Hasi Bagan

Land cover classification from MODIS EVI times-series data using SOM neural network

A high-dimensional dataset was built with time-series data of vegetation indexes derived from a Terra-Moderate Resolution Imaging Spectroradiometer (MODIS) sensor used for land use/cover classification. The self-organizing map (SOM) neural network technique can reduce the dimensionality of high-dimensional data, yet keep the same topological characters in the low-dimension space after dimension reduction. In this paper, we first employed the SOM neural network technique to classify land cover types using a 17-dimensional dataset that was generated from 16-day interval MODIS Enhanced Vegetation Index (EVI) data with a spatial resolution of 500 m in eastern China during the growing period of plants. Then, we defined an unlabelled class of neuron. Pixels matched to this type of neuron were regarded as unclassified land cover types, so that we could remove the poorly classified areas. Finally, the classification results were compared with those of the maximum likelihood classification (MLC) method. Comparison showed that the accuracy of the former exceeded that of the latter in classifying a high-dimensional dataset.

Land Cover Classification and Change Analysis in the Horqin Sandy Land From 1975 to 2007

Observations over the last three decades show that desertification poses a serious threat to the livelihood and productivity of inhabitants of the Horqin Sandy Land region of China. We evaluated the dynamics and trends of changes of land cover in the Horqin Sandy Land by using Landsat archive images from 1975, 1987, 1999, and 2007. We applied two supervised classification methods, the self-organizing map neural network method and the subspace method. Our analyses revealed significant changes to land cover over the period 1975-2007. The area of cropland doubled over the last three decades. This expansion was accompanied by large increases in water consumption and considerable loss of areas of grassland and woodland. Many lakes and rivers shrank rapidly or disappeared in this region between 1975 and 2007. The sandy area expanded rapidly from 1975 to 1987 but gradually slowed thereafter.

Land-cover change analysis in 50 global cities by using a combination of Landsat data and analysis of grid cells

Global urban expansion has created incentives to convert green spaces to urban/built-up area. Therefore, understanding the distribution and dynamics of the land-cover changes in cities is essential for better understanding of the cities fundamental characteristics and processes, and of the impact of changing land-cover on potential carbon storage. We present a grid square approach using multi-temporal Landsat data from around 1985–2010 to monitor the spatio-temporal land-cover dynamics of 50 global cities. The maximum-likelihood classification method is applied to Landsat data to define the cities urbanized areas at different points in time. Subsequently, 1 km2 grid squares with unique cell IDs are designed to link among land-cover maps for spatio-temporal land-cover change analysis. Then, we calculate land-cover category proportions for each map in 1 km2 grid cells. Statistical comparison of the land-cover changes in grid square cells shows that urban area expansion in 50 global cities was strongly negatively correlated with forest, cropland and grassland changes. The generated land-cover proportions in 1 km2 grid cells and the spatial relationships between the changes of land-cover classes are critical for understanding past patterns and the consequences of urban development so as to inform future urban planning, risk management and conservation strategies.

Classification of Airborne Hyperspectral Data Based on the Average Learning Subspace Method

This letter introduces the averaged learning subspace method (ALSM) that can be applied directly to original hyperspectral data for the purpose of classifying land cover. The ALSM algorithm of classification consists of the following iterative steps: (1) generate the initial appropriate feature subspace for each class in training datasets using the class-featuring information compression method, and (2) update the subspaces according to the maximum projection principle. We compare ALSM with the support vector machine classifier. By conducting experiments on two hyperspectral datasets (48 bands and 191 bands, respectively), we demonstrate that the ALSM can make dimensional reduction and classification simultaneously. When compared with the SVM classifier, it appears that the ALSM can achieve a higher accuracy on classification in some cases.

Combination of AVNIR-2, PALSAR, and Polarimetric Parameters for Land Cover Classification

We evaluate the potential of combined Advanced Land Observing Satellite Advanced Visible and Near-Infrared (AVNIR-2) and fully polarimetric Phased-Array-type L-band Synthetic Aperture Radar (PALSAR) data for land cover classification. Optical AVNIR-2 and fully polarimetric PALSAR can provide both surface spectral information and scattering information of the ground surface. The fully polarimetric PALSAR is particularly important for land cover classification because quad-polarization PALSAR data and its polarimetric parameters contain additional surface information. As a consequence, by combining optical AVNIR-2, PALSAR, and polarimetric parameters into a single data set, land cover classification accuracy may be further improved. For efficient and convenient handling of the combined multisource data, we used a subspace method for the classification and estimated its classification capability for various combinations of optical, PALSAR, and polarimetric parameter data sets in the Lake Kasumigaura region of Japan. We also compared the results obtained using the subspace method with those obtained by the support vector machine (SVM) and maximum-likelihood classification (MLC) methods. The classification results confirm that, when the combined optical AVNIR-2, PALSAR, and polarimetric coherency matrix data were used, the classification accuracy of the subspace method was better than that when other data combinations were used. The subspace method also performed better than the SVM or MLC method in high-dimensional data set classification. Moreover, the experimental results demonstrated that the proposed subspace method is robust for data classification when there is data redundancy and thus allows optimal feature selection procedures to be avoided.

Extended averaged learning subspace method for hyperspectral data classification.

Averaged learning subspace methods (ALSM) have the advantage of being easily implemented and appear to outperform in classification problems of hyperspectral images. However, there remain some open and challenging problems, which if addressed, could further improve their performance in terms of classification accuracy. We carried out experiments mainly by using two kinds of improved subspace methods (namely, dynamic and fixed subspace methods), in conjunction with the [0,1] and [-1,+1] normalization methods. We used different performance indicators to support our experimental studies: classification accuracy, computation time, and the stability of the parameter settings. Results are presented for the AVIRIS Indian Pines data set. Experimental analysis showed that the fixed subspace method combined with the [0,1] normalization method yielded higher classification accuracy than other subspace methods. Moreover, ALSMs are easily applied: only two parameters need to be set, and they can be applied directly to hyperspectral data. In addition, they can completely identify training samples in a finite number of iterations.

Land cover classification using moderate resolution imaging spectrometer-enhanced vegetation index time-series data and self-organizing map neural network in Inner Mongolia, China

The Moderate Resolution Imaging Spectroradiometer (MODIS) data offers a unique combination of spectral, temporal, and spatial resolution in comparison to other global sensors. The MODIS Enhanced Vegetation Index (EVI) product has several advantages, which make it suitable for regional land cover mapping. This paper investigates the application of MODIS EVI time-series data for mapping temperate arid and semi-arid land cover at a moderate resolution (500 m), for regional land-cover/land-use monitoring purposes. A 16-day composite EVI time-series data for 2003 (22 March 2003 - 30 September 2003) was adopted for the study. A land cover map was generated for the Inner Mongolia Autonomous Region using 7 tiles of MODIS EVI time-series data and Self-Organizing Map (SOM) neural network classification. Land-use GIS data, Landsat TM/ETM, and ASTER data were employed as reference data. The results show that the overall accuracy of land cover classification is about 84% with a Kappa coefficient of 0.8170. These results demonstrate that the SOM neural network model could work well for the multi-temporal MODIS EVI data, and suggest a potential of using MODIS EVI time-series remote sensing data to monitor desertification in Inner Mongolia with limited ancillary data and little labor-input in comparison with using high-spatial resolution remote sensing data.

Multispectral Land Cover Classification Using Averaged Learning Subspace Method

For the excellent appearances of Subspace methods in dimension reduction and classification, it is useful to introduce them into classification for multispectral remotely sensed data. This paper presents the first utilization of averaged learning subspace method (ALSM) for land cover classification using Landsat TM image. In particular, a comparative study was made about the classification performances of ALSM and maximum likelihood classification (MLC). ALSM yielded higher classification accuracies than MLC; the overall accuracy of the former algorithm was 99.00% while that of MLC was only 94.99%. The comparison of the classification performance in terms of training set size shows that ALSM outperformed MLC.

Analysis of spatiotemporal land cover changes in Inner Mongolia using self-organizing map neural network and grid cells method

Land use has changed dramatically in the Inner Mongolia Autonomous Region because of rapid economic growth and human disturbances. However, little information is available about the medium- and long-term land use changes in this region. The effects of ecological recovery policies have also been evaluated rarely. In this study, we employed the self-organizing map neural network method to identify the land cover changes in Inner Mongolia between 2000 and 2014. MOD13Q1, Landsat, and DMSP/OLS night-time light data were used as the data resources. The dynamic change map was characterized using the grid cell method. The results showed that urban area of Inner Mongolia increased by more than five times during the 15-year study period, while the mining area also increased. In addition, 35.3% of the farmland was changed into grassland, which may have been caused by the Grain to Green policy. The most significant environmental issue in Inner Mongolia is the loss of wetland. >40% of the wetland was converted into other land use types between 2000 and 2014. Grassland increased by 6.05%, but areas of open water and woodland remained about the same. In terms of the geographical distribution, cropland increased in the eastern and middle parts of the region. The transformation from wetland to grassland mainly occurred in the north. Grassland degradation occurred in the west. Thus, environmental policy has resulted in some ecological improvements in Inner Mongolia. However, new environmental problems associated with rapid economic development should be addressed in a timely manner.

Land-cover change in the Wulagai grassland, Inner Mongolia of China between 1986 and 2014 analysed using multi-temporal Landsat images

Abstract The Inner Mongolian steppe is a vast grassland ecosystem that has long been home to nomadic pastoralists. However, this steppe is experiencing grassland degradation as well as more frequent sand storms. The objective of this study was to detect land-cover changes in the Wulagai grassland of Inner Mongolia using multi-temporal Landsat images from 1986 to 2014, and to determine the factors driving these changes and their impacts. Land-cover maps for 1986, 1995, 2000, 2006 and 2014 were produced using the Support Vector Machine method. Subsequently, 300 m × 300 m grid-cell vector map which covered Wulagai grassland was made to detect land-cover changes and correlations between land-cover classes. The results show degradation trend from 1986 to 2014. Grid-cell-based spatial correlation analysis confirmed a strong negative correlation between grassland and barren, indicating that grassland degradation in this region is due to the regional modernization over the past 28 years.