Vegetation Index Time Series Data Research Articles

Time Series from Sentinel-2 for Organic Durum Wheat Yield Prediction Using Functional Data Analysis and Deep Learning

Yield prediction is a crucial activity in scheduling agronomic operations and in informing the management and financial decisions of a wide range of stakeholders of the organic durum wheat supply chain. This research aims to develop a yield forecasting system by combining vegetation index time-series data from Sentinel-2 L2A time-series data, field-measured yields, and deep learning techniques. Remotely sensed data over a season could be, in general, noisy and characterized by a variable density due to weather conditions. This problem was mitigated using Functional Principal Component Analysis (FPCA). We obtained a functional representation of acquired data, and starting from this, we tried to apply deep learning to predict the crop yield. We used a Convolutional Neural Network (CNN) approach, starting from images that embed temporal and spectral dimensions. This representation does not require one to a priori select a vegetation index that, typically, is task-dependent. The results have been also compared with classical approaches as Partial Least Squares (PLS) on the main reference vegetation indexes such as the Normalized Difference Vegetation Index (NDVI) and Normalized Difference Red Edge index (NDRE), considering both in-season and end-season scenarios. The obtained results show that the image-based representation of multi-spectral time series could be an effective method to estimate the yield, also, in the middle stage of cropping with R2 values greater than 0.83. The developed model could be used to estimate yield the neighbor fields characterized by similar setups in terms of the crop, variety, soil, and, of course, management.

Evaluation and quantitative characterization for the ecological environment impact of open pit mining on vegetation destruction from landsat time series: A case study of Wulishan limestone mine

Quantitatively evaluating the ecological environment impacts of vegetation destruction due to open-pit mining activities is vital for enhancing the green mining standard and cost management capabilities of mining enterprises. Based on the Landsat time series, this study proposes an ecological environment impact assessment and quantitative characterization method for vegetation destruction in mining areas resulting from open-pit mining activities. First, the modified normalized difference water index and the normalized difference vegetation index time series data were calculated. The water body thresholds and the fraction of vegetation coverage were ascertained using the K-means clustering algorithm and the dimidiate pixel model, respectively, to determine the area of direct vegetation destruction in mining areas. Second, utilizing the Theil-Sen Median trend analysis and the Mann-Kendall test, the indirect impact area of vegetation in the mining region was identified. Lastly, by integrating vegetation's net primary productivity with the Chinese Emission Allowance price index, the total carbon emission cost of vegetation destruction due to mining activities over 20 years was calculated to be about 2.122 million yuan. The findings indicated that the ecological environmental impact of open-pit mining activities on vegetation destruction cannot be ignored. From 2000 to 2020, open-pit mining at the Wulishan limestone mine in Anhui Province, China, increased the area of direct vegetation destruction by 9.072 × 105 m2, and the indirect impact area on vegetation was 7.371 × 105 m2. The carbon emission cost of vegetation destruction in the direct destruction area was about 104,000 yuan per year, and the carbon emission cost of vegetation damage in the indirect impact area was approximately 2,082.53 yuan per year. This research provides a scientific foundation for ecological environmental protection, regulations, green mining, and cost management for mining enterprises, promoting the harmonious progress of both the economy and environmental protection.

From frequency to intensity – A new index for annual large-scale cropping intensity mapping

Continuous population growth, global warming, extreme weather and local wars pose considerable challenges to agricultural production. Excessive agricultural intensification may lead to serious environmental problems such as water resources depletion and non-point source pollution. Sustainable agricultural intensification is a potential solution, and cropping intensity and its spatio-temporal patterns are of great significance to the planet hearth, human well-being and sustainable development of agriculture. However, traditional methods for cropping intensity mapping lack the deep analysis of crop phenology and failure to consider the spatial detail, spatial coverage and time continuity simultaneously. Existing cropping intensity indicators, such as cropping frequency or multiple crop index, have their limitations in China because of its diverse topographies and fragmented landscapes. In actual fact, few indicators for fine cropping intensity have been developed that address this problem. Time series vegetation index data contain meaningful vegetation phenology information. Information mining on these data can provide new insights for cropping intensity mapping. In this study, the concept of systematic mapping was put forward to measure cropping intensity with a fresh perspective. A new phenology-based cropping intensity index (PCII) was proposed to reach the goal of “from frequency to intensity” in cropping intensity mapping. Through thoroughly exploring the temporal spectral characteristics of croplands and considering the crop pattern as a whole, PCII set up the mapping between cropping intensity and land-cover types. The method can accurately map cropping intensity even though the study spanned a vast area, and can handle the effects of geographical differentiation caused by vegetation phenology differences. It can also precisely measure cropping intensity at the abundance level and reflect the heterogeneity within a pixel. It’s a novel method for annual fine cropping intensity mapping at large-scale and long-term.

Winter wheat yield prediction using integrated Landsat 8 and Sentinel-2 vegetation index time-series data and machine learning algorithms

Timely and accurate forecasting of winter wheat yield is important to crop management, food security, and sustainable agricultural development. Unfortunately, the process of predicting winter wheat yield using satellite time series data often fails to capture complete and critical information about the crop growth process, which can restrict the accuracy of crop yield predictions. To overcome this challenge, it is necessary to increase the frequency of monitoring crop growth dynamics, identify suitable vegetation index (VI), and determine the optimal prediction model for time-series remote sensing data. In this study, we propose proposes a novel method for predicting winter wheat yield using integrated Landsat 8 (L8) and Sentinel-2 (S2) vegetation index time-series data and machine learning algorithms. Firstly, the integrated L8 and S2 dataset was obtained through the steps of cloud masking, re-sampling, re-projection, BRDF correction, and band adjustment. Then, the optimal VI was determined based on the association between the growth characteristics of winter wheat and the time-series characteristic curves of each VI. Subsequently, we employed Bayesian optimized CatBoost (BO-CatBoost) regression model to predict winter wheat yield, and compared this method with three other data-driven methods, including least absolute shrinkage and selection operator (LASSO), support vector regression (SVM), and random forest (RF). Our results showed that the winter wheat yield prediction accuracies reached the best performance using integrated Landsat 8 and Sentinel-2 WDRVI (Wide Dynamic Range Vegetation Index) time-series data and BO-CatBoost model. The R2 values were 0.70, 0.63, and 0.68, and RMSE values were 0.62, 0.73, and 0.62 t/ha for the years 2019 to 2021, respectively. In addition, acceptable accuracy was obtained for yield prediction in 2021 based on the model trained with historical data from 2019 and 2020. Moreover, this study demonstrated the result that winter wheat yield could be predicted about 40 days earlier using the proposed method. Finally, results showed that the harmonized data improved the yield estimation accuracies by a factor of 1.52, 1.29, and 1.13 compared with single S2 dataset for years 2019–2021. Various experiments demonstrated that the proposed method could effectively estimate and predict winter wheat yield data with good accuracy and robustness. This study provides technical support for improving the accuracy of winter wheat yield prediction and has the potential to be extended to yield estimation for other crops.

Mapping Eucalyptus plantation in Guangxi, China by using knowledge-based algorithms and PALSAR-2, Sentinel-2, and Landsat images in 2020

Eucalyptus plantations promote the economic development of forestry in southern China, but many studies have reported their negative environmental impacts, such as high water resource usage of certain species of Eucalyptus plants and losses in biodiversity. To date, annual maps of Eucalyptus plantations at large scales with high spatial resolutions are not yet available. Here, we investigated the spectral properties of Eucalyptus plantations and developed a knowledge-based Eucalyptus plantation mapping algorithm. We produced annual maps of Eucalyptus plantation at 10-m spatial resolution in the Guangxi Zhuang Autonomous Region (Guangxi), China, using our proposed algorithm and images of ALOS PALSAR-2, Sentinel-2, and Landsat (ETM+/OLI) in a single year. First, we generated annual evergreen forest maps using PALSAR-2 and Landsat/Sentinel-2-based vegetation index time series data. Second, we distinguished Eucalyptus plantations from the evergreen forest layer using the unique biophysical features of Eucalyptus plantations, the Sentinel-2 red edge bands, and Landsat/Sentinel-2-based enhanced vegetation index (EVI). Our resultant 2020 Eucalyptus plantation map had high producer's, user's, and overall accuracies of 0.85, 0.89, and 0.96, respectively. There were 3.10 × 106 ha of Eucalyptus plantation in Guangxi in 2020. Among the 14 administrative units, Wuzhou City had the largest Eucalyptus plantation area in Guangxi, followed by Nanning, Baise, and Chongzuo cities. We demonstrated the potential of knowledge-based mapping approaches for identifying evergreen forest and Eucalyptus plantations in complex and fragmented landscapes where cloud cover is frequent. Our 10-m Eucalyptus plantation map is the most current dataset available and can be used to assist the sustainable production of Eucalyptus, ecological assessments, and conservation.

Mapping fine-spatial-resolution vegetation spring phenology from individual Landsat images using a convolutional neural network

ABSTRACT Monitoring and mapping vegetation dynamics using remote sensing data are essential for our understanding of land surface processes. Most current satellite-based methods process vegetation index time-series data from a series of images to retrieve key points that correspond to vegetation phenophases. As deep learning approaches have been found to be powerful in processing individual images, we tested the applicability of convolutional neural network (CNN) in mapping vegetation growth days (VGD) and the start of growing season (SOS) from each Landsat image at fine-spatial-resolution. To provide references for both model training and testing, we applied the Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model (ESTARFM) to fuse image pairs of Landsat 8 Operational Land Imager (OLI) and Moderate-resolution Imaging Spectroradiometer (MODIS) for four Landsat tiles in China. We then applied a first derivative method to retrieve VGD for the fused satellite data at fine-spatial-resolution. The CNN model was trained using each fused individual image as the model inputs and derived VGD as the targets. The trained model was further used to map VGD from individual Landsat image. The result could match the reference map well as indicated by the evaluation metrics. In terms of VGD, the method achieved a coefficient of determination of 0.85 and a root mean squared error of 8.17 days. In terms of SOS, the method achieved a coefficient of determination of 0.75 and a root mean squared error of 4.09 days. Compared with existing methods that require time series of satellite data spanning the entire growth cycles to retrieve phenological metrics, this study provides an alternative method to map VGD as well as SOS using individual Landsat image. Our study highlights the power of deep learning models in extracting phenological features from individual remote sensing images. Researchers can use our methods to predict near real-time VGD and SOS in the future.

Detecting crop phenology from vegetation index time-series data by improved shape model fitting in each phenological stage

Crop phenology provides important information for crop growth management and yield estimations. The popular shape model fitting (SMF) method detects crop phenology from vegetation index (VI) time-series data, but it has two limitations. First, SMF assumes the same “relative position” of phenological stages for the pixels of the same crop type. This assumption is valid only if all target pixels, relative to the shape model, display a synchronized increase (or decrease) in length between any two phenological stages, which is uncommon in practice. Second, the variance in the resulting phenology estimates for a particular phenological stage is related to the stage itself; this makes it challenging to simulate spatial and temporal variations in crop phenology using SMF. Here, we address both limitations by developing the shape model fitting by the Separate phenological stage method (“SMF-S"). SMF-S uses a modified fitting function and an iterative procedure to match the shape model with the VI time series for each phenological stage in an adaptive local window. Comparisons between SMF-S and SMF in simulation experiments show the superior performance of SMF-S in different scenarios, regardless of noise. Comparisons involving winter wheat field observations from the North China Plain showed that the RMSE values averaged over nine phenological stages were smaller for SMF-S (RMSE = 9.5 d) than for SMF (RMSE = 13.4 d) and one variant of SMF (the shape model with accumulated growing degree days (SM-AGDD); RMSE=33.6 d). Moreover, SMF-S better described the spatial variations (i.e., variance) in the results and captured the temporal shifts in multiple phenological stages. In the derived regional phenology maps of winter wheat on the North China Plain, SMF-S generated more reasonable spatial patterns, whereas SMF underestimated (overestimated) the variance in the early (late) phenological stages. We expect that the improved crop phenology estimates obtained with SMF-S could benefit various agricultural activities.

Functional Analysis for Habitat Mapping in a Special Area of Conservation Using Sentinel-2 Time-Series Data

The mapping and monitoring of natural and semi-natural habitats are crucial activities and are regulated by European policies and regulations, such as the 92/43/EEC. In the Mediterranean area, which is characterized by high vegetational and environmental diversity, the mapping and monitoring of habitats are particularly difficult and often exclusively based on in situ observations. In this scenario, it is necessary to automate the generation of updated maps to support the decisions of policy makers. At present, the availability of high spatiotemporal resolution data provides new possibilities for improving the mapping and monitoring of habitats. In this work, we present a methodology that, starting from remotely sensed time-series data, generates habitat maps using supervised classification supported by Functional Data Analysis. We constructed the methodology using Sentinel-2 data in the Mediterranean Special Area of Conservation “Gola di Frasassi” (Code: IT5320003). In particular, the training set uses 308 field plots with 11 target classes (five forests, two shrubs, one grassland, one mosaic, one extensive crop, and one urban land). Starting from vegetation index time-series data, Functional Principal Component Analysis was applied to derive FPCA scores and components. In particular, in the classification stage, the FPCA scores are considered as features. The obtained results out-performed a previous map derived from photo-interpretation by domain experts. We obtained an overall accuracy of 85.58% using vegetation index time-series, topography, and lithology data. The main advantages of the proposed approach are the capability to efficiently compress high dimensional data (dense remote-sensing time series) providing results in a compact way (e.g., FPCA scores and mean seasonal time profiles) that: (i) facilitate the link between remote sensing with habitat mapping and monitoring and their ecological interpretation and (ii) could be complementary to species-based approaches in plant community ecology and phytosociology.

Augmentation of Vegetation Index Curves Considering the Crop-Specific Phenological Characteristics

The state-of-the-art crop phenological classifiers use vegetation index (VI) time-series data and deep learning (DL) techniques. However, the scarcity of training samples limits the performance of these approaches. Unlike the conventional augmentation techniques, the data augmentation of VI curves should preserve the crop-specific phenological events. The DL-based augmentation approaches do not give good results when the training samples are limited. Also, the conventional approaches such as translation, rotation, scaling, and wrapping do not preserve the characteristic features of the index curves, thereby making them inappropriate for the VI-curve-based augmentations. This article proposes a non-DL-based data augmentation strategy that requires only a minimal number of actual training samples. In the proposed approach, the periodic phenological events and the underlying trend for each crop class are modeled to improve the augmentation. The trends of different crop classes are estimated by jointly maximizing the autocorrelation and variance, while the optimal subsequences are generalized as the phenological events. The proposed augmentation strategy of using Maximal overlap discrete wavelet transform for obtaining the surrogates that retain the crop-specific features and periodicities significantly improves the results. It may be noted that the proposed approach does not alter the wavelet coefficients that are characteristics of a given crop class. The experiments using time series VI data, covering 90 fields of wheat, and 60 fields of barley, confirm better accuracy of the proposed augmentation approaches as compared to the prominent approaches.

Ecological Effect of Ecological Engineering Projects on Low-Temperature Forest Cover in Great Khingan Mountain, China.

The evaluation of ecological restoration projects can provide support for further strengthening the efforts of ecological restoration work and implementing the strategic objectives of the ecological region. Considering the current problem of the single evaluation index, this study evaluated the implementation effect of ecological projects from different temporal and spatial dimensions. Based on the MODIS vegetation index time series data, this study first computed the Sustainable Development Goal (SDG) indicator 15.3.1 of Great Khingan Mountain (GKM) to evaluate the impact of ecological engineering on land use change and land productivity. As a common indicator, the Normalized Difference Vegetation Index (NDVI) values showed a trend of a decrease and then gradual increase after the start of the Natural Forest Protection Project (NFPP) II, which was related to the land use changes from the forest to the grassland during the implementation of the NFPP. However, land productivity maintained a steady trend because of the transition between the forest and grassland. Meanwhile, to detect changes in vegetation at a smaller scale, the LandTrendr algorithm was used to identify the magnitude of forest disturbance, the years when it occurred, and the year of restoration. After implementing the ecological project, the forests in the GKM region were only partially disturbed, and most of the forests in most areas maintained a stable trend. Our study highlighted the varying effectiveness of different indexes for NFPP and evaluated the ecological impact of ecological projects from multiple perspectives.

Improved Accuracy of Phenological Detection in Rice Breeding by Using Ensemble Models of Machine Learning Based on UAV-RGB Imagery

Accurate and timely detection of phenology at plot scale in rice breeding trails is crucial for understanding the heterogeneity of varieties and guiding field management. Traditionally, remote sensing studies of phenology detection have heavily relied on the time-series vegetation index (VI) data. However, the methodology based on time-series VI data was often limited by the temporal resolution. In this study, three types of ensemble models including hard voting (majority voting), soft voting (weighted majority voting) and model stacking, were proposed to identify the principal phenological stages of rice based on unmanned aerial vehicle (UAV) RGB imagery. These ensemble models combined RGB-VIs, color space (e.g., RGB and HSV) and textures derived from UAV-RGB imagery, and five machine learning algorithms (random forest; k-nearest neighbors; Gaussian naïve Bayes; support vector machine and logistic regression) as base models to estimate phenological stages in rice breeding. The phenological estimation models were trained on the dataset of late-maturity cultivars and tested independently on the dataset of early-medium-maturity cultivars. The results indicated that all ensemble models outperform individual machine learning models in all datasets. The soft voting strategy provided the best performance for identifying phenology with the overall accuracy of 90% and 93%, and the mean F1-scores of 0.79 and 0.81, respectively, in calibration and validation datasets, which meant that the overall accuracy and mean F1-scores improved by 5% and 7%, respectively, in comparison with those of the best individual model (GNB), tested in this study. Therefore, the ensemble models demonstrated great potential in improving the accuracy of phenology detection in rice breeding.

Cropland Extraction Using Water Mask and Phenological Features based on Landsat-8 Imagery

Accurate cropland information is crucial for food security assessment. At present, Landsat-8 remote sensing imagery and Google Earth Engine (GEE) cloud platform provide favorable conditions for mapping cropland. Despite the improved mapping conditions, the existence of mixed pixels is still one of the main difficulties in the cropland extraction. Especially in paddy fields, the influence of water bodies on farmland extraction should not be ignored. Also, the lack of time series data with temporal-spatial resolution impedes the accurate cultivated land mapping. Aiming at these problems, we proposed an effective method to extract cropland precisely based on the Landsat-8 imagery and GEE platform. First, we created cloud-free images and masked out water based on Landsat-8 images. Second, we constructed the normalized difference vegetation index time series data with 30-m resolution from Landsat-8 and MODIS images to obtain accurate phenological information. Third, other cropland features such as texture features, tasseled cap transformation features, terrain features and spectral features were acquired. Finally, we applied the Random Forest algorithm and obtained the cropland for 2015 in Qinzhou. The results show that terrain characteristics and phenological characteristics have the greatest influence on cropland extraction. Based on water mask, the overall accuracy of cropland improved by 6.9%. Overall, the product has high accuracy with the mean overall accuracy (OA) of 97.9%. Compared with MCD12Q1 and GFSAD30 cropland products, it is 6.2% and 24.3% higher in OA respectively. In brief, our method is helpful for the cropland extraction.

Estimation of Rape-Cultivated Area Based on Decision Tree and Mixed Pixel Decomposition

The timely and accurate information of rape-cultivated area in Jianghan Plain is of great significance for crop yield estimation and food security in China. MODIS-EVI time series images were used to estimate rape-cultivated area of Jianghan Plain in 2015–2017 based on decision tree and mixed pixel decomposition methods. By analyzing the features of EVI time series curves of different objects, rape distribution was preliminarily extracted by establishing the decision tree model; however, due to the problem of mixed pixels, there are heterogeneous objects in a single pixel, which will affect the extraction accuracy of crop area by decision tree method. To solve this problem, according to EVI spectral curve of the endmembers, the linear spectral mixture model was adopted to further improve the precision of crop extraction, and rape-cultivated area was extracted accurately by using computed rape endmember abundance. The results showed that the average estimation accuracy of rape area was 96.6% with city-level R2 (coefficient of determination) greater than 0.93. It demonstrated that the vegetation index time series data with medium resolution can accurately extract the rape-cultivated area, which can provide reference for rape monitoring over a large scale.

Ability of Remote Sensing Systems to Detect Bark Beetle Spots in the Southeastern US

Research Highlights: Sentinel-2 Normalized Difference Vegetation Index (NDVI) products show greater potential to detect indications of disturbance by bark beetles in the southeastern US than Moderate Resolution Imaging Spectroradiometer (MODIS), as the high spatiotemporal heterogeneity of the southeastern forest land prevents its deployment at the current resolution. Background and Objectives: Remote sensing technologies have been an essential tool to detect forest disturbances caused by insect pests through spectral trait variation. In the US, coordinated efforts such as ForWarn, led by the US Forest Service and based on MODIS satellite data, are used to monitor biotic and abiotic disturbances. Because of the particular characteristics of the southeastern US landscape, including forest fragmentation and rapid forest turnover due to management, detection and visualization of small bark beetle spots using remote sensing technology developed for more homogeneous landscapes has been challenging. Here, we assess the ability of MODIS and Sentinel-2 time-series vegetation index data products to detect bark beetle spots in the Florida Panhandle. Materials and Methods: We compared ForWarn’s detection ability (lower resolution images) with that of Sentinel-2 (higher resolution images) using bark beetle spots confirmed by aerial surveys and ground checks by the Florida Forest Service. Results: MODIS and Sentinel-2 can detect damage produced by bark beetles in the southeastern US, but MODIS detection via NDVI change exhibits a high degree of false negatives (30%). Sentinel-2 NDVI products show greater potential for identifying indications of disturbance by bark beetles than MODIS change maps, with Sentinel-2 capturing negative changes in NDVI for all spots. Conclusions: Our research shows that for practical bark beetle detection via remote sensing, higher spatial and temporal resolution will be needed.

Sensitivity of six typical spatiotemporal fusion methods to different influential factors: A comparative study for a normalized difference vegetation index time series reconstruction

Dozens of spatiotemporal fusion methods have been developed to reconstruct vegetation index time-series data with both high spatial resolution and frequent coverage for monitoring land surface dynamics. Although several studies comparing the different fusion methods have been conducted, selecting the suitable fusion methods is still challenging, as inevitable influential factors tend to be neglected. To address this problem, this study compared six typical spatiotemporal fusion methods, including the Unmixing-Based Data Fusion (UBDF), Linear Mixing Growth Model (LMGM), Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM), Fit-FC (regression model Fitting, spatial Filtering and residual Compensation), One Pair Dictionary-Learning method (OPDL), and Flexible Spatiotemporal DAta Fusion (FSDAF), based on simulation experiments and theoretical analysis considering three influential factors between sensors: geometric misregistration, radiometric inconsistency, and spatial resolution ratio. The results indicate that Fit-FC achieved the best performance with the strongest tolerance to geometric misregistration when radiometric inconsistency was negligible; thus, it is the first recommended algorithm for blending normalized difference vegetation index (NDVI) imagery. Instead, the FSDAF could generate the best results if radiometric inconsistency was non-negligible. These findings could help users determine the method that is appropriate for different remote sensing datasets, and provide guidelines for developers in the future development of novel methods.

Vegetation Carbon Sequestration Mapping in Herbaceous Wetlands by Using a MODIS EVI Time-Series Data Set: A Case in Poyang Lake Wetland, China

The carbon sequestration capacity of wetland vegetation determines carbon stocks and changes in wetlands. However, modeling vegetation carbon sequestration of herbaceous wetlands is still problematic due to complex hydroecological processes and rapidly changing biomass carbon stocks. Theoretically, a vegetation index (VI) time series can retrieve the dynamic of biomass carbon stocks and could be used to calculate the cumulative composite of biomass carbon stocks during a given interval, i.e., vegetation carbon sequestration. Hence, we explored the potential for mapping vegetation carbon sequestration in herbaceous wetlands in this study by using a combination of remotely sensed VI time series and field observation data. This method was exemplarily applied for Poyang Lake wetland in 2016 by using a 16-day Moderate Resolution Imaging Spectroradiometer (MODIS) enhanced vegetation index (EVI) time series. Results show that the vegetation carbon sequestration in this area was in the range of 193–1221 g C m−2 year−1 with a mean of 401 g C m−2 year−1 and a standard deviation of 172 g C m−2 year−1 in 2016. The approach has wider spatial applicability in wetlands than the currently used global map of vegetation production (MOD17A3) because our carbon estimation in areas depicted by ‘no data’ in the MOD17A3 product is considerable, which accounts for 91.2–91.5% of the total vegetation carbon sequestration of the wetland. Thus, we determined that VI time series data shows great potential for estimating vegetation carbon sequestration in herbaceous wetlands, especially with the continuously improving quality and frequency of satellite VI images.

Seasonal Comparisons of Himawari-8 AHI and MODIS Vegetation Indices over Latitudinal Australian Grassland Sites

The Advanced Himawari Imager (AHI) on board the Himawari-8 geostationary (GEO) satellite offers comparable spectral and spatial resolutions as low earth orbiting (LEO) sensors such as the Moderate Resolution Imaging Spectroradiometer (MODIS) and Visible Infrared Imaging Radiometer Suite (VIIRS) sensors, but with hypertemporal image acquisition capability. This raises the possibility of improved monitoring of highly dynamic ecosystems, such as grasslands, including fine-scale phenology retrievals from vegetation index (VI) time series. However, identifying and understanding how GEO VI temporal profiles would be different from traditional LEO VIs need to be evaluated, especially with the new generation of geostationary satellites, with unfamiliar observation geometries not experienced with MODIS, VIIRS, or Advanced Very High Resolution Radiometer (AVHRR) VI time series data. The objectives of this study were to investigate the variations in AHI reflectances and normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), and two-band EVI (EVI2) in relation to diurnal phase angle variations, and to compare AHI VI seasonal datasets with MODIS VIs (standard and sun and view angle-adjusted VIs) over a functional range of dry grassland sites in eastern Australia. Strong NDVI diurnal variations and negative NDVI hotspot effects were found due to differential red and NIR band sensitivities to diurnal phase angle changes. In contrast, EVI and EVI2 were nearly insensitive to diurnal phase angle variations and displayed nearly flat diurnal profiles without noticeable hotspot influences. At seasonal time scales, AHI NDVI values were consistently lower than MODIS NDVI values, while AHI EVI and EVI2 values were significantly higher than MODIS EVI and EVI2 values, respectively. We attributed the cross-sensor differences in VI patterns to the year-round smaller phase angles and backscatter observations from AHI, in which the sunlit canopies induced a positive EVI/ EVI2 response and negative NDVI response. BRDF adjustments of MODIS VIs to solar noon and to the oblique view zenith angle of AHI resulted in strong cross-sensor convergence of VI values (R2 > 0.94, mean absolute difference <0.02). These results highlight the importance of accounting for cross-sensor observation geometries for generating compatible AHI and MODIS annual VI time series. The strong agreement found in this study shows promise in cross-sensor applications and suggests that a denser time series can be formed through combined GEO and LEO measurement synergies.

Response Patterns of Vegetation Phenology along Urban-Rural Gradients in Urban Areas of Different Sizes

On the basis of MODIS Enhanced Vegetation Index time series data and multisource data, such as nighttime light data and China City Statistical Yearbook data, we investigated the differences in vegetation phenology along urban-rural gradients in urban areas of different sizes between coastal and inland cities in Liaoning Province, China. The results showed that the following: (1) the iterative extraction of urban built-up areas using the threshold method based on nighttime light data combined with the definition of urban built-up areas had high accuracy. (2) Additionally, we found that the start of the growing season (SOS) in Liaoning Province occurred between day 100 and day 180, while the end of the growing season (EOS) occurred between days 260 and 330. The difference in the SOS between coastal cities (i.e., Dalian, Yingkou, Panjin, Jinzhou, Huludao, and Dandong) and inland cities (i.e., Chaoyang, Fuxin, Tieling, Shenyang, Fushun, Liaoyang, Benxi, and Anshan) was 1.70 days. However, the difference in the EOS was more significant, i.e., the EOS in coastal cities occurred 4.47 days later than that in the inland cities. (3) In urban areas of different sizes, the ∆SOS and ∆EOS of inland cities had negative correlations with urban size. Specifically, when the urban size increased 10-fold, the ∆SOS and ∆EOS advanced by 10.03 and 5.71 days, respectively. In contrast, the ∆SOS and ∆EOS of coastal cities had positive and negative correlations with the urban size, respectively. Specifically, when urban size increased 10-fold, ∆SOS was delayed by 11.29 days while EOS was advanced by 8.83 days.

A near real-time deep learning approach for detecting rice phenology based on UAV images

Near real-time crop phenology detection is essential for crop management, estimation of harvest time and yield estimation. Previous approaches to crop phenology detection have relied on time-series (multi-temporal) vegetation index (VI) data, and have included threshold-based, phenometrics-based and shape-model-fitting-based (SMF) methods. However, the performance of these methods depends on the duration and temporal resolution of the time-series data. In this study, we propose a new approach which identifies the principal growth stages of rice (Oryza sativa L.) directly from RGB images. Only a mono-temporal unmanned aerial vehicle (UAV) imagery was required for a large-area phenology detection via the well-trained network. An efficient convolutional neural network (CNN) architecture was designed to estimate rice phenology. The CNN incorporated spatial pyramid pooling (SPP), transfer learning and an auxiliary branch with external data. A total of 82 plots across a 160-hectare rice cultivation area of Southern China were selected to evaluate the proposed network. CNN predictions were ground truthed using rice phenology measurements taken from each plot throughout the growing season. Aerial data were collected using a fixed-wing UAV equipped with multispectral and RGB cameras. The performance of traditional SMF methods deteriorated when time-series VI data were of short duration. In contrast, the phenological stage estimated by the proposed network showed good agreement with ground observations, with a top-1 accuracy rate of 83.9% and mean absolute error (MAE) of 0.18. The spatial distribution of harvest dates for 627 plots in the study area were computed from the phenological stage estimates. The estimates matched well with the observed harvest dates. The results demonstrated the excellent performance of the proposed deep learning approach in near real-time phenology detection and harvest time estimation.

Improved Characterisation of Vegetation and Land Surface Seasonal Dynamics in Central Japan with Himawari-8 Hypertemporal Data

Spectral vegetation index time series data, such as the normalized difference vegetation index (NDVI), from moderate resolution, polar-orbiting satellite sensors have widely been used for analysis of vegetation seasonal dynamics from regional to global scales. The utility of these datasets is often limited as frequent/persistent cloud occurrences reduce their effective temporal resolution. In this study, we evaluated improvements in capturing vegetation seasonal changes with 10-min resolution NDVI data derived from Advanced Himawari Imager (AHI), one of new-generation geostationary satellite sensors. Our analysis was focused on continuous monitoring sites, representing three major ecosystems in Central Japan, where in situ time-lapse digital images documenting sky and surface vegetation conditions were available. The very large number of observations available with AHI resulted in improved NDVI temporal signatures that were remarkably similar to those acquired with in situ spectrometers and captured seasonal changes in vegetation and snow cover conditions in finer detail with more certainty than those obtained from Visible Infrared Imaging Radiometer Suite (VIIRS), one of the latest polar-orbiting satellite sensors. With the ability to capture in situ-quality NDVI temporal signatures, AHI “hypertemporal” data have the potential to improve spring and autumn phenology characterisation as well as the classification of vegetation formations.