On the upstream inputs into the MODIS primary productivity products using biometric data from oil palm plantations

This study evaluated the influence of upstream inputs into the Moderate Resolution Imaging Spectroradiometer (MODIS) primary productivity products, termed the MOD17, at tropical oil palm plantations (Elaeis guineensis Jacq.). Evaluation of MOD17 using oil palm plantations as test sites is ideal because the plantations are cultivated on large areas which are comparable with the size of MODIS pixels. It is difficult to find test sites covered by other single species in a whole pixel. The upstream inputs studied included (1) MODIS land cover, (2) the National Centers for Environmental Prediction–Department of Energy (NCEP-DOE) Reanalysis 2 meteorological data set, (3) MODIS leaf area index/fraction of photosynthetically active radiation (LAI/fPAR), and (4) MODIS maximum light-use efficiency (maximum LUE). Oil palm biometric and local meteorological data were utilized as ground data. Furthermore, scaling up oil palm LAI and fPAR from plot scale to regional scale (Peninsular Malaysia) was done empirically by correlating oil palm LAI derived from the hemispherical photography technique with radiance information from the Disaster Monitoring Constellation 2 satellite (UK-DMC 2). The upscaled LAI/fPAR developed in this study was used to evaluate the MODIS LAI/fPAR. The results showed that the MODIS land-cover product has an overall accuracy of 78.8% when compared to the Peninsular Malaysia land-use map produced by the Department of Agriculture, Malaysia. Regarding the NCEP-DOE Reanalysis 2 data set, vapour pressure deficit (VPD) and photosynthetically active radiation (PAR) contain large uncertainties in our study area. However, MODIS LAI and fPAR were correlated relatively well with the upscaled LAI (R² = 0.50) and the upscaled fPAR (R² = 0.60), respectively. The constant values of maximum LUE for croplands and evergreen broadleaf forest ecosystems are lower than the maximum LUE of oil palm. The relative predictive error assessment showed that the MOD17 net primary productivity (NPP) overestimated oil palm NPP derived from biometric methods by 142–204%. We replaced the upstream inputs of MOD17 by the local inputs for estimating oil palm GPP and NPP in Peninsular Malaysia. This was done by (1) assigning maximum LUE for oil palm plantations as a constant at 1.68 g C m^?2 day^?1, (2) utilizing meteorological data from local meteorological stations, and (3) using the upscaled fPAR of oil palm plantations. The amount of oil palm GPP and NPP for Peninsular Malaysia in 2010 were estimated to be ~0.09 Pg C year^?1 (or equivalent to ~0.33 Pg CO₂ year^?1) and ~0.03 Pg C year^?1 (~0.11 Pg CO₂ year^?1), respectively, indicating that oil palm plantations in Peninsular Malaysia can play an important role in global carbon sequestration. In the future there is likely to be a demand for MODIS GPP and NPP products that are more accurate than those currently generated by MOD17. We recommend future developments of the MOD17 processing system to allow improvements in the upstream input parameters, in the manner described in this article, both for global processing and for the production of more accurate values for GPP and NPP at regional and local scales.     

Modelling and mapping of above ground biomass (AGB) of oil palm plantations in Malaysia using remotely-sensed data

Estimating accurate above ground biomass (AGB) of oil palm plantation in Malaysia is crucial as it serves as an important indicator to assess the role of oil palm plantations in the global carbon cycle, particularly whether it serves as carbon source or sink. Research on oil palm AGB in Malaysia using remote sensing is almost insignificant and it has known that remote sensing provides easy, inexpensive and less time consuming over larger areas. Therefore, this study focuses on evaluating the potential of Landsat Thematic Mapper (TM) data with combination of field data survey to predict AGB estimates and mapping the oil palm plantations. The relationships of AGB with individual TM bands and various selected vegetation indices were examined. In addition, various possibilities of data transform were explored in statistical analysis. The potential models selected were obtained using backward elimination method where R², adjusted R² (R²_adj), standard error of estimate (SE_E), root mean squared error (RMSE) and Mallows’s C_p criterion were examined in model development and validation. It was found that the most promising model provides moderately good prediction of about 62% of the variability of the AGB with RMSE value of 3.68 tonnes (t) ha^-1. In conclusion, Landsat TM offers the low cost AGB estimates and mapping of oil palm plantations with moderate accuracy in Malaysia.     

Canopy shadow in IKONOS satellite observations of tropical forests and savannas

The biological and structural complexity of tropical forests and savannas results in marked spatial variation in shadows inherent to remotely sensed measurements. While the biophysical and observational factors driving variations in apparent shadow are known, little quantitative information exists on the magnitude and variability of shadow in remotely sensed data acquired over tropical regions. Even less is known about shadow effects in multispectral observations from satellites (e.g., Landsat). The IKONOS satellite, with 1-m panchromatic and 4-m multispectral capabilities, provides an opportunity to observe tropical canopies and their shadows at spatial scales approaching the size of individual crowns and vegetation clusters.We used 44 IKONOS images from the Large-Scale Biosphere-Atmosphere Experiment in Amazonia (LBA) data archive to quantify the spatial variation of canopy shadow fraction across a broad range of forests in the Brazilian Amazon and savannas in the Brazilian Cerrado. Forests had substantial apparent shadow fractions as viewed from the satellite vantage point. The global mean (±S.D.) shadow fraction was 0.25±0.12, and within-scene (e.g., forest stand) variability was similar to interscene (e.g., regional) variation. The distribution of shadow fractions for forest stands was skewed, with 30% of pixels having fractional shadow values above 0.30. Shadow fractions in savannas increased from 0.0±0.01 to 0.12±0.04 to 0.16±0.05 for areas with woody vegetation at low (<25% cover), medium (25-75%), and high (>75%) density, respectively.Landsat-like observations using both red (0.63-0.70 μm) and near-infrared (NIR) (0.76-0.85 μm) wavelength regions were highly sensitive to sub-pixel shadow fractions in tropical forests, accounting for ∼30-50% of the variance in red and NIR responses. A 10% increase in shadow fraction resulted in a 3% and 10% decrease in red and NIR channel response, respectively. The normalized difference vegetation index (NDVI) of tropical forests was weakly sensitive to changes in shadow fraction. For low-, medium-, and high-density savannas, a 10% increase in shadow fraction resulted in a 5-7% decrease in red-channel response. Shadows accounted for ∼15-50% of the overall variance in red-wavelength responses in the savanna image archive. Weak to no relationship occurred between shadow fraction and either NIR reflectance or the NDVI of savannas. Quantitative information on shadowing is needed to validate or constrain radiative transfer, spectral mixture, and land-surface models used to estimate material and energy exchanges between the tropical biosphere and atmosphere.     

Mapping oil palm extent in Malaysia using ALOS-2 PALSAR-2 data

The extent of oil palm plantations has increased rapidly in Malaysia over the past few decades. To evaluate ecological effects and economic values, it is important to produce an accurate oil palm map for Malaysia. The Phased Array Type L-band Synthetic Aperture Radar (PALSAR) on the Advance Land Observing Satellite (ALOS) is useful in land-cover mapping in tropical regions under all-weather conditions. In this study, PALSAR-2 images from 2015 were used for oil palm mapping with maximum likelihood classifier (MLC)-based supervised classification. The processed PALSAR-2 data were resampled to multiple coarser resolutions (50, 100, 250, 500, and 1000 m), and then used to investigate the effect of speckle in oil palm mapping. Both independent testing samples and inventories from the Malaysia Palm Oil Board (MPOB) were used to evaluate the mapping accuracy. The oil palm mapping result indicates 50–500 m to be a good resolution for either retaining spatial details or reducing speckle noise of PALSAR-2 images. Among which, the best overall mapping accuracies and average oil palm accuracies reached 94.50% and 89.78%, respectively. Moreover, the oil palm area derived from the 100-m resolution map is 6.14 million hectares (Mha), which is the closest to the official MPOB inventories (~8.87% overestimation).     

Random Forest classification model of basal stem rot disease caused by Ganoderma boninense in oil palm plantations

Ganoderma boninense is a fungus that causes basal stem rot (BSR) disease in oil palm plantations. BSR is a major disease in oil palm plantations in both Indonesia and Malaysia. There is no effective treatment for curing BSR; current treatments only prolong the life of oil palms. One strategy to control BSR is early detection of G. boninense infection. Based on the infection symptoms, many researchers have applied remote-sensing techniques for early detection and mapping of BSR disease in oil palms. The main objectives of this article were to evaluate the potential of machine-learning models for predicting BSR disease in oil palm plantations and to produce maps of the distribution of BSR disease. QuickBird imagery archived on 4 August 2008 was applied in three classifier models: Support Vector Machine, Random Forest (RF), and classification and regression tree models The RF model was best at predicting, classifying, and mapping oil palm BSR in terms of overall accuracy (OA), producer accuracy, user accuracy, and kappa value. Using 75% of the data for training and 25% for testing, the RF classifier model achieved 91% OA. In addition, this model separated the healthy and unhealthy oil palms in the study sites into 37,617 (75%) and 12,320 (25%) individuals, respectively.     

Tree and building detection in dense urban environments using automated processing of IKONOS image and LiDAR data

The automated detection and reconstruction of artificial structures, larger than 10 m² in area using high resolution satellite images and Light Detection and Ranging (LiDAR) data through 3-dimensional shapes and/or 2-dimensional boundaries is described here. Additionally, it is demonstrated how individual tree crowns have been detected with more than 90% accuracy in very dense urban environments from very high-resolution images and range data. Pre-existing machine vision algorithms and techniques were modified and updated for this particular application to building detection within dense urban areas. All products from such procedures have not only been demonstrated with a significant areal coverage but have also been quantitatively assessed against manually obtained and third party mapping data. Accuracies of around 85% have been achieved for building detection and almost 95% for tree crown detection.     

Mapping growing stock volume and biomass carbon storage of larch plantations in Northeast China with L-band ALOS PALSAR backscatter mosaics

ABSTRACT

Reliable spatial information on growing stock volume (GSV) and biomass is critical for creating management strategies for plantation forests. This study developed empirical models to map the GSV and biomass of larch plantations (LPs) in Northeast China (1.25 million km² total area) by integrating L-band synthetic aperture radar (SAR) data with ground-based survey data. The best correlation model was used to map the GSVs and biomasses of LPs. The total GSV and biomass carbon storage were estimated at 224.3 ± 59.0 million m³ and 113.0 ± 29.7 × 10¹² g C with average densities of 85.1 m³ ha^?1 and 42.9 10⁶ g × C ha^?1, respectively, over a total area of 2.64 million ha. The saturation effect of SAR was determined beyond 260 m³ ha^?1, which was expected to influence the estimations for a small proportion of the study area. The accuracy of the estimations has limitations mainly due to the uncertainties in the GSV inventories, discrimination of natural larch and the SAR dataset. Based on the mapping results of the GSVs of LPs, a planning strategy for multipurpose management was tentatively proposed. This study can inform policies and management practices to assure broader and sustainable benefits from plantation forests in the future.     

Remote sensing of species mixtures in conifer plantations using LiDAR height and intensity data

In even-aged, single species conifer plantations LiDAR height data can be modelled to provide accurate estimates of tree height and volume. However, it is apparent that growth models developed for single species stands are not directly transferable to a more general situation of mixed species plantations. This paper evaluates the ability of small footprint, dual-return, pulsed airborne LiDAR data to estimate the proportion of the productive species when mixed with a nurse crop in closed canopy plantations. A study area located in Galloway Forest District in Scotland is used as an example of Lodgepole pine and Sitka spruce mixed plantation; this area contains good examples of a wide range of pure and mixed species plantation types. Three species groups are studied: areas of pure Sitka spruce, areas of pure Lodgepole pine and areas where the two species have been planted together. Two approaches are assessed for detection of plantation mixtures: the first uses LiDAR intensity data to separate spruce and pine species and the second uses LiDAR-derived canopy density measures, coefficient of variation, skewness, percent of ground returns (which provides a measure of canopy openness) and the mean canopy height, which enables areas with height variations to be identified. From analysis of LiDAR data extracted from 54 study plots using logistic regression, the coefficient of variation and LiDAR intensity data provide the most useful predictors of the proportion of spruce in a pine/spruce mixture with coefficients of determination (R²) of 0.914 and 0.930 respectively. The method could be developed as a mapping tool, which in combination with existing inventory data should help to improve timber volume forecasting for mixed species even-aged plantations.     

Hyperion, IKONOS, ALI, and ETM+ sensors in the study of African rainforests

The goal of this research was to compare narrowband hyperspectral Hyperion data with broadband hyperspatial IKONOS data and advanced multispectral Advanced Land Imager (ALI) and Landsat-7 Enhanced Thematic Mapper Plus (ETM+) data through modeling and classifying complex rainforest vegetation. For this purpose, Hyperion, ALI, IKONOS, and ETM+ data were acquired for southern Cameroon, a region considered to be a representative area for tropical moist evergreen and semi-deciduous forests. Field data, collected in near-real time to coincide with satellite sensor overpass, were used to (1) quantify and model the biomass of tree, shrub, and weed species; and (2) characterize forest land use/land cover (LULC) classes.The study established that even the most advanced broadband sensors (i.e., ETM+, IKONOS, and ALI) had serious limitations in modeling biomass and in classifying forest LULC classes. The broadband models explained only 13-60% of the variability in biomass across primary forests, secondary forests, and fallows. The overall accuracies were between 42% and 51% for classifying nine complex rainforest LULC classes using the broadband data of these sensors. Within individual vegetation types (e.g., primary or secondary forest), the overall accuracies increased slightly, but followed a similar trend. Among the broadband sensors, ALI sensor performed better than the IKONOS and ETM+ sensors.When compared to the three broadband sensors, Hyperion narrowband data produced (1) models that explained 36-83% more of the variability in rainforest biomass, and (2) LULC classifications with 45-52% higher overall accuracies. Twenty-three Hyperion narrowbands that were most sensitive in modeling forest biomass and in classifying forest LULC classes were identified and discussed.     

Towards refined estimation of vegetation carbon stock in Auckland,New Zealand using WorldView-2 and LiDAR data: the impact of scaling

ABSTRACT

It is necessary to estimate carbon (C) stored in urban vegetation for the purpose of carbon accounting and trading. This study aims to develop a refined method for reliably estimating above-ground carbon (AGC) stock of urban vegetation from integrated WorldView-2 imagery and Light Detection And Ranging (LiDAR) data in Auckland, New Zealand. Also assessed in this study is the impact of image resolution on regional AGC estimates by vegetation type. The integration of WorldView-2 imagery with a 2-m digital surface model produced from LiDAR data enables urban vegetation to be mapped into trees (101.5 km²), shrubs (64.9 km²), and grasses (172.2 km²) at a producer’s accuracy over 95.9%. The AGC stock of trees, shrubs and grasses is estimated at 1,134,287, 207,606, and 127,427 Mg C, respectively, from the vegetation map. Overall, the total AGC of all types of vegetation does not vary significantly with image spatial resolution over the range of 5 to 30 m if estimated using the same model. This is because high AGC densities are generalised at a coarser resolution, but the larger pixel size compensates for the decrease. Although the spatial resolution does not affect the most significant spectral predicators of plot-level AGC noticeably, it has an obvious effect on both model accuracy and complexity. Thus, the impact of image resolution on AGC would be pronounced if it were estimated using different models that were the best at a given resolution. Of the three vegetation types, the AGC of shrubs is the most variable with spatial resolution, followed by trees. Thus, the AGC of relatively small but more spatially fragmented vegetation parcels is more susceptible to change in image spatial resolution. The estimation model based on spectral features of vegetation has the lowest room-mean-square-error at 15 m. More research is needed to confirm whether it is true in other natural environments in future studies.     

Mangrove biomass carbon stock mapping of the Karimunjawa Islands using multispectral remote sensing

Among vegetated coastal habitats, mangrove forests are among the densest carbon pools. They store their organic carbon in the surrounding soil and thus the sequestered carbon stays in the sediment for a long time and cannot be easily returned to the atmosphere. Additionally, mangroves also provide various important ecosystem services in coastal areas and surroundings. Accordingly, it is important to understand the distribution of biomass carbon stock in mangrove habitats in a spatial and temporal context, not only to reduce CO₂ concentrations in the atmosphere, but also for their sustainability. The objectives of this research are to map the mangrove carbon stock and estimate the total biomass carbon stock sheltered by mangrove forests, with the Karimunjawa Islands as a study site, using the widely available passive remote sensing system ALOS AVNIR-2. The modelling and mapping of mangrove carbon stock incorporates the integration of image pixel values and mangroves field data via empirical modelling. Vegetation indices and PC bands at different levels of radiometric corrections were all used as the input in the mangrove carbon stock modelling so that the effectiveness and sensitivity of different image transformations to particular radiometric correction levels could be analysed and understood. Afterward, the accuracy and effectiveness of each mangrove carbon stock-mapping routine was compared and evaluated. The accuracy of the best mangrove above-ground carbon stock (AGC) map modelled from vegetation index is 77.1% (EVI1, SE 5.89 kg C m^?2), and for mangrove below-ground carbon stock (BGC) it is 60.0% (GEMI, SE 2.54 kg C m^?2). The mangrove carbon stock map from ALOS AVNIR-2 PC bands showed a maximum accuracy of 77.8% (PC2, SE 5.71 kg C m^?2) and 60.8% (PC2, SE 2.48 kg C m^?2) for AGC and BGC respectively. From the resulting maps, the Karimunjawa Islands are estimated to shelter 96,482 tonnes C of mangroves AGC with a mean value of 21.64 kg C m^?2 and 24,064 tonnes C of mangroves BGC with a mean value of 5.39 kg C m^?2. Potentially, there are approximately 120,546 tonnes C of mangrove biomass carbon stock in the Karimunjawa Islands. Remote-sensing reflectance can successfully model mangrove carbon stock based on the relationship between mangrove canopy properties, represented by leaf area index (LAI) and the tree or root biomass carbon stock. The accuracy of the mangrove carbon stock map is subject to errors, which are sourced mainly from: (1) the absence of a species-specific biomass allometric equation for several species present in the study area; (2) the generalized standard conversion value of mangrove biomass to mangrove carbon stock; (3) the relationship between mangrove reflectance and mangrove LAI; (4) the relationship between mangrove reflectance and above-ground mangrove biomass and carbon stock due to its relationship with LAI; (5) the relationship between mangrove LAI and mangrove below-ground parts; (6) the inability to perform mangrove carbon stock modelling at the species level due to the complexities of the mangrove forest in the study area; (7) background reflectance and atmospheric path radiance that could not be completely minimized using image radiometric corrections and transformations; and (8) spatial displacement between the actual location of the mangrove forest in the field and the corresponding pixel in the image. The availability of mangrove biomass carbon stock maps is beneficial for carrying out various management activities, and is also very important for the resilience of mangroves to changing environments.     

Towards the use of remote-sensing data for monitoring of abandoned oil palm lands in Malaysia: a semi-automatic approach

Oil palm is a commercial crop that is important for its food value and as a biofuel, along with its other benefits towards the economy and human health. Currently, Malaysia cultivates approximately 5.64 million ha of oil palm. To date, a study identifying abandoned oil palm areas using satellite images is almost non-existent. Conventionally, the monitoring of abandoned oil palm lands is tedious and time consuming, especially over large areas. Hence, in this article, the capability of high resolution satellite image via Satellite Pour I’Observation de la Terre-6 (SPOT-6) products to extract abandoned oil palm areas was explored, as was the use of multi-temporal Landsat Operational Land Imager (OLI) imagery to develop the phenology of abandoned oil palm sites. Homogeneity measures derived through SPOT images played a more important role to identify abandoned oil palm than crop phenology characteristics extracted from high spectral resolution of Landsat images. With the advancement of object-oriented classification, monitoring of abandoned oil palm areas can be done semi-automatically with an accuracy of 92±1%.     

A simple method for detection and counting of oil palm trees using high-resolution multispectral satellite imagery

In the past, oil palm density has been determined by manually counting trees every year in oil palm plantations. The measurement of density provides important data related to palm productivity, fertilizer needed, weed control costs in a circle around each tree, labourers needed, and needs for other activities. Manual counting requires many workers and has potential problems related to accuracy. Remote sensing provides a potential approach for counting oil palm trees. The main objective of this study is to build a robust and user-friendly method that will allow oil palm managers to count oil palm trees using a remote sensing technique. The oil palm trees analysed in this study have different ages and densities. QuickBird imagery was applied with the six pansharpening methods and was compared with panchromatic QuickBird imagery. The black and white imagery from a false colour composite of pansharpening imagery was processed in three ways: (1) oil palm tree detection, (2) delineation of the oil palm area using the red band, and (3) counting oil palm trees and accuracy assessment. For oil palm detection, we used several filters that contained a Sobel edge detector; texture analysis co-occurrence; and dilate, erode, high-pass, and opening filters. The results of this study improved upon the accuracy of several previous research studies that had an accuracy of about 90–95%. The results in this study show (1) modified intensity-hue-saturation (IHS) resolution merge is suitable for 16-year-old oil palm trees and have rather high density with 100% accuracy; (2) colour normalized (Brovey) is suitable for 21-year-old oil palm trees and have low density with 99.5% accuracy; (3) subtractive resolution merge is suitable for 15- and 18-year-old oil palm trees and have a rather high density with 99.8% accuracy; (4) PC spectral sharpening with 99.3% accuracy is suitable for 10-year-old oil palm trees and have low density; and (5) for all study object conditions, colour normalized (Brovey) and wavelet resolution merge are two pansharpening methods that are suitable for oil palm tree extraction and counting with 98.9% and 98.4% accuracy, respectively.     

Investigating the impact of discrete-return lidar point density on estimations of mean and dominant plot-level tree height in Eucalyptus grandis plantations

The accuracy of lidar remote sensing in characterizing three-dimensional forest structural attributes has encouraged foresters to integrate lidar approaches in routine inventories. However, lidar point density is an important consideration when assessing forest biophysical parameters, given the direct relationship between higher spatial resolution and lidar acquisition and processing costs. The aim of this study was to investigate the effect of point density on mean and dominant tree height estimates at plot level. The study was conducted in an intensively managed Eucalyptus grandis plantation. High point density (eight points/m²) discrete-return, small-footprint lidar data were used to generate point density simulations averaging 0.25, one, two, three, four, five, and six points/m². Field surveyed plot-level mean and dominant heights were regressed against metrics derived from lidar data at each simulated point density. Stepwise regression was used to identify which lidar metrics produced the best models. Mean height was estimated at accuracy of R² ranging between 0.93 and 0.94 while dominant height was estimated with an R² of 0.95. Root mean square error (RMSE) was also similar at all densities for mean height (～1.0 m) and dominant height (～1.2 m); the relative RMSE compared to field-measured mean was constant at approximately 5%. Analysis of bias showed that the estimation of both variables did not vary with density. The results indicated that all lidar point densities resulted in reliable models. It was concluded that plot-level height can be estimated with reliable accuracy using relatively low density lidar point spacing. Additional research is required to investigate the effect of low point density on estimation of other forest biophysical attributes.     

Inter-sensor relationships between IKONOS and Landsat-7 ETM+ NDVI data in three ecoregions of Africa

The goal of this research was to establish inter-sensor relationships between IKONOS and Landsat-7 ETM+ data. Dry and wet season images were acquired on the same date or about the same date from IKONOS and ETM+ sensors to enable direct comparison between the two distinctly different data types. The images were from three distinct ecoregions located in African rainforests and savannas that encompass a wide range of land use/land cover classes and ecological units. The IKONOS NDVI had a high degree of correlation with ETM+ NDVI with R ² values between 0.67 and 0.72. Inter-sensor model equations relating IKONOS NDVI with ETM+ NDVI were determined. The characteristics that contribute to the increased sensitivity in dynamic ranges of IKONOS NDVI relative to ETM+ NDVI were attributed to: (1) radiometric resolution that adds more bits per data point (11-bit IKONOS data as opposed to 8-bit ETM+); and (2) spatial resolution that helped in resolving spectral characteristics at micro landscape units. Spectral bandwidths of the two sensors had no effect on the dynamic ranges of NDVIs. Overall, the IKONOS data showed greater sensitivity to landscape units and ecological characteristics when compared with Landsat-7 ETM+ data. Across ecoregions and land use/land cover classes, the IKONOS NDVI dynamic range (?0.07 to 0.71) was considerably greater than the ETM+ NDVI dynamic range (?0.24 to 0.46). IKONOS data explained greater variability (R ²=0.73) in agroforest biomass when compared with ETM+ data (R ²=0.66). The inter-sensor relationships presented in this paper are expected to facilitate better understanding and proper interpretation of terrestrial characteristics studied using multiple sensors over time periods.     

A correction method for dynamic model calculations using observational data and its application in oceanography

A new data assimilation method for the correction of model calculations is developed and applied. The method is based on the least resistance principle and uses the theory of diffusion-type stochastic processes and stochastic differential equations. Application of the method requires solving a system of linear equations that is derived from this principle. The system can be considered as a generalization of the well-known Kalman scheme taking the model’s dynamics into account. The method is applied to the numerical experiments with the HYbrid Coordinate Ocean Model (HYCOM) and Archiving, Validating, and Interpolating Satellite Ocean (AVISO) data for the Atlantic. The skill of the method is assessed using the results of the experiments. The model’s output is compared with the twin experiments, namely, the model calculations without assimilation, which confirms the consistency and robustness of the proposed method.     

不均衡数据在股票研报分类中的应用

股票研报是由金融行业分析师对股票相关新闻做出的分析和评价。它从专业角度分析此类新闻是否会对某股票的未来走势产生影响,并提出专业投资建议,往往比论坛分析更具权威性。然而,各类别研报数量之间的严重不均衡性致使常规的SVM分类效果较差。为提高分类效果,提出一种新的不均衡数据分类方法,首先在文本特征项选择方面采用“组合特征”思想以选择更具语义信息的特征短语,并改进CHI统计以提高对少数类样本特征项的选择;然后设计一个基于K-means聚类的边界自适应层次欠采样算法对多数类样本进行层次欠采样。实验结果表明,该方法能够在不影响多数类分类的基础上对少数类的分类效果有较为明显地提升。     

Estimating and mapping grass phosphorus concentration in an African savanna using hyperspectral image data

We tested the utility of imaging spectroscopy and neural networks to map phosphorus concentration in savanna grass using airborne HyMAP image data. We also sought to ascertain the key wavelengths for phosphorus prediction using hyperspectral remote sensing. The remote sensing of foliar phosphorus has received very little attention as compared to nitrogen, yet it plays an equally important role in explaining the distribution and feeding patterns of herbivores. Band depths from two continuum‐removed absorption features as well as the red edge position (REP) were input into a backpropagation neural network. Following a series of experiments to ascertain the optimum wavelengths, the best trained neural network was used to predict and ultimately to map grass phosphorus concentration in the Kruger National Park. The results indicate that the best trained neural network could predict phosphorus distribution with a coefficient of determination of 0.63 and a root mean square error (RMSE) of 0.07 (28% of the mean observed phosphorus concentration) on an independent test data set. Our results also show that the absorption feature located in the shortwave infrared (R _2015–2199) contains more information on phosphorus distribution, a region that has hardly been explored before in most spectroscopic experiments for phosphorus as compared to the visible bands. Overall, the study demonstrates the potential of imaging spectroscopy in mapping grass phosphorus concentration in savanna rangelands.     

Wetland mapping in the Congo Basin using optical and radar remotely sensed data and derived topographical indices

This study reports results of a classification tree approach to mapping the wetlands of the Congo Basin, focusing on the Cuvette Centrale of the Congo River watershed, an area of 1,176,000 km². Regional expert knowledge was used to train passive optical remotely sensed imagery of the Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper Plus (ETM+) sensors, JERS-1 active radar L-band imagery, and topographical indices derived from 3 arc sec elevation data of the Shuttle Radar Topography Mission (SRTM). All data inputs were resampled to a common 57 m resolution grid. A classification tree bagging procedure was employed to produce a final map of per-grid cell wetland probability. Thirty bagged trees were ranked and the median result was selected to produce the final wetland probability map. Thresholding the probability map at < 0.5 yielded a proportion of wetland cover for the study area of 32%, equivalent to 360,000 km². Wetlands predominate in the CARPE Lake Tele-Lake Tumba landscape located in the western part of the Democratic Republic of the Congo and the south-eastern Republic of Congo, where they constitute 56% of the landscape. Local topography depicting relative elevation for sub-catchments proved to be the most valuable discriminator of wetland cover. However, all sources of information (i.e. optical, radar and topography) featured prominently in contributing to the classification tree procedure, reinforcing the idea that multi-source data are useful in the characterization of wetland land cover. The method employed freely available data and a fully automated process, except for training data collection. Comparisons to existing maps and in situ field observations indicate improvements compared to previous efforts.     

Estimating carbon and showing impacts of drought using satellite data in regression-tree models

Integrating spatially explicit biogeophysical and remotely sensed data into regression-tree models enables the spatial extrapolation of training data over large geographic spaces, allowing a better understanding of broad-scale ecosystem processes. The current study presents annual gross primary production (GPP) and annual ecosystem respiration (RE) for 2000–2013 in several short-statured vegetation types using carbon flux data from towers that are located strategically across the conterminous United States (CONUS). We calculate carbon fluxes (annual net ecosystem production [NEP]) for each year in our study period, which includes 2012 when drought and higher-than-normal temperatures influence vegetation productivity in large parts of the study area. We present and analyse carbon flux dynamics in the CONUS to better understand how drought affects GPP, RE, and NEP. Model accuracy metrics show strong correlation coefficients (r) (r ≥ 94%) between training and estimated data for both GPP and RE. Overall, average annual GPP, RE, and NEP are relatively constant throughout the study period except during 2012 when almost 60% less carbon is sequestered than normal. These results allow us to conclude that this modelling method effectively estimates carbon dynamics through time and allows the exploration of impacts of meteorological anomalies and vegetation types on carbon dynamics.