Mapping paddy rice fields by applying machine learning algorithms to multi-temporal Sentinel-1A and Landsat data

Sentinel-1A synthetic aperture radar (SAR) data present an opportunity for acquiring crop information without restrictions caused by weather and illumination conditions, at a spatial resolution appropriate for individual rice fields and a temporal resolution sufficient to capture the growth profiles of different crop species. This study investigated the use of multi-temporal Sentinel-1A SAR data and Landsat-derived normalized difference vegetation index (NDVI) data to map the spatial distribution of paddy rice fields across parts of the Sanjiang plain, in northeast China. The satellite sensor data were acquired throughout the rice crop-growing season (May–October). A co-registered set of 10 dual polarization (VH/VV) SAR and NDVI images depicting crop phenological development were used as inputs to Support Vector Machine (SVM) and Random Forest (RF) machine learning classification algorithms in order to map paddy rice fields. The results showed a significant increase in overall classification when the NDVI time-series data were integrated with the various combinations of multi-temporal polarization channels (i.e. VH, VV, and VH/VV). The highest classification accuracies overall (95.2%) and for paddy rice (96.7%) were generated using the RF algorithm applied to combined multi-temporal VH polarization and NDVI data. The SVM classifier was most effective when applied to the dual polarization (i.e. VH and VV) SAR data alone and this generated overall and paddy rice classification accuracies of 91.6% and 82.5%, respectively. The results demonstrate the practicality of implementing RF or SVM machine learning algorithms to produce 10 m spatial resolution maps of paddy rice fields with limited ground data using a combination of multi-temporal SAR and NDVI data, where available, or SAR data alone. The methodological framework developed in this study is apposite for large-scale implementation across China and other major rice-growing regions of the world.     

Assessment of grassland degradation near Lake Qinghai,West China,using Landsat TM and in situ reflectance spectra data

The severity of grassland degradation near Lake Qinghai, West China was assessed from a Landsat Thematic Mapper (TM) image in conjunction with in situ samples of per cent grass cover and proportion (by weight) of unpalatable grasses (PUG) collected over 1?m² sampling plots. Spectral reflectance at each sampling plot was measured with a spectrometer and its location determined with a Global Positioning System (GPS) receiver. After radiometric calibration, the TM image was geometrically rectified. Ten vegetation indices were derived from TM bands 3 and 4, and from the spectral reflectance data at wavelengths corresponding most closely to those of TM3 and TM4. Regression analyses showed that NDVI and SAVI are the most reliable indicators of grass cover and PUG, respectively. Significant relationships between TM bands-derived indices and in situ sampled grass parameters were established only after the former had been calibrated with in situ reflectance spectra data. Through the established regression models the TM image was converted into maps of grass cover parameters. These maps were merged to form a degradation map at an accuracy of 91.7%. It was concluded that TM imagery, in conjunction with in situ grass samples and reflectance spectra data, enabled the efficient and accurate assessment of grassland degradation inside the study area.     

Land-use/cover classification in a heterogeneous coastal landscape using RapidEye imagery: evaluating the performance of random forest and support vector machines classifiers

Mapping of patterns and spatial distribution of land-use/cover (LULC) has long been based on remotely sensed data. In the recent past, efforts to improve the reliability of LULC maps have seen a proliferation of image classification techniques. Despite these efforts, derived LULC maps are still often judged to be of insufficient quality for operational applications, due to disagreement between generated maps and reference data. In this study we sought to pursue two objectives: first, to test the new-generation multispectral RapidEye imagery classification output using machine-learning random forest (RF) and support vector machines (SVM) classifiers in a heterogeneous coastal landscape; and second, to determine the importance of different RapidEye bands on classification output. Accuracy of the derived thematic maps was assessed by computing confusion matrices of the classifiers’ cover maps with respective independent validation data sets. An overall classification accuracy of 93.07% with a kappa value of 0.92, and 91.80 with a kappa value of 0.92 was achieved using RF and SVM, respectively. In this study, RF and SVM classifiers performed comparatively similarly as demonstrated by the results of McNemer’s test (Z = 1.15). An evaluation of different RapidEye bands using the two classifiers showed that incorporation of the red-edge band has a significant effect on the overall classification accuracy in vegetation cover types. Consequently, pursuit of high classification accuracy using high-spatial resolution imagery on complex landscapes remains paramount.     

Guidance on and comparison of machine learning classifiers for Landsat-based land cover and land use mapping

Remote sensing scientists are increasingly adopting machine learning classifiers for land cover and land use (LCLU) mapping, but model selection, a critical step of the machine learning classification, has usually been ignored in the past research. In this paper, step-by-step guidance (for classifier training, model selection, and map production) with supervised learning model selection is first provided. Then, model selection is exhaustively applied to different machine learning (e.g. Artificial Neural Network (ANN), Decision Tree (DT), Support Vector Machine (SVM), and Random Forest (RF)) classifiers to identify optimal polynomial degree of input features (d) and hyperparameters with Landsat imagery of a study region in China and Ghana. We evaluated the map accuracy and computing time associated with different versions of machine learning classification software (i.e. ArcMap, ENVI, TerrSet, and R).

The optimal classifiers and their associated polynomial degree of input features and hyperparameters vary for the two image datasets that were tested. The optimum combination of d and hyperparameters for each type of classifier was used across software packages, but some classifiers (i.e. ENVI and TerrSet ANN) were customized due to the constraints of software packages. The LCLU map derived from ENVI SVM has the highest overall accuracy (72.6%) for the Ghana dataset, while the LCLU map derived from R DT has the highest overall accuracy (48.0%) for the FNNR dataset. All LCLU maps for the Ghana dataset are more accurate compared to those from the China dataset, likely due to more limited and uncertain training data for the China (FNNR) dataset. For the Ghana dataset, LCLU maps derived from tree-based classifiers (ArcMap RF, TerrSet DT, and R RF) routines are accurate, but these maps have artefacts resulting from model overfitting problems.     

遥感技术在西藏那曲地区草地退化评价中的应用

草地退化作为生态环境恶化的重要因素越来越引起人们的重视。在进行那曲地区草地退化评价中,应用遥感技术取得较好的结果。在调查中应用草地退化重要因子草地盖度与遥感图像植被指数之间较好的相关关系,利用野外调查测得的草地盖度样方资料,与在遥感图像上测出的各样方的植被指数的相关关系,进行监督分类,依据退化草地评价指标体系,对那曲地区草地进行科学评价并编制退化草地图。通过调查和测算,那曲地区草地退化面积占草地面积(不包括难利用草地)的48.8%。     

基于高分二号遥感影像的树种分类方法

为推广国产高分数据在森林树种分类方面的应用,以北京市延庆区八达岭国家森林公园主要区域的6期高分二号影像为数据源,在分层分类的基础上,利用支持向量机递归特征消除、C5.0决策树、FSO 3种特征优选方法,从4种特征维度下实现面向对象的支持向量机和随机森林的森林树种分类,最终取得总体精度平均为83.65%,特定树种生产者精度介于93.75%（山杏）和38.10%（刺槐）之间,特定树种用户精度介于100%（华北落叶松）和44.74%（榆树）之间的良好结果。结果表明：C5.0决策树耗时最短（0.01 h）且其所选特征应用于分类总体精度最高（86.90%）;在不同特征维度下支持向量机分类的总体精度比随机森林平均高出3.28%;支持向量机和随机森林均对特征维度不敏感,但良好的特征优选结果仍会对支持向量机的分类效率（最高提升86.98%）和随机森林的分类精度（最高提升9.22%）产生较大影响。     

基于高分二号遥感影像的树种分类方法

为推广国产高分数据在森林树种分类方面的应用，以北京市延庆区八达岭国家森林公园主要区域的6期高分二号影像为数据源，在分层分类的基础上，利用支持向量机递归特征消除、C5.0决策树、FSO 3种特征优选方法，从4种特征维度下实现面向对象的支持向量机和随机森林的森林树种分类，最终取得总体精度平均为83.65%，特定树种生产者精度介于93.75%（山杏）和38.10%（刺槐）之间，特定树种用户精度介于100%（华北落叶松）和44.74%（榆树）之间的良好结果。结果表明：C5.0决策树耗时最短（0.01 h）且其所选特征应用于分类总体精度最高（86.90%）；在不同特征维度下支持向量机分类的总体精度比随机森林平均高出3.28%；支持向量机和随机森林均对特征维度不敏感，但良好的特征优选结果仍会对支持向量机的分类效率（最高提升86.98%）和随机森林的分类精度（最高提升9.22%）产生较大影响。     

基于无人机高光谱影像的三江源草种精细识别研究

草种精细识别对三江源区草地生态系统退化监测具有重要意义。基于无人机高光谱遥感系统,获取三江源草地退化典型区的高光谱影像。在对原始光谱特征利用XGBoost进行优化选择的基础上,结合扩展形态学属性剖面特征,利用稀疏多项式逻辑回归与自适应稀疏表示两种分类方法分别对影像上的不同可食与毒杂草种进行精细识别,在此基础上提出形状自适应的后处理方法对识别结果进行平滑处理。结果表明：①利用XGBoost方法选择出重要性高的光谱特征能提升高光谱数据的识别效果并节省运行时间;②利用空间—光谱特征的识别方法相较于仅利用光谱特征的方法可以有效改善草种识别效果,使总体精度提升4%~5%;③利用两种稀疏表示方法在小样本的情况下对草种精细识别的精度分别达到94.07%、93.15%,利用形状自适应后处理方法能有效提高多种毒杂草种的识别精度,使得总体精度分别提升约1.64%和1.12%。基于特征挖掘的稀疏表示分类方法能实现高精度的无人机高光谱影像草种精细识别,为更大范围的草原物种精细识别提供了技术支撑。     

三江源蒸散发遥感估算及其时空分布特征研究

蒸散发是地表水热平衡的基本变量,也是衡量植被生长水分适应性的重要指标.针对三江源地面实测资料匮乏的现状,以MODIS系列产品为主要数据源,通过对地表温度—植被指数特征空间法的改进,在日尺度实现了该地区2011～2019年蒸散发的连续遥感估算,并进一步解析其时空变化特征与影响因子,揭示不同土地覆被类型的蒸散发差异,以期为...     

藏北中部地区草地退化遥感监测

为了探讨藏北中部地区近20年来的草地退化状况及主要驱动力,基于1990、2000及2010年的Landsat影像,采用混合分类方法对研究区主要景观类型进行分类,并利用景观指数分析景观格局及草地退化特征。另外,基于气象数据,利用灰色关联度方法分析草地退化的主要驱动力。结果表明:草地退化面积在1990~2000年呈增加趋势,在2000~2010年呈减少趋势,2000年之后研究区草地退化状况趋于缓和。造成草地退化的主要气候因素为降水量、气温、日照时数、平均相对湿度和蒸发量等,其中,高寒草甸草地主要受平均相对湿度及降水量的影响,高寒草原草地受蒸发量、日照时数、平均最低气温、降水量、平均相对湿度和平均最高气温的影响均较大,高寒荒漠草地与平均最高气温、日照时数和蒸发量的关系最为密切。     

Quantitative detection of change in grass cover from multi-temporal TM satellite data

In this paper, change in grassland cover near Lake Qinghai, west China was quantitatively detected from satellite remote-sensing data. Two Thematic Mapper images recorded in 1987 and 2000 were radiometrically corrected and used to derive the Normalized Difference Vegetation Index (NDVI). The NDVI image in 2000, after standardization via in situ measured spectra, was converted to a map of grass cover with the aid of in situ grass-cover samples. Another map was produced from the 1987 image after it was radiometrically benchmarked to the 2000 image using the calibration to like-values method. Comparison of these two maps revealed that a total of 36.28 km² of grassland had a higher cover, versus 44.72 km² that experienced grassland degradation in the study area. The absolute cover changed by a net value of??1.27%. The magnitude of change is related inversely to the value of the cover. The large majority of the area (82.6%), however, had a small change that was within ±20%. With this proposed method, it is possible to quantify changes in grassland cover from multi-temporal satellite data if one set of ground samples are concurrently collected with one of the satellite images.     

Improvement of land-cover classification over frequently cloud-covered areas using Landsat 8 time-series composites and an ensemble of supervised classifiers

Recent abundance of moderate-to-high spatial resolution satellite imagery has facilitated land-cover map production. However, in cloud-prone areas, building high-resolution land-cover maps is still challenging due to infrequent satellite revisits and lack of cloud-free data. We propose a classification method for cloud-persistent areas with high temporal dynamics of land-cover types. First, compositing techniques are employed to create dense time-series composite images from all available Landsat 8 images. Then, spectral–temporal features are extracted to train an ensemble of five supervised classifiers. The resulting composite images are clear with at least 99.78% cloud-free pixels and are 20.47% better than their original images on average. We classify seven land classes, including paddy rice, cropland, grass/shrub, trees, bare land, impervious area, and waterbody over Hanoi, Vietnam, in 2016. Using a time series of composites significantly improves the classification performance with 10.03% higher overall accuracy (OA) compared to single composite classifications. Additionally, using time series of composites and the ensemble technique, which combines the best of five experimented classifiers (eXtreme Gradient Boosting, logistic regression, Support Vector Machine (SVM) with Radial Basis Function (RBF) kernel – SVM–RBF and Linear kernel – SVM–Linear, multilayer perceptron), performed best with 84% OA and 0.79 kappa coefficient.     

Mapping an invasive plant, Phragmites australis, in coastal wetlands using the EO-1 Hyperion hyperspectral sensor

Mapping tools are needed to document the location and extent of Phragmites australis, a tall grass that invades coastal marshes throughout North America, displacing native plant species and degrading wetland habitat. Mapping Phragmites is particularly challenging in the freshwater Great Lakes coastal wetlands due to dynamic lake levels and vegetation diversity. We tested the applicability of Hyperion hyperspectral satellite imagery for mapping Phragmites in wetlands of the west coast of Green Bay in Wisconsin, U.S.A. A reference spectrum created using Hyperion data from several pure Phragmites stands within the image was used with a Spectral Correlation Mapper (SCM) algorithm to create a raster map with values ranging from 0 to 1, where 0 represented the greatest similarity between the reference spectrum and the image spectrum and 1 the least similarity. The final two-class thematic classification predicted monodominant Phragmites covering 3.4% of the study area. Most of this was concentrated in long linear features parallel to the Green Bay shoreline, particularly in areas that had been under water only six years earlier when lake levels were 66 cm higher. An error matrix using spring 2005 field validation points (n = 129) showed good overall accuracy—81.4%. The small size and linear arrangement of Phragmites stands was less than optimal relative to the sensor resolution, and Hyperion's 30 m resolution captured few if any pure pixels. Contemporary Phragmites maps prepared with Hyperion imagery would provide wetland managers with a tool that they currently lack, which could aid attempts to stem the spread of this invasive species.     

基于无人机高光谱影像和机器学习的红树林树种精细分类

利用海南省文昌市清澜港红树林保护区的无人机高光谱影像,采用递归特征消除的随机森林算法（Recursive Feature Elimination-Random Forest,RFE-RF）优选植被光谱特征和纹理特征,通过机器学习中的随机森林（Random Forest,RF）和支持向量机（Support Vector Machine,SVM）算法对研究区内的红树林树种进行精细分类,并对比分析和评价分类模型参数设置对总体精度的影响。结果表明：RF分类方法的总体精度为92.70%、 Kappa系数为0.91,与传统的SVM 分类方法相比,RF算法均提高了5类树种的生产者精度和使用者精度,能够有效地对红树林树种进行精细分类,可为种植资源规划和生态环境保护等方面提供技术支持。     

Land degradation monitoring using multi‐temporal Landsat TM/ETM data in a transition zone between grassland and cropland of northeast China

Land degradation is one of the most pressing problems of environments. This research presents a methodology to monitor land degradation in a transition zone between grassland and cropland of northeast China, where soil salinization and grassland degradation, even desertification, have been observed in the past few decades. Landsat TM/ETM data in 1988, 1996 and 2001 were selected to determine the rate and status of grassland degradation and soil salinization together based on both decision tree (DT) classifier and the field investigation. The thermal radiance values of TM/ETM 6 data, the Normalized Difference Vegetation Index (NDVI), and new variables (brightness, greenness, and wetness) generated by the Kauth–homas Transforms (KT) algorithms from Landsat TM/ETM data served as the feature nodes of a DT classifer and contributed to improving the classification results. It showed an overall accuracy of more than 85% and a Kappa statistic of agreement of about 0.79 in 1996 and 2001 with the exception of about 0.69 in 1988. The statistical areas of land degradation in the observation periods revealed that land degradation, especially the salt‐affected soil, is accelerating. The distribution maps of land degradation in the years of 1988, 1996 and 2001 were generated respectively based on the classification results. Their change maps were created by the difference between the distribution maps from 1988 to 1996 and from 1996 to 2001 respectively. The changes of salt‐affected soil occurred near the water bodies due to variations of water sizes, and most of the degraded grassland appeared around the salt‐affected soil. Although climate variations play an important role in this region, human activities are also crucial to land degradation.     

Finer resolution observation and monitoring of global land cover: first mapping results with Landsat TM and ETM+ data

We have produced the first 30 m resolution global land-cover maps using Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper Plus (ETM+) data. We have classified over 6600 scenes of Landsat TM data after 2006, and over 2300 scenes of Landsat TM and ETM+ data before 2006, all selected from the green season. These images cover most of the world's land surface except Antarctica and Greenland. Most of these images came from the United States Geological Survey in level L1T (orthorectified). Four classifiers that were freely available were employed, including the conventional maximum likelihood classifier (MLC), J4.8 decision tree classifier, Random Forest (RF) classifier and support vector machine (SVM) classifier. A total of 91,433 training samples were collected by traversing each scene and finding the most representative and homogeneous samples. A total of 38,664 test samples were collected at preset, fixed locations based on a globally systematic unaligned sampling strategy. Two software tools, Global Analyst and Global Mapper developed by extending the functionality of Google Earth, were used in developing the training and test sample databases by referencing the Moderate Resolution Imaging Spectroradiometer enhanced vegetation index (MODIS EVI) time series for 2010 and high resolution images from Google Earth. A unique land-cover classification system was developed that can be crosswalked to the existing United Nations Food and Agriculture Organization (FAO) land-cover classification system as well as the International Geosphere-Biosphere Programme (IGBP) system. Using the four classification algorithms, we obtained the initial set of global land-cover maps. The SVM produced the highest overall classification accuracy (OCA) of 64.9% assessed with our test samples, with RF (59.8%), J4.8 (57.9%), and MLC (53.9%) ranked from the second to the fourth. We also estimated the OCAs using a subset of our test samples (8629) each of which represented a homogeneous area greater than 500 m?×?500 m. Using this subset, we found the OCA for the SVM to be 71.5%. As a consistent source for estimating the coverage of global land-cover types in the world, estimation from the test samples shows that only 6.90% of the world is planted for agricultural production. The total area of cropland is 11.51% if unplanted croplands are included. The forests, grasslands, and shrublands cover 28.35%, 13.37%, and 11.49% of the world, respectively. The impervious surface covers only 0.66% of the world. Inland waterbodies, barren lands, and snow and ice cover 3.56%, 16.51%, and 12.81% of the world, respectively.     

Difficult to map regions in 30 m global land cover mapping determined with a common validation dataset

The free availability of decametre global satellite images and high-performance supercomputing provides opportunities for the development of many global products, including land cover, forest change, water, and cropland. However, some regions are particularly hard to map. Identification of these regions aids the understanding of map accuracy issues. In this study, we analysed seven maps produced with different algorithms/approaches but using the same classification system and training samples. A common validation dataset was used to identify regions incorrectly classified by all maps. These were defined as difficult to map regions (DMRs). They covered around 16% of the world’s ice-free terrestrial areas. Our analysis indicated that (1) grassland, shrubland, forest, and cropland were the most common land-cover types that could not be correctly classified, but impervious surfaces had the greatest proportion of misclassification; (2) incorrect classification mainly occurred in tropical/subtropical grassland/savanna/shrubland and desert/xeric shrubland; (3) the spatial distribution of DMRs was almost consistent with slope/elevation changes along latitude/longitude; and (4) the hotspot areas of land-cover mapping studies did not align with the DMRs. Our results suggest that there is a need for further work on DMRs to improve global land-cover mapping accuracy.     

Parameter tuning in the support vector machine and random forest and their performances in cross- and same-year crop classification using TerraSAR-X

This article describes the comparison of three different classification algorithms for mapping crops in Hokkaido, Japan, using TerraSAR-X data. In the study area, beans, beets, grasslands, maize, potatoes, and winter wheat were cultivated. Although classification maps are required for both management and estimation of agricultural disaster compensation, those techniques have yet to be established. Some supervised learning models may allow accurate classification. Therefore, comparisons among the classification and regression tree (CART), the support vector machine (SVM), and random forests (RF) were performed. SVM was the optimum algorithm in this study, achieving an overall accuracy of 89.1% for the same-year classification, which is the classification using the training data in 2009 to classify the test data in 2009, and 78.0% for the cross-year classification, which is the classification using the training data in 2009 to classify the data in 2012.     

Discriminating native and plantation forests in a Landsat time-series for land use policy design

The Warren River Catchment of south-western Australia is an area of high biodiversity threatened by the loss of native vegetation and dryland salinity. Over the last 20 years, it has been the target of a series of policies that encourage conversion of agricultural land to plantation forest. Remote sensing has a key role in measuring trends in the area of plantation forest observed across the landscape and hence the effectiveness of policy initiatives. Despite its importance to land use policy, accurate data on historical land use and land cover (LULC) dynamics of two spectrally similar but ecologically distinct forest types – such as native forest and plantation forest – are not readily available for south-western Australia, largely due to prohibitive data delivery costs. However, we argue that regular low-cost monitoring of long-term change in the spatial distribution of plantation forest through remote sensing is a critical input into environmental policy for the catchment. To this end, a 35-year time-series of Landsat imagery was acquired, and three different classifiers were tested (Support Vector Machines – SVM; Random Forests – RF; and Classification and Regression Trees – CART) on spectral and textural indices applied to four spectral bands. The six major LULC classes considered were agriculture, water, native forest, sand dunes, plantation forest and harvested native forest. In classifying the imagery the SVM and RF outperformed the CART across all classes. However, the SVM classifier gave a slightly higher F-score for most individual classes than the RF. Eucalypt dominated plantation forest reaching full canopy cover was subject to the highest rates of misclassification inasmuch as it shares spectral properties with the Eucalypt dominant native forest. When applied to Landsat time-series imagery, SVM classifier combined with four bands held in common between the four Landsat sensors, and derived textures metrics are valuable in classifying plantation and native forest, particularly where these have a similar species composition. The differences in prediction accuracy when including additional Landsat bands were not statistically significant, as demonstrated by the McNemar test. Thus, we achieved a trade-off in reducing processing time without significantly impacting on classification accuracy (≥86%). The relatively high accuracy of the proposed method enables the effects of past policy initiatives to be observed, and hence the efficient design of environmental and conservation policy in the future.     

基于高光谱影像的三江源区不同退化程度高寒草甸分类研究

三江源地区是我国最重要的生态功能区之一。近年来,受全球气候变暖及日趋频繁的人类活动的影响,三江源地区高寒草甸生态系统退化现象明显。以三江源称多县清水河镇东北部地区为实验区,基于环境小卫星HJ\|1A HSI高光谱数据,结合不同退化程度高寒草甸地面光谱采集和样方调查,采用MLC和SAM方法对不同退化程度的高寒草甸开展了分类研究。结果表明:基于高光谱数据的不同退化程度高寒草甸采用SAM方法分类总体精度达到75%以上,证实了分类方法的可行性,基于高光谱数据分类能有效区分盖度相近、退化程度不同的草地类型,其中SAM分类结果更加精细准确,优于MLC方法,SAM方法对中度退化草甸区分能力最高,对其他退化程度草甸区分能力稍弱。