Bambang H. Trisasongko

Mapping stand age of rubber plantation using ALOS-2 polarimetric SAR data

ABSTRACT This paper presents an evaluation on strategies for rubber plantation mapping employing SAR data coupled with Random Forest (RF) and Support Vector Machine (SVM). Linear backscatter coefficients achieved saturation point at about 10 years, making this form of polarimetric data being robust only for young to mature stands. This research found that the performance of both algorithms was comparable. The addition of texture features gave substantial impact to the overall accuracy. As indicated by the analysis of variable importance, only some texture features have contributed to higher overall accuracy. Classification using a subset of texture features pointed out that accuracy could be improved using dual polarimetric data, while trivial enhancement was seen in combined HH, HV and VV backscatter intensities. The research showed that classification accuracy could be further augmented by setting proper classification parameters. Nonetheless, it is argued that the level of improvement would greatly depend on selecting a proper dataset fed into classifier, rather than tuning classifier parameters.

Self-organizing Neural Networks for Unsupervised Classification of Polarimetric SAR Data on Complex Landscapes

This paper refers to a study on the pixel-by-pixel unsupervised classification of a polarimetric SAR image of a Central Italy landscape. The polarimetric data have been processed by self-organizing neural networks to test their performance in classifying a complex landscape. The discrimination accuracy attained by the self-organizing map method is compared both against that of H/A/alpha-Wishart unsupervised procedure and of a supervised scheme.

Unsupervised classification of a central italy landscape by polarimetric L-band SAR data

This paper reports on the classification of a Central Italy landscape imaged by the JPL/NASA AirSAR in the frame of the MAC-Europe 91 campaign. The coherence matrices have been computed from the scattering vector obtained through Pauli decomposition, and subsequently filtered. An H/A/α decomposition has been carried out preliminarily to the Wishart H/A/α classification procedure, fed by the coherence matrices, the results of the previous H/α classification and the anisotropy values. The iterative procedure has been terminated by a proper lower threshold on the percentage of pixels changing classes in subsequent steps. Given the nature of the imaged landscape, the number of classes has been suitably reduced to match that actually identified in the scene. Accuracy matrices have been computed from the obtained results and the ground truth map co-registered with the AirSAR image, and a comparison has been carried out with the results yielded for agricultural parcels by a classification scheme based on the physical modeling of the interaction between radar electromagnetic wave and crops.

Evaluating compact SAR polarimetry for tropical forest monitoring

Fully polarimetric Synthetic Aperture Radar (SAR) or PolSAR has been proven useful for diverse applications related to environment. Nevertheless, problems are arising since satellite-borne PolSAR requires special arrangements on data acquisition and consumes higher energy for signal transmission. Complexity of data acquisition and analysis can be reduced using compact polarimetry. The technique has been demonstrated to some extent; however, tests on various environments are still required. This paper assesses compact polarimetry on a tropical forest fringe, especially to monitor expanding oil palm estate and forest disturbance, in comparison to fully polarimetric mode. PALSAR data of Manokwari, Indonesia, were collected from JAXA through RA4.1029 project. In this paper, linear 45 degrees transmission is evaluated to detect various land cover classes using Wishart supervised classifier. Tonal discrepancies between both polarimetric modes are evident, suggesting compact polarimetry has limitation to recover information contained in fully polarimetric mode. However, Wishart classification procedure indicates that compact polarimetry is still useful for mapping.

Comparing six pixel-wise classifiers for tropical rural land cover mapping using four forms of fully polarimetric SAR data

ABSTRACT This study evaluates four commonly used forms of synthetic aperture radar (SAR) data for land-cover classification in tropical rural areas. The backscatter coefficient of linearly polarized L-band SAR was compared to two distinctive feature sets derived from Eigen-based and model-based decompositions. The performance of six classifiers available in Orfeo Toolbox (OTB), that is, Bayes, artificial neural networks (ANNs), Support Vector Machine (SVM), decision trees, Random Forests (RFs), and gradient boosting trees (GBTs), was investigated to distinguish five and seven land-cover classes, with particular attention given to several types of woody vegetation: forest, mixed garden, rubber, oil palm, and tea plantations. Classifiers reacted differently to ingested forms of SAR data, and careless use of data input yielded a negative impact. The results showed that SVM provided the highest overall accuracy although the performance was not significantly better than the others. Tuning the parameters, however, significantly improved the accuracy of ANN and SVM, while RF and GBT did not respond well. Responses of two SVM parameters (cost and kernel type) fluctuated somewhat, which required further attention. ANN accuracy was improved when the number of neurons in the hidden layer was set between 10 and 12. We found that accuracy imbalance existed between designated land-cover classes, especially in woody vegetation. Imbalance can partially be reduced by tuning specific classifiers. We showed that classifier tuning can lead to significantly improved accuracy, especially for classes having medium or low accuracies. This research also demonstrated that freely available toolkits such as OTB and QGIS can be beneficial for mapping activities in developing countries, achieving a reasonable accuracy if the classification parameters are tuned properly.

Potential use of hybrid synthetic aperture radar polarimetry in Earth surface monitoring

To observe delicate Earth surface continuously, satellite-based monitoring system is required. Especially in tropical region, Synthetic Aperture Radar (SAR) is necessitated considering its ability to penetrate cloud and other atmospheric attenuations. Recent fully polarimetric SAR has been exploited. Nonetheless, this mode of imaging consumes higher amount of energy, which is one of the main issues in satellite-based platform. In this paper, a study exploiting hybrid (also known as compact) polarization is presented. Comparison to fully polarimetric mode of SAR is made using polarimetric decomposition. This research indicates that single signal transmission in hybrid polarization cannot fully replace fully-polarized mode. This suggests that hybrid polarization should be limitedly applied to geo-biophysical applications such as biomass or soil moisture estimation. However, for general land cover discrimination and monitoring, hybrid polarimetry is fairly useful. Analysis of transformed divergence on decomp...

Interferometric processing of C-band SAR data for the improvement of stand age estimation in rubber plantation

Rubber ranks the second largest plantation in Indonesia after oil palm. While oil palm plantations have been exploited mainly by large companies, many rubber plantations are still managed by peasant farmers who maintain its biodiversity. Due to its broad and scattered location, monitoring tropical rubber plantation is a crucial application of active remote sensing. In this paper, the backscatter coefficient of Envisat Advanced Synthetic Aperture Radar (ASAR) is compared to interferometric coherence to study the relationship between stand age and SAR parameters. It is shown that VV polarized C-band SAR achieves its saturation level in plantations aged about 5-10 years. Extension of saturation level can be achieved by processing an interferometric pair of ASAR data, which results in interferometric coherence. In this paper, coherence can take up to 20 years stand age to achieve prior to saturation. Since stand age is highly related to biomass, this finding argues that the biomass can be best estimated using coherence.

A review of remote sensing applications in tropical forestry with a particular emphasis in the plantation sector

Abstract The aim of this article is to evaluate current achievements of remote sensing technologies in forest and plantation monitoring. Despite considerable efforts having been dedicated to monitor tropical forest, some issues remain open for further exploration, including forest type mapping, biomass estimation, change detection and the detection of invasive species. Large-scale forest conversion to plantations makes it necessary to assess applications and methodologies currently published with the aim to provide an outlook for future research. Multispectral datasets have been favoured in this domain, largely because of their long-term availability. Remote sensing applications in plantation forests are often perceived as less problematic than natural forests, perhaps due to their relatively homogenous cover. We present evidence that assumptions of homogeneity in canopy cover may not be fully satisfied. Vital aspects of plantation for management such as stand age mapping, detecting disturbance and productivity measurement have been understudied, which therefore warrant further investigation.

Characteristics of L-band backscatter coefficients of rubber plantation and their seasonal dynamics

As one of primary land uses in Indonesia, rubber plantation requires frequent, wide-scale monitoring. Due to the nature of tropical region, optical sensors are often inapplicable and therefore Synthetic Aperture Radar (SAR) plays a role. Dual-polarized SAR data have been a definitive imaging mode since fully polarimetric mode consumes higher energy. In this paper, characteristics of returning SAR signals from young rubber stands are investigated in terms of different polarization and time of acquisition. The research shows that strong ground attenuation is observed in very young plantation, which is similar to amplified Bragg scattering in rice field. Seasonal defoliation is also evident at this age, possibly due to limited root depth which reduces ability to obtain moisture in lower solum. Temporal change of canopy cover is detectable by HV polarization, which has been known sensitive to canopy structure. This research suggests that seasonal variation of HV backscatter coefficients may affect biophysical...

Seasonal Pattern of Vegetative Cover from NDVI Time-Series

Indonesia manages various forested land utilizations, for instance natural forest and plantations. In the past, natural forest had been exploited throughout the country, mainly in the islands of Sumatera and Borneo (Nawir & Rumboko, 2007). Greater criticisms on forest exploitation lead to a moratorium which needs to be monitored frequently. Most of forest concessions were converted into plantations (Kartodihardjo & Supriono, 2000). Popular plantations developed in the country have been rubber and oil palm. Successful management of the plantation as well as natural forest has been under careful examination. Assessment of woody vegetation could be taken using field surveys or remote sensing. Although having possibilities to capture detailed datasets, field survey is lacking in terms of implementation: lengthy data capture and least favorable to remote areas. In these situations, remotely sensed data play a crucial role.