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.     

Evaluation of Ku‐band QuikSCAT scatterometer data for rice crop growth stage assessment

A study has been carried out to analyse the high temporal Ku‐band scatterometer data from QuikSCAT with 4.45 km resolution for regional assessment of rice crop phenology. Four‐day composite data were used covering the two predominant rice‐growing states in India during the monsoon season of 2004. These data were registered with reference to a rice map derived from RADARSAT SAR data of the same season in order to select predominant rice sites. Analysis shows the dual peak backscatter profile of rice crop (at tillering stage and another at maturity). Minima of the backscatter profile were found to coincide with the heading. The derived heading stage using a lognormal curve fitting matched well with the observed dates. The slope of the second peak varied with crop variety, and shows the potential of correlation with the panicle characteristic.     

Towards an operational system for regional-scale rice yield estimation using a time-series of Radarsat ScanSAR images

This paper demonstrates that Radarsat ScanSAR data can be an important data source of radar remote sensing for monitoring crop systems and estimation of rice yield in large areas in tropical and sub-tropical regions. Experiments were carried out to show the effectiveness of Radarsat ScanSAR data for rice yield estimation in the whole province of Guangdong, South China. A methodology was developed to deal with a series of issues in extracting rice information from the ScanSAR data, such as topographic influences, levels of agro-management, irregular distribution of paddy fields and different rice cropping systems. A model was provided for rice yield estimation based on the relationship between the backscatter coefficient of multi-temporal SAR data and the biomass of rice. The study indicates that the whole procedure can become a low-cost and convenient operational system for large-scale rice yield estimation which is difficult for conventional methods.     

Predicting winter wheat condition,grain yield and protein content using multi‐temporal EnviSat‐ASAR and Landsat TM satellite images

Since optical and microwave sensors respond to very different target characteristics, their role in crop monitoring can be viewed as complementary. In particular, the all‐weather capability of Synthetic Aperture Radar (SAR) sensors can ensure that data gaps that often exist during monitoring with optical sensors are filled. There were three Landsat Thematic Mapper (TM) satellite images and three Envisat Advanced Synthetic Aperture Radar (ASAR) satellite images acquired from reviving stage to milking stage of winter wheat. These data were successfully used to monitor crop condition and forecast grain yield and protein content. Results from this study indicated that both multi‐temporal Envisat ASAR and Landsat TM imagery could provide accurate information about crop conditions. First, bivariate correlation results based on the linear regression of crop variables against backscatter suggested that the sensitivity of ASAR C‐HH backscatter image to crop or soil condition variation depends on growth stage and time of image acquisition. At the reviving stage, crop variables, such as biomass, Leaf Area Index (LAI) and plant water content (PWC), were significantly positively correlated with C‐HH backscatter (r = 0.65, 0.67 and 0.70, respectively), and soil water content at 5 cm, 10 cm and 20 cm depths were correlated significantly with C‐VV backscatter (r = 0.44, 0.49 and 0.46, respectively). At booting stage, only a significant and negative correlation was observed between biomass and C‐HH backscatter (r = ?0.44), and a saturation of the SAR signal to canopy LAI could explain the poor correlation between crop variables and C‐HH backscatter. Furthermore, C‐HH backscatter was correlated significantly with soil water content at booting and milking stage. Compared with ASAR backscatter data, the multi‐spectral Landsat TM images were more sensitive to crop variables. Secondly, a significant and negative correlation between grain yield and ASAR C‐HH & C‐VV backscatter at winter wheat booting stage was observed (r = ?0.73 and ?0.55, respectively) and a yield prediction model with a correlation coefficient of 0.91 was built based on the Normalized Difference Water Index (NDWI) data from Landsat TM on 17 April and ASAR C‐HH backscatter on 27 April. Finally, grain protein content was found to be correlated significantly with ASAR C‐HH backscatter at milking stage (r = ?0.61) and with Structure Insensitive Pigment Index (SIPI) data from Landsat TM at grain‐filling stage (r = 0.53), and a grain protein content prediction model with a correlation coefficient of 0.75 was built based on the C‐HH backscatter and SIPI data.     

Mapping rice areas with Sentinel-1 time series and superpixel segmentation

Rice is the single most important crop for food security in Asia. Knowledge about the distribution of rice fields is also relevant in the context of greenhouse-relevant methane emissions, disease transmission, and water resource management. Copernicus Sentinel-1 provides the first openly available archive of C-band SAR (synthetic aperture radar) data at high spatial and temporal resolution. We developed one of the first methods that shows the potential of this data for accurate and timely mapping of rice-growing areas. We used superpixel segmentation to create spatially averaged backscatter time series, which is robust to speckle and reduces the amount of data to process. This method has been applied to six study sites in different rice-growing regions of the world and achieved an average overall accuracy of 0.83.     

Effects of seasonal hydrologic patterns in south Florida wetlands on radar backscatter measured from ERS-2 SAR imagery

A multi-year study was carried out to evaluate ERS synthetic aperture radar (SAR) imagery for monitoring surface hydrologic conditions in wetlands of southern Florida. Surface conditions (water level, aboveground biomass, soil moisture) were measured in 13 study sites (representing three major wetland types) over a 25-month period. ERS SAR imagery was collected over these sites on 22 different occasions and correlated with the surface observations. The results show wide variation in ERS backscatter in individual sites when they were flooded and non-flooded. The range (minimum vs. maximum) in SAR backscatter for the sites when they were flooded was between 2.3 and 8.9 dB, and between 5.0 and 9.0 dB when they were not flooded. Variations in backscatter in the non-flooded sites were consistent with theoretical scattering models for the most part. Backscatter was positively correlated to field measurements of soil moisture. The MIchigan MIcrowave Canopy Scattering (MIMICS) model predicts that backscatter should decrease sharply when a site becomes inundated, but the data show that this drop is only 1-2 dB. This decrease was observed in both non-wooded and wooded sites. The drop in backscatter as water depth increases predicted by MIMICS was observed in the non-wooded wetland sites, and a similar decrease was observed in wooded wetlands as well. Finally, the sensitivity of backscatter and attenuation to variations in aboveground biomass predicted by MIMICS was not observed in the data.The results show that the inter- and intra-annual variations in ERS SAR image intensity in the study region are the result of changes in soil moisture and degree of inundation in the sites. The correlation between changes in SAR backscatter and water depth indicates the potential for using spaceborne SAR systems, such as the ERS for monitoring variations in flooding in south Florida wetlands.     

Monitoring leaf area of sugar beet using ERS-1 SAR data

This paper presents the results of field testing a radar model which relates leaf area index to radar backscatter for ERS-1 C-band VV polarization SAR data. Ground truth measurements of leaf area index and soil moisture content were made in selected sugar beet fields, with simultaneous acquisition of ERS-1 SAR image data. Radar backscatter coefficients were derived from the calibrated ERS-1 SAR data. The Leeuwen and Clevers expression of the water cloud model was fitted to determine the in situ relationship between radar back-scatter and leaf area index. The model can be inverted analytically to calculate leaf area index from radar backscatter. The results show considerable potential for the operational application of ERS-1 SAR data in crop monitoring.     

Evaluating ERS-1 SAR data for the discrimination of tropical forest from other tropical vegetation types in Papua New Guinea

ERS-1 Synthetic Aperture Radar (SAR) data over a study area located in Papua New Guinea, where there is a high probability of cloud cover, are evaluated on their information content for mapping tropical forest ecosystems. The feasibility of forest/non-forest discrimination using mono- and multi-temporal ERS-1 SAR data at 100m pixel size is investigated using two different classification methodologies. An assessment of the optimal acquisition period and number of acquisitions is undertaken. The automatic classification results are compared quantitatively with the aid of field observations in a comparative accuracy assessment methodology, and a comparison is made with Landsat Thematic Mapper (TM) data. Finally, the potential of ERS-1 SAR data for the discrimination of tropical forest types is investigated. The results showed that multi-temporal ERS-1 SAR data acquired at the appropriate times were found to have a high potential for forest/nonforest discrimination and achieved similar classification accuracies to the TM data. The discrimination of forest types proved difficult. However, discrimination was possible between dense and open forest types having different canopy structures.     

Effects of soil moisture and water depth on ERS SAR backscatter measurements from an Alaskan wetland complex

We conducted a preliminary investigation of the response of ERS C-band SAR backscatter to variations in soil moisture and surface inundation in wetlands of interior Alaska. Data were collected from 5 wetlands over a three-week period in 2007. Results showed a positive correlation between backscatter and soil moisture in sites dominated by herbaceous vegetation cover (r = 0.74, p < 0.04). ERS SAR backscatter was negatively correlated to water depth in all open (non-forested) wetlands when water table levels were more than 6 cm above the wetland surface (r = − 0.82, p < 0.001). There was no relationship between backscatter and soil moisture in the forested (black spruce-dominated) wetland site. Our preliminary results show that ERS SAR data can be used to monitor variations in hydrologic conditions in high northern latitude wetlands (including peatlands), particularly sites with sparse tree cover.     

微波遥感监测土壤水分的研究初探

在GPS定位的基础上，同步测量土攘水分、土壤后向散射系数，和同步获取的X波段、HH机化SAR图像进行了土攘水分监N.]的徽波遥感试验研究。结果表明，X波段SAR图像的灰度与表层土壤(0~10cm)水分有较好的相关性，35OHH极化的土峨后向散射系数与SAR图像灰度和土攘水分也有较好的相关性，由SAR图像及土攘的后向散射系数估算的土峨水分精度相近，相对误差均为12%左右，因而利用X波段、HH极化的机载SAR图像监浏土壤水分是可行的。雷达图像的穿透力一般在10cm以内，因此探讨了由表层土壤水分推求剖面土壤水分的可能性，并提出以土攘水分计法在浏童精度和速度上改进传统土壤水分测量的方法。     

Rice heading date retrieval based on multi-temporal MODIS data and polynomial fitting

The objective of the study is to identify the rice heading date and analyse its spatial characteristics on a regional scale using multi-temporal Moderate Resolution Imaging Spectroradiometer (MODIS) normalized differential vegetation index (NDVI) data and a new approach: quadratic polynomial fitting. The cloud-contaminated NDVI value was identified by reliability data and linearly interpolated with values before and after the cloudy one. The discrete Fourier transformation (DFT) and quadratic polynomial fitting were implemented to generate new time series curves. Rice heading date was retrieved by calculating the day for maximum NDVI. Comparing with DFT, the proposed quadratic polynomial fitting significantly improves the computation efficiency, while providing approximate precision of estimation. In regional analysis, the rice heading date retrieved from polynomial fitting is more consistent than that from DFT. The study also suggests that multi-temporal MODIS NDVI data combined with different methods can retrieve crop phenology information on a large scale.     

基于多时相Sentinel-2影像的黑河中游玉米种植面积提取研究

玉米是黑河中游种植面积最大的农作物,生长期需水量大、蒸散量高.准确获取玉米种植面积对该区域农作物种植结构调整、水资源合理规划有重要参考意义.基于2019年4月至9月Sentinel-2多时相影像,采用随机森林算法开展了黑河中游玉米种植面积提取研究.研究方法分为两类—直接提取法和两步提取法.进一步探讨了多时间信息量对玉米...     

Patterns of irrigated rice growth and malaria vector breeding in Mali using multi-temporal ERS-2 synthetic aperture radar

We explored the use of the European Remote Sensing Satellite 2 Synthetic Aperture Radar (ERS-2 SAR) to trace the development of rice plants in an irrigated area near Niono, Mali and relate that to the density of anopheline mosquitoes, especially An. gambiae. This is important because such mosquitoes are the major vectors of malaria in sub-Saharan Africa, and their development is often coupled to the cycle of rice development. We collected larval samples, mapped rice fields using GPS and recorded rice growth stages simultaneously with eight ERS-2 SAR acquisitions. We were able to discriminate among rice growth stages using ERS-2 SAR backscatter data, especially among the early stages of rice growth, which produce the largest numbers of larvae. We could also distinguish between basins that produced high and low numbers of anophelines within the stage of peak production. After the peak, larval numbers dropped as rice plants grew taller and thicker, reducing the amount of light reaching the water surface. ERS-2 SAR backscatter increased concomitantly. Our data support the belief that ERS-2 SAR data may be helpful for mapping the spatial patterns of rice growth, distinguishing different agricultural practices, and monitoring the abundance of vectors in nearby villages.     

Crop discrimination using multi-temporal SAR imagery

Seven ERS-1 SAR images obtained at different dates during the 1993 crop growing season are used in a study of the potential of multi-temporal SAR for agricultural crop discrimination for an area near Feltwell, Norfolk, UK. The study compares a per-pixel and a per-field approach. Pixel-based classification is based on raw intensity images, temporal subtraction images, filtered images, and texture features. Field-based classification uses the mean back-scatter coefficient derived for each field. Analysis of the contribution of each dataset uses statistical separability measures and confusion matrix methods. The classification algorithms used are maximum likelihood and Kohonen's self-organized feature map (SOM). We find that SAR-based texture features contribute nothing to crop discrimination. Filtered images produce the best result for the per-pixel approach, giving a classification accuracy of around 60%. The use of a SOM for field-based classification produces a classification accuracy greater than 75%. This is not a surprising result, as field-based classifications use averaged data, in which the noise effect is reduced.     

Accurate crop-type classification using multi-temporal optical and multi-polarization SAR data in an object-based image analysis framework

Accurate crop-type classification is a challenging task due, primarily, to the high within-class spectral variations of individual crops during the growing season (phenological development) and, second, to the high between-class spectral similarity of crop types. Utilizing within-season multi-temporal optical and multi-polarization synthetic aperture radar (SAR) data, this study introduces a combined object- and pixel-based image classification methodology for accurate crop-type classification. Particularly, the study investigates the improvement of crop-type classification by using the least number of multi-temporal RapidEye (RE) images and multi-polarization Radarsat-2 (RS-2) data utilized in an object- and pixel-based image analysis framework. The method was tested on a study area in Manitoba, Canada, using three different classifiers including the standard Maximum Likelihood (ML), Decision Tree (DT), and Random Forest (RF) classifiers. Using only two RE images of July and August, the proposed method results in overall accuracies (OAs) of about 95%, 78%, and 93% for the ML, DT, and RF classifiers, respectively. Moreover, the use of only two quad-pol images of RS-2 of June and September resulted in OAs of 92%, 75%, and 90% for the ML, DT, and RF classifiers, respectively. The best classification results were achieved by the synergistic use of two RE and two RS-2 images. In this case, the overall classification accuracies were 97% for both ML and RF classifiers. In addition, the average producer’s accuracies of 95% and 96% were achieved by the ML and RF classifiers, respectively, whereas the average user accuracy was 94% for both classifiers. The results indicated promising potentials for rapid and cost-effective local-scale crop-type classification using a limited number of high-resolution optical and multi-polarization SAR images. Very accurate classification results can be considered as a replacement for sampling the agricultural fields at the local scale. The result of this very accurate classification at discrete locations (approximately 25 × 25 km frames) can be applied in a separate procedure to increase the accuracy of crop area estimation at the regional to provincial scale by linking these local very accurate spatially discrete results to national wall-to-wall continuous crop classification maps.     

Classification of sea ice types in the Ross Sea,Antarctica from SAR and AVHRR imagery

ERS-2 synthetic aperture radar (SAR) and Advanced Very High Resolution Radiometer (AVHRR) imagery are used to examine spectral characteristics of late winter/early spring ice in the Ross Sea, Antarctica. The combined spectral signatures are used to distinguish six ice types: fast ice, new ice, smooth first year ice, rough first year ice, thin new ice/wind roughened open water and glacial ice. The procedure firstly involves 'picking' class boundaries from SAR imagery based on the morphology of a speckle reduced backscatter spectrum. These class boundaries are then used as input to an iterative segmentation procedure that involves the repeated application of a speckle reduction filter to the image. For an image from late September 1996 the segmentation procedure enabled separation of five general ice categories each with a characteristic backscatter range. However because of the combined contributions of ice thickness, surface roughness, salinity and water content to the SAR backscatter, further decision criteria are required to separate some physical ice types unable to be resolved individually using this method. Coincident and co-registered infrared data from the AVHRR sensor are used to extract spectral characteristics for the final ice classes. Using this procedure we were able to distinguish floating glacier ice from thin new ice/wind roughened open water and new ice from nearshore fast ice. These ice types were unable to be separated using SAR backscatter intensity alone. In addition image subtraction was also able to clearly delineate areas of shore fast ice.     

The Bothnian Experiment in Preparation for ERS-1, 1988 BEPERS-88) — an overview

Abstract

BEPERS-88 was an extensive field campaign on the use of Synthetic Aperture Radar (SAR) in sea ice remote sensing in the Baltic Sea. This experiment was performed in order to study the possibilities of using the ERS-1 satellite SAR (and radar altimeter) in connection with the brackish ice in the Baltic Sea. The Canada Centre for Remote Sensing CV-580 C/X-band SAR was flown and an extensive validation programme was carried out. The data have been used for SAR image analysis, backscatter investigations, geophysical validation of SAR over sea ice, and evaluation of the potentials of SAR in operational ice information services. The results indicate that SAR can be used to discriminate between ice and open water, classify ice types into thrcc categories, quantify ice ridging intensity, and determine the ice drift. As an operational tool SAR is expected to be an excellent complement to NOAA imagery and ground truth.     

Estimating crop and waveband specific water cloud model parameters using a theoretical backscatter model

Synthetic aperture radar (SAR) data are useful for monitoring various biophysical properties, and backscatter models are required to extract such information. The semi-empirical water cloud model is traditionally formulated using four parameters fitted using in situ data. The model becomes more accurate through the use of crop and soil specific values of these parameters, estimated using a robust theoretical second-order backscatter model. Methods are introduced in this letter for generating two of the four parameters specific to C-band data to enable greater transportability of the model.     

基于多极化星载SAR数据的水稻/旱田识别-以江苏省海安县为例 

合成孔径雷达(SAR)数据对于南方多云多雨天气的地表农作物类型的探测具有独特的优势。以江苏省海安县为例,基于多极化SAR数据,包括双极化ALOS PALSAR以及全极化Radarsat\|2数据,采用面向对象的方法,针对当地水稻/旱田进行识别。针对双极化SAR数据,利用了其强度信息进行分类识别;而基于全极化数据,除强度信息外,还利用了其SAR信号统计分布概率进行分类规则建立。结果表明:L波段的ALOS PALSAR在识别旱地的桑树方面具有很大的优势,而基于两种分类方法的C波段Radarsat\|2数据识别水稻的精度分别为85%和75%,略低于ALOS PALSAR的识别结果(87.5%)。     

Use of ENVISAT/ASAR wide-swath data for timely rice fields mapping in the Mekong River Delta

Because of the importance of rice for the global food security and because of the role of inundated paddy fields in greenhouse gases emissions, monitoring the rice production world-wide has become a challenging issue for the coming years. Local rice mapping methods have been developed previously in many studies by using the temporal change of the backscatter from C-band synthetic aperture radar (SAR) co-polarized data. The studies indicated in particular the need of a high observation frequency. In the past, the operational use of these methods has been limited by the small coverage and the poor acquisition frequency of the available data (ERS-1/2, Radarsat-1). In this paper, the method is adapted for the first time to map rice at large scale, by using wide-swath images of the Advanced SAR (ASAR) instrument onboard ENVISAT. To increase the observation frequency, data from different satellite tracks are combined. The detection of rice fields is achieved by exploiting the high backscatter increase at the beginning of the growing cycle, which allows the production of rice maps early in the season (in the first 50 days). The method is tested in the Mekong delta in Vietnam. The mapping results are compared to existing rice maps in the An Giang province, with a good agreement (higher than 81%). The rice planted areas are retrieved from the maps and successfully validated with the official statistics available at each province (R² = 0.92). These results show that the method is useful for large scale early mapping of rice areas, using current and future C band wide-swath SAR data.