The Classification Performance and Mechanism of Machine Learning Algorithms in Winter Wheat Mapping Using Sentinel-2 10 m Resolution Imagery

Discriminant Analysis as a Tool to Classify Grasslands Based on Near-Infrared Spectra

This study aims to classify plant communities by applying discriminant analysis based on principal components (DAPC) on near-infrared spectra (FT-NIRS) starting from fresh herbage samples. Grassland samples (n~156) belonged to (i) recent alfalfa pure crops (CAA), (ii) recent grass-legume mixtures (GLM), (iii) permanent meadows derived from old alfalfa stands that were re-colonized (PMA), and iv) permanent meadows originated from old grass-legume mixtures (PLM). Samples were scanned using FT-NIRS, and a multivariate exploration of the original spectra was performed using DAPC. The following two scenarios were proposed: (i) cross-validation, where all data were used for model training, and (ii) semi-external validation, where the group assignment was performed without samples of the training set. The first two components explained 98% of the total variability. The DAPC model resulted in an overall assignment success rate of 77%, and, from cross-validation, it emerged that it was possible to assign the CAA and PMA to their group with more than of 80% of success, which were different in botanical and chemical composition. In comparison, GLM and PLM obtained lower success of assignment (~52%). External validation suggested similarity between PLM and GLM groups (93%) and between GLM and PLM (77%). However, a dataset increase could improve group differentiation.

Extracting the winter wheat using the decision tree based on time series dual-polarization SAR feature and NDVI

Winter wheat is one of the most important crops in the world. It is great significance to obtain the planting area of winter wheat timely and accurately for formulating agricultural policies. Due to the limited resolution of single SAR data and the susceptibility of single optical data to weather conditions, it is difficult to accurately obtain the planting area of winter wheat using only SAR or optical data. To solve the problem of low accuracy of winter wheat extraction only using optical or SAR images, a decision tree classification method combining time series SAR backscattering feature and NDVI (Normalized Difference Vegetation Index) was constructed in this paper. By synergy using of SAR and optical data can compensate for their respective shortcomings. First, winter wheat was distinguished from other vegetation by NDVI at the maturity stage, and then it was extracted by SAR backscattering feature. This approach facilitates the semi-automated extraction of winter wheat. Taking Yucheng City of Shandong Province as study area, 9 Sentinel-1 images and one Sentinel-2 image were taken as the data sources, and the spatial distribution of winter wheat in 2022 was obtained. The results indicate that the overall accuracy (OA) and kappa coefficient (Kappa) of the proposed method are 96.10% and 0.94, respectively. Compared with the supervised classification of multi-temporal composite pseudocolor image and single Sentinel-2 image using Support Vector Machine (SVM) classifier, the OA are improved by 10.69% and 5.66%, respectively. Compared with using only SAR feature for decision tree classification, the producer accuracy (PA) and user accuracy (UA) for extracting the winter wheat are improved by 3.08% and 8.25%, respectively. The method proposed in this paper is rapid and accurate, and provide a new technical method for extracting winter wheat.

The rectangular tile classification model based on Sentinel integrated images enhances grassland mapping accuracy: A case study in Ordos, China

Arid zone grassland is a crucial component of terrestrial ecosystems and plays a significant role in ecosystem protection and soil erosion prevention. However, accurately mapping grassland spatial information in arid zones presents a great challenge. The accuracy of remote sensing grassland mapping in arid zones is affected by spectral variability caused by the highly diverse landscapes. In this study, we explored the potential of a rectangular tile classification model, constructed using the random forest algorithm and integrated images from Sentinel-1A (synthetic aperture radar imagery) and Sentinel-2 (optical imagery), to enhance the accuracy of grassland mapping in the semiarid to arid regions of Ordos, China. Monthly Sentinel-1A median value images were synthesised, and four MODIS vegetation index mean value curves (NDVI, MSAVI, NDWI and NDBI) were used to determine the optimal synthesis time window for Sentinel-2 images. Seven experimental groups, including 14 experimental schemes based on the rectangular tile classification model and the traditional global classification model, were designed. By applying the rectangular tile classification model and Sentinel-integrated images, we successfully identified and extracted grasslands. The results showed the integration of vegetation index features and texture features improved the accuracy of grassland mapping. The overall accuracy of the Sentinel-integrated images from EXP7-2 was 88.23%, which was higher than the accuracy of the single sensor Sentinel-1A (53.52%) in EXP2-2 and Sentinel-2 (86.53%) in EXP5-2. In all seven experimental groups, the rectangular tile classification model was found to improve overall accuracy (OA) by 1.20% to 13.99% compared to the traditional global classification model. This paper presents novel perspectives and guidance for improving the accuracy of remote sensing mapping for land cover classification in arid zones with highly diverse landscapes. The study presents a flexible and scalable model within the Google Earth Engine framework, which can be readily customized and implemented in various geographical locations and time periods.

Hierarchical Classification of Soybean in the Brazilian Savanna Based on Harmonized Landsat Sentinel Data

The Brazilian Savanna presents a complex agricultural dynamic and cloud cover issues; therefore, there is a need for new strategies for more detailed agricultural monitoring. Using a hierarchical classification system, we explored the Harmonized Landsat Sentinel-2 (HLS) dataset to detect soybean in western Bahia, Brazil. Multispectral bands (MS) and vegetation indices (VIs) from October 2021 to March 2022 were used as variables to feed Random Forest models, and the performances of the complete HLS time-series, HLSS30 (harmonized Sentinel), HLSL30 (harmonized Landsat), and Landsat 8 OLI (L8) were compared. At Level 1 (agricultural areas × native vegetation), HLS, HLSS30, and L8 produced identical models using MS + VIs, with 0.959 overall accuracies (OA) and Kappa of 0.917. At Level 2 (annual crops × perennial crops × pasturelands), HLS and L8 achieved an OA of 0.935 and Kappa > 0.89 using only VIs. At Level 3 (soybean × other annual crops), the HLS MS + VIs model achieved the best performance, with OA of 0.913 and Kappa of 0.808. Our results demonstrated the potential of the new HLS dataset for medium-resolution mapping initiatives at the crop level, which can impact decision-making processes involving large-scale soybean production and agricultural sustainability.

Prediction of Field-Scale Wheat Yield Using Machine Learning Method and Multi-Spectral UAV Data

Accurate prediction of food crop yield is of great significance for global food security and regional trade stability. Since remote sensing data collected from unmanned aerial vehicle (UAV) platforms have the features of flexibility and high resolution, these data can be used as samples to develop regional regression models for accurate prediction of crop yield at a field scale. The primary objective of this study was to construct regional prediction models for winter wheat yield based on multi-spectral UAV data and machine learning methods. Six machine learning methods including Gaussian process regression (GPR), support vector machine regression (SVR) and random forest regression (RFR) were used for the construction of the yield prediction models. Ten vegetation indices (VIs) extracted from canopy spectral images of winter wheat acquired from a multi-spectral UAV at five key growth stages in Xuzhou City, Jiangsu Province, China in 2021 were selected as the variables of the models. In addition, in situ measurements of wheat yield were obtained in a destructive sampling manner for prediction algorithm modeling and validation. Prediction results of single growth stages showed that the optimal model was GPR constructed from extremely strong correlated VIs (ESCVIs) at the filling stage (R2 = 0.87, RMSE = 49.22 g/m2, MAE = 42.74 g/m2). The results of multiple stages showed GPR achieved the highest accuracy (R2 = 0.88, RMSE = 49.18 g/m2, MAE = 42.57 g/m2) when the ESCVIs of the flowering and filling stages were used. Larger sampling plots were adopted to verify the accuracy of yield prediction; the results indicated that the GPR model has strong adaptability at different scales. These findings suggest that using machine learning methods and multi-spectral UAV data can accurately predict crop yield at the field scale and deliver a valuable application reference for farm-scale field crop management.

The Accuracy of Winter Wheat Identification at Different Growth Stages Using Remote Sensing

The aim of this study was to explore the differences in the accuracy of winter wheat identification using remote sensing data at different growth stages using the same methods. Part of northern Henan Province, China was taken as the study area, and the winter wheat growth cycle was divided into five periods (seeding‒tillering, overwintering, reviving, jointing‒heading, and flowering‒maturing) based on monitoring data obtained from agrometeorological stations. With the help of the Google Earth Engine (GEE) platform, the separability between winter wheat and other land cover types was analyzed and compared using the Jeffries‒Matusita (J‒M) distance method. Spectral features, vegetation index, water index, building index, texture features, and terrain features were generated from Sentinel-2 remote sensing images at different growth periods, and then were used to establish a random forest classification and extraction model. A deep U-Net semantic segmentation model based on the red, green, blue, and near-infrared bands of Sentinel-2 imagery was also established. By combining models with field data, the identification of winter wheat was carried out and the difference between the accuracy of the identification in the five growth periods was analyzed. The experimental results show that, using the random forest classification method, the best separability between winter wheat and the other land cover types was achieved during the jointing‒heading period: the overall identification accuracy for the winter wheat was then highest at 96.90% and the kappa coefficient was 0.96. Using the deep-learning classification method, it was also found that the semantic segmentation accuracy of winter wheat and the model performance were best during the jointing‒heading period: a precision, recall, F1 score, accuracy, and IoU of 0.94, 0.93, 0.93, and 0.88, respectively, were achieved for this period. Based on municipal statistical data for winter wheat, the accuracy of the extraction of the winter wheat area using the two methods was 96.72% and 88.44%, respectively. Both methods show that the jointing‒heading period is the best period for identifying winter wheat using remote sensing and that the identification made during this period is reliable. The results of this study provide a scientific basis for accurately obtaining the area planted with winter wheat and for further studies into winter wheat growth monitoring and yield estimation.

Comparison of Winter Wheat Extraction Methods Based on Different Time Series of Vegetation Indices in the Northeastern Margin of the Qinghai–Tibet Plateau: A Case Study of Minhe, China

The northeastern margin of the Qinghai–Tibet Plateau (QTP) is an agricultural protection area in China’s new development plan, and the primary region of winter wheat growth within QTP. Winter wheat monitoring is critical for understanding grain self-sufficiency, climate change, and sustainable socioeconomic and ecological development in the region. However, due to the complex terrain and high altitude of the region, with discontinuous arable land and the relatively low level of agricultural development, there are no effective localization methodologies for extracting and monitoring the detailed planting distribution information of winter wheat. In this study, Sentinel-2A/B data from 2019 to 2020, obtained through the Google Earth Engine platform, were used to build time series reference curves of vegetation indices in Minhe. Planting distribution information of winter wheat was extracted based on the phenology time-weighted dynamic time warping (PT-DTW) method, and the effects of different vegetation indices’ time series and their corresponding threshold parameters were compared. The results showed that: (1) the three vegetation indices—normalized difference vegetation index (NDVI), normalized differential phenology index (NDPI), and normalized difference greenness index (NDGI)—maintained high mapping potential; (2) under the optimal threshold, >88% accuracy of index identification for winter wheat extraction was achieved; (3) due to improved extraction accuracy and resulting boundary range, NDPI and its corresponding optimal parameter (T = 0.05) performed the best. The process and results of this study have certain reference value for the study of winter wheat planting information change and the formulation of dynamic monitoring schemes in agricultural areas of QTP.

Mapping Winter Wheat with Optical and SAR Images Based on Google Earth Engine in Henan Province, China

The timely and accurate acquisition of winter wheat acreage is crucial for food security. This study investigated the feasibility of extracting the spatial distribution map of winter wheat in Henan Province by using synthetic aperture radar (SAR, Sentinel-1A) and optical (Sentinel-2) images. Firstly, the SAR images were aggregated based on the growth period of winter wheat, and the optical images were aggregated based on the moderate resolution imaging spectroradiometer normalized difference vegetation index (MODIS-NDVI) curve. Then, five spectral features, two polarization features, and four texture features were selected as feature variables. Finally, the Google Earth Engine (GEE) cloud platform was employed to extract winter wheat acreage through the random forest (RF) algorithm. The results show that: (1) aggregated images based on the growth period of winter wheat and sensor characteristics can improve the mapping accuracy and efficiency; (2) the extraction accuracy of using only SAR images was improved with the accumulation of growth period. The extraction accuracy of using the SAR images in the full growth period reached 80.1%; and (3) the identification effect of integrated images was relatively good, which makes up for the shortcomings of SAR and optical images and improves the extraction accuracy of winter wheat.

Extraction of Kenyan Grassland Information Using PROBA-V Based on RFE-RF Algorithm

Africa has the largest grassland area among all grassland ecosystems in the world. As a typical agricultural and animal husbandry country in Africa, animal husbandry plays an important role in this region. The investigation of grassland resources and timely grasping the quantity and spatial distribution of grassland resources are of great significance to the stable development of local animal husbandry economy. Therefore, this paper uses Kenya as the study area to investigate the effective and fast approach for grassland mapping with 100-m resolution using the open resources in the Google Earth Engine cloud platform. The main conclusions are as follows. (1) In the feature combination optimization part of this paper, the machine learning algorithm is used to compare the scores and standard deviations of several common algorithms combined with RFE. It is concluded that the combination of RFE and random forest algorithm has the highest stability in modeling and the best feature optimization effect. (2) After feature optimization by the RFE-RF algorithm, the number of features is reduced from 12 to 8, which compressed the original feature space and reduced the redundancy of features. The optimal combination features are applied to random forest classification, and the overall accuracy and Kappa coefficient of classification are 0.87 and 0.85, respectively. The eight features are: elevation, NDVI, EVI, SWIR, RVI, BLUE, RED, and LSWI. (3) There are great differences in topographic features among the local land types in the study area, and the addition of topographic features is more conducive to the recognition and classification of various land types. There exists “salt-and-pepper phenomenon” in pixel-oriented classification. Later research focus will combine the RFE-RF algorithm and the segmentation algorithm to achieve object-oriented land cover classification.

Aboveground Biomass Mapping of Crops Supported by Improved CASA Model and Sentinel-2 Multispectral Imagery

The net primary productivity (NPP) and aboveground biomass mapping of crops based on remote sensing technology are not only conducive to understanding the growth and development of crops but can also be used to monitor timely agricultural information, thereby providing effective decision making for agricultural production management. To solve the saturation problem of the NDVI in the aboveground biomass mapping of crops, the original CASA model was improved using narrow-band red-edge information, which is sensitive to vegetation chlorophyll variation, and the fraction of photosynthetically active radiation (FPAR), NPP, and aboveground biomass of winter wheat and maize were mapped in the main growing seasons. Moreover, in this study, we deeply analyzed the seasonal change trends of crops’ biophysical parameters in terms of the NDVI, FPAR, actual light use efficiency (LUE), and their influence on aboveground biomass. Finally, to analyze the uncertainty of the aboveground biomass mapping of crops, we further discussed the inversion differences of FPAR with different vegetation indices. The results demonstrated that the inversion accuracies of the FPAR of the red-edge normalized vegetation index (NDVIred-edge) and red-edge simple ratio vegetation index (SRred-edge) were higher than those of the original CASA model. Compared with the reference data, the accuracy of aboveground biomass estimated by the improved CASA model was 0.73 and 0.70, respectively, which was 0.21 and 0.13 higher than that of the original CASA model. In addition, the analysis of the FPAR inversions of different vegetation indices showed that the inversion accuracies of the red-edge vegetation indices NDVIred-edge and SRred-edge were higher than those of the other vegetation indices, which confirmed that the vegetation indices involving red-edge information can more effectively retrieve FPAR and aboveground biomass of crops.

Machine-Learning Classification of Soil Bulk Density in Salt Marsh Environments

Remotely sensed data from both in situ and satellite platforms in visible, near-infrared, and shortwave infrared (VNIR–SWIR, 400–2500 nm) regions have been widely used to characterize and model soil properties in a direct, cost-effective, and rapid manner at different scales. In this study, we assess the performance of machine-learning algorithms including random forest (RF), extreme gradient boosting machines (XGBoost), and support vector machines (SVM) to model salt marsh soil bulk density using multispectral remote-sensing data from the Landsat-7 Enhanced Thematic Mapper Plus (ETM+) platform. To our knowledge, use of remote-sensing data for estimating salt marsh soil bulk density at the vegetation rooting zone has not been investigated before. Our study reveals that blue (band 1; 450–520 nm) and NIR (band 4; 770–900 nm) bands of Landsat-7 ETM+ ranked as the most important spectral features for bulk density prediction by XGBoost and RF, respectively. According to XGBoost, band 1 and band 4 had relative importance of around 41% and 39%, respectively. We tested two soil bulk density classes in order to differentiate salt marshes in terms of their capability to support vegetation that grows in either low (0.032 to 0.752 g/cm³) or high (0.752 g/cm³ to 1.893 g/cm³) bulk density areas. XGBoost produced a higher classification accuracy (88%) compared to RF (87%) and SVM (86%), although discrepancies in accuracy between these models were small (<2%). XGBoost correctly classified 178 out of 186 soil samples labeled as low bulk density and 37 out of 62 soil samples labeled as high bulk density. We conclude that remote-sensing-based machine-learning models can be a valuable tool for ecologists and engineers to map the soil bulk density in wetlands to select suitable sites for effective restoration and successful re-establishment practices.

Satellite imagery and machine learning for identification of aridity risk in central Java Indonesia

This study aims to develop a software framework for predicting aridity using vegetation indices (VI) from LANDSAT 8 OLI images. VI data are predicted using machine learning (ml): Random Forest (RF) and Correlation and Regression Trees (CART). Comparison of prediction using Artificial Neural Network (ANN), Support Vector Machine (SVM), k-nearest neighbors (k-nn) and Multivariate Adaptive Regression Spline (MARS). Prediction results are interpolated using Inverse Distance Weight (IDW). This study was conducted in stages: (1) Image preprocessing; (2) calculating numerical data extracted from the LANDSAT band imagery using vegetation indices; (3) analyzing correlation coefficients between VI; (4) prediction using RF and CART; (5) comparing performances between RF and CART using ANN, SVM, k-nn, and MARS; (6) testing the accuracy of prediction using Mean Square Error (MSE) and Mean Absolute Percentage Error (MAPE); (7) interpolating with IDW. Correlation coefficient of VI data shows a positive correlation, the lowest r (0.07) and the highest r (0.98). The experiments show that the RF and CART algorithms have efficiency and effectivity in determining the aridity areas better than the ANN, SVM, k-nn, and MARS algorithm. RF has a difference between the predicted results and 1.04% survey data MAPE and the smallest value close to zero is 0.05 MSE. CART has a difference between the predicted results and 1.05% survey data MAPE and the smallest value approaching to zero which is 0.05 MSE. The prediction results of VI show that in 2020 most of the study areas were low vegetation areas with the Normalized Difference Vegetation Index (NDVI) < 0.21, had an indication of drought with the Vegetation Health Index (VHI) < 31.10, had a Vegetation Condition Index (VCI) in some areas between 35%-50% (moderate drought) and < 35% (high drought). The Burn Area Index (dBAI) values are between -3, 971 and -2,376 that show the areas have a low fire risk, and index values are between -0, 208 and -0,412 that show the areas are starting vegetation growth. The result of this study shows that the machine learning algorithms is an accurate and stable algorithm in predicting the risks of drought and land fire based on the VI data extracted from the LANDSAT 8 OLL imagery. The VI data contain the record of vegetation condition and its environment, including humidity, temperatures, and the environmental vegetation health.

Machine Learning Comparison and Parameter Setting Methods for the Detection of Dump Sites for Construction and Demolition Waste Using the Google Earth Engine

Machine learning has been successfully used for object recognition within images. Due to the complexity of the spectrum and texture of construction and demolition waste (C&DW), it is difficult to construct an automatic identification method for C&DW based on machine learning and remote sensing data sources. Machine learning includes many types of algorithms; however, different algorithms and parameters have different identification effects on C&DW. Exploring the optimal method for automatic remote sensing identification of C&DW is an important approach for the intelligent supervision of C&DW. This study investigates the megacity of Beijing, which is facing high risk of C&DW pollution. To improve the classification accuracy of C&DW, buildings, vegetation, water, and crops were selected as comparative training samples based on the Google Earth Engine (GEE), and Sentinel-2 was used as the data source. Three classification methods of typical machine learning algorithms (classification and regression trees (CART), random forest (RF), and support vector machine (SVM)) were selected to classify the C&DW from remote sensing images. Using empirical methods, the experimental trial method, and the grid search method, the optimal parameterization scheme of the three classification methods was studied to determine the optimal method of remote sensing identification of C&DW based on machine learning. Through accuracy evaluation and ground verification, the overall recognition accuracies of CART, RF, and SVM for C&DW were 73.12%, 98.05%, and 85.62%, respectively, under the optimal parameterization scheme determined in this study. Among these algorithms, RF was a better C&DW identification method than were CART and SVM when the number of decision trees was 50. This study explores the robust machine learning method for automatic remote sensing identification of C&DW and provides a scientific basis for intelligent supervision and resource utilization of C&DW.

Susceptibility to Gully Erosion: Applying Random Forest (RF) and Frequency Ratio (FR) Approaches to a Small Catchment in Ethiopia

Soil erosion by gullies in Ethiopia is causing environmental and socioeconomic problems. A sound soil and water management plan requires accurately predicted gully erosion hotspot areas. Hence, this study develops a gully erosion susceptibility map (GESM) using frequency ratio (FR) and random forest (RF) algorithms. A total of 56 gullies were surveyed, and their extents were derived by digitizing Google Earth imagery. Literature review and a multicollinearity test resulted in 14 environmental variables for the final analysis. Model prediction potential was evaluated using the area under the curve (AUC) method. Results showed that the best prediction accuracy using the FR and RF models was obtained by using the top four most important gully predictor factors: drainage density, elevation, land use, and groundwater table. The notion that the groundwater table is one of the most important gully predictor factors in Ethiopia is a novel and significant quantifiable finding and is critical to the design of effective watershed management plans. Results from separate variable importance analyses showed land cover for Nitisols and drainage density for Vertisols as leading factors determining gully locations. Factors such as texture, stream power index, convergence index, slope length, and plan and profile curvatures were found to have little significance for gully formation in the studied catchment.