An effective implementation and assessment of a random forest classifier as a soil spatial predictive model

Mapping the spatial distribution of soil classes is important for informing soil use and management decisions. This study aimed to effectively implement Random Forest (RF) model and to evaluate the behaviour and performance of the model for soil classification of Indian districts. Soil-forming factors, known as ‘scorpan,’ are selected as environmental covariates to tune RF model to classify 11 different soil categories. Thirty-five digital layers are prepared using different satellite data [ALOS (Advanced Land Observing Satellite) digital elevation model, Landsat-8, Moderate Resolution Imaging Spectroradiometer normalized difference vegetation index product, RISAT-1 (Radar Imaging Satellite-1), Sentinel-1A] and climatic data (precipitation and temperature) to represent scorpan environmental covariates in the study area. The RF parameters corresponding to highest Cohen’s kappa coefficient (κ) value and lowest number of random split variables are considered optimum values for RF model. Model behaviour evaluation is based on mapping accuracy, sensitivity to data set size, and noise. Two other machine-learning methods, CART (Classification and Regression Tree) decision tree (CDT) and CART ensemble bagger (CEB), are used to provide the comparative study. To access behaviour of models to the false data set, noise in training set is produced by assigning a false class to the training set in 5% increment. Comparative performance of RF model is based on quality assessment measures. To evaluate the performance of models, marginal rates, F-measure, and Jaccard’s coefficient of the community, classification success index and agreement coefficients are selected under quality assessment measures. The score is calculated to rank the algorithm. RF model shows high stability against data set reduction in comparison to other methods. The results show that the abrupt change in accuracy is only observed after 60% training data reduction in RF model; however, significant decrease in accuracy can be noted after 45% and 25% data reduction in CEB and CDT, respectively. The RF model shows comparatively the greater resistance to noise. Overall, RF model has performed better than CDT and CEB to classify soil categories in the study area. The results of this research provide new insights into the performance of RF in the context of soil class mapping.     

The assessment of different vegetation indices for spatial disaggregating of thermal imagery over the humid agricultural region

ABSTRACT

Land surface temperature (LST) plays a significant role in surface water circulation and energy balance at both global and regional scales. Thermal disaggregation technique, which relies on vegetation indices, has been widely used due to its advantage in producing relatively high resolution LST data. However, the spatial enhancement of satellite LST using soil moisture delineated vegetation indices has not gained enough attention. Here we compared the performances of temperature vegetation dryness index (TVDI), normalized difference vegetation index (NDVI), and fractional vegetation coverage (FVC), in disaggregating LST over the humid agriculture region. The random forest (RF) regression was used to depict the relationship between LST and vegetation indices in implementing thermal disaggregating. To improve the model performance, we used the thin plate spline (TPS) approach to calibrate the RF residual estimation. Results suggested that the models based on TVDI performed better than those based on NDVI and FVC, with a reduced average root mean square error and mean absolute error of 0.20 K and 0.16 K, respectively. Moreover, based on the surface energy balance model, we found the surface evapotranspiration (ET) derived with the TVDI disaggregated LST as inputs achieved higher accuracy than those derived with NDVI and FVC disaggregated LST. It is indicated that TVDI, a soil moisture delineated vegetation indices, can improve the performance of LST enhancement and ET estimation over the humid agriculture region, when combining random forest regression and TPS calibration. This work is valuable for terrestrial hydrology related research.     

Mapping percentage tree cover from Envisat MERIS data using linear and nonlinear techniques

The aim of this study was to predict percentage tree cover from Envisat Medium Resolution Imaging Spectrometer (MERIS) imagery with a spatial resolution of 300 m by comparing four common models: a multiple linear regression (MLR) model, a linear mixture model (LMM), an artificial neural network (ANN) model and a regression tree (RT) model. The training data set was derived from a fine spatial resolution land cover classification of IKONOS imagery. Specifically, this classification was aggregated to predict percentage tree cover at the MERIS spatial resolution. The predictor variables included the MERIS wavebands plus biophysical variables (the normalized difference vegetation index (NDVI), leaf area index (LAI), fraction of photosynthetically active radiation (fPAR), fraction of green vegetation covering a unit area of horizontal soil (fCover) and MERIS terrestrial chlorophyll index (MTCI)) estimated from the MERIS data. An RT algorithm was the most accurate model to predict percentage tree cover based on the Envisat MERIS bands and vegetation biophysical variables. This study showed that Envisat MERIS data can be used to predict percentage tree cover with considerable spatial detail. Inclusion of the biophysical variables led to greater accuracy in predicting percentage tree cover. This finer-scale depiction should be useful for environmental monitoring purposes at the regional scale.     

Land-cover mapping using Random Forest classification and incorporating NDVI time-series and texture: a case study of central Shandong

ABSTRACT

Land-cover mapping in complex farming area is a difficult task because of the complex pattern of vegetation and rugged mountains with fast-flowing rivers, and it requires a method for accurate classification of complex land cover. Random Forest classification (RFC) has the advantages of high classification accuracy and the ability to measure variable importance in land-cover mapping. This study evaluates the addition of both normalized difference vegetation index (NDVI) time-series and the Grey Level Co-occurrence Matrix (GLCM) textural variables using the RFC for land-cover mapping in a complex farming region. On this basis, the best classification model is selected to extract the land-cover classification information in Central Shandong. To explore which input variables yield the best accuracy for land-cover classification in complex farming areas, we evaluate the importance of Random Forest variables. The results show that adding not only multi-temporal imagery and topographic variables but also GLCM textural variables and NDVI time-series variables achieved the highest overall accuracy of 89% and kappa coefficient (κ) of 0.81. The assessment of the importance of a Random Forest classifier indicates that the key input variables include the summer NDVI followed by the summer near-infrared band and the elevation, along with the GLCM-mean, GLCM-contrast.     

基于多时相Landsat8 OLI影像的作物种植结构提取

针对基于多时相遥感影像、多种特征量提取多种作物种植结构在我国研究较少的现状,利用多时相Landsat8OLI影像数据,根据温宿县不同作物的农事历,通过分析主要地物的光谱特征和归一化植被指数的时间变化信息,构建不同作物种植结构提取的决策树模型,实现了对温宿县多种作物种植结构信息的提取。结果表明:1水稻的最佳识别依据是5月20日影像的近红外波段和7月23日影像的NDVI值;棉花和春玉米的最佳识别依据是5月20日~9月9日影像的NDVI变化值;冬小麦—夏玉米和林果的最佳识别依据是5月20日~7月23日影像的NDVI变化值;2与单时相监督分类相比,多时相决策树法对多种作物种植结构的提取效果更理想,总体精度提高了7.90%,Kappa系数提高了0.10;3Landsat8OLI影像数据分辨率高、成本低、获取方便,是农作物遥感的良好数据源。     

Dynamic response of NDVI to soil moisture variations during different hydrological regimes in the Sahel region

Over the last few decades, the African Sahel has become the focus of many studies regarding vegetation dynamics and their relationships with climate and people. This is because rainfall limits the production of biomass in the region, a resource on which people are directly dependent for their livelihoods. In this study, we utilized a remote-sensing approach to answering the following two questions: (1) how does the dynamic relationship between soil moisture and plant growth vary across hydrological regimes, and (2) are vegetation-type-dependent responses to soil moisture availability detectable from satellite imagery? In order to answer these questions, we studied the relationship between monthly modelled soil moisture as an indicator for water availability and the remotely sensed normalized difference vegetation index (NDVI) as a proxy for vegetation growth between a “recovery rainfall period” (1982 to 1997) and a “stable rainfall period” (1998 to 2013), at different time lags across the Sahel region. Using windowed cross-correlation, we find a strong significant positive relationship between NDVI and soil moisture at a concurrent time and at NDVI lagging behind soil moisture by 1 month for grassland, cropland, and deciduous shrubland vegetation – the dominant vegetation classes in the Sahel. South of the Sahel (the Sudanian and Guinean areas), we find longer optimal lags (soil moisture lagged by 1–3 months) in association with mixed forest and deciduous shrubland. We find no major significant change in optimal lag between the recovery and stable periods in the Sahelian region; however, in the Sudanian and Guinean areas, we observe a trend towards shorter time lags. This change in optimal lag suggests a vegetation change, which may be a response to a climatic shift or land-use change. This approach of identifying spatiotemporal trends in optimal lag correlations between modelled soil moisture and NDVI could prove to be a useful tool for mapping vegetation change and ecosystem behaviour, in turn helping inform climate change mitigation approaches and agricultural planning.     

Spatiotemporal fusion through the best linear unbiased estimator to generate fine spatial resolution NDVI time series

Spatiotemporal fusion (STF) technologies are commonly used to acquire high spatiotemporal resolution remote sensing observations. However, most STF technologies fail to consider the nonlinear variation in vegetation in the time domain. Based on the Best Linear Unbiased Estimator (BLUE), this paper proposed a novel STF algorithm (referred to BLUE) which accounts for the phenological characteristics of vegetation. First, annual time series of normalized difference vegetation index (NDVI) data with high spatial resolution but low temporal resolution is fitted using a double logistic function and used as the background field. Then, NDVI data with low spatial resolution but high temporal resolution is used as the observation field. The information in the background and observation fields is fused using the BLUE to obtain high spatiotemporal resolution NDVI data. The proposed algorithm was used to produce dense time series of 30 m resolution NDVI data for a 10 km × 10 km experimental area in 2014. The experimental results demonstrate that the accuracy of fusion results from the proposed BLUE method are higher than those from the enhanced spatial and temporal adaptive reflectance fusion model (ESTARFM) and Linear Mixing Growth Model (LMGM), especially when the temporal component of surface heterogeneity is dominant. The proposed algorithm has broad prospects in vegetation monitoring at high spatiotemporal resolution.     

Mapping vegetation morphology types in a dry savanna ecosystem: integrating hierarchical object-based image analysis with Random Forest

Savanna ecosystems are geographically extensive and both ecologically and economically important, and require monitoring over large spatial extents. Remote-sensing-based characterization of vegetation properties in savannas is methodologically challenging, mainly due to high structural and functional heterogeneity. Recent advances in object-based image analysis (OBIA) and machine learning algorithms offer new opportunities to address these challenges. Focusing on the semi-arid savanna ecosystem in the central Kalahari, this study examined the suitability of a hierarchical OBIA approach combined with in situ data and an ensemble classification technique for mapping vegetation morphology types at landscape scale. A stack of Landsat TM imagery, NDVI, and topographic variables was segmented with six different scale factors resulting in a hierarchical network of image objects. Sample objects for each vegetation morphology class were selected at each segmentation scale and classification was performed using optimal features consisting of spectral and textural features. Overall and class-specific classification accuracies were compared across the six scales to examine the influence of segmentation scale on each. Results suggest that the highest overall classification accuracy (i.e. 85.59%) was observed not at the finest segmentation scale, but at coarse segmentation. Additionally, individual vegetation morphology classes differed in the segmentation scale at which they achieved highest classification accuracy, reflecting their unique ecology and physiognomic composition. While classes with high vegetation density/height attained higher accuracy at fine segmentation scale, those with lower vegetation density/height reached higher classification accuracy at coarse segmentation scales. Contrarily, for pans and bare areas, accuracy was relatively unaffected by changing segmentation scale. Variable importance plots suggested that spectral features were the most important, followed by textural variables. These results show the utility of the OBIA approach and emphasize the requirement of multi-scale analysis for accurately characterizing savanna systems.     

Estimation of land surface temperature-vegetation abundance relationship for urban heat island studies

Remote sensing of urban heat islands (UHIs) has traditionally used the Normalized Difference Vegetation Index (NDVI) as the indicator of vegetation abundance to estimate the land surface temperature (LST)-vegetation relationship. This study investigates the applicability of vegetation fraction derived from a spectral mixture model as an alternative indicator of vegetation abundance. This is based on examination of a Landsat Enhanced Thematic Mapper Plus (ETM+) image of Indianapolis City, IN, USA, acquired on June 22, 2002. The transformed ETM+ image was unmixed into three fraction images (green vegetation, dry soil, and shade) with a constrained least-square solution. These fraction images were then used for land cover classification based on a hybrid classification procedure that combined maximum likelihood and decision tree algorithms. Results demonstrate that LST possessed a slightly stronger negative correlation with the unmixed vegetation fraction than with NDVI for all land cover types across the spatial resolution (30 to 960 m). Correlations reached their strongest at the 120-m resolution, which is believed to be the operational scale of LST, NDVI, and vegetation fraction images. Fractal analysis of image texture shows that the complexity of these images increased initially with pixel aggregation and peaked around 120 m, but decreased with further aggregation. The spatial variability of texture in LST was positively correlated with those in NDVI and in vegetation fraction. The interplay between thermal and vegetation dynamics in the context of different land cover types leads to the variations in spectral radiance and texture in LST. These variations are also present in the other imagery, and are responsible for the spatial patterns of urban heat islands. It is suggested that the areal measure of vegetation abundance by unmixed vegetation fraction has a more direct correspondence with the radiative, thermal, and moisture properties of the Earth's surface that determine LST.     

Regional scale hierarchical classification of temporal series of AVHRR vegetation index

In this paper a hierarchical approach is taken to classify temporal sequences of images of the Normalized Difference Vegetation Index (NDVI) from the Advanced Very High Resolution Radiometer (AVHRR), using Iberia as an example. Iberia is a convenient area of study because it has a high environmental diversity and very strong environmental gradients, and yet a reduced size at the spatial resolution of current global data-sets. An Iberian subset of a global temporal series of AVHRR-NDVI images facilitates test and validation of different approaches while producing results that are likely to be valid over much larger areas. Our hierarchical clustering approach yields maps with nested legends. We compare these maps to a digitized map of potential natural vegetation, which reveals a clear bioclimatic control. The highest level of the hierarchical classification separates vegetation with a Summer peak of NDVI from vegetation with a Spring peak of NDVI. Such a discontinuity corresponds to the discontinuity between Atlantic and Submediterranean vegetation in the vegetation map. Lower levels in the hierarchical classification produce maps of increasing complexity but that keep a high degree of spatial continuity. A correspondence analysis between a 16-classes NDVI map and the digitized map of potential vegetation produces an ordination that is bioclimatically coherent. According to the known characteristics of the potential vegetation units, the two first correspondence axes can be interpreted, respectively, as water availability and temperature. These results are a consequence of the temporal NDVI series being an accurate signal of vegetative phenology, which in turn is a fundamental vegetation property. A comparison of our results with several global land cover digital maps by means of the Wilk's ratio indicates that the global maps do not produce an appropriate partition of the region in terms of the NDVI temporal course. We conclude that the analysis of temporal series of NDVI yield relevant ecological information at finer scales and with more detailed legends that had not been attempted until now, and, therefore, are suitable for regional scale applications. Our results also indicate the interest of a bioclimatic analysis and modeling of the NDVI signatures for their correct ecological understanding. Maps at a global scale can be produced based on such an understanding.     

A comparison of MODIS 250-m EVI and NDVI data for crop mapping: a case study for southwest Kansas

Multi-temporal vegetation index (VI) data from the Moderate Resolution Imaging Spectroradiometer (MODIS) are becoming widely used for large-area crop classification. Most crop-mapping studies have applied enhanced vegetation index (EVI) data from MODIS instead of the more traditional normalized difference vegetation index (NDVI) data because of atmospheric and background corrections incorporated into EVI's calculation and the index's sensitivity over high biomass areas. However, the actual differences in the classification results using EVI versus NDVI have not been thoroughly explored. This study evaluated time-series MODIS 250-m EVI and NDVI for crop-related land use/land cover (LULC) classification in the US Central Great Plains. EVI- and NDVI-derived maps classifying general crop types, summer crop types and irrigated/non-irrigated crops were produced for southwest Kansas. Qualitative and quantitative assessments were conducted to determine the thematic accuracy of the maps and summarize their classification differences. For the three crop maps, MODIS EVI and NDVI data produced equivalent classification results. High thematic accuracies were achieved with both indices (generally ranging from 85% to 90%) and classified cropping patterns were consistent with those reported for the study area (> 0.95 correlation between the classified and USDA-reported crop areas). Differences in thematic accuracy (< 3% difference), spatially depicted patterns (> 90% pixel-level thematic agreement) and classified crop areas between the series of EVI- and NDVI-derived maps were negligible. Most thematic disagreements were restricted to single pixels or small clumps of pixels in transitional areas between cover types. Analysis of MODIS composite period usage in the classification models also revealed that both VIs performed equally well when periods from a specific growing season phase (green, peak or senescence) were heavily utilized to generate a specific crop map.     

土壤背景对冠层NDVI的影响分析

归一化差值植被指数NDVI是植被遥感中应用最为广泛的指数之一, 但它受土壤背景等因素的干扰比较强烈。结合实测的土壤数据以及公式推导、PROSAIL 模型模拟等方法分析了这种影响。首先, 假定与土壤线性混合且叶片呈水平分布的植被冠层, 根据土壤与植被分别在红光、近红外波段处的反射率值、植被覆盖度等参数, 利用公式推导了土壤背景对不同覆盖度下冠层NDVI的影响。其次, 利用PROSAIL冠层光谱模拟模型, 模拟分析了土壤背景对不同LAI下冠层NDVI的影响。分析的结果表明:LAI 越小, 土壤背景的影响越大; 暗土壤背景下的冠层NDVI值大于亮土壤背景下冠层的NDVI值; 并且,暗土壤条件下,NDVI值对土壤亮度的变化更敏感,而亮土壤下,NDVI值则对LAI或覆盖度的变化更敏感。最后利用实测的不同土壤背景下的冬小麦冠层光谱数据, 验证了公式推导和模型模拟的结果。     

Classification of Landsat images based on spectral and topographic variables for land-cover change detection in Zagros forests

Detection of land-cover changes through time can be complicated because of sensor-specific differences in spatial and spectral resolutions; classified land-cover changes can be due to either real changes on the ground or a switch in sensors used to collect data. This study focused on two objectives: (1) selecting the best predictor variables for the classification of semi-arid Zagros forests given the characteristics of the study area and available data sets and (2) evaluating the application of the random forest (RF) algorithm as a unified technique for the classification of data sets acquired from different sensors. Three images of the same study area were acquired from the Landsat-5 Thematic Mapper (TM) sensor in 2009, the Landsat-7 Enhanced Thematic Mapper (ETM+) sensor with Scan Line Corrector (SLC) in 1999 and the Landsat-2 Multispectral Scanner (MSS) sensor in 1975. Following image preprocessing, the RF algorithm was applied for variable selection and classification. A test of equivalence was used to compare the overall accuracy of the classified maps from the three sensors. Slope, normalized difference vegetation index (NDVI) and elevation were determined to be the most important predictor variables for all three images. High overall classification accuracies were achieved for all three images (97.90% for MSS, 95.43% for TM and 95.29% for ETM). The ETM- and TM-derived maps had equivalent overall accuracy and even significantly higher overall accuracy was obtained for the MSS-derived map. The post-classification comparison showed an increase in agriculture and a decrease in forest cover. The selected predictor variables were consistent with ecological reality and showed more details on the changes of the land-cover classes across biophysical variables of the study area through time.     

Simulation and Analysis on the Influence of Soil Background of Wheat Farmland on NDVI under Different Vegetation Coverage

It is quite confusing to effectively monitor and precisely evaluate growing conditions of wheat by using normalized differential vegetation index (NDVI)which is based on pixel scale as they are significantly different when acquired by the same growth status wheat with different background of soil types.This paper selects 9 typical soil types in our country as background with the wheat canopy spectrum is fixed which means the NDVIc is a constant value to study the influence of different soil background types on NDVI of wheat and analyze the sensitivity of NDVI of wheat to the vegetation coverage simulated by diverse liner mixed ratio of wheat canopy and soil background.The results show that：(1)wheat NDVI of farmland increases along with the increase of vegetation coverage under the same of soil background type,and vice versa;(2)wheat NDVI of farmland vary greatly with different soil background types,and the difference decrease while the vegetation coverage exceed 25%;(3)NDVI sensitivity also shows a quite difference to vegetation coverage under the diverse soil background types.With the increase of vegetation coverage,NDVI sensitivity decreases with the lower\|reflectance soil background while it increases monotonously with the higher reflectance soil background.It provides the foundation for the times of calculating the remote sensing’s NDVI information of all wheat growing periods under different types of soil background.     

多云雾地区高时空分辨率植被覆盖度构建方法研究

针对多云雾地区高时空分辨率数据缺乏现状,提出了一套区域尺度高时空分辨率植被覆盖度数据构建方法.首先,通过时空适应反射率融合模型(STARFM)有效地将TM 的较高空间分辨率与MODIS的高时间分辨率融合在一起,构建了研究区植被生长峰值阶段的NDVI数据;然后,以植被生长峰值阶段的NDVI为输入,基于地表覆被类型,综合应用等密度和非密度亚像元模型对研究区的植被覆盖度进行估算.结果表明:①即使数据源存在大量的云雾,且存在一定的时相差异,研究区植被覆盖度的估算结果过渡自然,不存在明显的不接边效应;②以植被生长峰值阶段的NDVI数据为输入进行植被覆盖度估算,有效拉开了同一地表覆被类型不同覆盖度像元的NDVI梯度,提高了亚像元估算模型对输入数据的抗扰动性;③基于地表覆被类型,应用亚像元混合模型,能够提高植被覆盖度的估算精度.经野外实测数据验证,总体约８５％的估算精度表明,针对高时空分辨率遥感数据缺乏的多云雾区域,本研究提出的方法能够实现区域尺度植被覆盖度数据的构建.     

基于多时序特征和卷积神经网络的农作物分类

近年来,以卷积神经网络为主的深度学习模型在各种遥感应用中都显示出巨大的潜力。以加州帝国郡为研究区,以Landsat 8 OLI年内时序遥感影像计算时序植被指数NDVI、EVI、RVI以及TVI,组合后输入到构建的一维卷积神经网络 模型,以实现作物的高精度精细分类。为了验证卷积模型的优越性,另搭建了基于递归神经网络及其变体的深度学习模型。结果表明：①引入其他时序特征后,能够有效地提高卷积神经网络的分类精度。NDVI+EVI+TVI+RVI组合特征总体精度和Kappa系数最高,分别是89.667 4%和0.856 0,对比NDVI时序特征总体精度和Kappa系数提高了近4%和0.6。②在与其他深度学习模型的对比中,一维卷积神经网络分类精度最高,能够从时序数据中较为准确捕捉作物时序特征信息,尽管递归神经网络被广泛应用于序列数据的研究,但分类结果要略差于卷积神经网络。实验表明在NDVI的基础上引入其他植被指数辅助,能够有效地提高分类精度。基于一维卷积神经网络的深度学习框架为长时间序列分类任务提供了一种有效且高效的方法。     

Soil moisture estimation using MODIS and ground measurements in eastern China

Recent technological advances in remote sensing have shown that soil moisture can be measured by microwave remote sensing under some topographic and vegetation cover conditions. However, current microwave technology limits the spatial resolution of soil moisture data. It has been found that the Normalized Difference Vegetation Index (NDVI) and Land Surface Temperature (LST) are related to surface soil moisture; therefore, a relationship between ground observed soil moisture and satellite NDVI and LST products can be developed. Three years of 1 km NDVI and LST products from the Moderate Resolution Imaging Spectroradiometer (MODIS) have been combined with ground measured soil moisture to determine regression relationships at a 1 km scale. Results show that MODIS NDVI and LST are strongly correlated with the ground measured soil moisture, and regression relationships are land cover and soil type dependent. These regression relationships can be used to generate soil moisture estimates at moderate resolution for study area.     

Analysis of NDVI and scaled difference vegetation index retrievals of vegetation fraction

The normalized difference vegetation index (NDVI) is the most widely used vegetation index for retrieval of vegetation canopy biophysical properties. Several studies have investigated the spatial scale dependencies of NDVI and the relationship between NDVI and fractional vegetation cover, but without any consensus on the two issues. The objectives of this paper are to analyze the spatial scale dependencies of NDVI and to analyze the relationship between NDVI and fractional vegetation cover at different resolutions based on linear spectral mixing models. Our results show strong spatial scale dependencies of NDVI over heterogeneous surfaces, indicating that NDVI values at different resolutions may not be comparable. The nonlinearity of NDVI over partially vegetated surfaces becomes prominent with darker soil backgrounds and with presence of shadow. Thus, the NDVI may not be suitable to infer vegetation fraction because of its nonlinearity and scale effects. We found that the scaled difference vegetation index (SDVI), a scale-invariant index based on linear spectral mixing of red and near-infrared reflectances, is a more suitable and robust approach for retrieval of vegetation fraction with remote sensing data, particularly over heterogeneous surfaces. The proposed method was validated with experimental field data, but further validation at the satellite level would be needed.     

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.     

沂蒙山区植被NDVI的时空特征及其对水热条件的响应

植被是生态环境变化的综合指示器,研究其对水热条件的响应已成为当前气候变化研究中的主要内容之一。选取北方土石山区典型代表--沂蒙山区为研究对象,基于沂蒙山区1980~2010年的气温、降水和2001~2010年MODIS\|NDVI数据,结合相关分析和最小二乘法,定量分析该区植被指数的年际、年内的时空变化及其对水热条件的响应。结果表明:①近10 a沂蒙山区NDVI max的变化斜率为0.0026;②植被显著退化区和良好区分别占研究区总面积的10.52%和28.62%;③不同季节(主要是春、夏和秋季)植被状况均呈现良性发展趋势;④台站数据显示植被年际变化与年降水和年均气温的关系并不密切,而在月时间尺度上植被与气温的相关性要强于与降水的相关性。综上所述,沂蒙山区植被状况总体呈良性发展趋势,气温可能是影响该区植被生长的主导因子。