Crop classification methods and influencing factors of reusing historical samples based on 2D-CNN

Validation of synthetic daily Landsat NDVI time series data generated by the improved spatial and temporal data fusion approach

Normalized Difference Vegetation Index (NDVI) time series is one of the most important instruments in precision agriculture. Forecasting of this index in precision agriculture allows us to define problems related to growth rates of agricultural crops in time. This Doctoral Thesis is devoted to the analysis and forecasting of nonlinear and nonstationary NDVI index time series with the use of data pre-processing, signal processing, linear algebra and machine learning methods. The aim of Doctoral Thesis is to develop a forecasting system of normalized difference vegetation index time series based on signal decomposition and sub-signal approximation approach, specialized data processing methods and machine learning. In the Thesis, NDVI time series forecasting system (NDVI FS) is proposed that makes short-term time series value forecasting for the next period using historical observations. In the developed system framework, an innovative approach to using variational mode decomposition method in NDVI time series forecasting task is realized. There is also offered an innovative approach to trained forecasting model transfer, which allows using this model to forecast other NDVI time series obtained from neighbourhood pixels.

MODerate-resolution Imaging Spectroradiometer (MODIS) vegetation index (VI) time-series data are among the most widely utilized remote sensing datasets. To improve the quality of MODIS VI time-series data, most prior methods have focused on correcting negatively biased VI noise by approaching the upper envelope of the VI time series. Such treatment, however, may cause overcorrections on some true local low VI values, resulting in inaccurate simulations of vegetation phenological characteristics. In addition, another challenge in reconstructing MODIS VI time series is to fill temporally continuous gaps. The earlier spatiotemporal Savitzky-Golay (STSG) method tackled this problem by utilizing multiyear VI data, but its performance heavily relies on the consistency of data across different years. In this study, we proposed an improved STSG (iSTSG) method. The new method accounts for the autocorrelation within the VI time series and fills missing values in the VI time series by leveraging spatiotemporal VI data from the current year alone. Furthermore, iSTSG incorporates an indicator to quantify potential overcorrections in the VI time series, aiming to more accurately simulate phenological characteristics. The experiments to reconstruct MODIS normalized difference VI (NDVI) time-series product (MOD13A2) at four typical sites (a million square kilometers for each site) suggest two clear advantages in iSTSG over the iterative SG (called Chen-SG) and STSG methods. First, iSTSG more accurately reconstructs the annual NDVI time series, exhibiting the smallest mean absolute differences (MADs) between the smoothed and the simulated reference NDVI time series (0.012, 0.018, and 0.020 for iSTSG, STSG, and Chen-SG, respectively). Second, iSTSG more effectively simulates phenological characteristics in the NDVI time series, including the onset dates for vegetation greenup and dormancy, as well as the crop harvest period. The advantages of iSTSG were also demonstrated when applied to the successor of MODIS, Visible Infrared Imaging Radiometer Suite (VIIRS) VI time-series product (VNP13A1). iSTSG can be implemented on the Google Earth Engine (GEE), offering significant benefits for various applications, particularly in crop mapping and vegetation/crop phenology studies.

Timely and accurate crop type distribution maps are an important inputs for crop yield estimation and production forecasting as multi-temporal images can observe phenological differences among crops. Therefore, time series remote sensing data are essential for crop type mapping, and image composition has commonly been used to improve the quality of the image time series. However, the optimal composition period is unclear as long composition periods (such as compositions lasting half a year) are less informative and short composition periods lead to information redundancy and missing pixels. In this study, we initially acquired daily 30 m Normalized Difference Vegetation Index (NDVI) time series by fusing MODIS, Landsat, Gaofen and Huanjing (HJ) NDVI, and then composited the NDVI time series using four strategies (daily, 8-day, 16-day, and 32-day). We used Random Forest to identify crop types and evaluated the classification performances of the NDVI time series generated from four composition strategies in two studies regions from Xinjiang, China. Results indicated that crop classification performance improved as crop separabilities and classification accuracies increased, and classification uncertainties dropped in the green-up stage of the crops. When using daily NDVI time series, overall accuracies saturated at 113-day and 116-day in Bole and Luntai, and the saturated overall accuracies (OAs) were 86.13% and 91.89%, respectively. Cotton could be identified 40∼60 days and 35∼45 days earlier than the harvest in Bole and Luntai when using daily, 8-day and 16-day composition NDVI time series since both producer’s accuracies (PAs) and user’s accuracies (UAs) were higher than 85%. Among the four compositions, the daily NDVI time series generated the highest classification accuracies. Although the 8-day, 16-day and 32-day compositions had similar saturated overall accuracies (around 85% in Bole and 83% in Luntai), the 8-day and 16-day compositions achieved these accuracies around 155-day in Bole and 133-day in Luntai, which were earlier than the 32-day composition (170-day in both Bole and Luntai). Therefore, when the daily NDVI time series cannot be acquired, the 16-day composition is recommended in this study.

Predicting seismic interaction effect between soil and structure group using convolutional neural network

Improved and up-to-date land use/land cover (LULC) data sets are needed over intensively land use/cover change area in the Amur River Basin (ARB) to support science and policy applications focused on understanding of the role and response of the LULC to environmental change issues. The main goal of this study was to map LULC in the Amur River Basin using MODIS 250 m Normalized Difference Vegetation Index (NDVI) and Land Surface Water Index (LSWI) time series data in 2001 and 2007. A combination of unsupervised ISODATA and hierarchical decision tree classification were performed on 12-month time-series of MODIS NDVI data over the study region. The MODIS land cover result of Northeast China was evaluated using existing land use/cover data, and the rest part was evaluated by LULC information derived from LANDSAT-TM. MODIS 250m NDVI, LSWI and reflectance datasets were found to have sufficient spatial, spectral, and temporal resolutions to detect unique multitemporal signatures of the major land cover types over the region. The overall classification accuracy was 0.81 and the kappa coefficient is 0.64. In conclusion, this method has been used successively for LULC change monitoring in the year 2001 and 2007. The result indicate that MODIS 250 NDVI time series data can derive relatively accurate LULC information for hydrological and climate modeling.

HJ-1A/B is the first small satellite constellation built by China for environmental and disaster monitoring and forecasting. The satellite group has a 2-day repetition cycle and 30 m spatial resolution (charge coupled device camera). Thus, HJ-1A/B can provide hyper-temporal normalized difference vegetation index (NDVI) time series with a medium–high spatial resolution. However, the quality of the HJ NDVI time series can be abnormally low due to a number of factors, such as cloud cover, continuous fog, and haze. In the rainy season or in areas with serious atmospheric pollution, low-quality series often appear in succession, which is referred to as an abnormal segment. Neither the composition method nor quality flags satisfactorily solve this problem; therefore, a large amount of noise and long periods of abnormally low values often remain in HJ NDVI time series. This article presents a method to reconstruct the abnormal segments in HJ NDVI time series with the assistance of Moderate Resolution Imaging Spectroradiometer (MODIS) NDVI time series. The cointegration test was adopted to decide whether MODIS can be used for the reconstruction of NDVI time series for the corresponding HJ image pixels. Statistical quality control methods were used for singling out the abnormal segments in the HJ NDVI time series and establishing an error correction model that combines MODIS and HJ NDVI time series to perform the reconstruction. The study area is located in Jiangsu Province, China. Four-year (2009–2012) HJ multispectral images that cover the study area were used. The results show that abnormal segments in the HJ NDVI time series can be corrected using the proposed method. In a particular year, this method can decrease the root mean square error between the HJ NDVI time series and the reference sequence by 52.5%.

Normalized difference vegetation index (NDVI) time series data, derived from optical images, play a crucial role for crop mapping and growth monitoring. Nevertheless, optical images frequently exhibit spatial and temporal discontinuities due to cloudy and rainy weather conditions. Existing algorithms for reconstructing NDVI time series using multi-source remote sensing data still face several challenges. In this study, we proposed a novel method, an enhanced gap-filling and Whittaker smoothing (EGF-WS), to reconstruct NDVI time series (EGF-NDVI) using Google Earth Engine. In EGF-WS, NDVI calculated from MODIS, Landsat-8, and Sentinel-2 satellites were combined to generate high-resolution and continuous NDVI time series data. The MODIS NDVI was employed as reference data to fill missing pixels in the Sentinel–Landsat NDVI (SL-NDVI) using the gap-filling method. Subsequently, the filled NDVI was smoothed using a Whittaker smoothing filter to reduce residual noise in the SL-NDVI time series. With reference to the all-round performance assessment (APA) metrics, the performance of EGF-WS was compared with the conventional gap-filling and Savitzky–Golay filter approach (GF-SG) in Fusui County of Guangxi Zhuang Autonomous Region. The experimental results have demonstrated that the EGF-WS can capture more accurate spatial details compared with GF-SG. Moreover, EGF-NDVI of Fusui County exhibited a low root mean square error (RMSE) and a high coefficient of determination (R2). In conclusion, EGF-WS holds significant promise in providing NDVI time series images with a spatial resolution of 10 m and a temporal resolution of 8 days, thereby benefiting crop mapping, land use change monitoring, and various ecosystems, among other applications.

High quality Normalized Difference Vegetation Index (NDVI) time series data are important for regional and global ecological applications. However, missing observations in cloudy and shaded areas in current high-resolution satellite data severely limit their use. Traditional temporal filtering reconstruction methods are easily affected by frequent fluctuations, large data gaps and irregular noise in the observations. Meanwhile, satellite images often lack complete and accurate real data as training sets, which limits deep learning-based reconstruction methods. To address these problems, this paper proposes a reconstruction method with Landsat-8 NDVI time-series images based on bidirectional long-short memory networks (Bi-LSTM). The core of the method is a sequence-to-sequence time series reconstruction model using Bi-LSTM as an encoder, decoder, and additionally adding a mask layer to locate irregular gaps in the time series data. To train the model, a multi-scale continuous NDVI time series dataset was constructed using near-surface observations and satellite fused images. Simulation experiments were conducted in 48 areas covering PhenoCam sites in North America. The coefficient of determination (R2) between the simulation results of this method and the real observations reached 0.7024 with an absolute bias of -0.0080, which outperformed the filtering method both in predicting missing cloud-covered areas and in extracting key nodes for vegetation growth. This method is pre-trained on a large-scale ground sample dataset before fine-tuning the parameters for the scenarios, which can be easily and quickly applied to generate long-term continuous NDVI datasets at a large spatial scale. In addition, the analysis of the phenology monitoring in the experimental areas shows that the results extracted from the reconstructed results of this method is very close to the PhenoCam-measured values, which is significantly improved compared to the extraction results using original Landsat data only, and at the same time shows more spatial detail compared to the MODIS climate products. This indicates that the 30m gap-free NDVI time series is sufficient for analysis and understanding of climate dynamics in heterogeneous areas with multiple vegetation growth patterns, mixing plant components and microclimates variation, which are difficult to be identified by the coarse spatial resolution phenology product.

International audience

Accurate mapping of the spatio-temporal dynamic information of conversion from -ropland into forest/grassland is necessary for studying hydro-ecological effects of implementation of Grain to Green Project (GTGP) on the Loess Plateau, China. Currently, the accuracy of change detection of farmland and forest/grassland at 30-m scale in this area is seriously affected by insufficient temporal information from observations and irregular fluctuations in vegetation greenness caused by precipitation and human activities. In this study, an innovative method for continuous change detection of cropland and forest/grassland using all available Landsat time-series data. The period with vegetation coverage is firstly identified using normalized difference vegetation index (NDVI) time series. The intra-annual NDVI time series is then developed at a 1-day resolution based on linear interpolation and S-G filtering using all available NDVI data during the period when vegetation types are stable. Vegetation type change is initially detected by comparing the NDVI of intra-annual composites and the newly observed NDVI. Finally, the time of change and classification for vegetation types are determined using decision tree rules developed using a combination of inter-annual and intra-annual NDVI temporal metrics. Validation results showed that the change detection was accurate, with an overall accuracy of 88.9% ± 1.0%, and a kappa coefficient of 0.86, and the time of change was successfully retrieved, with 85.2% of the change pixels attributed to within a 2-year deviation.

Crop type mapping provides essential information to control and make decisions related to agricultural practices and their regulations. To map crop types accurately, it is important to capture their phenological stages and fine spatial details, especially in a temporally and spatially heterogeneous landscape. The data availability of new generation multispectral sensors of Landsat-8 (L8) and Sentinel-2 (S2) satellites offers unprecedented options for such applications. Given this, our study aims to display how the synergistic use of these optical sensors can efficiently support crop type mapping research while integrating an object-based image analysis (OBIA). Through the applied methods, we used the enhanced spatial and temporal adaptive reflectance fusion model (ESTARFM) and the flexible spatiotemporal data-fusion model (FSDAF) to blend L8 and S2 data and obtain reliable normalized difference vegetation index (NDVI) datasets with fine spatial and temporal resolution. Then the crop phenological information was extracted using a Savitzky–Golay filter and fused NDVI time series. Finally, a model combining phenological metrics and fused reconstructed NDVI as classification features was developed using a random forest (RF) classifier/OBIA approach. The results show that the FSDAF method creates more accurate fused NDVI and keeps more spatial details than ESTARFM. The FSDAF model was then used to create fused, high-resolution time-series products that were able to extract crop phenology in single-crop fields while providing a very detailed pattern relative to that from individual sensor time-series data. Moreover, combined L8 and S2 data by FSDAF produced highly significant overall classification accuracies (90.03% for pixel-based RF to 93.12% OBIA RF), outperforming individual sensor use (82.57% for L8-only; 88.45% for S2-only). Our proposed workflow highlights the advantage of spatiotemporal fusing and OBIA environment in spatiotemporally heterogeneous areas and fragmented landscapes, which represents a promising step toward generating fast, accurate, and ready-to-use agricultural data products.

Abstract: Increasing regional vegetation activity has been frequently found especially in middle and high latitude and alpine areas, but the effects of which on regional hydrology is still highly uncertain. The Upstream Catchment of Minjiang River is a large mountainous catchment covering 22,919 km2 with a diverse vegetation distribution pattern, including alpine group (A), subalpine group (SA), and temperate and subtropical group (T/ST). The Seasonal Mann‐Kendall test, a nonparametric trend test method, detected consistent upward trends in all groups in monthly accumulated growing degree days (AGDDM) time series from 1982 to 2003, but no significant trend in mean monthly precipitation (MMP) time series in any group. The alpine group had a significant (p = 0.024) upward trend in monthly Normalized Difference of Vegetation Index (NDVI) time series from 1982 to 2003, in contrast, the SA and T/ST groups had decreasing (although not significant) trends. AGDDM plays more important role than MMP in affecting NDVI change in alpine areas, indicating temperature was the main climatic driver. In contrast, water was the main driver for the T/ST group, as indicated by the significant correlation between NDVI and MMP and a weak correlation with AGDDM. Correlation coefficients of NDVI and river flow varied with seasons, mostly negative, especially during the growing season (April to October). A significant (p = 0.025) correlation was found only in August, indicating that an increase in peak‐NDVI decreased high flow significantly. TI‐NDVIc, which was developed in an attempt to track the vegetation change at the catchment scale, accounted for more than 40% of the evapotranspiration increase (r2 = 0.43).

Different studies on predicting future green cover changes exist with various success levels. Each one focuses on a different story about which model is more appropriate. Therefore, finding a suitable model remains a difficult task due to the remotely sensed data issues and the complexity of the vegetation cover change process. Despite the unicity of vegetation indices time series, the forecasting assessment relies basically on the commonly used forecast error measurements (e.g. Mean Square Error (MSE), Root MSE (RMSE), and the symmetric Mean Absolute Percentage Error (sMAPE), etc.) which may not reflect the real potential of the fitted forecasting model. Herein, the experimentation of forecasting vegetation indices employing two univariate time series models and new accuracy metrics is discussed. Box Jenkins models (Seasonal AutoRegressive Integrated Moving Average (ARIMA) model) and neural network model (nonlinear autoregressive (NAR) model) are applied individually and coupled (NAR-NAR and NAR-ARIMA) based on a multi-resolution analysis-wavelet transform. These models’ forecasting ability is evaluated using 16-days Moderate Resolution Imaging Spectroradiometer Normalized Difference Vegetation Index (NDVI) time series data of two different vegetation cover types of northwestern area in Tunisia. The major finding highlights firstly the importance of integration of the decomposition step to show off the hidden components of vegetation indices. The commonly used accuracy measures show that the coupled neural networks model outperforms other models. Then, interesting conclusions were drawn when phenological metrics are used as performance measures. Herein, Box Jenkins forecasting model generates a better NDVI curve shape than hybrid models despite their low RMSE measures. This is mainly due to good estimation of some phenological events, namely, the amplitude, the peak and the season’s length. Generally, Box Jenkin model excels at handling quick variations. By contrast, combined models show better phenological metrics’ estimation when the observations are complex or describe long time periods. Based on these findings, we suggest that the choice of the evaluation metric must be related to the future forecaster interest.

Croplands are commonly mapped using time series of remotely sensed images. The dynamic time warping (DTW) algorithm is an effective method for realizing this. However, DTW algorithm faces the challenge of capturing complete and accurate representative cropland time series on a national scale, especially in Asian countries where climatic and topographic conditions, cropland types, and crop growth patterns vary significantly. This study proposes an automatic cropland extraction method based on the DTW algorithm and density-based spatial clustering of applications with noise (DBSCAN), hereinafter referred to as ACE-DTW, to map croplands in Pakistan in 2015. First, 422 frames of multispectral Landsat-8 satellite images were selected from the Google Earth Engine to construct monthly normalized difference vegetation index (NDVI) time series. Next, a total of 2409 training samples of six land cover types were generated randomly and explained visually using high-resolution remotely sensed images. Then, a multi-layer DBSCAN was used to classify NDVI time series of training samples into different categories automatically based on their pairwise DTW distances, and the mean NDVI time series of each category was used as the standard time series to represent the characteristics of that category. These standard time series attempted to represent cropland information and maximally distinguished croplands from other possible interference land cover types. Finally, image pixels were classified as cropland or non-cropland based on their DTW distances to the standard time series of the six land cover types. The overall cropland extraction accuracy of ACE-DTW was 89.7%, which exceeded those of other supervised classifiers (classification and regression trees: 78.2%; support vector machines: 78.8%) and existing global cropland datasets (Finer Resolution Observation and Monitoring of Global Land Cover: 87.1%; Global Food Security Support Analysis Data: 83.1%). Further, ACE-DTW could produce relatively complete time series of variable cropland types, and thereby provide a significant advantage in mountain regions with small, fragmented croplands and plain regions with large, high-density patches of croplands.