Recent Multimodal Large Language Models (MLLMs) exhibit strong zero-shot abilities but struggle with complex Grounded Situation Recognition (GSR) and are resource-intensive for edge device deployment. Meanwhile, conventional GSR models often lack generalization ability, falling short in recognizing unseen and rare situations. In this paper, we exploit transferring knowledge from a teacher MLLM to a small GSR model to enhance its generalization and zero-shot abilities, thereby introducing the task of Open-vocabulary Grounded Situation Recognition (Ov-GSR). To achieve this, we propose Multimodal Interactive Prompt Distillation (MIPD), a novel framework that distills enriched multimodal knowledge from the foundation model, enabling the student Ov-GSR model to recognize unseen situations and be better aware of rare situations. Specifically, the MIPD framework first leverages the LLM-based Judgmental Rationales Generator (JRG) to construct positive and negative glimpse and gaze rationales enriched with contextual semantic information. The proposed scene-aware and instance-perception prompts are then introduced to align rationales with visual information from the MLLM teacher via the Negative-Guided Multimodal Prompting Alignment (NMPA) module, effectively capturing holistic and perceptual multimodal knowledge. Finally, the aligned multimodal knowledge is distilled into the student Ov-GSR model, providing a stronger foundation for generalization that enhances situation understanding, bridges the gap between seen and unseen scenarios, and mitigates prediction bias in rare cases. We evaluate MIPD on the refined Ov-SWiG dataset, achieving superior performance on seen, rare, and unseen situations, and further demonstrate improved unseen detection on the HICO-DET dataset.

Segment Anything Model (SAM) is an innovative image segmentation model by Meta AI. It has demonstrated generalization capabilities across a diverse range of image datasets with zero shot. The objective of this study is to advance SAM applications in building detection for high resolution optical satellite imagery. The automated mask produced by SAM for RGB 3 channels of satellite imagery is initially introduced as an additional channel. Subsequently, the DeeplabV3+ deep learning model with Dilated ResNet C42 backbone is employed to train the combination of these four channels along with the ground truth of building masks for effective building detection. The SpaceNet dataset serves as the foundational training dataset, while a new dataset specific to Southeast Asia is incorporated for fine-tuning the model. Results demonstrate the model's notable success in building detection across various satellite images.

Methane is a major greenhouse gas besides carbon dioxide. It is important to monitor methane emission from natural and anthropogenic sources. In this study, we tested an inverse modeling algorithm for retrieving methane column density of a near-ground methane cloud. A synthetic dataset of top-of-atmosphere reflectance is generated using a radiative-transfer based forward model. In the retrieval procedure, the linearized reflectance model is solved by inverting the linear matrix equation with the pseudo-inverse technique. The logarithm of the surface reflectance is modeled by a third order polynomial within a narrow spectral window. For the test data set with column density up to 5000 ppmm, the retrieved column density correlates highly with the true values with R2=0.9834, average bias –290 ppmm and rms error 551 ppmm. The relative error is below 20% for column density above 1400 ppmm. Considering the sensor’s SNR and the detection model, the detection limit is about 1300 ppmm for a sensor with SNR 100, spectral bandwidth 10 nm and a spectral window for detection spanning from 2310 nm to 2380 nm.

Satellite and ground level observations during the period of August to October 2023 captured an elevated number of fire hot spots and biomass burning (smoke) activity over the South-East Asia landmasses of Sumatra and Borneo. Over this region, fire activity can be a yearly occurrence due to prevalent regional dryness and low rainfall, such activity can also be enhanced by other tropical phenomena like El Nino Southern Oscillation. In this paper, we present a multi—site photometric analysis of smoke emissions during this period with emphasis on aerosol physical and optical characteristics that would allow us to differentiate the source levels of emissions, aerosol composition and possibly transport pathways to downwind receptor sites.

Sentinel-1 Synthetic Aperture Radar (SAR) data proves valuable in detecting crop phenological stages. This study explores the effectiveness of time series SAR indices from Sentinel-1 in monitoring cotton phenology, an aspect not extensively investigated. During the process, seasonal extrema and breakpoints of SAR indices were derived over a cotton plantation in Brazil from December 2021 to July 2023. The results indicate that the seasonal maximum and minimum of Dual-polarization Radar Vegetation Index (DpRVIc) align with cotton emergence and flowering dates, respectively. Additionally, the seasonal minimum of Cross-Ratio (CR) and Gamma0_VH aligns with cotton boll formation and square formation dates, respectively. During evaluation, most detected phenology stages are within ±12 days of field observation dates. However, some emergence and square formation detections exhibit large outliers, necessitating further investigation. This study underscores the significant potential of time-series Sentinel-1 data for cotton phenology monitoring, providing valuable insights for improved agricultural practices.

Both PlanetScope and Sentinel-1 synthetic aperture radar (SAR) data have proven valuable in detecting crop phenological stages. However, the use of time series data from Sentinel-1, particularly derived SAR indices, for phenology detection has been underexplored. Additionally, leveraging phenology information from both Sentinel-1 and PlanetScope time series data has not been extensively investigated. This study examines the time series of Sentinel-1 backscatter coefficient, interferometric coherence, and several SAR indices, integrating them with PlanetScope data to improve the accuracy of cotton phenology detection. Seasonal extrema and breakpoints of backscattering coefficient, interferometric coherence, and SAR indices from Sentinel-1 and normalized difference vegetation index (NDVI) from PlanetScope over a cotton plantation in Brazil from December 2021 to July 2023. These retrievals were directly correlated with in situ observations and applied on a large scale for phenology detection. The results indicate that time series cross-ratio (CR), coherence VV, dual-polarization RVI (DpRVIc), and Gamma0_VH are useful for detecting cotton emergence, vegetative growth, square formation, flowering, and boll formation, respectively. Furthermore, integrating PlanetScope and Sentinel-1 significantly enhances phenology detection accuracy [Bias = 0.87 and root mean square error (RMSE) = 17.3] compared to using only Sentinel-1 (Bias = 16.05 and RMSE = 71.98). Generally, the RMSE for all four phenology stages is around one or two times the revisit frequency of Sentinel-1. This study underscores the potential of integrating PlanetScope and Sentinel-1 for large-scale phenology monitoring.

This paper examines our previously proposed gravitational filter and iterative multistage model-based polarimetric decomposition for an improved scattering retrieval from multi-look RADARSAT-2 C-band and ALOS-2 PALSAR-2 L-band data over a coastal region in the tropics. The decomposition results showed that a large amount of negative power pixels over vegetated areas were reduced. The retrieved scattering properties for different land cover types, namely, mangrove forest, oil palm, and rice paddy, were also analyzed and are discussed in this paper.

As an effective tool to investigate time-series of ground de-formation resulting from human activities or natural disas-ters, persistent scatterer (PS) interferometry requires a large amount of computing resources especially for continuous rou-tine monitoring over large areas. Powered by open-source software packages including the SentiNel Application Plat-form (SNAP) from ESA and the Stanford Method of Persis-tent Scatterer (StaMPS), a fully automatic PS-InSAR frame-work is built to process large-volume Sentinel-I data. The processing time is reduced tremendously with the paralleliz-able procedures in the end-to-end processing pipeline boosted by High Performance Computing (HPC) clusters. Embedding in the web GIS developed in CRISP, the framework also provides interfaces in browsers to visualize the resulting defor-mation time-series and ground displacement rate maps. The automatic framework executable in HPC also enables routine monitoring of ground subsidence over major cities in South-east Asia.

We present a new algorithm for retrieving fire temperature and radiative power with the short-wave infrared (SWIR) bands of high resolution satellite sensors. At the SWIR bands, solar input is usually large in comparison to the thermal emission from fires. We overcome this challenge by modeling the unknown ground reflectance with a few basis functions constructed from a spectral library. The algorithm is evaluated using a synthetic dataset generated by forward modeling of the solar reflectance and thermal emission from fires, including the atmospheric effects. The retrieved fire temperature and fire radiative power show high correlation with the actual values.

Speckle filtering is often carried out to improve the interpretation of polarimetric synthetic aperture radar (PolSAR) data for terrain classification, retrieval of physical scattering properties etc. As inspired by Newton's law of universal gravitation, a force of attraction between two image pixels is formulated in polarimetric-spatial domain. It is directly proportional to polarimetric similarity but inversely proportional to spatial distance. By using the attractive force as a filtering weight, this paper presents another variant of bilateral filter, called gravitational filter, for speckle reduction in PolSAR data. The performance of the gravitational filter was examined with ALOS-2 PALSAR-2 polarimetric data. Four aspects of the filtering performance were investigated: 1) speckle reduction, 2) image feature retention, 3) polarimetric property preservation, and 4) computational efficiency. The experimental results confirmed the effectiveness of the gravitational filter in the first two aspects, which was archived at the cost of an increased computational load. Applying to the four-look ALOS-2 PALSAR-2 test data, a bias in radiometric estimation was observed and can be reduced by using only intensity components when computing the attractive force.

This paper examines two test statistics, namely, likelihood ratio statistic and Roy's largest eigenvalue, for detecting changes in multi-look bi-temporal ALOS-2 PALSAR-2 fully polarimetric data. The latter turns into the well-known intensity ratio when applying to single-polarization data. The experimental results showed a comparable good performance of both the statistics in change detection. A more comprehensive change map was obtained from the fully polarimetric data compared with the use of only single-polarization data.

In this paper, we evaluate the performance of the power law model and the exponential model for retrieving CDOM absorption coefficient at 440 nm from 5 reflectance band ratios. Simulated water reflectance for the Sentinel-2 MSI sensor were computed using a semi-empirical water reflectance model. The models are evaluated in Four water types with various levels of CDOM, phytoplankton and suspended sediment. It was found that the power law applied to the band ratio of band 2 (443 nm) to band 4 (665 nm) has the best performance across all water types, except that the performance deteriorates for the water type with low CDOM but high chlorophyll content.

On 23rdJuly 2018 during an intense heat wave, two large fires broke out in Attika, a region in the central-southern part of Greece. Attika Regional Authorities declared a state of emergency and the European Union Civil Protection Mechanism was activated to oversee and manage the effects of this event. Subsequently, burn area delineation maps were produced as a way to evaluate the damage caused by these fires. For example, for July 26thmaps, which is near the fire episode's end, approximately 41 Km2were burned during this event. A side effect of these fires are the large emission of aerosol pollutants (smoke) and other gases generated during the burning process much of which can be very detrimental to human health. In this work and as part of an international collaborative validation effort on air quality methodologies; a WRF-Chem and source generation model was tasked to model the amount of gases and particles emitted during this event. Here in, we present a model perspective of the evolution and transport characteristics of aerosols particles and other gases (CO, NOx) generated during this intensive period of burning over Attika Greece.

A large-scale episode of biomass burning was observed during the months of August to October 2019. Satellite and in-situ observations reported scores of fires over the regions of central Sumatra and southern Kalimantan of Indonesia. Prevailing meteorological conditions contributed to long range transport of these aerosol particles and they were observed over several cities in Malaysia, Indonesia and Singapore. In this work, we report in-situ observations at two AERONET sites located in Singapore and Kuching (Sarawak, East Malaysia) complemented by satellite imagery (MODIS) over the above mentioned regions.

In this paper, a band-ratio model is developed and validated using data collected in the coastal waters of Singapore to obtain the absorption coefficient (at 440nm) of colored dissolved organic matter (CDOM) from Sentinel-2 images. Sentinel-2 is a suitable sensor for the monitoring of Singapore's coastal waters given the high temporal and spatial resolution which minimizes land and cloud contamination. A simple band-ratio model is used such that the model could be easily operationalized and allow for the useful monitoring of CDOM in this region. The leave-one-out-cross-validation scheme was used to validate the model due to the small number of data points. This model is further tested with coincident Sentinel-2 images within ± 24 hours from the in-situ measurements.

The recently developed Bruhn optical flow method has been found to improve imagery analysis results but has yet to be tested for cloud tracking purposes. We propose a cloud tracking model that is based on the Bruhn method and compare its performance with models based on more conventional optical flow methods. For lead-times spanning from 5s to 600s, the Bruhn based model has a comparable tracking accuracy when contrasted with current methods yet achieves a better than 30% improvement in computing processing time. We conclude that the Bruhn method represents a promising method for use in cloud tracking models.

We developed an algorithm for active fire detection using the Advanced Himawari Imager (AHI) on the Himawari-8 geostationary satellite. The algorithm is based on a stochastic detection model that aims to minimize a cost function derived from the probability density functions of the fire and background pixels. Using MODIS active fire product as the reference data, the optimal threshold was derived and the omission and commission errors were found to be 0.66 and 0.16, respectively.

Automated individual tree crown detection and delineation (ITCD) is essential for forest mapping, sustainable urban planning and 3D city model. In this paper, ITCD using high-resolution optical Worldview-2 satellite imagery over Singapore is presented. The object-based multiresolution approach for tree mapping including tree positions, crown sizes and canopy gaps is described. Accuracy assessment is performed with ground truth data.

A multi-stage four-component decomposition, which is inspired by the generalized odd-bounce and double-bounce scattering models, is presented in this paper for reducing negative power pixels. Its performance was evaluated by using NASA/JPL AIRSAR PolSAR C- and L-band data. The experimental results showed that a large amount of negative power pixels over vegetated areas were reduced.

An automated procedure to operationally derive water parameters from high spatial resolution coastal images is presented. The choice of imagery is Planet's 3 m resolution images with a potential daily revisit time. A semi-analytical model is adopted and modified using a 2-step process to retrieve the water parameters from Planet's 4-band images. Initial validation between the retrieved water depth and nautical chart depths show good results with on-going field measurements for validation of the retrieved water optical properties.

Mangrove forests constitute an important coastal ecosystem, which provides valuable ecosystem services such as coastal erosion protection, water filtration and shelters for a wide variety of plants and animals. Remote sensing satellite imagery provides valuable information for mangrove mapping and monitoring. In this study, we use high resolution images from SPOT-5, Landsat-7 and Landsat-8 satellites to perform change detection on an area of interest that covers the mangrove forests at the Ramsar sites of Pulau Kukup, Tanjung Piai and Sungai Pulai in the southwestern part of Johor State of Malaysia. Land use/land cover maps and the mangrove change map are generated to identify the mangrove distribution and temporal variation between 2000 and 2016. A hierarchical object-based classification method was used to produce the land use/land cover map in 2000, 2005, 2013 and 2016. The WorldView2 data were also used for validation of the classification results. This study presented an approach for mangrove mapping and change detection analysis. The map provides an up to date information for the study area and can be used for future comparative study.

Absolution radiometric calibration is the process of obtaining the relation between the at-sensor radiance and the pixel digital number recorded by the sensor. It is important to be able to calculate the radiance for every pixel in a satellite imagery, for comparison of data collected at different time, and for monitoring the radiometric stability of the sensor. We successfully performed radiometric calibration of TeLEOS-1 imaging sensor by inter-comparison with coincidental images collected by MODIS.

In the present study, the surface meteorological parameters over Singapore are simulated using WRF-ARW mesoscale model by varying the planetary boundary layer (PBL) parameterization schemes, horizontal resolutions and two land cover data sets (USGS and MODIS). Simulations are conducted with four nested domains having horizontal resolution of 27, 9, 3 and 1 km; 51 vertical levels by using the 1° × 1° NCEP final analysis meteorological fields for initial and boundary conditions. Eight days (20-28 January 2015) are selected for simulating various surface meteorological parameters. The model-simulated parameters of surface temperature, relative humidity, wind speed and wind direction are validated with the available observations over Singapore. It has been found that, improvements in predicting surface meteorological parameters with the increase in model resolution up to 3 km. The experiment with the 3 km grid resolution showed better simulated surface meteorological variables than that of 1 km resolution grid. Further, MODIS land cover data considerably improved the prediction of surface meteorological variables compare to the USGS. The surface meteorological variables simulated using the ACM2 PBL scheme with MODIS data are in better agreement with the observations showing least error statistics than the other PBL schemes used in the study. The better performance by ACM2 could be due to the non-local turbulence closure during unstable conditions and local-closure during stable conditions formulated in this scheme.

The empirical orthogonal function (EOF) technique is used to investigate the spatial and temporal distributions of fires in insular Southeast Asia. Different fire regimes were extracted by unsupervised classification of the second to fifth EOF components. In regions undergoing rapid land conversion processes, fires activities seem to occur all year round with no distinct fire seasons, with no correlation with climatic factor such as El-Nino.

We describe here the basic concepts for an operational system for deriving water depth and sea bottom albedo from high resolution multispectral satellite images. The uncertainties in depth derived from different satellites and at different time are examined. The uncertainty is generally within 20% of the depth value.

Over the last two decades and due to rapid economic growth, the South-East Asia region experiences mild to severe smoke episodes generated from annual biomass burning events, usually occurring during the regions's dry season that can last several months (June to October). Such a events poses a hazard to the region itself and specially to Singapore due to close proximity to the land masses of Sumatra and Borneo. In-situ aerosol measurements are now performed by our AERONET link photometer (Singapore and Kuching) sites. Such measurements readily provide optical characterization of atmospheric aerosols when biomass burning events occur. In this work, we present an analysis of three months of AERONET level 1.5 data recorded during August-October 2015 period. Besides the classical aerosol optical depth and Angstrom exponent number classifications, a special emphasis is paid to separate the fine/coarse mode part of the particle size distribution and by applying strict constrains to avoid cloud contamination and other artifacts.

A set of optimized parameters to compute the underwater remote sensing reflectance from the inherent optical properties covering a wide range of geometries was derived from Hydrolight simulations. The optimized values reduced the errors by 30 % at very steep zenith angles (40°-50°) as compared to published values that were derived for zenith angles of 10°.

Extraction of 3D building information from high resolution satellite imagery is generally carried out manually. Automatic algorithms to extract such information are highly desirable, but have so far met with limited success. Previously, we reported a highly automated method of extracting 3D building information from stereo pair high resolution satellite imagery by using normalized cross correlation (NCC) template matching technique. But the computation of depth is generally prone to error close to surface discontinuity with NCC template matching. In this paper, we present our results of using a semi-global matching (SGM) instead of NCC template matching technique to solve the problem.

The characteristics of sub-pixel fires in Suomi-NPP VIIRS data are studied using the Nightfire algorithm which attempts to fit the spectral radiance to a blackbody function. For small and/or low intensity fires, the fire temperature tends to be underestimated while the fire area are overestimated. The accuracy is improved when only the shorter wavelength bands are used in the fitting or when the whole spectrum is fitted to two blackbody functions representing the fire and the background components.

Over the last two decades and due to rapid economic growth, South-East Asia experiences mild to severe smoke episodes generated from annual biomass burning, usually occurring during the dry season period (from June to October). In Singapore, in-situ aerosol measurements are now performed by our AERONET sun photometer site since 2006. Such measurements readily provide optical characterization of atmospheric aerosols, such as optical depth and Angstrom number, as well as other physical properties such as refractive index, single scattering albedo, particle size distribution and scattering phase function when available. However, realistic aerosol phase functions can be quite complex and sometimes prohibitive to use in full radiative transfer calculations. This is due to the fact that many radiative transfer codes use Legendre polynomials and/or generalized spherical functions to represent the aerosol phase function. In such a cases, the number of expansion terms, for realistic aerosol particles, could be very large. In this work, a double Henyey-Greenstein function approximation is adopted to generate aerosol phase functions from level 1.5 AERONET measurements during smoke/smoke-free periods.

An object oriented approach was developed to derive land use/land cover classification maps over the island of Sumatra, Indonesia. The approach was developed to be used for Landsat 8 data over Sumatra. The accuracy assessment of the classified image used for development is about 90% while the test data yielded 80 %. The method will be applied to the data from Landsat 8 over the Sumatra to generate a 30 m resolution land cover classification map.

In this paper, the extreme precipitation events in the Southeast Asia region were studied using the TRMM 3B42v7 gridded daily precipitation product. The daily rain rate data were used to construct intensity-probability distribution curves for daily rainfall. Our results indicate that extreme precipitation events seem to occur more frequently than predicted by the intensity-probability curves during El-Nino years.

Persistent Scatterer InSAR (PS-InSAR) technique provides ground deformation monitoring in an accuracy of millimeter, due to its capability of overcoming the atmospheric disturbance, geometrical and temporal decorrelation. The Phased Array L-band Synthetic Aperture Radar (PALSAR) of ALOS (Advanced Land Observation Satellite) provides better capability to penetrate vegetation, which is especially useful in tropical area. In this paper, we have conducted the preliminary study on ground surface change monitoring in Singapore. 18 ascending PALSAR FBS/FBD level 1.1 data are used for this study, and the first result has identified several spots that might have small deformation, which need to be further investigated.

Classification of Very high-resolution optical sensors imagery into Land Cover/Land use map is an important and challenging task. The detailed visual information of land targets makes the classification task difficult and methods difficult to standardize. In this work, we use an objected orientated approach to classify worldview-2 images into Land Cover maps. The approach applies atmospheric correction to worldview-2 image, segmenting into various polygons and a rule-based decision table is applied to process the image into a land cover map.

The Straits of Malacca has rich biodiversity, but it experienced a dramatic decline due to the anthropogenic activities for economical development and natural disasters. The objective of this study was to generate the coastal and marine habitat map for the Straits of Malacca using the SPOT and Landsat data acquired at low tide in 2010 ~ 2012. Images were first processed with atmospheric and geometric correction. The hierarchical unsupervised classification and contextual editing were then used to produce the final habitat map. The WorldView2 and GeoEye data were also used for evaluating purposes in some studied area. Our classification scheme included sandy beach, mangrove forest, mud /sand flat, seagrass /algae, coral reefs, water and others. The produced habitat map provides the most up to date habitat information in the Straits of Malacca and will aid for the regional coastal ecosystems assessment as well as for future comparative study.

In this paper, we use the MODIS thermal anomaly product (hotspots data) to study the temporal and spatial patterns of vegetation fires in the western part of Insular Southeast Asia for a decade from 2001 to 2010. Fire occurrence exhibits a negative correlation with rainfall, and is more severe overall during the El-Nino periods. However, not all regions are equally affected by El-Nino. In Southern Sumatra and Southern Borneo the correlation with El-Nino is high. However, fires in some regions such as Riau and Jambi in the central part of Sumatra do not appear to be influenced by El-Nino. These regions are also experiencing rapid conversion of forest, especially those in peat areas, to large scale plantations.

The concentration of total suspended sediments (TSS) has often been empirically related to the above-surface remote-sensing reflectance, Rrs(λ). However, since a wide range of particle types exists in natural waters, a given TSS can often be the result of different particle size distributions (PSD). Directly relating TSS to Rrs(λ) disregards the effects of the PSD and may introduce inaccuracies in estimating TSS from Rrs(λ) measurements. This paper describes how the effects of the PSD on Rrs(λ) can be quantified for waters of various TSS, through numerical simulations. Results indicate that depending on the water types under investigation, these inaccuracies may be significant, exceeding 30% in the 500–600nm range in cleaner waters.

We report our first photometric measurements of aerosol optical depth from AERONET's mini-DRAGON sites at Singapore performed over the months of August and September 2012. Multi-spectral measurements of aerosol optical depth provide essential spectral information to obtain and/or retrieve the so-called Angstrom exponent number which is an essential parameter for inferring aerosol particle size regime. Based on the range of variability of Angstrom exponent number and aerosol optical depth, various aerosol types present in the local environment, can be identified. Special emphasis is placed on detecting the possible presence of external sources of aerosols such as from trans-boundary smoke originating from regional biomass burning episodes which is prevalent during this time of the year.

Above water reflectance measurements are usually made laboriously with handheld sensors. In recent years, autonomous robotic systems or fixed spectro-radiometers are used. In this work, we used off-the shelf optical systems with standard netbooks to make autonomous measurements of water reflectance. The optical systems are calibrated radiometrically and spectrally.

In this paper, we investigate the effects of water turbidity and water depth on the task of detecting submerged aquatic vegetation, such as seagrass, using WorldView-2 satellite images. We computed the reflectance spectra of submerged aquatic vegetation using a shallow water reflectance model for various water depth, water turbidity and benthic vegetation cover fraction on sandy substrate. The computed spectra were added with noise and used in a inverse modeling algorithm for retrieving the water depth, optical properties and water column corrected sea bottom reflectance. For low turbidity water (up to 25 NTU), the retrieved water column corrected NDVI values correlate very well with the vegetation cover fraction.

Over the past two decades, the Asia and South-Asia region has experienced a dramatic economic, industrial and population grow. Big cities are becoming large emission sources of anthropogenic aerosols resulting from the incessant industrial activity. In rural areas, clearance of large forested areas, via the method of burning, has resulted in severe smoke emission episodes during the 1996-2006 decade and has become an annual phenomenon specially in the South-East-Asia region. Large emission episodes can occur during periods of severe draught and exacerbated by the inter- annual El Ninõ events. Depending on weather patterns, smoke events can evolve into a persistent trans-boundary smoke with aerosol concentrations high enough to significantly reduce visibility and become a health hazard for local populations. During the month of October 2010, elevated levels of fire activity was detected by remote sensing satellites as well as by local in-situ measurements of fine particulate matter. In this work, we investigate the radiative impact of this smoke episode by firstly, analyzing the physical and optical properties of smoke particles with the aid of passive (Sun-photometer, AERONET), active and in-situ sampling of fine particulate (PM2.5) at our Singapore receptor site. Secondly, inversions of particle size distribution as well as single scattering albedo were used to evaluate the radiative impact of this biomass burning episode.

Lusi is a mud volcano in East Java, Indonesia, which started its eruption on 29thMay 2006 and never stopped. To study this volcano, we used SAR interferometry with ALOS/Palsar satellite images from 2006 to 2010. By creation of a set of interferograms, and suppression of unwanted components, such as Earth curvature and elevation, we were able to compute the motion that occurred between two dates. The obtained set of motions shows that the region of Lusi is undergoing a general subsidence of several metres. However, as this region is very wet, more precise results, on the volcano itself, can be obtained by GPS campaigns.

The recently launched Worldview-2 satellite has 8 high resolution (2-m) spectral bands in the visible to near-infrared region. The additional spectral bands provide an opportunity to test the spectral matching algorithm for retrieving the water depth, bottom albedo and intrinsic optical properties of coastal sea water. Our test area is an intertidal zone and reefs near an off-shore island of Singapore consisting of seagrass, corals and mangrove habitats. Examples of applying the spectral matching method to submerged coral reefs are shown. The retrieved water depth and bottom albedo are utilized together with the reflectance spectra of various habitat classes for better classification and mapping of the coastal habitats.

Mangroves are an important part of the ecosystem in tropical region. Unfortunately, they are also under intense ecological pressure from fishing, tourism or logging. As they are often in not easily accessible places and scattered over large areas, satellite observation is an ideal solution to monitor the mangrove evolution over the past few years. However, the mapping of mangrove from satellite images is a difficult task and mostly done manually. Here we propose a detection method based on support vector machine, exploring more than 100 features, providing a good accurary, enabling the mangrove expert to focus on the most difficult areas.

The empirical orthogonal function (EOF) analysis is used to investigate the spatial and temporal variations of precipitation rate measured by the Tropical Rainfall Measuring Mission (TRMM) satellite in the Southeast Asian region. The EOF analysis decomposes the rainfall data into several modes of variations. Each mode can be associated with one or several mechanisms of variations. The most dominant mode is associated with annual seasonal monsoons while the second and third modes have moderate El-Nino influences. The precipitation rate does not seem to have significant correlation with the global warming index.

Baseline calibration is needed in most of SAR interferometry processing. An iterative optimization of baseline with constrain of relative platform position is presented in this paper. The SAR passes which gives inaccurate platform position is successfully detected and calibrated using this algorithm. After processing, new estimated baseline improves the quality of interferogram. Existence of reference Digital Elevation Model (DEM) error and atmospheric phase screen (APS) can also be detected from the convergence value. This method is based on a reversed concept of platform position estimation from interferometric result. Validation of method performed on multiple SAR images over Singapore.

Optical remote sensing images of urban areas usually contain large amount of line-segments that represent the boundaries of man-made objects. The extraction and vectorization of these line-segments are desirable for many applications, such as digital city reconstruction. This paper presents a “Sweep & Scan” algorithm for the extraction and vectorization of these line-segments. The “Sweep” part of the algorithm uses the Wallace parameterization of the Muff Transform to identify straight lines globally, and the “Scan” part uses edge continuation and intensity correlation to identify line-segments locally.

A method for retrieving water turbidity and suspended sediment concentration of inland and near-shore coastal waters using high resolution remote sensing satellite data is described. Backscattering coefficient at the near infrared band is first derived from the satellite images using a quasi-analytical algorithm after applying a simple atmospheric correction routine. A turbidity map is obtained by applying an empirical relation derived from laboratory experiments to convert the backscattering coefficient to nephelometric turbidity units (NTU). This method can be operationalized for global water quality monitoring using currently operating satellite sensors such as those on Landsat, SPOT and Terra / Aqua MODIS.

In this paper we first present the methodology and accuracy assessment of a 500 m spatial resolution land cover map m insular Southeast Asia. This is followed by an analysis of the usability of the map for monitoring of deforestation/forest degradation. The map was produced using Moderate Resolution Imaging Spectroradiometer (MODIS) images (acquired 1stJan-2ndM 2007), elevation information and peatland maps. The map covers the Malaysian Peninsula and the major islands of Sumatra, Java and Borneo, m addition to numerous smaller islands. The classification scheme of 12 classes reflects the special characteristics of land cover m insular Southeast Asia. We conclude that the map is suitable for monitoring the extent of forest cover m this region, but it misses the important aspect of forest degradation due to its inability to present varying forest degradation levels. Further development of the methodology is m progress to overcome this limitation.

Since October 2006, the Center for Remote Imaging, Sensing and Processing (CRISP) of the National University of Singapore has joined the Aerosol Robotic Network, AERONET, to monitoring aerosol optical properties in this region. This initiative complements the already existing activities on satellite remote sensing performed at CRISP. In Singapore, the sources of aerosols are mostly from fossil fuel burning (energy stations, incinerators, cars etc.) and from the industrial and urban areas. The proximity to the sea adds a possible oceanic source and depending on the time of the year, there can be a strong bio-mass component coming from forest fires from various regions of the neighboring countries. In this work, we present a survey of aerosol optical depth (AOD) and Angstrom exponent retrievals at several regions of Singapore. AOD measurements were performed with the help of AERONET's CIMEL Sun-photometer (located at CRISP) and a versatile hand held Microtops II Sun-photometer. In order to identify the degree of geographical variability of aerosol optical properties, we collected data samples over selected sites across the island. Furthermore, we extracted concurrent data points from our own AERONET site in order to check for consistency and validation of our results. This study has enabled us to identify possible aerosol sources and understand the nature and physical characteristics of atmospheric aerosols across Singapore.

In this paper, we implemented a spectrum matching method for retrieving the inherent optical properties, water depth and sea bottom reflectance from measurements of remote sensing reflectance for turbid coastal waters. In the spectrum matching method for inversion, a database of reflectance spectra for various water conditions is first generated. The measured reflectance spectrum is then compared with each computed spectrum in the database. The water conditions corresponding to the best-match spectrum in the database is taken as the solution of the inverse problem. For water depths between 3.0 and 5.0 m, the absolute error of depth retrieval was about 0.35 m. The retrieved absorption coefficients at 412 nm were within 32% of the measured ones while the scattering coefficients at 555 nm were within 34%. The retrieved bottom reflectance spectra was similar to that of a sandy surface.

Traditional building height estimation from a stereo pair of satellite images relies heavily on the selection of building primitives by the human operator. It is labor-intensive and highly inefficient. In this work, we propose a system to automate the estimation of 3-D building height from an IKONOS stereo pair. Minimum user intervention is required to outline the building top in the left image. Stereo correspondence is automatically established through a multi-scale feature (coarse-to-fine) matching, which is capable of achieving sub-pixel accuracy. The height of the building top is computed using the Rational Polynomial Coefficient (RPC) model. Experimental results show that the proposed method is efficient, robust and it produces accurate building height estimation.

The December 2004 Sumatra earthquake triggered a massively destructive tsunami that affected several coastal areas in the Indian Ocean. In this study, very high resolution IKONOS satellite images acquired before and after the tsunami were used to study the changes of the coastal morphology in the Northwestern coast of Sumatra. The coasts were eroded during the tsunami. At some stretches of the coast, the coastline receded by as much as 500 m. About 13 months after the tsunami, new beaches were formed at many areas along the coastline. The widths of the newly formed beaches exceeded those of the pre-tsunami coast. In areas where anthropogenic modifications have occurred before the tsunami, the coastline remained eroded after 13 months and no evidence of recovery was observed.

Analysis of burn scar pattern variability in insular Southeast Asia was carried out to evaluate its effect on regional burnt area mapping with medium resolution (250-500 m) satellite imagery. Burnt area maps based on 16 high resolution SPOT 4 and 5 images were overlaid on land cover and soil maps. Although the number of small burn scars (<25 ha) was found to be high throughout the region, the proportion of total burnt area found in small scars varied remarkably (from 3% to 97%) between study sites. Unlike land cover, soil type was found to be a good indicator for usability of medium resolution burnt area mapping. A medium resolution burnt area mapping simulation (where 500times500 m grid cells containing at least 50% burnt were considered detected as burnt) resulted in 86% detected burnt area in wetlands (peat and alluvial soils) as opposed to only 33% in other areas.

Satellite remote measurements taken over the ocean surface, often contain information of near-surface or sub-surface processes, which occur on, or over the ocean. These remote measurements, can enable as to retrieve important parameters such as near surface wind, sea surface temperature and salinity and underwater bathymetry, which, all, in one way or another, depend on how well we understand sea surface reflection and roughness. However, the quality of these retrievals are often limited by the amount of scattering, emission and absorption of electromagnetic radiation caused by atmospheric aerosols which often shows dependency on polarization. From a remote sensing perspective, it is very well known that at off-nadir incidence angles and wavelenghts longer than 3.6 mum, the radiance originated from the sea surface becomes partially vertically polarized. Similarly, at shorter wavelenghts (shorter than 3.6 mum), the total radiance at off-nadir angles can become strongly horizontally polarized , . Since at shorter wavelengths, the reflected skylight becomes more significant; it is important to quantify the contribution that atmospheric polarization has on the top of the atmosphere remotely observed sea surface reflection, specially near areas of enhanced surface reflection such as from the Sun glitter. Furthermore, surface phenomena, such as wind-generated waves, gravity and capillary waves, internal waves, surface currents, density fluctuations etc., modulate the ocean's surface roughness which has a direct effect on sea surface reflection. This modulating effect is represented by the so called the mean square slope (MSS) of ocean surface waves. In general, the MSS is mainly related to short wind waves which are strongly dependent on wind speed, wind direction, surface waves and air sea temperature. To understand these wind and wave related effects on the Sun glint pattern, Salinas [3] have developed an atmospheric radiative transfer model that incorporates a new wind-wave description for the mean square slope of the sea surface. This model solves the scalar radiative transfer equation and includes wind interaction and wave states, such as wave age, as the main factors contributing to surface roughness. In this paper, we extend the above mentioned work and include light polarization effects originating from aerosol and molecular scattering and from the roughness of the surface. Polarization is considered by using the 4-Stokes vector formalism and the 4x4 scattering phase matrix for aerosols and molecules. Simulation results showing polarization effects on the simulated, top of the atmosphere, Sun glint pattern are presented. Initial simulations performed at low wind speed show increasing levels of linear polarization with increasing values for the wave age parameter. However, at higher wind speed, lower levels of linear polarization is obtained for similar values of the wave age parameter. On the other hand, calculations performed at near surface levels, show low linear polarization and a gradual shift of the Brewster angle with increasing wind speed and wave age. These results, although initial, will be carefully analyzed and compared with limiting cases where available. Finally, the present model can have possible applications to the retrieval of wind and wave states, such as wave age, near a Sun glint region and for the study of of ocean surface roughness characteristics arising from air-sea interaction.

The 2004 Sumatra earthquake, together with the tsunami that followed, caused massive destruction in many coastal areas of the Indian Ocean. Uplift and subsidence of land near the epicenter of the earthquake in the Andaman Sea have been reported. In this letter, we report observations of the uplift in high-resolution SPOT-5 satellite images in a coral island near the Andaman Islands. The depth of the submerged reefs was measured using water reflectance models commonly employed in the ocean color research community. The uplift magnitude was estimated by the difference in water depth measured in the pre- and post-earthquake images, taking into account the difference in tide levels. The uplift was found to be 1.6 m with an uncertainty of about 0.3 m. The methodology could be widely used for efficient estimation of shallow water depth of broad areas, particularly over shallow coral reefs.

In this paper, we present a method for atmospheric correction that uses the cloud and shadow image features. An iterative scheme is formulated to computes the ratio of diffuse to direct irradiance as well as the path radiance using the radiance detected over the open water and shadow pixels. These parameters are then used to compute the cloud and water reflectance. We implemented this method on IKONOS images which are known to have inferior signal to noise ratio compared to sensors specially designed for ocean color measurements. The corrected image reflectance over water pixels are compared to field measurements.

The 2004 Sumatra earthquake, together with the tsunami that followed, caused massive destruction in many coastal areas of the Indian Ocean. Uplift and subsidence of land near to the epicenter of the earthquake in the Andaman Sea have been reported. In this paper, we report observations of the uplift in high resolution SPOT-5 satellite images in a coral island near to the Andaman Islands. The depth of the submerged reefs can be measured using water reflectance models commonly employed in the ocean color research community. The uplift magnitude is estimated by the difference in water depth measured in the pre- and post-earthquake images, taking into account the difference in tide levels. The uplift is found to be 1.6 m with an uncertainty of about 0.3 m.

Texture features have often been used in remote sensing applications to retrieve spatial regularities for images. Among existent texture analytical tools, grey-level occurrence matrix (GLCM) is effective for high resolution optical imagery (e.g. Ikonos) and widely accepted for the classification and analysis of texture features. On May 27th 2006, a Mw 6.3 earthquake took place in Java, Indonesia, which destroyed 135,000 homes. Texture information was retrieved using GLCM from Ikonos panchromatic images, which were taken in the earthquake-affected areas both before and after the earthquake, to evaluate the damage of local residences.

We present a method to retrieve the absorption and backscattering coefficients of water from remote-sensing reflectance. Our method is based on semi-empirical modeling but does not require non-linear optimization methods which are sensitive to initial conditions. Instead, the equations are linearized and so can be solved using matrix inversion. The linearization is accomplished by two ways. Firstly, a linearized form of the basis function for phytoplankton pigment absorption is created. Secondly, the equations are solved for different spectral slopes, s, in the exponential function model for the absorption coefficient of dissolved organic matter and detritus, as well as, different spectral shapes, y, in the backscattering model. The linearized model is applied to the blue-green hyperspectral channels between 470 and 520 nm. The retrieved parameters are used to predict the remote-sensing reflectance for the channels in the wavelength bands between 600 and 650, and between 750 and 800 nm. The modeled values are compared to the actual values. In this way, the best fit s and y can be determined. The spectral bands 470-520, 600-650 and 750-800 nm have been chosen so as to minimize the effects of bottom reflectance in coastal waters. We tested our method on both synthetic data and field measurements.

We used high-resolution SPOT 5 images acquired before and after the tsunami occurred in the Indian Ocean on December 26, 2004 to perform the damage assessment using image processing and GIS techniques to study the effects of the tsunami on coastal features. Our study area covers all tsunami affected regions along the western coast of Sumatra. A landuse map was created based on the pre-tsunami SPOT 5 images and a damage map was produced from pre and post-tsunami SPOT 5 images. IKONOS images of the affected areas were used as ground truth data. Our damage assessment map shows that much of the low land area, such as aquaculture ponds, cities and villages near the coast, paddy fields and mangrove etc., were severely damaged. Coastline features were significantly eroded and destroyed. This paper demonstrates that the integration of remote sensing and GIS is an effective approach for tsunami damage assessment.

10.1109/IGARSS.2005.1525386

10.1109/IGARSS.2005.1526364

For many oceanic studies, it is required to know the distribution of visible solar radiation (EPAR) in the upper water column. One way to reach this is by remote sensing. This includes two components: First, EPAR at surface is calculated based on atmosphere properties along with the position of the Sun. Second, the vertical attenuation of EPAR (KPAR) is derived from products of ocean-color remote sensing. Currently, KPAR is estimated based on chlorophyll concentration ([C]) from ocean color. This kind of approach works well for waters where all optical properties can be adequately described by values of [C], but will result in large uncertainties for coastal waters where [C] alone cannot accurately describe the optical properties. In this paper, we present an innovative model that describes KPAR as a function of water's inherent optical properties (IOP).

10.1109/IGARSS.2005.1526635

Active fire detection using satellite thermal sensors usually involves thresholding the detected brightness temperature in several bands. Most frequently used features for fire detection are the brightness temperature in the 4-μm wavelength band (T4) and the brightness temperature difference between 4- and 11-μm bands (/spl Delta/T=T4-T/sub 11/). The task of active fire detection is examined in the context of a stochastic model for target detection. The proposed fire detection method consists of applying a decorrelation transform in the (T4,/spl Delta/T) space. Probability density functions for the fire and background pixels are then computed in the transformed variable space using simulated Moderate Resolution Imaging Spectroradiometer (MODIS) thermal data under different atmospheric humidity conditions and for cases of flaming and smoldering fires. The Pareto curve for each detection case is constructed. Optimal thresholds are derived by minimizing a cost function, which is a weighted sum of the omission and commission errors. The method has also been tested on a MODIS reference dataset validated using high-resolution SPOT images. The results show that the detection errors are comparable with the expected values, and the proposed method performs slightly better than the standard MODIS absolute detection method in terms of the lower cost function.

Many studies have been carried out to retrieve water quality measures like concentrations of suspended matter and chlorophyll from Landsat TM images. Such studies typically use extensive in-situ measurements and empirical data fitting (usually correlations) to retrieve concentrations of the water constituent under investigation. One disadvantage of such empirically derived algorithms is that they tend to be applicable only to the water constituent, the particular location and water type investigated. In addition, the atmospheric effects are either not taken into account or a uniform atmosphere is assumed. On the other hand, the backscattering and absorption coefficients, have been shown to determine the water color and hence, the measured reflectance. We report on our attempt to retrieve these inherent optical properties from Landsat ETM+ data. We show how we can correct for the atmospheric effects using Band 4. Using current ocean color models, we retrieved the backscattering and absorption coefficients from Bands 1, 2 and 3 for the coastal waters around Singapore. The derived backscattering and absorption coefficients are within the range of values retrieved using Hyperion data and those from in-situ measurements. The advantage of this method is that minimal empirical data are needed and so it can be easily applied to other similar data sets. However, the precision of the retrieved coefficients is necessarily coarse.

Active fire detection using satellite thermal sensors usually involves thresholding the detected brightness temperature in several bands. Most frequently used features for fire detection are the brightness temperature in the 4-micron wavelength band (T4) and the brightness temperature difference between the 4-micron and 11-micron bands (dT=T4-T11). In this paper, the task of active fire detection is examined in the context of a stochastic model for target detection. Probability density functions for the fire and background pixels are computed in the (T4, dT) feature space using simulated MODIS thermal data under different atmospheric humidity conditions and for cases of flaming and smoldering fires. The Pareto curve for each detection case is constructed. Optimal thresholds are derived by minimizing a cost function, which is a weighted sum of the omission and commission errors

In the humid tropics, humidity is high implying a high water vapour content in the atmosphere. This causes strong attenuation in the radiation measured in the thermal infrared as the main absorption in this region is due to water vapour. In this study Land Surface Temperature (LST) is retrieved from single band infrared (IR) band through the modeling of atmospheric transmission and thermal emission given only surface parameters. Comparison of the LST retrieved from single IR bands within the same dataset and cross comparison with the standard MODIS LST Product gives high correlation indicating the validity of the models adopted for the atmosphere

In developing a compositing methodology specifically designed for the humid tropics, targeting the removal of cloud and haze effected pixels, a new method was devised, maximum Normalized Difference Blue Index (NDBI). This method uses the blue band in place of the red band of the normal NDVI equation. The NDBI technique provided a compositing result with considerably less speckle (noise) than the maximum NDVI composite product. The maximum NDBI result also had less variation present within a given vegetation type, allowing one to distinguish more easily between different vegetation classes. This clarity between vegetation types is an essential property of the final composite product, as it will be used for vegetation and land use classification.

The Southeast Asia region is always under partial or complete cloud covers. As a result, optical remote sensing images of this region always contain cloud covers and associated shadows. In this paper, we report an operational method for generating a "cloud-free" image mosaic of 1-m resolution IKONOS images. The inputs are several cloudy images acquired by the IKONOS satellite over an area of interest. This technique is especially useful in tropical regions with persistent and extensive cloud cover.

This study employs a spectral fitting technique to retrieve optical parameters of coastal seawaters from hyperspectral reflectance spectra. The model used includes both the atmospheric effects and the scattering and absorption of light by sea water constituents. The problem of low SNR in Hyperion data can be overcome by averaging over a larger area (10 /spl times/ 10 pixels). The retrieved water reflectance correlates well with the MODIS data. However, chlorophyll absorption correlates poorly with both the MODIS derived chlorophyll concentration and OC4 concentration. The results support the general observations that the band ratio based algorithms do not work well in coastal regions due to interference from colored dissolved organic matter (CDOM) absorption.

A study on the relationship between fire and land cover change was carried out in Riau province, Sumatra, Indonesia. Multispectral full resolution SPOT mosaics were used in visual land cover classification and burnt area estimation for 1998, 2000 and 2002. Preliminary analysis of the data with active fires detected on SPOT images show that fire is very strongly connected with land cover change in primary vegetation and somewhat less in secondary growth. However, this connection can not be seen in managed land cover types. The primary vegetation on the study area diminished about 5% per year. The main directions of land cover change went via secondary growth towards plantations and small-holder dominated mosaic.

The Landsat MSS, TM and ETM+ data acquired in the year of 1975, 1989 and 2001 were used to quantify land cover change in the Minjiang area. Knowledge-based stratified classification was performed for each image and change detection was analysed. Profound changes in land cover, including widespread deforestation and extension of agricultural land, have been observed, notably along the river valley. Such a large reduction in forestland cover might be expected to lead to a significant impact on hydro-geomorphic processes. An attempt was made to identify gully erosion, one of the severe geomorphic processes after reduction in vegetation cover, using the Landsat image.

The use of C- and L-band polarimetric synthetic aperture radar (POLSAR) data for classifying land cover features in a tropical area is investigated in this study. The POLSAR data were acquired during the NASA/JPL PACRIM-1 mission over the northern part of Peninsular Malaysia on 3rd December 1996. Prior to classification, the Lee polarimetric filter was applied to the complex covariance matrix for speckle suppression. In unsupervised classification, the scattering mechanism of each pixel in the speckle-suppressed images was analyzed and grouped into one of the three categories: (1) odd-bounce, (2) even-bounce, or (3) diffuse scattering. Training samples were then generated from the outputs of the unsupervised classification, to be used in subsequent supervised classifications of various frequency and polarization combinations. The Kappa statistics computed for classification using single-frequency fully polarized C- and L-band data were 0.69 and 0.73, respectively. An improvement to 0.79 was achieved by using the dual-frequency (combined C and L bands), fully polarized data in the classification.

A method is explicated for producing cloud free and cloud-shadow free images from SPOT panchromatic images over a given region. It is well known that optical remote sensing images often encounter the problem of cloud covers, especially over the humid tropical areas. Cloud shadow accompanied with cloud is also the problem. The problem can be partially overcome by acquisition of multiple images at different time over a given region. By mosaicking the cloud free and cloud-shadow free areas in the set of images, a reasonably cloud free and cloud-shadow free composite scene can be made. This paper relates to an algorithm for generating cloud free and cloud-shadow free multi-scene mosaics of SPOT panchromatic images.

This work aims to extract optical parameters of coastal sea water by fitting the hyperspectral reflectance spectra acquired from the Hyperion instrument on-board the EO-1 satellite to a coupled sea water reflectance and atmospheric transmission model. This method is able to retrieve the absorption and scattering coefficients of the various components of coastal sea water.

In this paper, we attempt to perform land cover classification using hyperspectral data acquired by the EO-1 Hyperion instrument over two test sites in the tropical region: one in Singapore and the other one in coastal Jambi on the Sumatra island of Indonesia. Atmospheric correction on the hyperspectral imagery was first performed using a commercial package. Principal component decomposition was then performed and an unsupervised ISODATA classification was carried out on the dominant components to produce a land cover classification map for each test site. Classification using the pixel unmixing method as implemented in the ENVI package was also performed. The results of classification were compared with existing land cover maps.

The ability to retrieve information at subpixel spatial resolution from a satellite with more than one channel in the thermal infrared spectral range has benefited the monitoring of forest fires. With better sensors, the severity of fire and the degree of damage can be estimated by retrieving fire fraction and fire temperature using the Dozier procedure. A sensitivity analysis was also performed to estimate the uncertainty in the retrieval of fire fraction and fire temperature as a result of inaccurate determination of emissivity and background temperature.

The authors report the use of multitemporal ERS-2 satellite synthetic aperture radar (SAR) images in delineating and mapping areas under different rice cropping systems in the Mekong River Delta, Vietnam. Change index maps were generated from seven images acquired between May and December 1996. Using a 3-dB threshold, the pixels in each change index (CI) map were classified into one of three classes: Increasing, decreasing, or constant backscattering. Five of the CI maps were used to generate a composite map with 243 possible change classes. These change classes were grouped into thematic categories of rice cropping systems using two methods: human visual inspection and a semiautomatic hierarchical clustering algorithm. The derived thematic maps were compared with SPOT scenes acquired during the same rice seasons.