Error Matrix Method Research Articles

Enhancing precision in coastal dunes vegetation mapping: ultra-high resolution hierarchical classification at the individual plant level

ABSTRACT The classification of ultra-high-resolution (UHR) imagery, characterized by spatial resolutions exceeding 10 cm, presents opportunities and challenges distinct from lower-resolution counterparts. Particularly, challenges are pronounced in some scenarios, such as mapping plant species in coastal environments, where similar vegetation responses and small plant sizes pose additional difficulties. The present work addressed such issues by developing a UHR vegetation cover classification model at the single plant level using data from uncrewed aerial systems (UASs) equipped with a multispectral optical sensor. The model was tested across the San Rossore Regional Park (Italy), where three pilot areas were defined as training-test-validation sites. The proposed solution consists of a hierarchical two-level-of-detail machine learning model based on object-based image analysis (OBIA) and random forest. This model considers spectral features and indices, elevation, and texture and can classify twelve plant species and two service classes (debris and sand) within the study areas. Train and test were carried out utilizing UAS flight data collected during two specific phenological periods and precise field data derived from in-situ vegetation surveys, which provided 937 herbaceous and shrub samples. The model performance was evaluated based on the error matrix and 50-fold stratified cross-validation method, obtaining an overall accuracy (OA) of 0.76 and a standard deviation of 0.08. Such assessment underscored the crucial role of texture information, in addition to radiometric and elevation. Finally, the model was tested against an unseen dataset, proving its transferability (OA equal to 0.62). Although the discussion highlights some aspects to be further improved and claims for future research, the first version of this hierarchical classification model demonstrated its potential for mapping and monitoring coastal sand dune ecosystems, providing data for understanding and, eventually, modeling ecological and biogeomorphological dynamics.

Spatiotemporal Variability Analysis of Glaciers in the Hindukush Region of Pakistan Using Remote Sensing Data

Headwater in the Indus River in Pakistan is largely dependent on the glaciers located in the northern part of the country, along with other sources such as direct precipitation. Glaciers are a major source of freshwater that provides agriculture and livelihood to millions of people. The hydro-climatic variations in the Gilgit watershed of the Upper Indus basin are poorly investigated scientifically due to high topographical differences, geography, remoteness of the region, and larger variations in climatic conditions. These glaciers are continuously changing due to melting as a consequence of global warming or accumulation due to snowfall/precipitation at higher altitude regions. The study is carried out using remote sensing data to quantify glacier changes in spatiotemporal variability in the past three decades. Five glaciers in the Gilgit region (near the junction of the Hindukush and Karakoram Mountains) with an area of more than 5 square kilometers were selected, namely Phakor, Karamber, East Gammu, Bhort, and Bad-e-Swat glaciers. These glaciers were monitored for changes in their sizes through a cloud-free continuous series of Landsat satellite imagery. The annual climatic trends were studied through spatially interpolated gridded climate data WοrldClim version-1 climate database for 1970–2000, utilized for assessment of meteorological condition by analyzing the variations of minimum and maximum temperature, solar radiation, and precipitation. The temporal variations in five glaciers in the Gilgit watershed are found to be minimal and, thus, are rather stable and show no sign of rapid melting or diminishing. The little variability of glaciers’ extent may be attributed to their geographic condition, altitude, topography, and orientation. The mapped glacier classes have been validated to check the accuracy assessment through an error matrix method. The kappa coefficient from the error matrix has been calculated as 84%, which shows a good agreement. The study makes a critical input towards understanding the dynamics of the glacier in the upper Indus catchment’s Gilgit watershed.

SOIL SALINITY MAPPING USING SATELLITE REMOTE SENSING: A CASE STUDY OF LOWER CHENAB CANAL SYSTEM, PUNJAB

Salinity is the most important factor of consideration for the water management policies. The water availability from the rootzone reduced with the increase in the soil salinity due to the increase in the osmatic pressure. In Pakistan, salinity is the major threat to the agriculture land due to the tradition practices of irrigation and extensive utilization of the groundwater to meet the cope the irrigation water requirement of high intensity cropping system. The salinity impact is spatially variable on the canal commands area of the irrigation system. There is dire need to map the spatially distributed soil salinity with the high resolution. Landsat satellite imagery provides an opportunity to have 30m pixel information in seven spectral wavelength ranges. In this study, the soil salinity mapping was performed using pixel information on visible and infrared bands for 2015. These bands were also used to infer Normalized Difference Vegetation Index (NDVI). The raw digital numbers were converted into soil salinity information. The accuracy assessment was carried out using ground trothing information obtained using the error matrix method. Four major classes of non-saline, marginal saline, moderate saline and strongly, saline area was mapped. The overall accuracy of the classified map was found 83%. These maps can be helpful to delineate hot spots with severe problem of soil salinity in order to prepare reciprocate measures for improvement.

Comparison of Logistic Regression and Geomod Approaches to Forest Change Modeling in the Period of 1988 – 2025

Spatial modelling of land use change is a technique for understanding changes in terms of the location and amount. In this study, logistic regression and Geomod approaches were used for modelling forest change in Gorgan area in Northern Iran in the time period of 1988-2007. To do this, at first, remotely sensed imagery data of the years 1988, 1998 and 2007 were used to produce land use maps. Land use maps accuracy assessments were achieved using Error matrix method and then the maps were used to implement change detection process in two time periods of 1988-1998 and 1998-2007. Results indicated a reduction in forest areas during the mentioned time period. Next, the independent variables were extracted in order to land use change modeling. The Results of the models implementation showed the ability of both models for forest change modeling in this region. Also, the models were used to predict the future condition of forest area in the years 2016 and 2025. The results revealed that the forest area would be associated with a reduction in the future. Comparison of the results of the models using kappa indices showed the successful implementation of both models for forest change modelling in this region. The results of this research reveal the need for appropriate applications of the proper plans to control land use change in order to preserve the environment and ecological balance of the area. Therefore, careful planning can reduce the land use change and its impacts in the future in this region.

Analysis of land use/land cover changes using geospatial techniques in Salem district, Tamil Nadu, South India

Urbanization is a global phenomenon, but the rate of urban growth is relatively faster in developing countries like India. The drastic urban growth in India has resulted in a huge loss of agricultural land and water bodies, which needs to be quantified for future planning and development. In this context, the present study depicts a combined approach of remote sensing and GIS techniques for delineating the land use/land cover dynamics of Salem district of Tamil Nadu in South India. The Landsat Thematic Mapper data of the years 1992, 2001 and 2010 and Operational Land Imager data of the year 2015 were used to assess the changes in the land use and land cover. The supervised classification technique-maximum likelihood algorithm, was used in this study to classify the remote sensing data. The results indicate that the deciduous forests, croplands, cultivation and plantation lands and water bodies have decreased from 1992 to 2015 by about 398 km2, 250 km2, 45 km2 and 16 km2, respectively. At the same time, evergreen and semi-evergreen forests, built-up lands and mines and barren lands have increased by about 288 km2, 293 km2 and 128 km2, respectively. The accuracy of the present study has been assessed by using error matrix method. The overall accuracy of the land use/land cover classification of the years 1992, 2001, 2010 and 2015 is about 88%, 95%, 93% and 93%, respectively. The study showed that the combination of remote sensing and GIS is an efficient tool to detect the changes in land use/land cover. The land use/land cover information thus obtained can be used to the decision-makers for effective land use planning and management.

Heuristic Evaluation of Groundwater in Arid Zones Using Remote Sensing and Geographic Information System

The purpose of this study is to identify potential areas of groundwater using remote sensing and GIS. Therefore, some of the effective layers were extracted such as geology, fractures and faults, geomorphology, slope, land use and regional drainage density using ETM sensing imagery processes such as producing various feature class codes, edge detection filters, regulated classification and imposing vegetation indices conducted by topographic maps of 1:50,000, geology and DEM. All layers were classified depending on their effectiveness based on the expert opinions, using analytic hierarchy process. They were also weighted in different classes. After modeling in GIS, groundwater’s potential of Semnan plains was determined in Iran. The results showed that the existence of fractures and fault in sandstone formations of Shemshak and gunpowder as well as thick limestone of the layer-formation of Lar have led to name these areas as highly well potential areas of groundwater. Then after, alluvial valleys, Stream sediments, foothills alluvial fans and alluvial plains have formed the well potential areas. Using the discharge of the 54 fountains, aqueduct and wells; the accuracy of the map was estimated through an error matrix method with an overall accuracy of 79.63 and Kappa coefficient of 0.702, which implies the good accuracy of this model.

Assessment of trends in inundation extent in the Barotse Floodplain, upper Zambezi River Basin: A remote sensing-based approach

Study regionThe annually flooded Barotse Floodplain in the upper Zambezi River Basin in the Western Province of Zambia, Southern Africa. Study focusDischarge variability plays a significant role in inundation extent and thus it controls habitat conditions of river channels and the linked wetlands. The linkage between discharge and inundation extent in the Barotse Floodplain allowed us to analyse the trends in extent overtime using optical satellite imagery MODIS. The Desert Flood Index, a surface water extraction algorithm, was used to generate time series of inundation extent. For validation of the inundation extent we used a flood mask extracted from a supervised classification land cover map using Landsat imagery. The land cover map was validated using the error matrix method with ground truthed data. The estimated inundation extent time series enabled us to test the inundation correlation with discharge and water level using Pearson r correlation, a parametric statistical test. Based on the established correlation we used the Mann–Kendall, a non-parametric test, to analyse trends in the inundation extent and discharge and water level time series from which we made inferences on the direction of the historical trend in inundation extent. New hydrological insights for the regionThe results revealed that there is observable inter-annual variability in inundation extent in the Barotse Floodplain with prominent differences demonstrated in both the flood ascending/peak and receding period. For the period 2003–2013 the results indicated a rising trend in inundation extent with a Mann–Kendall Z statistic of 1.71 and increase in magnitude of 33.1 km2 at significance level alpha of 0.05. Strong correlations between inundation extent and water level and between inundation extent and discharge with correlation coefficients of determination of 0.86 and 0.89 respectively were observed. For the period 2000–2011 water level time series showed a rising trend with the Mann–Kendall Z statistic of 2.97 and increase in magnitude of 0.1 m at significance level alpha of 0.05. Overall, during the period 1952–2004 discharge in the floodplain showed a declining trend with Mann–Kendall Z statistics of −2.88 and −3.38 at the inlet and outlet of the floodplain respectively. By correlation inference, the overall inundation extent trend in the floodplain was in a downward movement. Rainfall and discharge variability, high evapotranspiration and the changes in the land cover-use in the catchment of the floodplain are largely the factors affecting the observed variability and trends in inundation extent in the floodplain. The presented remote sensing based approach significantly reduces the need for the expensive and time limiting traditional physical field based wetland inundation mapping methods that form a limitation for achieving progress in wetland monitoring especially in open and sparsely gauged floodplains such as the Barotse.

Detecting peatland drains with Object Based Image Analysis and Geoeye-1 imagery

BackgroundPeatlands play an important role in the global carbon cycle. They provide important ecosystem services including carbon sequestration and storage. Drainage disturbs peatland ecosystem services. Mapping drains is difficult and expensive and their spatial extent is, in many cases, unknown. An object based image analysis (OBIA) was performed on a very high resolution satellite image (Geoeye-1) to extract information about drain location and extent on a blanket peatland in Ireland. Two accuracy assessment methods: Error matrix and the completeness, correctness and quality (CCQ) were used to assess the extracted data across the peatland and at several sub sites. The cost of the OBIA method was compared with manual digitisation and field survey. The drain maps were also used to assess the costs relating to blocking drains vs. a business-as-usual scenario and estimating the impact of each on carbon fluxes at the study site.ResultsThe OBIA method performed well at almost all sites. Almost 500 km of drains were detected within the peatland. In the error matrix method, overall accuracy (OA) of detecting the drains was 94% and the kappa statistic was 0.66. The OA for all sub-areas, except one, was 95–97%. The CCQ was 85%, 85% and 71% respectively. The OBIA method was the most cost effective way to map peatland drains and was at least 55% cheaper than either field survey or manual digitisation, respectively. The extracted drain maps were used constrain the study area CO2 flux which was 19% smaller than the prescribed Peatland Code value for drained peatlands.ConclusionsThe OBIA method used in this study showed that it is possible to accurately extract maps of fine scale peatland drains over large areas in a cost effective manner. The development of methods to map the spatial extent of drains is important as they play a critical role in peatland carbon dynamics. The objective of this study was to extract data on the spatial extent of drains on a blanket bog in the west of Ireland. The results show that information on drain extent and location can be extracted from high resolution imagery and mapped with a high degree of accuracy. Under Article 3.4 of the Kyoto Protocol Annex 1 parties can account for greenhouse gas emission by sources and removals by sinks resulting from “wetlands drainage and rewetting”. The ability to map the spatial extent, density and location of peatlands drains means that Annex 1 parties can develop strategies for drain blocking to aid reduction of CO2 emissions, DOC runoff and water discoloration. This paper highlights some uncertainty around using one-size-fits-all emission factors for GHG in drained peatlands and re-wetting scenarios. However, the OBIA method is robust and accurate and could be used to assess the extent of drains in peatlands across the globe aiding the refinement of peatland carbon dynamics .

Crop Dominance Mapping with IRS-P6 and MODIS 250-m Time Series Data

This paper describes an approach to accurately separate out and quantify crop dominance areas in the major command area in the Krishna River Basin. Classification was performed using IRS-P6 (Indian Remote Sensing Satellite, series P6) and MODIS eight-day time series remote sensing images with a spatial resolution of 23.6 m, 250 m for the year 2005. Temporal variations in the NDVI (Normalized Difference Vegetation Index) pattern obtained in crop dominance classes enables a demarcation between long duration crops and short duration crops. The NDVI pattern was found to be more consistent in long duration crops than in short duration crops due to the continuity of the water supply. Surface water availability, on the other hand, was dependent on canal water release, which affected the time of crop sowing and growth stages, which was, in turn, reflected in the NDVI pattern. The identified crop-wise classes were tested and verified using ground-truth data and state-level census data. The accuracy assessment was performed based on ground-truth data through the error matrix method, with accuracies from 67% to 100% for individual crop dominance classes, with an overall accuracy of 79% for all classes. The derived major crop land areas were highly correlated with the sub-national statistics with R2 values of 87% at the mandal (sub-district) level for 2005–2006. These results suggest that the methods, approaches, algorithms and datasets used in this study are ideal for rapid, accurate and large-scale mapping of paddy rice, as well as for generating their statistics over large areas. This study demonstrates that IRS-P6 23.6-m one-time data fusion with MODIS 250-m time series data is very useful for identifying crop type, the source of irrigation water and, in the case of surface water irrigation, the way in which it is applied. The results from this study have assisted in improving surface water and groundwater irrigated areas of the command area and also provide the basis for better water resource assessments at the basin scale.

FUZZY C-MEANS IN FINDING SUBTYPES OF CANCERS IN CANCER DATABASE

Finding subtypes of cancer in breast cancer database is an extremely difficult task because of heavy noise by measurement error. Most of the recent clustering techniques for breast cancer database to achieve cancerous and noncancerous often weigh down the interpretability of the structure. Hence, this paper tries to find effective Fuzzy C-Means-based clustering techniques to identify the proper subtypes of cancer in breast cancer database. This paper obtains the objective function of effective Fuzzy C-Means clustering techniques by incorporating the kernel induced distance function, Renyi's entropy function, weighted distance measure and neighborhood terms-based spatial context. The effectiveness of the proposed methods are proved through the experimental works on Lung cancer database, IRIS dataset, Wine dataset, Checkerboard dataset, Time Series dataset and Yeast dataset. Finally, the proposed methods are implemented successfully to cluster the breast cancer database into cancerous and noncancerous. The clustering accuracy has been validated through error matrix and silhouette method.

Development of an IKONOS image classification rule-set for multi-scale mapping of Mediterranean rural landscapes

In this research, a rule-set of object-based classification of IKONOS imagery for fine-scale mapping of Mediterranean rural landscapes was developed. This study was conducted on the Mediterranean island of Crete (Greece). A three-level classification hierarchy was designed in a bottom-up approach containing a total number of 22 classes. The first level was associated with vegetation physiognomy (6 classes), the second level with linear features (6 classes) and the third level with land uses existing in the area (10 classes). Image objects were created with multiresolution segmentation, an algorithm supplied by eCognition software. The segmentation parameters were selected through a trial-and-error approach after visual evaluation of the resulting image objects. The rule-set comprised 100 classification rules described with the ‘Membership Function’ classifier. The classification stability was found to lie between 0.59 and 0.77, inversely proportional to the complexity of each level's classification. For an accuracy assessment, the error matrix method was used in a set of 250 randomly selected points. The overall classification accuracy achieved at the first level was 74%, at the second level 50% and at the third level 64%. The geometric accuracy of the classification was beyond the scope of this research; and moreover, consistent reference data sets were not available. The conclusion is that the use of rules in an object-based image analysis (OBIA) process has the potential to produce accurate landscape maps even in the case of complex environments, in which ancillary data are not available. Future work should focus on testing the transferability of the rule-set in different Mediterranean study sites, in order to draw a conclusion in relation to its potential operational use.

Effects of band combinations and GIS masking on fire-scar mapping at local scales in east-central Florida, USA

The fire-adapted vegetation in east-central Florida provides habitat for many threatened and endangered species, such as the Florida scrub-jay (Aphelocoma coerulescens). Accurate fire occurrence records are critically important for better understanding the relationship between fire and vegetation structure. The rapid growth rates of fire-adapted vegetation in east-central Florida make it difficult to capture detailed fire scars with remote sensing data acquired weeks after the fires. The objective of this study is to develop a reliable remote sensing approach for accurately mapping burned areas in Florida scrub vegetation at the National Aeronautics and Space Administration (NASA) John F. Kennedy Space Center (KSC) and Merritt Island National Wildlife Refuge (MINWR). Landsat thematic mapper (TM) data acquired on 21 April 1987 were used for classification experiments. Geographic information system (GIS) data layers of fire management units (FMUs) with known fire occurrence (presence or absence) were used to mask the original remote sensing data or thematic maps following classification. A separation index (SI) was used to evaluate each individual band for its power to discriminate unburned and burned areas. Twelve classifications with selected band groups derived from Landsat TM data with different geographic extents were compared using an error matrix method. The classification of the four most suitable bands derived for the entire KSC-MINWR area resulted in the highest accuracy. The final map product was generated by overlaying the classified map with the FMU data layer and masking out FMUs that did not burn. This paper addresses a number of issues relevant to the classification of burned areas and includes the effect of geographic extent (GE effect) of remote sensing data on classification, determining the best bands for classification, and cleaning classification results using GIS masking. It also serves as the first published effort to map fire scars in the Florida scrub and flatwoods vegetative communities of the southeastern US using image processing techniques.

FE Model Updating for Health Monitoring of Structures and its Experimental Verification by Damage Detection

This paper presents an effective method of FE model updating for health monitoring of structures by applying ambient vibration. And this method is experimented through damage detection and proved to be valid. Experiment about ambient vibration is performed on cantilever beam, and the dynamic characteristics are analyzed by NExT and ERA. The results of such experiments are compared to those of FE analysis, and this comparison enables us to overcome some errors in experiments and analysis. On the basis of improved results by the comparison, model updating is performed in order to construct a basic structure for health monitoring. For model updating, we employ direct matrix updating method (DMUM) and Error matrix method (EMM) in which ambient vibration is easily applied. The model updating by the methods are again evaluated in terms of error ratio of natural frequency, comparing each result before and after updating. Finally, we perform experiments on damage detection to verify the method of updating presented here, and evaluate its performance by eigen-parameter change method. The evaluation proves that the method of FE model updating using ambient vibration is effective for health monitoring of structure, and some further application of this method is suggested.

Accuracy of discontinuous binary surfaces: a case study using boreal forest fires

Confidence in the conclusions of GIS and remote sensing analyses depends on our ability to specify their accuracy. The square error matrix, which is commonly used for accuracy assessment when the database contains continuous digital choropleth data, allows computation of user's, producer's, overall, and Kappa coefficient accuracy values. However, when the underlying assumptions of the square error matrix method are violated, these accuracy values may become unreliable. We demonstrate how traditional forms of accuracy assessment can fail for discontinuous binary surfaces and provide an improved spatial assessment method. We developed 15 new accuracy measures that allow analysts to assess scene- and feature-level accuracy from many new perspectives.

Multivariate fitting and the error matrix in global analysis of data

When a large body of data from diverse experiments is analyzed using a theoretical model with many parameters, the standard error-matrix method and the general tools for evaluating errors may become inadequate. We present an iterative method that significantly improves the reliability of the error matrix calculation. To obtain even better estimates of the uncertainties on predictions of physical observables, we also present a Lagrange multiplier method that explores the entire parameter space and avoids the linear approximations assumed in conventional error propagation calculations. These methods are illustrated by an example from the global analysis of parton distribution functions.

Analysis of density matrix reconstruction in NMR quantum computing

Reconstruction of density matrices is important in NMR quantum computing. An analysis is made for a 2-qubit system by using the error matrix method. It is found that the state tomography method determines well the parameters that are necessary for reconstructing the density matrix in NMR quantum computations. Analysis is also made for a simplified state tomography procedure that uses fewer read-outs. The result of this analysis with the error matrix method demonstrates that a satisfactory accuracy in density matrix reconstruction can be achieved even in a measurement with the number of read-outs being largely reduced.

A fuzzy set-based accuracy assessment of soft classification

Despite the sizable achievements obtained, the use of soft classifiers is still limited by the lack of well-assessed and adequate methods for evaluating the accuracy of their outputs. This paper proposes a new method that uses the fuzzy set theory to extend the applicability of the traditional error matrix method to the evaluation of soft classifiers. It is designed to cope with those situations in which classification and/or reference data are expressed in multimembership form and the grades of membership represent different levels of approximation to intrinsically vague classes.

A Higher Order Error Matrix Method for Error Localization*

A higher order version of the Error Matrix Method (EMM) is proposed to increase the accuracy of finite element error localization. The method retains a user specified number of terms from the appropriate binomial expansion. Jacobi's iterative method is then used to solve the set of nonlinear equations. It is hypothesized that keeping the higher order terms will improve the error identification for the same number of coordinate degrees-of-freedom and modes used in first order EMM.The method is implemented on nine-degree-of-freedom and Euier-Bemoulli beam numerical examples. Although a large number of measured coordinates and modes are needed, the magnitude of the errors is more accurately identified.

Controller Reduction Using Normal Coordinates of Reconstruction Error Matrix and Component Cost Analysis Method

Controller Reduction Using Normal Coordinates of Reconstruction Error Matrix and Component Cost Analysis Method

Verification of unfold error estimates in the unfold operator code

Spectral unfolding is an inverse mathematical operation that attempts to obtain spectral source information from a set of response functions and data measurements. Several unfold algorithms have appeared over the past 30 years; among them is the unfold operator (UFO) code written at Sandia National Laboratories. In addition to an unfolded spectrum, the UFO code also estimates the unfold uncertainty (error) induced by estimated random uncertainties in the data. In UFO the unfold uncertainty is obtained from the error matrix. This built-in estimate has now been compared to error estimates obtained by running the code in a Monte Carlo fashion with prescribed data distributions (Gaussian deviates). In the test problem studied, data were simulated from an arbitrarily chosen blackbody spectrum (10 keV) and a set of overlapping response functions. The data were assumed to have an imprecision of 5% (standard deviation). One hundred random data sets were generated. The built-in estimate of unfold uncertainty agreed with the Monte Carlo estimate to within the statistical resolution of this relatively small sample size (95% confidence level). A possible 10% bias between the two methods was unresolved. The Monte Carlo technique is also useful in underdetermined problems, for which the error matrix method does not apply. UFO has been applied to the diagnosis of low energy x rays emitted by Z-pinch and ion-beam driven hohlraums.