Operational deforestation detection and forest early warning is one hot topic in synthetic aperture radar (SAR) Earth observation (EO) today. Especially in tropical regions where persistent cloud coverage is incapacitating optical sensors during large parts of the year, active microwave is regarded as the most promising technology to combat illegal and unsustainable deforestation activities. The key aspect for a true EWS is the timely detections of fresh and ongoing deforestation activities in a matter of days. allowing authorities to act on a given warning in meaningful fashion, e.g., inspecting detected sites, detain deforesters and pretend the remaining forest from further destruction. The development of JAXA’s next-generation algorithm (NGA) for deforestation detection builds upon the important breakthroughs in long-term pantropical forest monitoring made by the ALOS-2 mission. A field survey conducted in the Brazilian Amazon verified the unparalleled performance of the NGA to reliably detect ongoing deforestation.

Forest monitoring is a prime objective for orbital synthetic aperture radar (SAR) missions. Especially in tropical regions where persistent cloud coverage incapacitates optical satellite observations for large parts of the year, radar remote sensing provides a unique tool for consistent all-year round data acquisition. In contrast to high-frequencies SARs at C- and X-band, L-band radar constitutes a powerful tool to obtain direct information from the full forest strata ranging from the top canopy all the way down to the forest ground. However, despite important advantages, SAR faces various challenges when it comes to potential error sources. As radar backscattering from distributed natural forest targets is profoundly complicated in nature, ill-defined signal fluctuations often limited the reliability of models and applications. To overcome these challenges by improved understanding of the spatial-temporal fluctuations in forest backscatter, coherent, multitemporal and polarimetric observations covering all kinds of environment conditions are required. Nearing 9 years in orbit since its launch in 2014, ALOS-2 has acquired the world’s first long-term pantropical L-band SAR archive fulfilling exactly these requirements.

Nearing 8 years in orbit, ALOS-2/PALSAR-2 has acquired the world's first long-term pantropical L-band SAR forest monitoring data archive. The unprecedented dual-polarimetric time-series data allow to revolutionize our knowledge of the complex spatial-temporal processes governing the backscattering behaviors in tropical forests from local to global scale. The paper discusses ALOS-2's tremendous contributions to fundamental SAR forest research, forest cover mapping, biomass estimation, and deforestation monitoring, from better forest/non-forest maps and improved forest type classifications to all-new detailed flooded forest maps and groundbreaking insights on the rainfall impacts on L-band SAR imaging shedding light on the often overlooked relation between precipitation and forest backscatter. Based on these pioneering achievements a novel AI-supported false alarm suppression method was developed for the JICA-JAXA forest early warning system in the tropics (JJ-FAST) enabling the first operational fully automatic near real-time deforestation detection service with overall accuracies above 80% in the Amazon basin.

Forest observation is one prime objective for current and future L-band SAR missions. Despite the tremendous achievements made in the last 3 decades, the reliability of derived products, e.g., forest cover and biomass maps, is still largely hampered by ill-defined backscatter fluctuations. To better understand the rain-induced noise-like variations in tropical forest backscatter timeseries, we demonstrate big data deep learning applications exploiting the unprecedented long-term data sets provided by ALOS-2/PALSAR-2 from 2016–2021. Robust fully automatic deforestation detection is achieved in an operational forest monitoring system by using a CNN-based false alarm suppression method. Using 400 TB of ScanSAR images together with 50 billion auxiliary rainfall data an automatic rainfall prediction and correction procedure for L-band SAR is proposed based on LSTM artificial neural network. First results show the approach's potential to reliably reduce rain related disturbances in L-band forest timeseries in an unsupervised fashion.

JICA-JAXA Forest Early Warning System in the Tropics (JJ-FAST) monitors tropical forest in 77 countries by using PALSAR-2/ScanSAR time-series data. The deforestation detection algorithm used in JJ-FAST shows overall user's accuracies of 71.1%, but the user's accuracies was 53.3 % for Latin America. Four convolutional neural network models (CNN) have been developed and tested to improve deforestation detection accuracies in Peru and Brazil. By applying the suggested CNN models, averaged user's accuracies are improved from 45 % to 77 - 85 % in dry season, and from 34 % to 48 - 57 % in rainy season. Averaged relative F1 are improved from 0.6 to 0.67 - 0.75 in dry season, and from 0.49 to 0.52 - 0.63 for rainy season. It is shown that adding CNN operation improves the deforestation detection accuracies. Intermediate CNN outputs were visualized to gain insights about what features of the image does each layer learn for improving the accuracies.

One of the prime objectives for orbital civilian SARs is forest observation. Though tremendous effort has been devoted to improving applications like biomass estimation and deforestation detection, the reliability has been impeded by insufficient knowledge about spatial-temporal variabilities of forest backscatter. In the tropics, these backscatter fluctuations are largely controlled by the rainfall regime. This paper investigates the impact of rainfall on L-band forest observation using 300 TB pan-tropical ALOS-2 ScanSAR timeseries with 1.5 billion TRMM based GSMaP precipitation datasets. The results show that the correlation between forest gamma naught and antecedent rainfall can exceed R=0.9 in many areas outside the rainforest. In the rainforest strong negative correlation below R=-0.8 can occur in the severely flooded forest. We assess the impact of rainfall on deforestation detection accuracies and discuss a new classification scheme for tropical forests using statistics of seasonal backscattering behaviors together with response to precipitation.

JICA and JAXA have operated a tropical forest early warning system (JJ-FAST) since 2016 and ALOS-2 L-band SAR system (PALSAR-2/ScanSAR) is used for forest monitoring and deforestation detection. There are many factors to cause commission errors in SAR image and ground moisture is one of them. To reduce such commission errors, seasonal change analysis is effective. In this study, a time-series analysis method using Savitzky-Golay (SG) filtering was developed. The method of speckle noise filtering by frost and 2-step thresholding and time-series analysis using SG was good method. The accuracy of forest change detection was improved from 19% of the threshold method to 80%.

JJ-FAST is the SAR-satellite based early deforestation monitoring system, and deforestation information obtained by using PALSAR-2/ScanSAR data are released on the JJ-FAST web site, 3 to 4 days after the observation. ScanSAR/50m spatial resolution data is currently used for the deforestation detection, and the minimum detection size is 3ha. In this paper, ScanSAR/25m spatial resolution data is used for the deforestation detection, and the detection accuracies are compared with the one obtained with the 50m spatial resolution data for four test sites. The minimum deforestation detection size is set to be 1ha in this time. The results show that the 25m spatial resolution data always shows higher detection accuracies. The results obtained in Cambodia test site show that user's accuracies are 21.4 % higher than the case of 50m resolution, and number of correctly detected polygon is 2.3 times more than the one for the 50m resolution case. It is indicated that the 25m spatial resolution ScanSAR data improves the detection accuracies with minimum detection size of 1 ha.

Debris estimation is one of the most important initial challenges after a disaster like the Great East Japan Earthquake and Tsunami. Reasonable estimates of the debris must be made available to decision makers as quickly as possible. Classical approaches to obtain this information are far from being optimal, usually relying on manual interpretation of optical imagery. We have developed a novel approach for the estimation of tsunami debris pile heights and volumes for improved emergency response. The method is based on a stereo-synthetic aperture radar (stereo-SAR) approach for very high-resolution airborne polarimetric SAR. An advanced gradient-based optical-flow estimation technique is applied for optimal image coregistration of the low-coherence non-interferometric data. Its suitability to combine multiresolution data allows generating stereo SAR data by combining two different sensors. Based on model-based decomposition of the PolSAR data, only the odd bounce scattering contributions are used to optimize echo time computation. In this paper, we propose the further development of the method by combining multiresolution data from i) air-/spaceborne SAR and spaceborne/spaceborne SAR with various illumination geometries. The proposed technique is validated using in situ data of real tsunami debris taken on a temporary debris management site in the tsunami affected area near Sendai city, Japan. The estimated height error is in the order of 0.7 m RMSE. The good quality of derived pile heights allows estimating debris volume with an RMSE of 2500 m3corresponding to <; 10% of the total debris volume. Advantages of the proposed method are fast computation time, and robust height and volume estimation of debris piles without the need for pre-event data or auxiliary information like DEM, topographic maps or GCPs

Insufficient knowledge about spatial-temporal forest heterogeneities in the tropics is a major impediment to better estimation of carbon storage and prevention of deforestation by remote sensing. While the differences between the dense evergreen rainforest and the open (semi-) deciduous dry forest is obviously large, variations caused by seasonal changes can easily introduce fluctuations in the same order of magnitude within the same forest class. In this study, we present a comprehensive analysis of the variability of tropical forests based on homogeneous big data analysis on multitemporal ALOS-2 dual-polarized ScanSAR data. The first, easy to understand global forest variability maps provide unseen insights into forest structures for the entire tropical belt. Based on these results, we discuss the development of a new global classification scheme for tropical forests. Preliminary results demonstrate how the use of various statistical parameters obtained from the long-term systematic L-band SAR forest monitoring data, including the average γ0, its temporal standard deviation and the γ0range, can improve the classical global-scale forest classifications. In addition to the basic separation into the three main forest types i) rainforest, ii) moist forest and iii) dry forest, the results provide a detailed mapping of the seasonally flooded forest areas.

In order to reduce speckle noise effects in forest cover change analysis, two types of SAR processing approach have been proposed so far. One is a type of object-based segmentation approaches and another is pixel-based processing approaches using speckle noise filtering such as Lee and Frost filter. This paper proposes a methodology of the pixel-based algorithm for forest change detection using PALSAR-2 ScanSAR data. Frost filter is applied to reduce speckle noise and normalization filter is implemented to reduce ground moisture effects. In case of Peru and Paraguay, the user's accuracy of developed algorithm was almost same as those of JJ-FAST version-2 while the producer's accuracy was 3 times better than JJ-FAST latest version (version-2).

JJ-FAST is the first SAR-based global and early deforestation monitoring system, and started the operation from November 2016. In this paper, improved deforestation detection algorithm were suggested for using JJ-FAST system. One of the improvement is to increase maximum number of data used from 15 to 20 to suppress error detection occurred in mountain area and swamp forest. The other improvement is to adopt 0.5 dB lower threshold level to increase number of deforestation detection. Suggested algorithm were evaluated by using optical satellite data (GLAD), obtained in one of the active deforestation spots in Peru, Brazil, Cambodia, and Mozambique. User's accuracies increases by 6.7%, and number of correctly detected deforestation polygon make it 1.3 times as large.

In the dry tropical forest and semi-arid savanna information about the soil moisture dynamics and distribution are vital to better understand and mitigate damages caused by drought. To estimate the soil moisture over large areas from 50-m resolution ALOS-2 ScanSAR data, we adopt dual-polarization L-band algorithms originally developed for high resolution ALOS data. Based on the partial-polarimetric eigenvector decomposition, the scattering entropy and alpha angles are employed to correct for the above ground vegetation effects. The large incidence angle variation along the 350-km ScanSAR swath requires the use of an incidence angle correction factor. However, due to the relative poor data quality of the ScanSAR products suffering from scalloping and near-range/far-range striping along the subswathes, we observe lower accuracies than Stripmap mode. While in the dry season soil moisture retrieval in the open forest is well possible, in the rainy season the entropy exceeds the validity range due to the larger amounts of canopy biomass resulting in unreliable soil moisture estimates. Savanna vegetation allows a biomass-corrected soil moisture estimation throughout the whole year. Due to the lack of distributed in-situ data, we use coarse scale SMOS and SMAP soil moisture data to validate the accuracy of dual-polarization ScanSAR soil moisture estimates. Comparisons between the kilometer-scale radiometer derived soil moisture estimates and 50-m ALOS-2 results show a good agreement in the dry season with RMSEs smaller 5 Vol.-%, while in the rainy season the RMSEs can be as high as 15 Vol.-%.

JJ-FAST is an operational forest early warning system using PALSAR-2/ScanSAR mode data. Its deforestation detection algorithm has been updated since 2016 and the current is version 1 (as of January, 2018) and the algorithm version 2 is preparing to implement in March 2018. Overall user's accuracy of version-1 products was estimated from several field experiments, and achieved as 83.3%. There are omission errors in the version 1 products because of its simple HV change detection approach.

A preliminary version of the semi-automatic deforestation detection algorithm was developed for the JJ-FAST system, which monitors deforestation and forest changes in tropical regions for approximately every 1.5 months and covers 77 countries. PALSAR-2/ScanSAR HH, HV, HH and HV ratio were used to detect various deforestation stages. Multitemporal data were used to suppress the effect of seasonality and rainfall. Moreover, the deforestation accuracies for the active deforestation spots in Brazil and Peru were estimated. The user's accuracies achieved as 80.5% in the dry season, but these decreased to 11.1% in the rainy season.

ALOS-2 ScanSAR data was used for examining the effect of precipitation and soil moisture variations on L-band forest backscatter in tropical regions. Multitemporal data spanning from observation cycle 39 to 62 acquired over tropical forests in South America and Africa has been used. Precipitation data provided through JAXA's global rainfall watch was used to examine the temporal change of L-band σ0for forests as well as deforestation areas. The best choice for temporal integration of rainfall prior to the image acquisitions was investigated. The results show that σ0HHvalues show an average increase smaller 1 dB, whereas σ0HVvalues can increase by more than 2 dB in dense forests after heavy rain. Our findings also indicate that dense forest and open/moderately dense forests can show quite different rainfall effects. The latter may even exhibit a decrease in both σ0HHand σ0HVafter heavy rain due to standing water. Implementations for the ongoing development of an advanced tropical forest monitoring system are discussed.

Time series PALSAR-2/ScanSAR data and Landsat data were used for examining the differences in detection timing of deforestation. Optical sensor-based (Landsat) deforestation information taken about every 16 days and SAR data taken about every 1.5 months were used, and the temporal change of L-band γ0was examined for the deforestation areas. The γ0HHvalue increased by 1.2 dB on average for areas undergoing the early stages of deforestation, where fallen trees were left on the ground. The detection timing was almost same as that using the optical sensor. On the other hand, the γ0HVvalue decreased by 1.2 dB on average for areas undergoing the late stages of deforestation, where fallen trees were removed. The detection timing using γ0HVwas about a few month after the detection using γ0HH, or optical sensor. It is concluded that γ0HHis useful for early-stage deforestation detection, and γ0HVis useful for late-stage deforestation detection.

The polarimetric characteristics of early- and late-stage deforestation areas were examined using PALSAR-2. An early-stage deforestation site, where the fallen trees had been left on the ground, showed an increase in σsurface0of 2.9 dB, and decrease in σvolume0of 1.4 dB. This induced the 0.9 dB increase in σHH0. A late-stage deforestation site, where the fallen tree had been removed, showed an increase in σsurface0of 2.3 dB, and a decrease in σvolume0of 6.4 dB. This induced the 6.1 dB decrease in σHV0. Polarimetric analysis clarified the scattering mechanism for the early- and late-stage deforestation areas.

Mapping of the urban damage levels with synthetic aperture radar (SAR) is still challenging. Fully polarimetric SAR (PolSAR) has the potential to identify the type of scattering mechanism changes induced by urban damage. In radar polarimetry, dominant double-bounce scattering mechanism in urban areas is primarily induced by the ground-wall structures. Thereby, within an urban patch, the reduction of the dominant double-bounce scattering mechanism reflects the urban damage level in terms of destroyed ground-wall structures, which is the basis of this study. Based on our previous study, the proposed polarimetric index is further investigated. A rapid urban damage mapping technique including mainly two steps of urban area extraction and damage level index estimation is proposed. The 3.11 East Japan Earthquake and Tsunami is adopted as the study case using multi-temporal ALOS/PALSAR PolSAR data. Experimental studies demonstrate that the estimated damage levels are closely consistent to the ground-truth.

A full polarimetric ground-based SAR system for subsurface damage assessment of wooden building structures has been developed. The system is based on a 4-channel VNA connected to a polarimetric antenna array. The array elements consist of 4 newly developed circular polarization cavity-backed spiral antennas, which have very good wideband characteristics and excellent polarimetric performance. By 2 dimensional scanning the system can achieve polarimetric 3D imaging with super high-resolution of 1 cm in x-/y-direction and 2 cm in z-direction. Using the polarimetric phase information, deformations of internal wooden structures like beams and bars with inclination angles <;1 deg. can be detected, which is not possible with any other non-destructive testing technique.

An empirical soil moisture retrieval model for dual-polarimetric L-band SAR imaging, the ALOS-2 standard observation, is being developed. The model, based on empirical relations between ALOS/PALSAR and Pi-SAR-L2 observations and ground measurements, can correct for the disturbing effects caused by surface roughness or vegetation cover. In this paper we focus on airborne campaigns carried out over the Sendai area, Japan, in 2013 and 2014. In situ surface parameters (soil moisture and surface roughness) were measured on bare soil and under rice canopy. Using roughness and biomass corrections the SAR derived soil moisture estimates yield an accuracy of 5.9 Vol.-% RMSE. The results indicate that the proposed approach, which was mainly parameterized in test sites in Germany, can also work in different geographic regions.

Full polarimetric airborne SAR data acquired over the Tsunami affected coastal area in Northeastern Japan is used to estimate soil moisture and debris volumes. Using empirical surface roughness calibration of the I2EM soil moisture is estimated with a RMSE of 5.2 Vol.-%. The heights of Tsunami debris piles are estimated with a RMSE of 0.71 m by using a reference point aided stereo-SAR approach. Debris pile volumes are estimated with an overall RMSE of 2478 m3. The accuracy increases with increasing size of the piles. While the error ranges from 50% to 90% for the smallest piles, it is in the order of 20% to 30% for the large piles.

We are developing radar technology to inspect the inner structure of wooden buildings suffered from strong quake by earthquake. GB-SAR (Ground Based Synthetic Aperture Radar) for inspection of wooden and concrete walls and structures has been developed and the system was evaluated by test measurements. The GB-SAR system uses frequency bandwidth 1-20GHz, and it can acquire full polarimetric radar signal. We found that frequency up to 20GHz can be usefully used for inspection, and show that the radar could achieve the resolution about 1.5cm, and found that t can be used for detection of crack inside concrete structure. Then Synthetic Aperture (SAR) Radar signal processing is used to reconstruct 3-dimenatioal images of inner structure of the targets. We found that the radar polarimetry gives us very precise information of the damaged structures, and demonstrated that radar polarimetry is a useful tool for detecting fractures inside a concrete structures, and detection of small deformation of wooden structures.

Apart from surface roughness effects the major impediment to accurate quantitative estimations of surface soil moisture content from SAR data is probably the presence of a vegetation cover attenuating the radar backscatter. In this paper, we report observations of vegetation effects on ALOS soil moisture products. A multi-temporal set of (quad/dual) polarimetric PALSAR images is used to investigate the effects of different agricultural crops on the L-band soil moisture retrieval. The vegetation effects on co-(HH) and cross-polarized (HV) backscattering coefficients, total power, as well as on the scattering entropy and alpha angle were analyzed based on correlation analysis between PALSAR observables and geophysical parameters. The impact on the soil moisture retrieval is investigated using the IEM and an empirical approach.

Disturbing effects caused by vegetation and surface roughness are major impediments to accurate quantitative retrievals of soil moisture from SAR. In this study we use PALSAR FBD images to investigate the potential of dual polarized L-band data to derive information on biomass and surface roughness. In correlation analyses between radar observables and in-situ measurements high sensitivities towards surface roughness and crop biomass could be ascertained. Based on these findings, we estimate surface roughness ks and vegetation biomass. The good quality of the estimates allows correcting the backscattering coefficients for the surface roughness and vegetation effects. The results give a promising outlook in terms of the possibility to develop an operational soil moisture retrieval model for PALSAR data collected in the FBD mode.