is of great importance in improving the accuracy of data assimilation of crop model. The optical remote sensing data are susceptible to cloud and rain, so the amount of clear optical data is very limited in cloudy weather or rainy day. Synthetic Aperture Radar (SAR) is not dependent on cloud cover or light conditions, it can penetrate through clouds and have all-weather capabilities. This allows for a more reliable and consistent crop monitoring and yield estimation in terms of radar sensor data. So, the aim of this article is to improve the accuracy for winter wheat yield estimation by joint assimilation of SAR and optical satellite images into crop model. In this study, SAR images are acquired by C-band SAR sensor boarded on Sentinel-1 satellites, and optical images are obtained from Sentinel-2 satellites. Remote sensing data and ground data are all collected during the main growth and development stages of winter wheat. Both normalized difference vegetation index (NDVI) derived from Sentinel-2 images and backscattering coefficients and polarimetric indicators computed from Sentinel-1 images are used in water cloud model to derive soil moisture (SM) time series images. To improve the prediction of crop yields at filed scale, we incorporate remotely sensed soil moisture into the WOrld FOod STudies (WOFOST) model using Ensemble Kalman filter (EnKF) algorithm. In general, the results show that data assimilation schemes of remotely sensed soil moisture slightly improved the correlation of observed and simulated yields (R2 = 0.30; RMSE =782 kg ha−1) compared to the situation without data assimilation (R2 = 0.14; RMSE = 1398 kg ha−1). Results of this study indicate that the potential for assimilation SAR and optical remote sensing data to improve field yield estimates is relatively low, limitations are due to insufficient no-cloud optical remotely sensed data and root zone soil moisture information. Consequently, the results of this study demonstrate the potential and usefulness of assimilating both SAR and optical remote sensing data into crop model for crop monitoring and yield estimation. Moreover, this also provides the reference for crop yield estimation with data assimilation in other agricultural landscapes.

Modulation of surface capillary waves by tidal current flow over submarine relief features causes roughness variations on the ocean surface that are detected by Synthetic Aperture Radar (SAR). Previous studies have demonstrated the presence of features in SAR images that correspond with submarine topography under appropriate sea-state and imaging conditions. In this study, bathymetric signatures observed on two European Remote Sensing Satellite (ERS) SAR images and a near-coincident Landsat Thematic Mapper (TM) image over an area west of Melville Island in the Timor Sea were investigated. Bathymetric features visible on the SAR and optical satellite images correspond well with isobaths of the study area. Submarine relief features seen on the satellite images of the study area were analysed in conjunction with wind and sea-state data. Based on the analysis, we highlight some factors governing the expression of submarine relief features seen in satellite images of the study area. Submarine relief signatures are sensed by optical and SAR sensors through different mechanisms, however similarities in signatures observed in the near-coincident optical and SAR images indicate that the underlying mechanism is common to optical and SAR imaging. Sun glint resulting from specular reflection and modulated by ocean surface roughness is the predominant mechanism producing features on the Landsat TM image that correspond with bathymetric signatures seen on the ERS SAR images. This study demonstrates the potential to extract information on sea bottom topography from sun glint in optical images under specific sun-sensor-target geometries and sea-state conditions, to complement similar information derived from SAR.

Abstract. Aboveground tropical tree biomass and carbon storage estimates commonly ignore tree height (H). We estimate the effect of incorporating H on tropics-wide forest biomass estimates in 327 plots across four continents using 42 656 H and diameter measurements and harvested trees from 20 sites to answer the following questions: 1. What is the best H-model form and geographic unit to include in biomass models to minimise site-level uncertainty in estimates of destructive biomass? 2. To what extent does including H estimates derived in (1) reduce uncertainty in biomass estimates across all 327 plots? 3. What effect does accounting for H have on plot- and continental-scale forest biomass estimates? The mean relative error in biomass estimates of destructively harvested trees when including H (mean 0.06), was half that when excluding H (mean 0.13). Power- and Weibull-H models provided the greatest reduction in uncertainty, with regional Weibull-H models preferred because they reduce uncertainty in smaller-diameter classes (≤40 cm D) that store about one-third of biomass per hectare in most forests. Propagating the relationships from destructively harvested tree biomass to each of the 327 plots from across the tropics shows that including H reduces errors from 41.8 Mg ha−1 (range 6.6 to 112.4) to 8.0 Mg ha−1 (−2.5 to 23.0). For all plots, aboveground live biomass was −52.2 Mg ha−1 (−82.0 to −20.3 bootstrapped 95% CI), or 13%, lower when including H estimates, with the greatest relative reductions in estimated biomass in forests of the Brazilian Shield, east Africa, and Australia, and relatively little change in the Guiana Shield, central Africa and southeast Asia. Appreciably different stand structure was observed among regions across the tropical continents, with some storing significantly more biomass in small diameter stems, which affects selection of the best height models to reduce uncertainty and biomass reductions due to H. After accounting for variation in H, total biomass per hectare is greatest in Australia, the Guiana Shield, Asia, central and east Africa, and lowest in east-central Amazonia, W. Africa, W. Amazonia, and the Brazilian Shield (descending order). Thus, if tropical forests span 1668 million km2 and store 285 Pg C (estimate including H), then applying our regional relationships implies that carbon storage is overestimated by 35 Pg C (31–39 bootstrapped 95% CI) if H is ignored, assuming that the sampled plots are an unbiased statistical representation of all tropical forest in terms of biomass and height factors. Our results show that tree H is an important allometric factor that needs to be included in future forest biomass estimates to reduce error in estimates of tropical carbon stocks and emissions due to deforestation.

In this thesis the modeling of SAR (Synthetic Aperture Radar) images of natural surfaces described via fractal models is dealt with. A complete theoretical forward model linking the parameters describing the scene observed by the sensor to the stochastic characterization of the relevant SAR image is provided. The inverse problem is treated as well: a SAR image post-processing able to automatically retrieve - operating on an amplitude single SAR image - the fractal parameters of the scene, is presented. The developed imaging model is based on sound geometrical and electromagnetic models that are combined according to the SAR impulse response function. The power spectral densities of appropriate cuts of the SAR image are evaluated in closed form in terms of the surface fractal parameters. The theoretical results are here conceptually assessed, analytically derived, graphically validated and numerically verified. Moreover, based on the inversion of the forward theoretical model, an innovative SAR image post-processing for the fractal parameters estimation is implemented. It is firstly tested on simulated SAR images, then it is applied to different types of new generation (i.e. high resolution) SAR images. The generated fractal maps show themselves to be very useful for a wide range of application, e.g. prevention and monitoring of environmental disasters, geodynamic processes interpretation, land classification, rural planning, and so on.

The net annual flux of carbon from south and southeast Asia as a result of changes in the area of forests was calculated for the period 1850 to 1985. The total net flux ranged from 14.4 to 24.0 Pg of carbon, depending on the estimates of biomass used in the calculations. High estimates of biomass, based on direct measurement of a few stands, and low estimates of biomass, based on volumes of merchantable wood surveyed over large areas, differ by a factor of almost 2. These and previous estimates of the release of carbon from terrestrial ecosystems to the atmosphere have been based on changes in the area of forests, or rates of deforestation. Recent studies have shown, however, that the loss of carbon from forests in tropical Asia is greater than would be expected on the basis of deforestation alone. This loss of carbon from within forests (degradation) also releases carbon to the atmosphere when the products removed from the forest burn or decay. Thus, degradation should be included in analyses of the net flux of carbon from terrestrial ecosystems. Degradation may also explain some of the difference between estimates of tropical forest biomass if the higher estimates are based on undisturbed forests and the lower estimates are more representative of the region. The implication of degradation for estimates of the release of carbon from terrestrial ecosystems is explored. When degradation was included in the analyses, the net flux of carbon between 1850 and 1985 was 30.2 Pg of carbon, about 25% above that calculated on the basis of deforestation alone (with high estimates of biomass), and about 110% above that calculated with low estimates of biomass. Thus, lower estimates of biomass for contemporary tropical forests do not necessarily result in lower estimates of flux.

Forest plays a vital role in regulating climate through carbon sequestration in its biomass. Biomass reflects the health and environmental conditions of a forest ecosystem. In context to the climate change mitigation mechanisms like REDD (reducing emissions from deforestation and forest degradation), an extensive forest monitoring campaign is especially important. Remote sensing of forest structure and biomass with synthetic aperture radar (SAR) bears significant potential for mapping and understanding forest ecological processes. Limitations of the conventional forest inventory procedures, like the extensive cost, labor and time, can be overcome through integrated geospatial techniques. Optical sensor or SAR data are suitable for extracting information about simple and homogeneous forest stand sites. However, optical sensors face serious limitations, specifically in tropical regions, like the cloud cover that SAR can overcome along with targeting saturation and penetration aspects. Simultaneous use of spectral information and image texture parameters improves the biomass assessment over undulating terrain and in radical conditions. Also, synergic use of multi-sensor optical and SAR has better potential than single sensor. Interferometric (InSAR) and polarimetric (PolSAR) SAR or a combination of the both (PolInSAR) serves as effective alternatives. These techniques could serve as valuable methods for biomass assessment of heterogeneous complex biophysical environments. However, SAR data have its own limitations and complexities. Identifying, understanding and solving major uncertainties in different stages of the biomass estimation procedure are critical. In this regard, the current study provides a review of radar remote sensing-based studies in forest biomass estimation.