No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Multi-sensor Approach for the Estimation of Above-Ground Biomass of Mangroves
Abstract
Mangroves are woody halophytes thriving in muddy substratum along the coastal areas of the tropics and sub-tropics. They are often credited for their exceptional carbon sequestration capability. Estimating above-ground biomass (AGB) through field survey is tedious, particularly in a hostile environment like a mangrove ecosystem. However, the quantification of AGB is made possible with the help of continued advancements in sensor technology and computational algorithms. This research attempts to model the AGB of mangroves present in Bhitarkanika, Odisha, using a multi-sensor approach. We utilized multispectral Sentinel-2 (SM) and Landsat-8 (LO), and hyperspectral Airborne Visible Infra-Red Imaging Spectrometer—Next Generation (AN) datasets in our analysis. The mangrove biomass was calculated for 42 sample plots from a field survey using species specific and common allometric equations. After data-specific preprocessing; six feature sets namely reflectance bands, band ratios, vegetation indices (VIs), texture-based Gray Level Co-occurrence Matrix (GLCM) features of reflectance, band ratios and VIs were extracted for each dataset. The co-located set of features derived from each dataset were regressed against the AGB estimated using field methods of 42 sample plots (1) independently for each feature set, (2) in a combination of feature sets for each dataset and (3) in a combination of the feature sets of all three datasets as a multi-sensor approach. Feature selection techniques were used to get the best possible output of combined AN, SM and LO datasets. The results show that the combination of textural features gave better prediction models than independent sets of features. Also, Genetic Algorithm (GA) and Recursive Feature Elimination CV (RFECV) proved to be better feature selectors than other classical approaches. AN, SM and LO resulted in the R2 value of 0.41, 0.85 and 0.35 with RMSE of 356.81, 195.49 and 366.84 t/ha, respectively; while, the multisensory approach yielded a maximum R2 value of 0.7 and RMSE of 244.86 t/ha. The results show that the structural information of vegetation canopy obtained from textural parameters of different input bands has improved the regression model to predict the biomass.
Mangroves provide multiple valuable environmental services, including their high potential for the sequestration of blue carbon. However, mangrove conservation is socially demanding. Through a spatiotemporal analysis of the inhabitants of this dynamic coastal ecosystem in Mexico, we found that the numbers of small settlements within mangroves increased during the study period despite a decline in mangrove stands and in the human populations inhabiting them. Extended rural developments have been banished to external buffer areas, whereas near-subsistence hamlets are increasingly occupying mangrove forest. The ability of mangroves to sequester atmospheric blue carbon relies on regional socioeconomic patterns and stressors that should be considered in the development of mangrove conservation initiatives. Our study enables identification of potential areas of research — e.g., forestry buffer zones, social exclusion, permanent settlements, and integral monitoring — that can incorporate the social dimensions of blue carbon conservation into the political and research agendas.
Mangroves play a major role in supporting biodiversity , providing economic and ecological security to the coastal communities, mitigating the effects of climate change and global warming. Species level classification of mangrove forest, understanding physical as well as chemical properties of mangrove vegetation, mangrove health, pigments, and levels of stress are some of the key issues for making scientific and management decisions. Hyperspectral remote sensing owing to its narrow bands, yield information on structural details and canopy parameters. Hyperspectral data over Sundarban and Bhitarkanika mangrove forests are analyzed for species discrimination and forest health assessment. In all, 15 mangrove species in Sundarban and 7 mangrove species in Bhitarkanika have been identified and classified using Spectral Angle Mapper technique. In-situ spectro-radiometer data has been used along with AVIRIS-NG hyperspectral data. Based on response of vegetation in blue, red and near-infrared regions, combination of vegetation indices are used to assess mangrove forest's health. Reduction in NIR reflectance with shift towards lower wavelength has been observed in less healthy groups.
Tactile texture refers to the tangible feel of a surface and visual texture refers to see the shape or contents of the image. In the image processing, the texture can be defined as a function of spatial variation of the brightness intensity of the pixels. Texture is the main term used to define objects or concepts of a given image. Texture analysis plays an important role in computer vision cases such as object recognition, surface defect detection, pattern recognition, medical image analysis, etc. Since now many approaches have been proposed to describe texture images accurately. Texture analysis methods usually are classified into four categories: statistical methods, structural, model-based and transform-based methods. This paper discusses the various methods used for texture or analysis in details. New researches shows the power of combinational methods for texture analysis, which can't be in specific category. This paper provides a review on well known combinational methods in a specific section with details. This paper counts advantages and disadvantages of well-known texture image descriptors in the result part. Main focus in all of the survived methods is on discrimination performance, computational complexity and resistance to challenges such as noise, rotation, etc. A brief review is also made on the common classifiers used for texture image classification. Also, a survey on texture image benchmark datasets is included.
Nowadays, being in digital era the data generated by various applications are increasing drastically both row-wise and column wise; this creates a bottleneck for analytics and also increases the burden of machine learning algorithms that work for pattern recognition. This cause of dimensionality can be handled through reduction techniques. The Dimensionality Reduction (DR) can be handled in two ways namely Feature Selection (FS) and Feature Extraction (FE). This paper focuses on a survey of feature selection methods, from this extensive survey we can conclude that most of the FS methods use static data. However, after the emergence of IoT and web-based applications, the data are generated dynamically and grow in a fast rate, so it is likely to have noisy data, it also hinders the performance of the algorithm. With the increase in the size of the data set, the scalability of the FS methods becomes jeopardized. So the existing DR algorithms do not address the issues with the dynamic data. Using FS methods not only reduces the burden of the data but also avoids overfitting of the model.
We demonstrate real-time model-based atmospheric correction onboard the Next Generation Airborne Visible/Infrared Imaging Spectrometer. We achieve a reduction in processing time from hours or days to seconds by modifying a standard physics-based atmospheric correction algorithm to support real-time execution. We achieved this reduction by modifying the physics-based ATmospheric REMoval algorithm to leverage a large lookup table of precomputed scattering and transmission coefficients, indexed by parameters specifying the aircraft operating conditions at capture time. Interpolation among the precomputed coefficients allows surface reflectance retrieval at the sensor acquisition rate of 500 Mb/s. Our system produced science-quality reflectance products during over 30 test flights and, to our knowledge, is the first reported demonstration of real-time model-driven visible shortwave infrared atmospheric correction onboard an aircraft.
Previous research studies have demonstrated that the relationship between remote sensing-derived parameters and aboveground biomass (AGB) could vary across different species types. However, there are few studies that calibrate reliable statistical models for mangrove AGB. This study quantifies the differences of accuracy in AGB estimation between the results obtained with and without the consideration of species types using Worldview-2 images and field surveys. A Back Propagation Artificial Neural Network (BP ANN) based model is developed for the accurate estimation of uneven-aged and dense mangrove forest biomass. The contributions of the input variables are further quantified using a "Weights" method based on BP ANN model. Two types of mangrove species, Sonneratia apetala (S. apetala) and Kandelia candel (K. candel), are examined in this study. Results show that the species type information is the most important variable for AGB estimation, and the red edge band and the associated vegetation indices from WorldView-2 images are more sensitive to mangrove AGB than other bands and vegetation indices. The RMSE of biomass estimation at the incorporation of species as a dummy variable is 19.17% lower than that of the mixed species level. The results demonstrate that species type information obtained from the WorldView-2 images can significantly improve of the accuracy of the biomass estimation.
This chapter reviews and discusses various aspects of texture analysis. The concentration is on the various methods of extracting textural features from images. The geometric, random field, fractal, and signal processing models of texture are presented. The major classes of texture processing problems such as segmentation, classification, and shape from texture are discussed. The possible application areas of texture such as automated inspection, document processing, and remote sensing are summarized. A bibliography is provided at the end for further reading.
Mangroves are salt-tolerant trees or shrubs commonly seen in mudflats of intertidal regions of tropical and subtropical coastlines. Recent advances in field spectroscopic techniques enabled the species level discrimination among closely related vegetation types. In this study, we have analysed the laboratory spectroscopy data collected from eight species of Rhizophoraceaea family of mangroves. The spectral data ranges between the wavelength of 350 nm and 2500 nm at a very narrow bandwidth of 1 nm. Pre-processing techniques including smoothing were done on the spectra to remove the noise before compiling it to a spectral library. Derivative analysis of the spectra was done and its corresponding first and second derivatives were obtained. Statistical analysis such as parametric and non-parametric tests were implemented on the original processed spectra as well as their respective first and second order derivatives for the identification of significant bands for species discrimination. Results have shown that red-edge region (680 nm – 720 nm) and water vapour absorption region around 1150 nm and 1400 nm are optimal as they were consistent in discriminating species in reflectance spectra as well as in its first and second derivative spectra. C. decandra species is found to be discriminable from other species while reflectance and its derivative spectra were used. Non-parametric statistical analysis gave better results than that of parametric statistical analysis especially in SWIR 2 spectral region (1831 nm – 2500 nm).
The contribution of mangrove carbon to the coastal ocean in low latitudes was evaluated. Mangrove forests occupy only 2% of the world's coastal ocean area yet they account for about 5% of net primary production, 12% of ecosystem respiration and about 30% of carbon burial on all continental margins in subtropical and tropical seas. Mangroves also account for nearly one-third of all riverine DIC discharging into low latitude coastal waters. Mangrove forests fix, release and sequester more carbon by area than all other coastal ecosystem types, except perhaps for subtropical and tropical seagrass meadows for which data are lacking. Globally, mangrove waters release to the atmosphere more than 2.5 times (-42.8TgCy-1) the amount of CO2 emitted from all other subtropical and tropical coastal waters. The global destruction of the large carbon stocks (956MgCha-1) of mangroves at the current annual rate of about 1% results in an additional annual release of roughly 133TgCy-1 to the atmosphere. Mangroves account for only 0.7% of tropical forest area globally, but their destruction currently adds another 10% to global CO2 release from tropical deforestation. Despite considerable uncertainty upscaling small numbers of measurements with large coefficents of variation, our calculations suggest that mangroves are a globally significant contributor to the carbon cycle in low latitude seas, and to greenhouse emissions resulting from tropical deforestation.
The paper suggests a framework for the estimation of biomass and, ultimately, the amount of carbon sequestered in Avicennia marina stands of Mumbai region in Thane creek. The literature-adopted framework for estimating biomass uses a direct (Carbon, Hydrogen and Nitrogen (CHN) analysis), an indirect (allometry and remote sensing) and a hybrid approach. The study area has Avicennia marina as a dominant species. The carbon content was estimated using allometry and CHN analysis and further refined with GIS techniques. The Diameter at Breast Height (DBH) of 1,726 trees at 49 different locations in the study area was measured for estimating total biomass. A CHN analysis of leaf and soil samples was performed to determine the carbon content and the data extrapolated using geostatistical analysis. Using allometry and a GIS hybrid approach, the estimated carbon stock was 34.14769 tons per hectare. The regression model developed may be useful for estimating the carbon stock of areas that are inaccessible for conducting allometric measurements or gathering samples for CHN analysis. The study attempts to improve the estimates of the carbon content by including geostatistical analysis as a final step for allometry and combines CHN analysis for better conversion of biomass to carbon content.
The significance of forested areas in carbon sequestration is conventional, and well renowned. But, hardly any attempts have been made to study the potential of trees in carbon sequestration from urban areas. Andhra University was selected for the study in Visakhapatnam city with the objectives of quantifying the total carbon sequestration by Pongamia pinata. Stratified random sampling was used for assessing biomass in two site and about 230 P. pinnata trees were taken. Biomass was calculated using Non-destructive allometric models. The biomass carbon content was taken as 55% of the tree biomass. Soil samples were taken from soil profile up to 40 cm depth for deep soils and up to bedrock for shallow soils at an interval of 10 and 20 cm for top and sub-soil respectively. Walkley‐Black Wet Oxidation method was applied for measuring soil organic carbon. Belowground biomass was estimated by the Root:Shoot ratio relationship. Total biomass and soil carbon was higher in Site-2 than in Site-1. Total carbon sequestration in Site-2 was found 1.59 times higher compared to Site-1 but the mean SOC stored was found higher in Site-1 than in Site-2, i.e.,14.48 tC/ha and 10.33 tC/ha, respectively.
Hyperspectral indices of plant water stress were examined at leaf and canopy scales over a rainfed corn crop. The relationship between Water Index (WI) and plant water status was examined spatially and temporally. It was found that WI was not sensitive to the low ater contetn of the individual leaf. At canopy scale, WI was strongly influenced by canopy structure, aand this impact must be accounted for to identify water stress.
Accurate estimation of aboveground biomass and carbon stock has gained importance in the context of the United Nations Framework Convention on Climate Change (UNFCCC) and the Kyoto Protocol. In order to develop improved forest stratum–specific aboveground biomass and carbon estimation models for humid rainforest in northeast Madagascar, this study analyzed texture measures derived from WorldView-2 satellite data. A forest inventory was conducted to develop stratum-specific allometric equations for dry biomass. On this basis, carbon was calculated by applying a conversion factor.
After satellite data preprocessing, vegetation indices, principal components, and texture measures were calculated. The strength of their relationships with the stratum-specific plot data was analyzed using Pearson’s correlation. Biomass and carbon estimation models were developed by performing stepwise multiple linear regression.
Pearson’s correlation coefficients revealed that (a) texture measures correlated more with biomass and carbon than spectral parameters, and (b) correlations were stronger for degraded forest than for non-degraded forest. For degraded forest, the texture measures of Correlation, Angular Second Moment, and Contrast, derived from the red band, contributed to the best estimation model, which explained 84% of the variability in the field data (relative RMSE = 6.8%). For non-degraded forest, the vegetation index EVI and the texture measures of Variance, Mean, and Correlation, derived from the newly introduced coastal blue band, both NIR bands, and the red band, contributed to the best model, which explained 81% of the variability in the field data (relative RMSE = 11.8%).
These results indicate that estimation of tropical rainforest biomass/carbon, based on very high resolution satellite data, can be improved by (a) developing and applying forest stratum–specific models, and (b) including textural information in addition to spectral information.
Mangroves play important roles in providing a range of ecosystem services, mitigation of strong waves, protection of coastlines against erosion, maintenance of water quality, and carbon sink in the context of global warming. For trees in mangrove forests in southern Ranong Province, Thailand, we investigated the allometric relationship between crown area derived from high-resolution satellite data and stem diameter and used the resulting model to estimate aboveground biomass. We used QuickBird panchromatic and multispectral data acquired for the study area on 15 October 2006 as the high-resolution satellite data. Individual tree crowns were extracted from the satellite image of panchromatic data by using the watershed method, and the species were identified by using the maximum-likelihood method for the multispectral data. Overall classification accuracy for species identification was 88.5 %. The biomass derived from our field survey was plotted against aboveground biomass in the sample plots, estimated from the QuickBird data. The regression line through the origin between the satellite-estimated biomass and biomass based on the field data had a slope of 1.26 (R
2 = 0.65). Stand aboveground biomass estimated from the high-resolution satellite data was underestimated because of a lack of data on the biomass of suppressed trees and inappropriate segmentation of crowns of large trees into two or more trees.
Mangroves are salt tolerant plants that grow within the intertidal zone along tropical and subtropical coasts. They are important barriers for mitigating coastal disturbances, provide habitat for over 1300 animal species and are one of the most productive ecosystems. Mozambique's mangroves extend along 2700 km and cover one of the largest areas in Africa. The purpose of this study was to determine the countrywide mean tree height spatial distribution and biomass of Mozambique's mangrove forests using Landsat ETM+ and Shuttle Radar Topography Mission (SRTM) data. The SRTM data were calibrated using the Landsat derived land-cover map and height calibration equations. Stand-specific canopy height-biomass allometric equations developed from field measurements and published height-biomass equations were used to calculate aboveground biomass of the mangrove forests on a landscape scale. The results showed that mangrove forests covered a total of 2909 km2 in Mozambique, a 27% smaller area than previously estimated. The SRTM calibration indicated that average tree heights changed with geographical settings. Even though the coast of Mozambique spans across 16 degrees latitude, we did not find a relationship between latitude and biomass. These results confirm that geological setting has a greater influence than latitude alone on mangrove production. The total mangrove dry aboveground biomass in Mozambique was 23.6 million tons and the total carbon was 11.8 million tons.
The purpose of this paper is to evaluate spatial resolution effects on image classification. Classification maps were generated with a maximum likelihood (ML) classifier applied to three multi-spectral bands and variance texture images. A total of eight urban land use/cover classes were obtained at six spatial resolution levels based on a series of aggregated Colour Infrared Digital Orthophoto Quarter Quadrangle (DOQQ) subsets in urban and rural fringe areas of the San Diego metropolitan area. The classification results were compared using overall and individual classification accuracies. Classification accuracies were shown to be influenced by image spatial resolution, window size used in texture extraction and differences in spatial structure within and between categories. The more heterogeneous are the land use/cover units and the more fragmented are the landscapes, the finer the resolution required. Texture was more effective for improving the classification accuracy of land use classes at finer resolution levels. For spectrally homogeneous classes, a small window is preferable. But for spectrally heterogeneous classes, a large window size is required.
Three decades have passed since the launch of the first international satellite sensor programme designed for monitoring Earth's resources. Over this period, forest resources have come under increasing pressure, thus their management and use should be underpinned by information on their properties at a number of levels. This paper provides a comprehensive review of how satellite remote sensing has been used in forest resource assessment since the launch of the first Earth resources satellite sensor (ERTS) in 1972. The use of remote sensing in forest resource assessment provides three levels of information; namely (1) the spatial extent of forest cover, which can be used to assess the spatial dynamics of forest cover; (2) forest type and (3) biophysical and biochemical properties of forests. The assessment of forest information over time enables the comprehensive monitoring of forest resources. This paper provides a comprehensive review of how satellite remote sensing has been used to date and, building on these experiences, the future potential of satellite remote sensing of forest resources is highlighted.
The use of multiscale textural features for urban satellite Synthetic Aperture Radar (SAR) image characterization is introduced. The multiscale nature of urban environments requires that no single scale of analysis is exclusively considered. An accurate texture‐based discrimination of land use/land cover classes needs, for instance, the computation of multiscale textural features for a wide range of parameters of the co‐occurrence algorithm. The technique proposed in this paper shows how to reduce the full multiscale feature set to a subset, the most suitable for classification using a fuzzy ARTMAP neural network. This is done by analysing the relevance of each feature for this particular classifier by means of the Histogram Distance Index (HDI). We validate the procedure by providing results of the classification of several satellite SAR data of the same urban test site. The results are encouraging. They show the potential of this technique for automatic extraction of the best texture and scale subset, suitable for efficient urban mapping using SAR satellite data.This work is an enlarged version of the paper ‘Discriminating urban environments using multiscale texture and multiple SAR images’, presented at WARSD'03.
Mangrove forests are an integral part of tropical coastal ecosystems, particularly in northern Australia. In the Northern Territory, studies have determined the extent and species diversity of these associations but little is known of biomass or productivity. We sampled the above- and below-ground biomass of the four most abundant species, Avicennia marina, Bruguiera exaristata, Ceriops australis and Rhizophora stylosa, developed allometric relationships and examined partitioning. Unlike many other studies, we sampled below-ground biomass, which constituted a substantial proportion (0.29-0.57) of the total dry weight. Our results should be valuable in modelling potential changes in carbon allocation resulting from small- and large-scale ecosystem changes.
The storage and flux of various mineral and trace elements in soils (0–30?cm depth) were examined in relation to monsoonal rains and fine root biomass in four mangrove forests of different age and type in southern Thailand. The onset of the wet SW monsoon resulted in the percolation and dilution of porewater solutes by rainwater and by less saline tidal water, as indicated by shifts in Eh, pH and porewater SO4/Cl ratios. This is contrary to temperate intertidal environments where seasonal patterns of porewater constituents, and biological and biogeochemical activities, are strongly cued to temperature. Fluxes across the soil–water interface were most often not statistically significant. Concentration of dissolved porewater metals were dominated by Fe, Mn, Al, Mo and Zn. The decreasing order of solid-phase element inventories in these soils, on average, was: Al, S, Fe, Na, Mg, K, Ca, N, P, Mn, V, Zn, Cr, Ni, As, Co, Cu, Pb, Mo, Cd and Hg. There were no gradients in concentrations of dissolved or solid-phase elements with increasing soil depth. This phenomenon was attributed to physical and biological processes, including the presence and activities of roots and tidal recharge of soil water. Fine dead roots were storage sites for most mineral and trace elements, as some elements in roots composed a significant fraction (=5%) of the total soil pool. Analysis of S and Fe concentration differences between live and dead roots suggested extensive formation of pyrite associated with dead roots; correlation analysis suggested that trace metals coprecipitated with pyrite. An analysis of inventories and release/uptake rates indicate turnover of the N, P, Na and Ca soil pools equivalent to other tropical forests; turnover was slow (decades to centuries) for S, Fe, K and trace elements. Our results indicate that mineral and trace element cycling in these soils are characterized by net storage, with net accumulation of most elements much greater than uptake and release by tree roots.
An accurate global carbon budget requires information on terrestrial carbon sink strength. Regenerating tropical forests are known to be important terrestrial carbon sinks but information on their location, extent and biomass (from which carbon content can be estimated) is incomplete. The use of remotely sensed data in optical wavelengths has been of limited use due to both the weak relationship between optical radiation and forest biomass and near‐constant cloud cover in the tropics. L‐band Synthetic Aperture Radar (SAR) backscatter, however, is related positively to biomass (but only up to an asymptote of around 40–90 T ha −1 ) and can be obtained independently of cloud cover. Both canopy structure and biomass change over time as pioneer species are replaced by early and late regenerating species. These structural changes are related to an increase in (i) tree height, (ii) tree species richness and (iii) canopy thickness and influence the roughness of the canopy surface and consequently SAR image texture. Therefore, we investigated the degree to which textural information could be used to increase the correlation between image tone (backscatter) and biomass. Field data were used to estimate the biomass of 37 regenerating forests plots in Brazilian Amazonia. Texture measures derived from local statistics, the grey level co‐occurrence matrix (GLCM) and the variogram were evaluated using simulated images on the basis of their ability to identify significant differences in image texture independently of image contrast. The selected texture measures were applied to L‐band JERS‐1 (Japanese Earth Resources Satellite) SAR images and the correlation between backscatter and biomass was determined for regenerating tropical forests. A strong correlation was found for the texture measures and biomass. The r a 2 (adjusted coefficient of determination), measuring the correlation between backscatter and biomass, increased from 0.74 to 0.82 with the addition of GLCM‐derived contrast. The addition of image texture (GLCM‐derived contrast) to image tone (backscatter) potentially increases the accuracy with which JERS‐1 SAR data can be used to estimate biomass in tropical forests.
Mangroves, the only woody halophytes living at the confluence of land and sea, have been heavily used traditionally for food, timber, fuel and medicine, and presently occupy about 181 000 km2 of tropical and subtropical coastline. Over the past 50 years, approximately one-third of the world's mangrove forests have been lost, but most data show very variable loss rates and there is considerable margin of error in most estimates. Mangroves are a valuable ecological and economic resource, being important nursery grounds and breeding sites for birds, fish, crustaceans, shellfish, reptiles and mammals; a renewable source of wood; accumulation sites for sediment, contaminants, carbon and nutrients; and offer protection against coastal erosion. The destruction of mangroves is usually positively related to human population density. Major reasons for destruction are urban development, aquaculture, mining and overexploitation for timber, fish, crustaceans and shellfish. Over the next 25 years, unrestricted clear felling, aquaculture, and overexploitation of fisheries will be the greatest threats, with lesser problems being alteration of hydrology, pollution and global warming. Loss of biodiversity is, and will continue to be, a severe problem as even pristine mangroves are species-poor compared with other tropical ecosystems. The future is not entirely bleak. The number of rehabilitation and restoration projects is increasing worldwide with some countries showing increases in mangrove area. The intensity of coastal aquaculture appears to have levelled off in some parts of the world. Some commercial projects and economic models indicate that mangroves can be used as a sustainable resource, especially for wood. The brightest note is that the rate of population growth is projected to slow during the next 50 years, with a gradual decline thereafter to the end of the century. Mangrove forests will continue to be exploited at current rates to 2025, unless they are seen as a valuable resource to be managed on a sustainable basis. After 2025, the future of mangroves will depend on technological and ecological advances in multi-species silviculture, genetics, and forestry modelling, but the greatest hope for their future is for a reduction in human population growth.
In this paper, the use of airborne polarimetric Synthetic Aperture Radar (SAR) for retreiving information on the structure and biomass of mangroves in the Pacific (Australia) and Atlantic (French Guiana) oceanic regions is considered. The study makes a comparison between the biomass and structure of the different communities and how these impact on the SAR response at different frequencies and polarisations. The future potential of SAR for generating baseline datasets of mangrove biomass and structure against which to evaluate change is suggested.
1] Leaf pigment content and composition provide important information about plant physiological status. Reflectance measurements offer a rapid, nondestructive technique to estimate pigment content. This paper describes a recently developed three-band conceptual model capable of remotely estimating total of chlorophylls, carotenoids and anthocyanins contents in leaves from many tree and crop species. We tuned the spectral regions used in the model in accord with pigment of interest and the optical characteristics of the leaves studied, and showed that the developed technique allowed accurate estimation of total chlorophylls, carotenoids and anthocyanins, explaining more than 91%, 70% and 93% of pigment variation, respectively. This new technique shows a great potential for noninvasive tracking of the physiological status of vegetation and the impact of environmental changes. (2006), Three-band model for noninvasive estimation of chlorophyll, carotenoids, and anthocyanin contents in higher plant leaves, Geophys. Res. Lett., 33, L11402, doi:10.1029/ 2006GL026457.
Accurate forest biomass estimation is essential for greenhouse gas inventories, terrestrial carbon accounting, and climate change modeling studies. Unfortunately, no universal and transferable technique has been developed so far to quantify biomass carbon sources and sinks over large areas because of the environmental, topographic, and biophysical complexity of forest ecosystems. Among the remote sensing techniques tested, the use of multisensors and the spatial as well as the spectral characteristics of the data have demonstrated a strong potential for forest biomass estimation. However, the use of multisensor data accompanied by spatial data processing has not been fully investigated because of the unavailability of appropriate data sets and the complexity of image processing techniques in combining multisensor data with the analysis of the spatial characteristics. This paper investigates the texture parameters of two high resolution (10 m) optical sensors (Advanced Visible and Near Infrared Radiometer type 2 (AVNIR-2) and SPOT-5) in different processing combinations for biomass estimation. Multiple regression models are developed between image parameters extracted from the different stages of image processing and the biomass of 50 field plots, which was estimated using a newly developed "allometric model" for the study region. The results demonstrate a clear improvement in biomass estimation using the texture parameters of a single sensor (r<sup>2</sup> = 0.854 and rmse = 38.54) compared to the best result obtained from simple spectral reflectance (r<sup>2</sup> = 0.494) and simple spectral band ratios (r2 = 0.59). This was further improved to obtain a very promising result using the texture parameter of both sensors together (r<sup>2</sup> = 0.897 and rmse = 32.38), the texture parameters from the principal component analysis of both sensors (r<sup>2</sup> = 0.851 and rmse = 38.80), and the texture parameters from the av eraging of both sensors (r<sup>2</sup> = -
-
0.911 and rmse = 30.10). Improvement was also observed using the simple ratio of the texture parameters of AVNIR-2 (r<sup>2</sup> = 0.899 and rmse = 32.04) and SPOT-5 (r<sup>2</sup> = 0.916), and finally, the most promising result (r<sup>2</sup> = 0.939 and rmse = 24.77) was achieved using the ratios of the texture parameters of both sensors together. This high level of agreement between the field and image data derived from the two novel techniques (i.e., combination/fusion of the multisensor data and the ratio of the texture parameters) is a very significant improvement over previous work where agreement not exceeding r<sup>2</sup> = 0.65 has been achieved using optical sensors. Furthermore, biomass estimates of up to 500 t/ha in our study area far exceed the saturation levels observed in other studies using optical sensors.
There is currently a great deal of interest in the quantitative characterization of temporal and spatial vegetation patterns with remotely sensed data for the study of earth system science and global change. Spectral models and indices are being developed to improve vegetation sensitivity by accounting for atmosphere and soil effects. The soil-adjusted vegetation index (SAVI) was developed to minimize soil influences on canopy spectra by incorporating a soil adjustment factor L into the denominator of the normalized difference vegetation index (NDVI) equation. For optimal adjustment of the soil effect, however, the L factor should vary inversely with the amount of vegetation present. A modified SAVI (MSAVI) that replaces the constant L in the SAVI equation with a variable L function is presented in this article. The L function may be derived by induction or by using the product of the NDVI and weighted difference vegetation index (WDVI). Results based on ground and aircraft-measured cotton canopies are presented. The MSAVI is shown to increase the dynamic range of the vegetation signal while further minimizing the soil background influences, resulting in greater vegetation sensitivity as defined by a “vegetation signal” to “soil noise” ratio.
The sensitivity of the normalized difference vegetation index (NDVI) to soil background and atmospheric effects has generated an increasing interest in the development of new indices, such as the soil-adjusted vegetation index (SAVI), transformed soil-adjusted vegetation index (TSAVI), atmospherically resistant vegetation index (AR VI), global environment monitoring index (GEMI), modified soil-adjusted vegetation index (MSAVI), which are less sensitive to these external influences. These indices are theoretically more reliable than NDVI, although they are not yet widely used with satellite data. This article focuses on testing and comparing the sensitivity of NDVI, SAVI, TSAVI, MSAVI and GEMI to soil background effects. Indices are simulated with the SAIL model for a large range of soil reflectances, including sand, clay, and dark peat, with additional variations induced by moisture and roughness. The general formulation of the SAVI family of indices with the form VI = (NIR - R) / (NIR + R + X) is also reexamined. The value of the parameter X is critical in the minimization of soil effects. A value of X = 0.16 is found as the optimized value. Index performances are compared by means of an analysis of variance.
Farmers must balance the competing goals of supplying adequate N for their crops while minimizing N losses to the environment. To characterize the spatial variability of N over large fields, traditional methods (soil testing, plant tissue analysis, and chlorophyll meters) require many point samples. Because of the close link between leaf chlorophyll and leaf N concentration, remote sensing techniques have the potential to evaluate the N variability over large fields quickly. Our objectives were to (1) select wavelengths sensitive to leaf chlorophyll concentration, (2) simulate canopy reflectance using a radiative transfer model, and (3) propose a strategy for detecting leaf chlorophyll status of plants using remotely sensed data. A wide range of leaf chlorophyll levels was established in field-grown corn (Zea mays L.) with the application of 8 N levels: 0%, 12.5%, 25%, 50%, 75%, 100%, 125%, and 150% of the recommended rate. Reflectance and transmittance spectra of fully expanded upper leaves were acquired over the 400-nm to 1,000-nm wavelength range shortly after anthesis with a spectroradiometer and integrating sphere. Broad-band differences in leaf spectra were observed near 550 nm, 715 nm, and >750 nm. Crop canopy reflectance was simulated using the SAIL (Scattering by Arbitrarily Inclined Leaves) canopy reflectance model for a wide range of background reflectances, leaf area indices (LAI), and leaf chlorophyll concentrations. Variations in background reflectance and LAI confounded the detection of the relatively subtle differences in canopy reflectance due to changes in leaf chlorophyll concentration. Spectral vegetation indices that combined near-infrared reflectance and red reflectance (e.g., OSAVI and NIR/Red) minimized contributions of background reflectance, while spectral vegetation indices that combined reflectances of near-infrared and other visible bands (MCARI and NIR/Green) were responsive to both leaf chlorophyll concentrations and background reflectance. Pairs of these spectral vegetation indices plotted together produced isolines of leaf chlorophyll concentrations. The slopes of these isolines were linearly related to leaf chlorophyll concentration. A limited test with measured canopy reflectance and leaf chlorophyll data confirmed these results. The characterization of leaf chlorophyll concentrations at the field scale without the confounding problem of background reflectance and LAI variability holds promise as a valuable aid for decision making in managing N applications.
Usage of remote sensing technology has increased for estimation of Above Ground Biomass (AGB) due to accessibility issues within the forests. Current study is piloted for mangroves of Mundra taluka in Kachchh district, western India. The study aims to validate and compare various allometric models for mangrove Above Ground Biomass (AGB) estimation to find out best suitable model. Models are validated by extensive ground truth. The most suitable allometric model is then used to develop a novel methodology to estimate AGB using dual-pol Sentinel-1A Synthetic Aperture Radar (SAR) data. This study is first of its kind, implemented for Indian mangroves, using Polarimetric Radar Vegetation Index (PRVI) for modelling AGB. The study also incorporates time series change detection in AGB within the study area. Our results suggest that amalgamation of field measurements, allometric equations and remote sensing technology can be the vital solution for modelling Above Ground Biomass of mangrove forest.
Different approaches to the classification of remotely sensed data of mangroves are reviewed, and five different methodologies identified. Landsat TM, SPOT XS and CASI data of mangroves from the Turks and Caicos Islands, were classified using each method. All classifications of SPOT XS data failed to discriminate satisfactorily between mangrove and non-mangrove vegetation. Classification accuracy of CASI data was higher than Landsat TM for all methods, and more mangrove classes could be discriminated. Merging Landsat TM and SPOT XP data improved visual interpretation of images, but did not enhance discrimination of different mangrove categories. The most accurate combination of sensor and image processing method for mapping the mangroves of the eastern Caribbean islands is identified.
Plenty of feature selection methods are available in literature due to the availability of data with hundreds of variables leading to data with very high dimension. Feature selection methods provides us a way of reducing computation time, improving prediction performance, and a better understanding of the data in machine learning or pattern recognition applications. In this paper we provide an overview of some of the methods present in literature. The objective is to provide a generic introduction to variable elimination which can be applied to a wide array of machine learning problems. We focus on Filter, Wrapper and Embedded methods. We also apply some of the feature selection techniques on standard datasets to demonstrate the applicability of feature selection techniques.
Inventory data on tree weights of 104 individual trees representing 10 mangrove species were collected from mangrove forests in South-East Asia to establish common allometric equations for the trunk, leaf, above-ground and root weight. We used the measurable tree dimensions, such as dbh (trunk diameter at breast height), DR0.3 (trunk diameter at 30 cm above the highest prop root of Rhizophora species), DB (trunk diameter at lowest living branch), and H (tree height) for the independent variable of equations. Among the mangrove species studied, the trunk shape was statistically identical regardless of site and species. However, of each species, when DB2 or dbh2 or DR0.32 was selected as the independent variable. For the root weight, the common equation was derived from the allometric relationship between root weight and above-ground weight, since these two partial weights significantly correlated with each other. Based on these physical and biological parameters, we have proposed four common allometric equations for estimating the mangrove tree weight of trunk, leaf, above-ground part and root.
Allometric relationships are described for estimating leaf biomass, branch biomass, stem biomass and total above-ground biomass from measurements of stem diameter (DBH) in the mangrove species Rhizophora apiculata, R. stylosa, Bruguiera gymnorrhiza, B. parviflora, Ceriops tagal var. australis and Xylocarpus granatum. A linear relationship was found when the biomass of each above-ground component was plotted against DBH on a log-log scale. The two Rhizophora species were found to have the greatest stem and total above-ground biomass for a given DBH, followed by B. parviflora, B. gymnorrhiza, C. tagal var. australis, and X. granatum, the last having a significantly lower biomass for a given DBH than the other five species. However, there was much less variation in stem volume for a given DBH amongst the six species, owing to differences in the specific gravity of their stems.
In this paper, a new methodology to estimate the biomass (organic matter) of conifer-dominated boreal forests is developed. It aims to estimate biomass of extensive areas where ground data are limited. First, the principal models are computed using ground measurements and high resolution satellite images. Spectral models are then applied directly to a calibrated AVHRR image mosaic covering the entire area of interest. This methodology was tested quantitatively in Finland, where detailed forest measurement data are available, on an area reaching from the west coast of Norway to the Ural mountains. The methodology appeared to perform beyond pre-test expectation.
In this paper, a multiscale texture-based classifier for mapping tropical forest land cover types is discussed. The classifier was implemented using the Japanese Earth Remote Sensing Satellite (JERS-1) 100 m resolution radar data acquired over the Amazon Rainforest as part of the Global Rainforest Mapping (GRFM) Project. Demonstrated here is the use of the information content present in different texture measurements at different scales to separate three categories of land cover types: forest from nonforest, terre firme from floodplain vegetation, and grassland from woodland savanna. Various combinations of first-order image statistics known as texture measures were used at different scales as feature dimensions to aid the class discrimination. Eight of the most common first-order texture measures found in the literature were used. The best combination of texture measures at each scale were determined by employing a class separability test using the Bhattachuryya distance. The results were then used as input images into a supervised multiscale maximum likelihood estimation classifier. The classified maps were validated against independent test sites, and by comparison with a Landsat Thematic Mapper (TM) classification. It was found that JERS-1 backscatter and texture measures can discriminate forest from nonforest types with very high accuracy (above 90%). Old secondary forest or regrowth areas were often mixed with forest. Radar backscatter alone was able to separate terre firme and floodplain vegetation. However, texture measures were important in separating open from dense floodplain vegetation. Similarly, the backscatter sensitivity to low biomass values was instrumental in separating woodland from grassland savanna. Texture had a lesser role in separating these two vegetation types but was important to separate the woodland savanna from dense evergreen forest and secondary forests.
Studies are needed to evaluate the ability of present or future SAR data to extract forest attributes over various sites. This study focuses on large unmanaged pine plantations in a vast flat area of ca. 500,000 ha where the tree biomass ranges from 0 to 200 m3/ha corresponding to different forest canopy structures. Results show a good correlation between the backscattering coefficient σ0 (with a 6-dB dynamic range and R2≥.8), the stand timber volume and the stand density. The trend is mainly driven by stand density and different relationships are observed according to age class, which explicitly points out the effects of canopy structure on the backscattering level. Stem volume is derived from the inversion of statistical and semiempirical models, which take these effects into account. Inversion results show that forest biomass attributes can be estimated with relatively small errors commensurate with those achieved by field measurements. Best overall accuracy of ca. 21 m3/ha is reached with the semiempirical model. Error decomposition as a function of age classes shows that, for the same biomass range, errors are higher for old stands than for young stands. Finally, the results indicate that (1) JERS-1 data can be used in an operational way for estimating the biomass of such plantations and (2) it is necessary to take forest stand structure into account. In order to develop reliable biomass-retrieval schemes, future research should focus on examining in a more mechanistic way the relationship between canopy structure and radar signature.
A Modified Simple Ratio (MSR) Is proposed for retrieving biophysical parameters of boreal forests using remote sensing data. This vegetation index is formulated based on an evaluation of several two-band vegetation indices, including the Normalized Difference Vegetation Index (NDVI), Simple Ratio (SR), Soil Adjusted Vegetation Indices (SAVI, SAVI1, SAVI2), Weighted Difference Vegetation Index (WDVI), Global Environment Monitoring Index (GEMI), Non-Linear Index (NLI), and Renormalized Difference Vegetation Index (RDVI). MSR is an improved version of RDVI for the purpose of linearizing their relationships with biophysical parameters. All indices were obtained from Landsat-5 TM band 3 (visible) and band 4 (near infrared) images after atmospheric corrections (except for GEMI) and were correlated with ground-based measurements made in 20 Jack Pine (Pinus banksiana) and Black Spruce (Picea mariana) stands during the BOREAS field experiment in 1994. The measurements include Leaf Area Index (LAI) and the Fraction of Photosynthetically Active Radiation (FPAR) absorbed by the forest canopies. Among these vegetation indices, SR, MSR, and NDVI were found to be best correlated with LAI and FPAR in both spring and summer. All other indices performed poorly. Both NDVI and MSR can be expressed as a function of SR. Measurement errors in remote sensing data often occur due to changes in solar zenith angle, subpixel contamination of clouds, or dissimilar surface features and the variation in the local topography and other environmental factors. These errors generally cause simultaneous increases or decreases in the red and near infrared reflectances, and their effects can be greatly reduced by taking the ratio. All other indices involving mathematical operations other than ratioing could retain the errors or even amplify them. The major problem in using the vegetation indices obtained from red and near infrared bands is the small sensitivity to the overstorey vegetation conditions. Although many of the vegetation indices such as SAVI, SAVI1, and SAVI2 are developed to minimize the effect of the background on retrieving the vegetation information, they also reduce their sensitivity to the changes in the overstorey conditions.
This paper reports on a test of the ability to estimate above-ground biomass of tropical secondary forest from canopy spectral reè ectance using satel- lite optical data. Landsat Thematic Mapper data were acquired concurrent with é eld surveys conducted in secondary forest fallows near Manaus, Brazil and Santa Cruz de la Sierra, Bolivia. Measurements of age and above-ground live biomass were made in 34 regrowth stands. Satellite data were converted to surface reè ectances and compared with regrowth stand age, biomass and structural variables. Among the Brazilian stands, signié cant relationships were observed between middle-infrared reè ectance and stand age, height, volume and biomass. The canopy reè ectance- biomass relationship saturated at around 15.0kgm Õ 2, or over 15 years of age ( r>0.80, p
An experiment has been conducted in which narrow-band field reflectance spectra were acquired of a roofed pinyon pine canopy with Fee different gravel backgrounds. Leaf area teas successively removed as the measurements were repeated. From these reflectance spectra, narrow-band and broad-band (AVHRR, TM, MSS) red and near-infrared (NIR) vegetation index values were calculated. The performance of the vegetation indices was evaluated based on their capability to accurately estimate leaf area index (LAI) and percent green cover. Background effects were found for each of the tested vegetation indices. However the background effects are most pronounced in the normalized difference vegetation index (NDVI) and ratio vegetation index (RVI). Background effects can be reduced using either the perpendicular vegetation index (PVI) or soil adjusted vegetation index (SAVI) formulations. The narrow-band versions of these vegetation indices had only slightly better accuracy than their broad-band counterparts. The background effects were minimized using derivative based vegetation indices, which measure the amplitude of the chlorophyll red-edge using continuous narrow-band spectra from 626 nm to 795 nm.
Data available from here:
https://datadryad.org/stash/dataset/doi:10.5061/dryad.234
Remotely sensed vegetation indices such as NDVI, computed using the red and near infrared bands have been used to estimate pasture biomass. These indices are of limited value since they saturate in dense vegetation. In this study, we evaluated the potential of narrow band vegetation indices for characterizing the biomass of Cenchrus ciliaris grass measured at high canopy density. Three indices were tested: Modified Normalized Difference Vegetation Index (MNDVI), Simple Ratio (SR) and Transformed Vegetation Index (TVI) involving all possible two band combinations between 350 nm and 2500 nm. In addition, we evaluated the potential of the red edge position in estimating biomass at full canopy cover. Results indicated that the standard NDVI involving a strong chlorophyll absorption band in the red region and a near infrared band performed poorly in estimating biomass (R2=0.26). The MNDVIs involving a combination of narrow bands in the shorter wavelengths of the red edge (700-750 nm) and longer wavelengths of the red edge (750-780 nm), yielded higher correlations with biomass (mean R2=0.77 for the highest 20 narrow band NDVIs). When the three vegetation indices were compared, SR yielded the highest correlation coefficients with biomass as compared to narrow band NDVI and TVI (average R2=0.80, 0.77 and 0.77 for the first 20 ranked SR, NDVI and TVI respectively). The red edge position yielded comparable results to the narrow band vegetation indices involving the red edge bands. These results indicate that at high canopy density, pasture biomass may be more accurately estimated by vegetation indices based on wavelengths located in the red edge than the standard NDVI.
The biomass and biomass dynamics of forests are major uncertainties in our understanding of tropical environments. Remote sensing is often the only practical means of acquiring information on forest biomass but has not always been used successfully. Here the conventional approaches to the estimation of forest biomass from remotely sensed data were evaluated relative to techniques based on the application of artificial neural networks. Together these approaches were used to estimate and map the biomass of tropical forests in north-eastern Borneo from Landsat TM data. The neural networks were found to be particularly suited to the application. A basic multi-layer perceptron network, for example, provided estimates of biomass that were strongly correlated with those measured in the field (r = 0.80). Moreover, these estimates were more strongly correlated with biomass than those derived from 230 conventional vegetation indices, including the widely used normalized difference vegetation index (NDVI).
Reflectance spectra in the visible and near infra-red range of the spectrum, acquired for maple (Acer platanoides L.), chestnut (Aesculus hippocastanum L.), potato (Solanum tuberosum L.), coleus (Coleus blumei Benth.), leaves and lemon (Citrus limon L.) and apple (Malus domestica Borkh.) fruits were studied. An increase of reflectance between 550 and 740 nm accompanied senescence-induced degradation of chlorophyll (Chl), whereas in the range 400–500 nm it remained low, due to retention of carotenoids (Car). It was found that both leaf senescence and fruit ripening affect the difference between reflectance (R) near 670 and 500 nm (R678−R500), depending on pigment composition. The plant senescing reflectance index in the form (R678−R500)/R750 was found to be sensitive to the Car/Chl ratio, and was used as a quantitative measure of leaf senescence and fruit ripening. The changes in the index were followed during leaf senescence, and natural and ethylene-induced fruit ripening. This novel index can be used for estimating the onset, the stage, relative rates and kinetics of senescence/ripening processes.
Scanning Light Detecting and Ranging (LiDAR), Synthetic Aperture Radar (SAR) and Interferometric SAR (InSAR) were analyzed to determine (1) which of the three sensor systems most accurately predicted forest biomass, and (2) if LiDAR and SAR/InSAR data sets, jointly considered, produced more accurate, precise results relative to those same data sets considered separately. LiDAR ranging measurements, VHF–SAR cross-sectional returns, and X- and P-band cross-sectional returns and interferometric ranges were regressed with ground-estimated (from dbh) forest biomass in ponderosa pine forests in the southwestern United States. All models were cross-validated. Results indicated that the average canopy height measured by the scanning LiDAR produced the best predictive equation. The simple linear LiDAR equation explained 83% of the biomass variability (n = 52 plots) with a cross-validated root mean square error of 26.0 t/ha. Additional LiDAR metrics were not significant to the model. The GeoSAR P-band (λ = 86 cm) cross-sectional return and the GeoSAR/InSAR canopy height (X–P) captured 30% of the forest biomass variation with an average predictive error of 52.5 t/ha. A second RaDAR–FOPEN collected VHF (λ ∼ 7.8 m) and cross-polarized P-band (λ = 88 cm) cross-sectional returns, none of which proved useful for forest biomass estimation (cross-validated R2 = 0.09, RMSE = 63.7 t/ha). Joint consideration of LiDAR and RaDAR measurements produced a statistically significant, albeit small improvement in biomass estimation precision. The cross-validated R2 increased from 83% to 84% and the prediction error decreased from 26.0 t/ha to 24.9 t/ha when the GeoSAR X–P interferometric height is considered along with the average LiDAR canopy height. Inclusion of a third LiDAR metric, the 60th decile height, further increased the R2 to 85% and decreased the RMSE to 24.1 t/ha. On this 11 km2 ponderosa pine study area, LiDAR data proved most useful for predicting forest biomass. RaDAR ranging measurements did not improve the LiDAR estimates.
This study was part of an interdisciplinary research project on soil carbon and phytomass dynamics of boreal and arctic permafrost landscapes. The 45 ha study area was a catchment located in the forest tundra in northern Siberia, approximately 100 km north of the Arctic Circle.The objective of this study was to estimate aboveground carbon (AGC) and assess and model its spatial variability. We combined multi-spectral high resolution remote sensing imagery and sample based field inventory data by means of the k-nearest neighbor (k-NN) technique and linear regression.Field data was collected by stratified systematic sampling in August 2006 with a total sample size of n = 31 circular nested sample plots of 154 m2 for trees and shrubs and 1 m2 for ground vegetation. Destructive biomass samples were taken on a sub-sample for fresh weight and moisture content. Species-specific allometric biomass models were constructed to predict dry biomass from diameter at breast height (dbh) for trees and from elliptic projection areas for shrubs.Quickbird data (standard imagery product), acquired shortly before the field campaign and archived ASTER data (Level-1B product) of 2001 were geo-referenced, converted to calibrated radiances at sensor and used as carrier data. Spectral information of the pixels which were located in the inventory plots were extracted and analyzed as reference set. Stepwise multiple linear regression was applied to identify suitable predictors from the set of variables of the original satellite bands, vegetation indices and texture metrics. To produce thematic carbon maps, carbon values were predicted for all pixels of the investigated satellite scenes. For this prediction, we compared the kNN distance-weighted classifier and multiple linear regression with respect to their predictions.The estimated mean value of aboveground carbon from stratified sampling in the field is 15.3 t/ha (standard error SE = 1.50 t/ha, SE% = 9.8%). Zonal prediction from the k-NN method for the Quickbird image as carrier is 14.7 t/ha with a root mean square error RMSE = 6.42 t/ha, RMSEr = 44%) resulting from leave-one-out cross-validation. The k-NN-approach allows mapping and analysis of the spatial variability of AGC. The results show high spatial variability with AGC predictions ranging from 4.3 t/ha to 28.8 t/ha, reflecting the highly heterogeneous conditions in those permafrost-influenced landscapes. The means and totals of linear regression and k-NN predictions revealed only small differences but some regional distinctions were recognized in the maps.
In this study, a hybrid genetic algorithm is adopted to find a subset of features that are most relevant to the classification task. Two stages of optimization are involved. The outer optimization stage completes the global search for the best subset of features in a wrapper way, in which the mutual information between the predictive labels of a trained classifier and the true classes serves as the fitness function for the genetic algorithm. The inner optimization performs the local search in a filter manner, in which an improved estimation of the conditional mutual information acts as an independent measure for feature ranking taking account of not only the relevance of the candidate feature to the output classes but also the redundancy to the already-selected features. The inner and outer optimizations cooperate with each other and achieve the high global predictive accuracy as well as the high local search efficiency. Experimental results demonstrate both parsimonious feature selection and excellent classification accuracy of the method on a range of benchmark data sets.