Search

Early prediction of winter wheat yield at the regional scale is essential for food policy making and food security, especially in the context of population growth and climate change. Agricultural big data and artificial intelligence (AI) are key technologies for smart agriculture, bringing cost-effective solutions to the agricultural sector. Deep learning-based crop yield forecast has currently emerged as one of the key methods for guiding agricultural production. In this study, we proposed a Bayesian optimization-based long- and short-term memory model (BO-LSTM) to construct a multi-source data fusion-driven crop growth feature extraction algorithm for winter wheat yield prediction. The yield prediction performance of BO-LSTM, support vector machine (SVM), and least absolute shrinkage and selection operator (Lasso) was then compared with multi-source data as input variables. The results showed that effective deep learning hyperparameter optimization is made possible by Bayesian optimization. The BO-LSTM (RMSE = 177.84 kg/ha, R2 = 0.82) model had the highest accuracy of yield prediction with the input combination of "GPP + Climate + LAI + VIs". BO-LSTM and SVM (RMSE = 185.7 kg/ha, R2 = 0.80) methods outperformed linear regression Lasso (RMSE = 214.5 kg/ha, R2 = 0.76) for winter wheat yield estimation. There were also differences between machine learning and deep learning, BO-LSTM outperformed SVM. indicating that the BO-LSTM model was more effective at capturing data correlations. In order to further verify the robustness of the BO-LSTM method, we explored the performance estimation performance of BO-LSTM in different regions. The results demonstrated that the BO-LSTM model could obtain higher estimation accuracy in regions with concentrated distribution of winter wheat cultivation and less influence of human factors. The approach used in this study can be expected to forecast crop yields, both in regions with a deficit of data and globally; it can also simply and effectively forecast winter wheat yields in a timely way utilizing publicly available multi-source data. [ABSTRACT FROM AUTHOR]

Monitoring agricultural drought via ground hyper-spectral remote sensing has always been a hot topic in the fields of agriculture and meteorology. In this study, a greenhouse experiment was conducted on wheat subjected to water stress during its different growth stages, namely tillering, jointing, and milk maturity. An instrument (HOBO ware PRO) used to continuously measure soil moisture was employed to measure the soil water content (SWC). An analytical spectral device (ASD) was utilized to obtain the spectral curve of wheat subject to different water treatment methods. The canopy temperature was obtained using thermal infrared sensors (METER SI-400). The relationships between the SWC, wheat drought stage, canopy temperature, and spectral response characteristics were elucidated. The results showed that the significant differences in spectral characteristics were due to water stress during the different growth stages of wheat. Red-edge parameters of red-valley position (RVP) and red-edge position (REP) both changed by 21 nm for the tillering-stage drought and the jointing-stage drought; however, the RVP and REP values for the milk maturity stage drought and the treatment under no water stress changed by 2 nm. Further, it was proved that the red-edge blue-shift phenomenon was affected not only by the different wheat growth processes, but also by the water stress at different growth stages. Red-edge reflectance clearly reflects wheat water stress at different growth stages. From SWC and canopy temperature analysis results, SWC and canopy temperature had a significant difference between wheat drought at different growth stages, and the canopy temperature at the jointing stage drought had the strongest change. The water index (WI) based on eleven vegetation water indexes exhibited a good performance for distinguishing wheat water stress at different growth stages. In conclusion, ground-based hyperspectral remote sensing can provide a large amount of high temporal and spectral resolution data on vegetation and its surrounding environment, making it an important technical tool for wheat drought monitoring, which has a great significance on the monitoring and early warning of wheat drought, reducing drought-related yield losses, and ensuring food security. [ABSTRACT FROM AUTHOR]

• The MP-Net model aggregates context information. • The MP-Net model reduces the feature losses. • The MP-Net model efficiently and precisely classifies crops. Accurate crop classification map is of great significance in various fields such as the survey of agricultural resource, the analysis of existing circumstance on land application, the yield estimation of crop and the disaster warning. The methods based on machine learning and deep learning are popularly used in crop classification and recognition of remote sensing images. However, the crop classification task based on neural networks still faces significant challenges due to the spatial and temporal distribution of crops and the inherent characteristics of remote sensing images. Therefore, this study proposes the multi-layer pyramid crop classification network (MP-Net) to solve the above problems. To reduce the feature loss during the crop extraction, the proposed model uses the pyramid pooling module to improve the ability of global information acquisition, and the information concatenation module to retain the upper features. Using the GF-6 and Sentinel-2 satellite data, the proposed model was tested in Erhai Lake Basin and Beian City. Compared with other five deep learning models, such as FCN, SegNet, U-Net, PSPNet and DeepLabv3+, the experimental results indicate that the proposed model achieves the highest accuracy in both study areas. Meanwhile, the proposed model has the advantages of short training time and high efficiency under the same running conditions. Overall, this study is beneficial to improve the efficiency and accuracy of crop classification task in the unbalanced temporal and spatial distribution. It also brings a feasible scheme for crop classification tasks in complex growing areas. The code has been publicly available at https://github.com/Xu-Chang-Hong/MP-Net. [ABSTRACT FROM AUTHOR]

Accurate retrieval of land surface temperature (LST) over urban areas is of great significance for urban thermal environment monitoring. In previous studies, most of the urban LST retrieval methods were developed based on the assumption of a flat surface without considering the influence of urban 3-D geometry structure, which has a significant impact on the retrieval accuracy of LST over urban areas. In this study, a radiative transfer equation (RTE)-based single-channel method was developed to retrieve LST with urban geometry effect correction from the Landsat 8 thermal infrared (TIR) data in band 10. The increase in adjacent radiance from the surrounding pixels and the decrease in atmospheric downwelling radiance caused by urban geometry structure were taken into account in this method. Because it is difficult to directly validate the retrieval accuracy of LST over urban areas using in situ LST measurements, the performance of the RTE-based LST retrieval method was evaluated via comparing brightness temperature (BT) at the top of the atmosphere (TOA) simulated by the discrete anisotropic radiative transfer (DART) model and the urban RTE over three subregions. There is a good agreement between BT at the TOA simulated by the DART model and the urban RTE, with a root-mean-squared error (RMSE) of less than 0.25 K. The variations in LST retrieved with urban geometry effect correction over different local climate zones (LCZs) were analyzed. In general, built-up LCZs have relatively higher LST than land cover LCZs. The differences between LST retrieved without/with urban geometry effect correction over different LCZs are greater than 0.2 K. The largest average LST difference over built-up LCZs is approximately 0.9 K, whereas that over land cover LCZs is approximately 0.65 K. LST retrieved without/with urban geometry effect correction was used to calculate urban heat island intensity (UHII) in terms of the LCZ-based method. The results indicate that UHII calculated from LST with urban geometry effect correction is lower than that calculated from LST without urban geometry effect correction, with an average difference of approximately 0.5 K. [ABSTRACT FROM AUTHOR]

Accurate estimation of land surface temperature (LST) is crucial for ecological environment monitoring and climate change studies in mountainous areas. The current LST retrieval algorithms were developed without accounting for the topographic effect, which can only be used to retrieve LST over relatively flat surfaces. Due to the impact of 3-D structure of mountainous surfaces, rugged terrain makes the processes of thermal radiation more complex. In this study, a radiative transfer equation (RTE)-based single-channel algorithm was proposed to retrieve LST with topographic effect correction from the Landsat 8 thermal infrared (TIR) data in mountainous areas. This algorithm accounts for the changes in the thermal radiation components in the TIR RTE caused by the topographic effect. According to the analysis of simulation data, sky-view factor (SVF), atmospheric water vapor content, surface emissivity of target pixel, and average LST of the surrounding terrain have significant influence on the magnitude of the topographic effect. The differences between the LST retrieved without/with topographic effect correction from the Landsat 8 TIR data are related to SVF. The topographic effect should be taken into account in the LST retrieval algorithm when SVF is smaller than 0.7. The largest LST difference of approximately 1 K occurs in the deep valley. The results indicate that LST without topographic effect correction could be overestimated to be as high as 1 K. Due to a lack of in situ LST measurements, the performance of the LST retrieval algorithm in mountainous areas was only evaluated by comparing the brightness temperature (BT) at the top of the atmosphere (TOA) simulated by the DART+MODTRAN model and the TIR RTE over mountainous surfaces at three subregions. There is a good consistency between BT at the TOA simulated by the DART+MODTRAN model and the TIR RTE over mountainous surfaces at the three subregions, with a root-mean-squared error (RMSE) of less than 0.23 K. [ABSTRACT FROM AUTHOR]