Search

The seasonal effect on land cover classification has been widely recognized. It is important to use the imagery acquired at key points of vegetation phenological development to obtain a higher classification accuracy for land cover. This study compared the effect of seasons on landscape classification and the quasi-circular vegetation patches (QVPs) detection from four fused Gaofen 1 images acquired in the different seasons by using the pixel-based random forest (RF) and object-based support vector machine (SVM) methods over the Yellow River Delta, China. The results from this study demonstrated that the seasonal effect on classifying landscapes and detecting the QVPs is significant, especially for the pixel-based RF method. The object-based SVM method was more appropriate for classifying landscape from the non-growing season images, while the pixel-based RF approach was more suitable for classifying the growing-season images. The spring data (April imagery; overall accuracy = 99.8%) and the winter data (February imagery; F measure = 65.9%) yielded the best results for landscape classification and QVP detection, respectively, by using the object-based SVM approach. Therefore, in practice, we recommend the use of February to April imagery with the object-based SVM approach to map the QVPs in the future.

A band selection model was described for efficient and accurate remotely-sensed vegetation mapping in cloudy mixed-vegetation areas, demonstrated with an application on mapping mangroves in Southeast Asia (SE Asia). We show how to use multi-source satellite imagery and Cloud Computing Platforms to improve mapping and computational efficiency in complex environments. A key element of the method relies upon field surveys to establish a detailed sample database that includes easily-confused land cover. The Maximal Separability and Information (MSI) model was developed to select key bands for target land cover classification from multiple satellite imagery based on two principles: 1. maximize separability of the target cover from other land cover; and 2. maximize and prioritize information from band combinations. Application of the MSI model to map mangroves in SE Asia using three optical and SAR data systems (Landsat OLI, Sentinel-2 and Sentinel-1) showed: 1. Sentinel-2 is better at classifying mangrove than Landsat and Sentinel-1; and 2. SWIR, NIR and Red bands (with SWIR in particular) are effective in separating mangrove from other vegetation. The MSI-mapped mangroves showed lower computation cost compared to using all bands from individual satellites, and higher accuracy (above 90%) when applied to SE Asia. It was robust in tolerating smaller sample sizes, thereby demonstrating computational feasibility and substantial improvements with the MSI model for large-scale land cover mapping in complex environments.

Because of a decrease in water provision from the Yellow River and the influence of various human actions, the Shandong Yellow River Delta Wetland, selected a Ramsar Wetland of International Importance in 2013, has undergone substantial degradation. Although some restoration work has been undertaken, more adaptive active restoration strategies should be developed based on researches on spontaneous revegetation in this region. In this regard, the quasi-circular vegetation patches (QVPs), characterized by a rapid rate of succession, are thus considered as ideal research targets. The spatial and temporal quantification of QVPs is not only fundamental to determining QVP formation mechanism but also contributes to the development of adaptive restoration strategies for the degraded ecosystems. Therefore, we examined QVP dynamics over seven time periods spanning 22 years using one aerial photograph and seven high-spatial resolution satellite images. We found that although the total number of QVPs remained almost unchanged, the total area had an increasing trend, and that the shape index (SIi=Peremeteri2πAreai) of QVPs tended to initially increase and subsequently decrease, with QVPs finally attaining an almost circular shape. Observations for the period between 2000 and 2016 indicated that the area and shape of QVPs could reach maximum extents within about three years. This knowledge could provide a valuable basis for studying the interactions of QVPs with environment, and the implications/consequences to ecosystems, e.g., mechanisms of the evolution of ecosystems, ecological services etc. We accordingly suggest that developing a long-term remote sensing monitoring and ground-based survey program would make a significant contribution to developing a nature-based solution for the degraded ecosystems in the Yellow River Delta, China.

Vegetation, hydrology and geomorphology are three major elements of the floodplain ecosystem on Earth. Although the normalized difference vegetation index (NDVI) has been used extensively to characterize floodplain vegetation growth, vigour and biomass, methods for quantifying the various distinct responses of floodplain vegetation to hydro-geomorphological changes in different lateral belts in arid regions are still needed. In this study, the Linhe reach was divided into four lateral belts based on their hydro-geomorphological characteristics, and the Moderate Resolution Imaging Spectroradiometer (MODIS)-NDVI time series statistical indicators were used to characterise the distinct changing the patterns of vegetation growth in different belts. The response of floodplain vegetation to river hydro-geomorphology in each belt was analysed. The result showed that the average maximum NDVI value in the regular inundation area was 0.23 and higher than that in the other lateral belts. The correlation between the water persistence time and peak NDVI value in the regular water inundation area was significant (ρ = 0.84), indicating that in contrast to highly frequent or extremely rare water inundation, regular water inundation provides significant benefits to floodplains. Continuous or highly frequent inundation may cause decreased vegetation productivity. Overall, our results suggest that the vegetation greenness response to the river hydro-geomorphology is different from the river to the edge of the floodplain. Thus, a better understanding of the interactions between the floodplain vegetation and river hydro-morphology and river water resource management in arid-region floodplains.

The Mun River Basin is one of Thailand’s major grain-producing areas, but the production is insufficient, and most of the cultivated lands are rain-fed and always unused in the dry season. All this makes it necessary to determine the status of soil nutrients and soil quality in the dry season to improve soil conditions, which will be useful for cultivation in the farming period. The aim of this study was to construct a soil-quality assessment based on soil samples, and in the process the minimum data set theory was introduced to screen the assessment indicators. The geographically weighted regression method was used to complete the spatial interpolation process of indicators, and the fuzzy logic model was constructed to evaluate the soil quality. The results showed that the spatial distributions of soil quality and indicators were similar. The soil quality was the best in the upstream while poor in the downstream, and the dry fields in the west and the forests in the east of the basin were better than other areas nearby. However; the soil qualities of paddy fields in the middle and east of the basin were poor due to the lack of soil nutrient supply when the fields were unused

Timely and accurate regional rice paddy monitoring plays a significant role in maintaining the sustainable rice production, food security, and agricultural development. This study proposes an operational automatic approach to mapping rice paddies using time-series SAR data. The proposed method integrates time-series Sentinel-1 data, auxiliary data of global surface water, and rice phenological characteristics with Google Earth Engine cloud computing platform. A total of 402 Sentinel-1 scenes from 2017 were used for mapping rice paddies extent in the Mun River basin. First, the calculated minimum and maximum values of the backscattering coefficient of permanent water (a classification type within global surface water data) in a year was used as the threshold range for extracting the potential extent. Then, three rice phenological characteristics were extracted based on the time-series curve of each pixel, namely the date of the beginning of the season (DBS), date of maximum backscatter during the peak growing season (DMP), and length of the vegetative stage (LVS). After setting a threshold for each phenological parameter, the final rice paddy extent was identified. Rice paddy map produced in this study was highly accurate and agreed well with field plot data and rice map products from the International Rice Research Institute (IRRI). The results had a total accuracy of 89.52% and an F1 score of 0.91, showing that the spatiotemporal pattern of extracted rice cover was consistent with ground truth samples in the Mun River basin. This approach could be expanded to other rice-growing regions at the national scale, or even the entire Indochina Peninsula and Southeast Asia.

Timely understanding of the coastal accretion–erosion dynamics of the Yellow River Delta (YRD) can not only deepen the understanding of the evolution of the delta but also provide scientific support for water-sediment regulation (WSR) in the lower reaches of Yellow River and the implementation of a protection strategy for the Yellow River Estuary. In this long-term study, Landsat images from 1976 to 2018 were acquired, and the cloud processing platform of the Google Earth Engine was used for extraction of coastlines. On the basis of these coastlines, the area and accretion–erosion dynamics were analyzed. Then, after statistical analysis of the interannual and intra-annual variations in runoff and sediment, we discuss the relationship between the accretion–erosion dynamics and the annual runoff and sediment. The results show that (1) the coastline of the YRD lengthened first and then shortened, and the average annual growth rate was 1.48 km/a. (2) The land area of the YRD showed a significant accretionary trend before 1996, with an average annual growth rate of 28.60 km2/a. Then, the area gradually decreased from 1997 to 2001. After WSR was implemented in 2002, the accretion–erosion dynamics gradually became smooth, with an annual growth rate of 0.31 km2/a. (3) After WSR, the maximum annual sedimentation decreased by 79.70%. The average annual sediment discharge accounted for only 6.69% from November to March of the following year during the non-flood season. (4) With the continuous decrease in sediment discharge, the determination coefficient (R2) between the cumulative accretion–erosion area of the estuary and the annual sedimentation decreased from 0.98 in 1976–1996 to 0.77 after 2002. Overall, although WSR has maintained a steady increase in delta land area, it cannot change the long-term decrease in the land area of the delta. The insights gained from our study can provide some references for related coastline research, and will be useful to science community and decision makers for coastal environmental monitoring, management, protection, and sustainable development of the YRD.

Accurate remote sensing and mapping of land cover in the tropics remain difficult tasks since data gaps and a heterogenic landscape make it challenging to perform land cover classification. In this paper, we proposed a multi-feature classification method to integrate temporal statistical features with spectral and textural features. This method is designed to improve the accuracy of land cover classification in cloud-prone tropical regions. Sentinel-2 images were used to construct an NDVI stack for a time-series statistical analysis to characterize the temporal variance of land cover. Two statistical indices were calculated and used to represent the variation in annual vegetation. These indices included the mean (NDVI_mean) and coefficient of variation (NDVI_cv) for the NDVI time series. The temporal statistical features were then integrated with spectral and textural features extracted from high-quality Sentinel-2 imagery for Random Forest classification. The performance and contribution of different combinations were assessed based on their classification accuracies. Our results show that the time-series statistical analysis is an effective way to represent land cover category information contained in annual NDVI variance. The method uses clear pixels from dense low-quality images to obtain the NDVI statistical characteristics, thus, to reduce the influence of random factors such as weather conditions on single-date image. The addition of NDVI_mean and NDVI_cv can improve the separability among most types of land cover. The overall accuracy and the kappa coefficient reached values of 0.8913 and 0.8514 when NDVI_mean and NDVI_cv were integrated. Furthermore, the time-series statistical analysis has less stringent requirements regarding image quality and features a high computational efficiency, which shows its great potential to improve the overall accuracy of land cover classification at regional scales in cloud-prone tropical regions.

The expansion of rubber (Hevea brasiliensis) plantations has been a critical driver for the rapid transformation of tropical forests, especially in Thailand. Rubber plantation mapping provides basic information for surveying resources, updating forest subplot information, logging, and managing the forest. However, due to the diversity of stand structure, complexity of the forest growth environment, and the similarity of spectral characteristics between rubber trees and natural forests, it is difficult to discriminate rubber plantation from natural forest using only spectral information. This study evaluated the validity of textural features for rubber plantation recognition at different spatial resolutions using GaoFen-1 (GF-1), Sentinel-2, and Landsat 8 optical data. C-band Sentinel-1 10 m imagery was first used to map forests (including both rubber plantations and natural forests) and non-forests, then the pixels identified as forests in the Sentinel-1 imagery were compared with GF-1, Sentinel-2, and Landsat 8 images to separate rubber plantations and natural forest using two different approaches: a method based on spectral information characteristics only and a method combining spectral and textural features. In addition, we extracted textural features of different window sizes (3 × 3 to 31 × 31) and analyzed the influence of window size on the separability of rubber plantations and natural forests. Our major findings include: (1) the suitable texture extraction window sizes of GF-1, Sentinel-2, and Landsat 8 are 31 × 31, 11 × 11 to 15 × 15, and 3 × 3 to 7 × 7, respectively; (2) correlation (COR) is a robust textural feature in remote sensing images with different resolutions; and (3) compared with classification by spectral information only, the producer’s accuracy of rubber plantations based on GF-1, Sentinel-2, and Landsat 8 was improved by 8.04%, 9.44%, and 8.74%, respectively, and the user’s accuracy was increased by 4.63%, 4.54%, and 6.75%, respectively, when the textural features were introduced. These results demonstrate that the method combining textural features has great potential in delineating rubber plantations.

Machine learning comprises a group of powerful state-of-the-art techniques for land cover classification and cropland identification. In this paper, we proposed and evaluated two models based on random forest (RF) and attention-based long short-term memory (A-LSTM) networks that can learn directly from the raw surface reflectance of remote sensing (RS) images for large-scale winter wheat identification in Huanghuaihai Region (North-Central China). We used a time series of Moderate Resolution Imaging Spectroradiometer (MODIS) images over one growing season and the corresponding winter wheat distribution map for the experiments. Each training sample was derived from the raw surface reflectance of MODIS time-series images. Both models achieved state-of-the-art performance in identifying winter wheat, and the F1 scores of RF and A-LSTM were 0.72 and 0.71, respectively. We also analyzed the impact of the pixel-mixing effect. Training with pure-mixed-pixel samples (the training set consists of pure and mixed cells and thus retains the original distribution of data) was more precise than training with only pure-pixel samples (the entire pixel area belongs to one class). We also analyzed the variable importance along the temporal series, and the data acquired in March or April contributed more than the data acquired at other times. Both models could predict winter wheat coverage in past years or in other regions with similar winter wheat growing seasons. The experiments in this paper showed the effectiveness and significance of our methods.

Land cover is a fundamental component of crucial importance in the earth sciences. To date, many excellent international teams have created a variety of land cover products covering the entire globe. To provide a reference for researchers studying the Arctic, this paper evaluates four commonly used land cover products. First, we compare and analyze the four land cover products from the perspectives of land cover type, distribution and spatial heterogeneity. Second, we evaluate the accuracy of such products by using two sets of sample points collected from the Arctic region. Finally, we obtain the spatial consistency distribution of the products by means of superposition analysis. The results show the following: (a) among the four land cover products, Climate Change Initiative Land Cover (CCI-LC) has the highest overall accuracy (63.5%) in the Arctic region, GlobeLand30 has an overall accuracy of 62.2% and the overall accuracy of the Global Land Cover by the National Mapping Organization (GLCNMO) is only 48.8%. When applied in the Arctic region, the overall accuracy of the Moderate Resolution Imaging Spectroradiometer (MODIS) is only 29.5% due to significant variances. Therefore, MODIS and GLCNMO are not recommended in Arctic-related research as their use may lead to major errors. (b) An evaluation of the consistency of the four products indicates that the classification of the large-scale homogeneous regions in the Arctic yields satisfactory results, whereas the classification results in the forest−tundra ecotone are unsatisfactory. The results serve as a reference for future research. (c) Among the four products, GlobeLand30 is the best choice for analyzing finely divided and unevenly distributed surface features such as waters, urban areas and cropland. Climate Change Initiative Land Cover (CCI-LC) has the highest overall accuracy, and its classification accuracy is relatively higher for forests, shrubs, sparse vegetation, snow/ice and water. GlobeLand30 and CCI-LC do not vary much from each other in terms of overall accuracy. They differ the most in the classification accuracy of shrub-covered land; CCI-LC performed better than GlobeLand30 in the classification of shrub-covered land, whereas the latter obtained higher accuracy than that of the former in the classification of urban areas and cropland.

High-resolution satellite imagery enables decametric-scale quasi-circular vegetation patch (QVP) mapping, which greatly aids the monitoring of vegetation restoration projects and the development of theories in pattern evolution and maintenance research. This study analyzed the potential of employing five seasonal fused 5 m spatial resolution CBERS-04 satellite images to map QVPs in the Yellow River Delta, China, using the Random Forest (RF) classifier. The classification accuracies corresponding to individual and multi-season combined images were compared to understand the seasonal effect and the importance of optimal image timing and acquisition frequency for QVP mapping. For classification based on single season imagery, the early spring March imagery, with an overall accuracy (OA) of 98.1%, was proven to be more adequate than the other four individual seasonal images. The early spring (March) and winter (December) combined dataset produced the most accurate QVP detection results, with a precision rate of 66.3%, a recall rate of 43.9%, and an F measure of 0.528. For larger study areas, the gain in accuracy should be balanced against the increase in processing time and space when including the derived spectral indices in the RF classification model. Future research should focus on applying higher resolution imagery to QVP mapping.

Vegetation in arid and semi-arid regions frequently exists in patches, which can be effectively mapped by remote sensing. However, not all satellite images are suitable to detect the decametric-scale vegetation patches because of low spatial resolution. This study compared the capability of the first Gaofen Satellite (GF-1), the second Gaofen Satellite (GF-2), and China-Brazil Earth Resource Satellite 4 (CBERS-04) panchromatic images for mapping quasi-circular vegetation patches (QVPs) with K-Means (KM) and object-based example-based feature extraction with support vector machine classification (OEFE) in the Yellow River Delta, China. Both approaches provide relatively high classification accuracy with GF-2. For all five images, the root mean square errors (RMSEs) for area, perimeter, and perimeter/area ratio were smaller using the KM than the OEFE, indicating that the results from the KM are more similar to ground truth. Although the mapped results of the QVPs from finer-spatial resolution images appeared more accurate, accuracy improvement in terms of QVP area, perimeter, and perimeter/area ratio was limited, and most of the QVPs detected only by finer-spatial resolution imagery had a more than 40% difference with the actual QVPs in these three parameters. Compared with the KM approach, the OEFE approach performed better for vegetation patch shape description. Coupling the CBERS-04 with the OEFE approach could suitably map the QVPs (overall accuracy 75.3%). This is important for ecological protection managers concerned about cost-effectiveness between image spatial resolution and mapping the QVPs.

The Chinese Loess Plateau is an ecologically fragile and sensitive area. The carbon storage dynamics in this region and the contributions from land use/land cover change (LUCC) and carbon density from 2000 to 2010 were analyzed in this paper. Normalized difference vegetation index (NDVI), biomass and soil carbon data in 2000 were used for regression analysis of biomass and soil carbon, and an inversion analysis was used to estimate biomass and soil carbon in 2005 and 2010. Quadrat data, including aboveground biomass and soil organic carbon, were used to calibrate the model output. Carbon storage and sequestration were calculated by the InVEST toolset with four carbon pools, including aboveground biomass, belowground biomass, dead wood and soil carbon. The results showed that carbon storage increased steadily from 2000 to 2010, increasing by 0.260 billion tons, and that woodland area increased and arable land decreased; the overall trend in land cover improved, but the improvement was not pronounced. Carbon storage in the Loess Plateau was correlated with geographical factors. We found that when assuming a constant carbon density, carbon storage decreased, accounting for −1% of the carbon storage dynamics. When assuming no land conversion, carbon storage increased, accounting for 101% of the carbon storage dynamics.

Leaf area index (LAI) is one of the key biophysical parameters in crop structure. The accurate quantitative estimation of crop LAI is essential to verify crop growth and health. The PROSAIL radiative transfer model (RTM) is one of the most established methods for estimating crop LAI. In this study, a look-up table (LUT) based on the PROSAIL RTM was first used to estimate winter wheat LAI from GF-1 data, which accounted for some available prior knowledge relating to the distribution of winter wheat characteristics. Next, the effects of 15 LAI-LUT strategies with reflectance bands and 10 LAI-LUT strategies with vegetation indexes on the accuracy of the winter wheat LAI retrieval with different phenological stages were evaluated against in situ LAI measurements. The results showed that the LUT strategies of LAI-GNDVI were optimal and had the highest accuracy with a root mean squared error (RMSE) value of 0.34, and a coefficient of determination (R2) of 0.61 during the elongation stages, and the LUT strategies of LAI-Green were optimal with a RMSE of 0.74, and R2 of 0.20 during the grain-filling stages. The results demonstrated that the PROSAIL RTM had great potential in winter wheat LAI inversion with GF-1 satellite data and the performance could be improved by selecting the appropriate LUT inversion strategies in different growth periods.

This paper proposes an open-boundary locally weighted dynamic time warping (OLWDTW) method using MODIS Normalized Difference Vegetation Index (NDVI) time-series data for cropland recognition. The method solves the problem of flexible planting times for crops in Southeast Asia, which has sufficient thermal and water conditions. For NDVI time series starting at the beginning of the year and terminating at the end of the year, the method can separate the non-growing season cycle and growing season cycle for crops. The non-growing season cycle may provide some useful information for crop recognition, such as soil conditions. However, the shape of the growing season’s NDVI time series for crops is the key to separating cropland from other land cover types because the shape contains all of the crop growth information. The principle of the OLWDTW method is to enhance the effects of the growing season cycle on the NDVI time series by adding a local weight to the growing season when comparing the similarity of time series based on the open-boundary dynamic time warping (DTW) method. Experiments with two satellite datasets located near the Khorat Plateau in the Lower Mekong Basin validate that OLWDTW effectively improves the precision of cropland recognition compared to a non-weighted open-boundary DTW method in terms of overall accuracy. The method’s classification accuracy on cropland exceeds the non-weighted open-boundary DTW by 5–7%. In future studies, an open-boundary self-adaption locally weighted DTW and a more effective combination rule for different crop types should be explored for the method’s best performance and highest extraction accuracy for cropland.

To improve the accuracy of winter wheat yield estimation, the Crop Environment Resource Synthesis for Wheat (CERES-Wheat) model with an assimilation strategy was performed by assimilating measured or remotely-sensed leaf area index (LAI) values. The performances of the crop model for two different assimilation methods were compared by employing particle filters (PF) and the proper orthogonal decomposition-based ensemble four-dimensional variational (POD4DVar) strategies. The uncertainties of wheat yield estimates due to different assimilation temporal scales (phenological stages and temporal frequencies) and spatial scale were also analyzed. The results showed that, compared with the crop model without assimilation and with PF-based assimilation, a better yield estimate performance resulted when the POD4DVar-based strategy was used at the field scale. When using this strategy, root mean square errors (RMSE) of 523 kg·ha−1, 543 kg·ha−1 and 172 kg·ha−1 and relative errors (RE) of 5.65%, 5.91% and 7.77% were obtained at the field plot scale, a pixel scale of 1 km and the county scale, respectively. Although the best yield estimates were obtained when all of the observed LAIs were assimilated into the crop model, an acceptable estimate of crop yield could also be achieved by assimilating fewer observations between jointing and anthesis periods of the crop growth season. With decreasing assimilation frequency and pixel resolution, the accuracy of the crop yield estimates decreased; however, the computation time decreased. Thus, it is important to consider reasonable spatio-temporal scales to obtain tradeoffs between accuracy and effectiveness in regional wheat estimates.

Normalized Difference Vegetation Index (NDVI) derived from Moderate Resolution Imaging Spectroradiometer (MODIS) time-series data has been widely used in the fields of crop and rice classification. The cloudy and rainy weather characteristics of the monsoon season greatly reduce the likelihood of obtaining high-quality optical remote sensing images. In addition, the diverse crop-planting system in Vietnam also hinders the comparison of NDVI among different crop stages. To address these problems, we apply a Dynamic Time Warping (DTW) distance-based similarity measure approach and use the entire yearly NDVI time series to reduce the inaccuracy of classification using a single image. We first de-noise the NDVI time series using S-G filtering based on the TIMESAT software. Then, a standard NDVI time-series base for rice growth is established based on field survey data and Google Earth sample data. NDVI time-series data for each pixel are constructed and the DTW distance with the standard rice growth NDVI time series is calculated. Then, we apply thresholds to extract rice growth areas. A qualitative assessment using statistical data and a spatial assessment using sampled data from the rice-cropping map reveal a high mapping accuracy at the national scale between the statistical data, with the corresponding R2 being as high as 0.809; however, the mapped rice accuracy decreased at the provincial scale due to the reduced number of rice planting areas per province. An analysis of the results indicates that the 500-m resolution MODIS data are limited in terms of mapping scattered rice parcels. The results demonstrate that the DTW-based similarity measure of the NDVI time series can be effectively used to map large-area rice cropping systems with diverse cultivation processes.