No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Deep Transfer Learning for Crop Yield Prediction with Remote Sensing Data
Abstract
Accurate prediction of crop yields in developing countries in advance of harvest time is central to preventing famine, improving food security, and sustainable development of agriculture. Existing techniques are expensive and difficult to scale as they require locally collected survey data. Approaches utilizing remotely sensed data, such as satellite imagery, potentially provide a cheap, equally effective alternative. Our work shows promising results in predicting soybean crop yields in Argentina using deep learning techniques. We also achieve satisfactory results with a transfer learning approach to predict Brazil soybean harvests with a smaller amount of data. The motivation for transfer learning is that the success of deep learning models is largely dependent on abundant ground truth training data. Successful crop yield prediction with deep learning in regions with little training data relies on the ability to fine-tune pre-trained models.
... Similar [27], [108], [109], [110], [111], [112] [113], [114] [115], [116], [117], [118] [119], [120] Different [23], [121], [122], [123], [124], [126], [127], [128], [129] [130], [131], [132] [133], [134], [135], [136] [137], [138], [139], [140] [90], [91], [92], [93], [94], [95], [96], [97], [98], [99], [100], [101], [102], [103], [104], [105], [106], [107] -Similar [27], [108], [114], [115], [116], [119], [120] [109], [110], [111], [112], [113], [117], [118] Different [23], [121], [122], [123], [124], [126], [127], [128], [129], [130], [131], [132], [133], [134], [135] [136], [137], [138], [139], [140] This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. ...
... Similar [27], [108], [109], [110], [111], [112] [113], [114] [115], [116], [117], [118] [119], [120] Different [23], [121], [122], [123], [124], [126], [127], [128], [129] [130], [131], [132] [133], [134], [135], [136] [137], [138], [139], [140] [90], [91], [92], [93], [94], [95], [96], [97], [98], [99], [100], [101], [102], [103], [104], [105], [106], [107] -Similar [27], [108], [114], [115], [116], [119], [120] [109], [110], [111], [112], [113], [117], [118] Different [23], [121], [122], [123], [124], [126], [127], [128], [129], [130], [131], [132], [133], [134], [135] [136], [137], [138], [139], [140] This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. ...
... Homogeneous Heterogeneous Same [93], [94], [96], [97], [98], [101], [102], [104], [105] [90], [91], [92], [95], [99], [100], [103], [106], [107] Similar [27], [114], [116] [108], [109], [110], [111], [112], [113], [115], [117], [118], [119], [120], [133] Different [121], [128], [131] [23], [122], [123], [124], [126], [127], [129], [130], [132], [134], [135], [136], [137], [138], [139], [140] This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. ...
The significant increase in the amount of satellite data in recent years along with the increase in computing resources has opened up new possibilities for Earth Observation (EO) data analysis. However, the main obstacle to obtaining highquality products is the limited amount of labeled data. Among the methods that can help solve this type of problem, transfer learning (TL) has gained enormous interest in recent years. It involves using a model trained on one problem to solve another problem. This article reviews TL methods as applied to EO data analyses. It provides a structured overview of methods in relation to applications, publication statistics, and summarizes the advantages and disadvantages noted of previous research.
... The hybrid model achieved a high prediction accuracy of over 90%, demonstrating the effectiveness of combining ML and DL techniques [32]. Hybrid ML/DL models can also integrate DL techniques with crop simulation models, offering insights into complex rice growth dynamics [33]. Sharma et al. (2021) developed a hybrid model that utilized remote sensing data and weather data to train a DL model. ...
... The output from the DL model was then used as input to a crop simulation model, which captured the interactions between environmental conditions and crop growth. This integrated approach achieved a high prediction accuracy of over 93% for rice yield prediction in India [33]. ...
... Moreover, hybrid models can also reduce the risk of overfitting and improve the generalizability of the models [33]. ...
Maximizing rice yield is critical for ensuring food security and sustainable agriculture in Malaysia. This research investigates the impact of environmental conditions and management methods on crop yields, focusing on accurate predictions to inform decision-making by farmers. Utilizing machine learning algorithms as decision-support tools, the study analyses commonly used models-Linear Regression, Support Vector Machines, Random Forest, and Artificial Neural Networks-alongside key environmental factors such as temperature, rainfall, and historical yield data. A comprehensive dataset for rice yield prediction in Malaysia was constructed, encompassing yield data from 2014 to 2018. To elucidate the influence of climatic factors, long-term rainfall records spanning 1981 to 2018 were incorporated into the analysis. This extensive dataset facilitates the exploration of recent agricultural trends in Malaysia and their relationship to rice yield. The study specifically evaluates the performance of Random Forest, Support Vector Machine (SVM), and Neural Network (NN) models using metrics like Correlation Coefficient, Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Squared Error (MSE), and Mean Absolute Percentage Error (MAPE). Results reveal Random Forest as the standout performer with a Correlation Coefficient of 0.954, indicating a robust positive linear relationship between predictions and actual yield data. SVM and NN also exhibit respectable Correlation Coefficients of 0.767 and 0.791, respectively, making them effective tools for rice yield prediction in Malaysia. By integrating diverse environmental and management factors, the proposed methodology enhances prediction accuracy, enabling farmers to optimize practices for better economic outcomes. This approach holds significant potential for contributing to sustainable agriculture, improved food security, and enhanced economic efficiency in Malaysia's rice farming sector. Leveraging machine learning, the research aims to transform rice yield prediction into a proactive decision-making tool, fostering a resilient and productive agrarian ecosystem in Malaysia.
... This limitation restricts their use in areas with limited historical yield data. Moreover, a model trained on data from one region may not perform well in entirely new locations because of the domain shift [13]. One of the reasons existing deep-learning-based crop yield prediction research has predominantly focused on specific regions of the world is the availability of abundant historical data in those areas [6]. ...
... The transfer learning model with BiLSTM achieved a 16% and 23% reduction in MAE compared to the transfer learning model with MLP for the two test years. Advanced deep learning models, such as LSTM and 1D-CNN, have already demonstrated their effectiveness in yield prediction, outperforming MLP based approaches [6,9,13,53]. The findings of our study suggest that these techniques can provide superior feature representation, in the context of transfer learning for yield prediction as well. ...
The timely and reliable prediction of crop yields on a larger scale is crucial for ensuring a stable food supply and food security. In the last few years, many studies have demonstrated that deep learning can offer reliable solutions for crop yield prediction. However, a key challenge in applying deep-learning models to crop yield prediction is their reliance on extensive training data, which are often lacking in many parts of the world. To address this challenge, this study introduces TrAdaBoost.R2, along with fine-tuning and domain-adversarial neural network deep-transfer-learning strategies, for predicting the winter wheat yield across diverse climatic zones in the USA. All methods used the bidirectional LSTM (BiLSTM) architecture to leverage its sequential feature extraction capabilities. The proposed transfer-learning approaches outperformed the baseline deep-learning model, with mean absolute error reductions ranging from 9% to 28%, demonstrating the effectiveness of these methods. Furthermore, the results demonstrate that the semi-supervised transfer-learning approach using the two-stage version of TrAdaBoost.R2 and fine-tuning achieved a superior performance compared to the domain-adversarial neural network and standard TrAdaBoost.R2. Additionally, the study offers insights for improving the accuracy and generalizability of crop yield prediction models in diverse agricultural landscapes across different regions.
... This would enable worldwide breeding programs to reduce the costs of harvesting procedure while predicting yield with largely pre-trained models. Few attempts have been made regarding this topic at a field scale (Wang et al., 2018) but not for plot-level yield prediction to our knowledge. ...
... When performing transfer learning within locations in ARG and when performing transfer learning from ARG to USA, the best results were achieved by first training the model with the initial dataset and then retraining the model with the weights of the previous model for the initialization of it, avoiding random initialization but keeping all the layers retrainable (without freezing any layer) and adding a new batch normalization layer and a new dense layer for the prediction layer (Wang et al., 2018). The decision to keep all layers trainable was influenced by the variability in the number of samples and yield means across the datasets (Table 3). ...
Despite significant progress has been made towards crop yield prediction with remote sensing, there exist knowledge gaps on (1) the impacts of temporal resolution of imaging frequencies on yield prediction, (2) transferability of the models among different genotypes and test sites, and (3) translation of these research developments to crop breeding that benefit farmers. Existing research predominantly provides an on-site perspective, frequently missing the complexities of real-world applications. The objectives of this paper are to investigate the transferability and generalization capabilities of yield prediction models for crop breeding across test sites located in North and South Americas. Toward that goal, we tested different machine learning techniques including Random Forest Regressor (RF), Gradient Boosting Regression (GB), and Deep Neural Networks (DNN) for soybean yield prediction with experiments conducted in different climate and growth conditions. A novel transfer learning approach was proposed for genotype selection and categorizing soybean yield for screening high-yield varieties. Furthermore, we studied the effect of temporal resolution on yield prediction, focusing on the critical development stages and optimal aerial survey frequencies for precise yield prediction using large 31,404 sample data. Results demonstrated that the combined dataset of Argentina and United States representing different climate regimes provided the highest performance with an R2 of 0.76 using RF and GB algorithms. The classification approach was proven to be most useful for crop breeding as demonstrated by accurately identifying the high-yielding genotypes. Increasing temporal sampling of key phenological stages significantly improved yield prediction. Although transfer learning yielded promising outcomes across trials within Argentina the efficacy of transferring models from Argentina to the United States was limited, attributed to significant seasonal and climate variations. This study pioneered the use of transfer learning for model adaptability in real-world breeding scenarios, training and transferring models within the South and North Americas, providing actionable insights and strategies for the breeding community, aiming to facilitate improved decision-making for agricultural productivity.
... This would enable worldwide breeding programs to reduce the costs of harvesting procedure while predicting yield with largely pre-trained models. Few attempts have been made regarding this topic at a field scale (Wang et al., 2018) but not for plot-level yield prediction to our knowledge. ...
... When performing transfer learning within locations in ARG and when performing transfer learning from ARG to USA, the best results were achieved by first training the model with the initial dataset and then retraining the model with the weights of the previous model for the initialization of it, avoiding random initialization but keeping all the layers retrainable (without freezing any layer) and adding a new batch normalization layer and a new dense layer for the prediction layer (Wang et al., 2018). The decision to keep all layers trainable was influenced by the variability in the number of samples and yield means across the datasets (Table 3). ...
... You et al. (2017) introduced a scalable approach to crop yield prediction using modern representation learning rather than traditional hand-crafted features, achieving impressive results. Wang et al. (2018) demonstrated soybean yield predictions in Argentina and obtained reliable results in Brazil using transfer learning, even with limited data. Ma et al. (2021a) developed a county-level corn yield prediction model using a Bayesian Neural Network across 12 states in the Corn Belt, showing the potential for large-scale corn yield prediction tasks. ...
Remote sensing (RS) techniques, by enabling non-contact acquisition of extensive ground observations, have become a valuable tool for corn yield prediction. Traditional process-based (PB) models are limited by fixed input features and struggle to incorporate large volumes of RS data. In contrast, machine learning (ML) models are often criticized for being ``black boxes'' with limited interpretability. To address these limitations, we used Knowledge-Guided Machine Learning (KGML), which combined the strengths of both approaches and fully used RS data. However, previous KGML methods overlooked the crucial role of soil moisture in plant growth. To bridge this gap, we proposed the Knowledge-Guided Machine Learning with Soil Moisture (KGML-SM) framework, using soil moisture as an intermediate variable to emphasize its key role in plant development. Additionally, based on the prior knowledge that the model may overestimate under drought conditions, we designed a drought-aware loss function that penalizes predicted yield in drought-affected areas. Our experiments showed that the KGML-SM model outperformed other ML models. Finally, we explored the relationships between drought, soil moisture, and corn yield prediction, assessing the importance of various features and analyzing how soil moisture impacts corn yield predictions across different regions and time periods.
... The MTL method combined with neural networks has been widely applied in remote sensing studies, which have mostly focused on feature extraction and classification [33][34][35], expanding the estimations of one variable to other similar variables [36,37], extending the simulated data to real-world data [13,28,38,39], and estimating different sensor data [27]. In addition, previous studies showed that the MTL method could be used to train DL models in a country or region with sufficient data, which was hoped to be appropriate for other countries or regions with small training samples to predict crop yield [40,41]. However, there is no similar study on grassland GPP estimates over IMG. ...
It is significant to simulate grassland gross primary production (GPP) to understand the terrestrial carbon budget over Inner Mongolia (IMG), China. Nevertheless, there is not sufficient in situ GPP data over this region. In this study, we proposed a novel model-based transfer learning (MTL) approach with generative adversarial networks-long short-term memory (GAN-LSTM) and light use efficiency (LUE) models to derive grassland GPP over IMG, China. We first used 25 grassland eddy covariance (EC) sites over the conterminous United States to establish the GAN-LSTM model, and then fine-tuned it with six sites over IMG to estimate water constraints (Ws) that were embedded into the LUE model to predict GPP. We then compared it with an instance-based transfer learning (ITL) and non-transfer learning approaches. Against the six IMG EC sites, the GPP estimates of MTL-LUE outperformed the other approaches with a lower root-mean-square error median (1.35 g C m-2 d-1) and a higher Kling-Gupta efficiency of 0.54. An innovation of this approach is that MTL-LUE mitigates the effect of limited training samples on the machine learning-based LUE hybrid model for GPP estimates over IMG.
... In the field of remote sensing, transfer learning has also been applied in diverse applications, including scene classification Zeng et al., 2021), crop yield production (A. X. Wang et al., 2018), airplane detection (Chen et al., 2018), Land Use Land Cover (Alem and Kumar, 2022), roads extraction (Senthilnath et al., 2020), and rice seedling analysis (Tseng et al., 2022). Alipour et al., 2023 utilized transfer learning for fuel mapping using multimodal data (Landsat-8 and NAIP imagery) and LANDFIRE samples as ground truth. ...
... More recently, motivated by the success of deep learning in computer vision, deep convolutional neural networks, which are highly nonlinear and scalable, have been used for improved crop yield prediction. For example, Wang et al. [26] leveraged deep transfer learning techniques to achieve promising results in predicting soybean crop yields in Argentina. Khaki et al. [27] developed a combination of convolutional neural networks (CNNs) and recurrent neural networks (RNNs), called CNN-RNN, for corn and soybean yield prediction based on environmental data and management practices. ...
Accurate crop yield prediction and modeling are essential for ensuring food security, optimizing resource allocation, and guiding policy decisions in agriculture, ultimately benefiting society at large. With the increasing threat of weather change, it is important to understand the impacts of weather dynamics on agricultural productivity, particularly for crucial crops like soybeans. This study considers the study area of the Greater Mississippi River Basin, where most soybeans are typically planted, with a large variety of weather across from the North to the South in the US. Leveraging the greenness and density measured by the normalized difference vegetation index (NDVI) from NASA’s Moderate Resolution Imaging Spectroradiometer (MODIS) satellite images, along with weather variables including mean precipitation, minimum temperature, and maximum temperature, we aim to uncover the relationships between these variables and soybean yield for different geographical and weather regions. Our analysis focuses on the four weather regions within the US: Very Cold, Cold, Mixed Humid, and Hot Humid, where most soybeans are planted in the Mississippi River Basin. The findings reveal that soybean yield in the Cold and Very Cold regions is positively correlated with minimum temperatures, whereas in the Mixed Humid and Hot Humid regions, negative correlations between maximum temperatures and yields are found. We identify a significant positive correlation between precipitation and soybean yield across all regions. In addition, the NDVI shows significant positive correlations with the soybean yield. Both linear and nonlinear regression models, including support vector machine and random forest models, are trained with remotely sensed data and weather data, showing a reliable and improved crop yield prediction. The findings of this study contribute to a better understanding of how soybean yield responds to climatic variations and will help the national agricultural management system in better monitoring and predicting crop yield when facing the increasing challenge of weather dynamics.
... oExplainability of deep learning models used to predict the crop yield can address the "black box" nature [98]. oTo increase the scalability of predictive machine learning models, transfer learning can help find patterns in regions with less data availabilities [99,100] (continued on next page) [54]. oUse of ANN and novel texture transfer method to predict thermal changes in urban areas due to land use modifications [53]. ...
This systematic review examines the application of artificial intelligence (AI), including machine learning (ML) and deep learning (DL), for climate change adaptation and mitigation in Iran, Pakistan, and Turkey. These three nations—key Economic Cooperation Organization (ECO) members and a nexus between Europe and South Asia—are experiencing diverse environmental challenges due to varying climatic conditions. Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, we conducted a comprehensive search in the Scopus database, ultimately identifying 76 relevant articles out of an initial 492. Although some articles utilized multiple techniques, classical ML methods appeared in approximately 37.3 % of the studies, neural network paradigms in about 57.5 %, and optimization or meta-heuristic algorithms in around 5.0 %. Regarding thematic focus, about 33.3 % of the articles addressed water resource management, 22.2 % focused on climate prediction, 11.1 % on land and agriculture, 9 % on ecosystem modeling, and 24.2 % on natural disaster preparedness and response. The analysis reveals a growing but uneven body of research utilizing AI across the ECO countries. By highlighting successful applications, identifying key gaps—such as limited cross-border collaboration and inconsistent data availability—and proposing a framework for more integrated research, this review aims to guide future initiatives that leverage AI's potential to improve climate resilience and sustainability in the region.
... There are three main limitations on processing remote sensing data. First, high-resolution satellite remote sensing images are expensive, and most accessible open-source datasets are in low-resolution [7]. The poor quality of these images constrains model performance. ...
Accurate vehicle type classification plays a significant role in intelligent transportation systems. It is critical to understand the road conditions and usually contributive for the traffic light control system to respond correspondingly to alleviate traffic congestion. New technologies and comprehensive data sources, such as aerial photos and remote sensing data, provide richer and higher-dimensional information. In addition, due to the rapid development of deep neural network technology, image-based vehicle classification methods can better extract underlying objective features when processing data. Recently, several deep learning models have been proposed to solve this problem. However, traditional purely convolution-based approaches have constraints on global information extraction, and complex environments such as bad weather seriously limit their recognition capability. To improve vehicle type classification capability under complex environments, this study proposes a novel Densely Connected Convolutional Transformer-in-Transformer Neural Network (Dense-TNT) framework for vehicle type classification by stacking Densely Connected Convolutional Network (DenseNet) and Transformer-in-Transformer (TNT) layers. Vehicle data for three regions under four different weather conditions were deployed to evaluate the recognition capability. Our experimental findings validate the recognition ability of the proposed vehicle classification model, showing little decay even under heavy fog.
... A one-dimensional convolutional neural network (1D-CNN) is the base architecture for developing TL-based SOM satellite mapping models [31], [51]. Within the 1D-CNN, the convolutional layer is employed to extract features from onedimensional spectral data. ...
Calibrating an accurate soil organic matter (SOM) satellite mapping model requires a sufficient number of representative satellite samples (including actual ground-based SOM values and corresponding satellite spectral data). However, collecting these samples is challenging due to complex geological environments and inconvenient transportation conditions across large areas. To address this issue, a satellite sample simulation strategy was developed to transform ground-based local-area soil spectral samples into satellite-simulated spectral samples. Subsequently, a transfer learning (TL) approach was implemented to encode the spectra-SOM relationship learned from the satellite-simulated sample set into a basic neural network model. Finally, by fine-tuning this model with only a small number of satellite spectral samples, a robust SOM satellite mapping model was established. The results indicated that a total of 846 satellite-simulated spectral samples were generated, and the differences between the satellite-simulated spectral samples and the satellite spectral samples were minimized. The optimal TL-based SOM satellite mapping model (TL-half model) had an R2 of 0.90 and RPIQ of 4.02, representing a 12.50% improvement over the traditional SOM mapping model (R² = 0.80; RPIQ = 3.10). The TL-based SOM satellite mapping method proposed in this study offers an effective technical approach for regional SOM monitoring and global carbon storage management.
... Yield prediction is one of the core components of precision agriculture [7,8]. Real-time monitoring of soil temperature, humidity, nutrient content, and other factors can be achieved through sensor and IoT technology. ...
Precision agriculture is an important field that aims to optimize crop yields and quality, and it is crucial for ensuring food security and sustainability. With the increasing demand for food and the limited resources available for farming, precision agriculture has become more important than ever. In this paper, we will focus on the application of deep learning techniques, such as CNN and RNN models, in precision agriculture. Our primary objectives are to provide a comprehensive overview of artificial neural networks (ANNs), including Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), and to explore their applications in precision agriculture. Deep learning has shown great potential in accurately predicting crop yields and quality, and in optimizing the use of resources in farming. By achieving these objectives, we aim to contribute to the growing body of literature on the applications of deep learning in agriculture and to inspire further research in this promising field.
... • Dimensionality reduction and data transformation: You et al. [1] employed histogram mapping and Gaussian processes to reduce data dimensionality while preserving essential information. Wang et al. [21] utilized band mode and NDVI calculations to transform raw spectral data into more meaningful features. • Basic RS data pre-processing: Cheng et al. [9] applied standard techniques such as split, calibrate, and deburst to improve data quality. ...
The use of high-altitude remote sensing (RS) data from aerial and satellite platforms presents considerable challenges for agricultural monitoring and crop yield estimation due to the presence of noise caused by atmospheric interference, sensor anomalies, and outlier pixel values. This paper introduces a "Quartile Clean Image" pre-processing technique to address these data issues by analyzing quartile pixel values in local neighborhoods to identify and adjust outliers. Applying this technique to 20,946 Moderate Resolution Imaging Spectroradiometer (MODIS) images from 2002 to 2015, improved the mean peak signal-to-noise ratio (PSNR) to 40.91 dB. Integrating Quartile Clean data with Convolutional Neural Networks (CNN) models with exponential decay learning rate scheduling achieved RMSE improvements up to 5.88% for soybeans and 21.85% for corn, while Long Short-Term Memory (LSTM) models demonstrated RMSE reductions up to 11.52% for soybeans and 29.92% for corn using exponential decay learning rates. To compare the proposed method with state-of-the-art technique, we introduce the Vision Transformer (ViT) model for crop yield estimation. The ViT model, applied to the same dataset, achieves remarkable performance without explicit pre-processing, with R² scores ranging from 0.9752 to 0.9875 for soybean and 0.9540 to 0.9888 for corn yield estimation. The RMSE values range from 7.75086 to 9.76838 for soybean and 26.25265 to 34.20382 for corn, demonstrating the ViT model's robustness. This research contributes by (1) introducing the Quartile Clean Image method for enhancing RS data quality and improving crop yield estimation accuracy, and (2) comparing it with the state-of-the-art ViT model. The results demonstrate the effectiveness of the proposed approach and highlight the potential of the ViT model for crop yield estimation, representing a valuable advancement in processing high-altitude imagery for precision agriculture applications.
... A DL architecture for agricultural yield prediction 46 . This model demonstrated significant transfer learning capabilities and achieved outstanding accuracy by using a unique feature representation built from raw picture histograms 48 . Even with its effectiveness, the method's applicability to low spatial resolution pictures or yield prediction on a smaller scale is limited since it needs a significant number of pixels in a particular region to produce meaningful histograms. ...
Wheat commands attention due to its significant impact on culture, nutrition, the economy, and the guarantee of food security. The anticipated rise in temperatures resulting from climate change is a key factor contributing to food insecurity, as it markedly reduces wheat harvests. Terminal heat stress mostly affects spike fertility in wheat, specifically influencing pollen fertility and anther morphology. This research especially focuses on the shape of anthers and examines the effects of heat stress. The DinoLite Microscope’s high-resolution images are used to measure the length and width of wheat anthers. By using object identification techniques, the research accurately measures the length and width of each anther in images, offering valuable insights into the differences between various wheat varieties. Furthermore, Deep Learning (DL) methodologies are utilized to enhance agriculture, specifically employing record categorization to advance plant breeding management. Given the ongoing challenges in agriculture, there is a belief that incorporating the latest technologies is crucial. The primary objective of this study is to explore how Deep Learning algorithms can be beneficial in categorizing agricultural records, particularly in monitoring and identifying variations in spring wheat germplasm. Various Deep Learning algorithms, including Convolution Neural Network (CNN), LeNet, and Inception-V3 are implemented to classify the records and extract various patterns. LeNet demonstrates optimized accuracy in classifying the records, outperforming CNN by 52% and Inception-V3 by 70%. Moreover, Precision, Recall, and F1 Measure are utilized to ascertain accuracy levels. The investigation also enhances our comprehension of the distinct roles played by various genes in abiotic stress tolerance among diverse wheat varieties. The outcomes of the research hold the potential to transform agricultural practices by introducing a more effective, data-driven approach to plant breeding management.
... In addition, Asseng et al [37] and Wall et al [38] have shown a robust linear relationship between wheat yield and seasonal mean temperature. Other crop yield models for Argentina used additional variables like soil moisture [30], large-scale climate indices like the El Niño-Southern Oscillation index [21,39] or remotely sensed quantities like land surface temperature and vegetation indices [28,40]. Our objective here is an isolated analysis of the inseason forecast accuracy coming from a MME (see section 2.3). ...
While multi-model ensembles (MMEs) of seasonal climate models (SCMs) have been used for crop yield forecasting, there has not been a systematic attempt to select the most skillful SCMs to optimize the performance of a MME and improve in-season yield forecasts. Here, we propose a statistical model to forecast regional and national wheat yield variability from 1993–2016 over the main wheat production area in Argentina. Monthly mean temperature and precipitation from the four months (August–November) before harvest were used as features. The model was validated for end-of-season estimation in December using reanalysis data (ERA) from the European Centre for Medium-Range Weather Forecasts (ECMWF) as well as for in-season forecasts from June to November using a MME of three SCMs from 10 SCMs analyzed. A benchmark model for end-of-season yield estimation using ERA data achieved a R² of 0.33, a root-mean-square error (RMSE) of 9.8% and a receiver operating characteristic (ROC) score of 0.8 on national level. On regional level, the model demonstrated the best estimation accuracy in the northern sub-humid Pampas with a R² of 0.5, a RMSE of 12.6% and a ROC score of 0.9. Across all months of initialization, SCMs from the National Centers for Environmental Prediction, the National Center for Atmospheric Research and the Geophysical Fluid Dynamics Laboratory had the highest mean absolute error of forecasted features compared to ERA data. The most skillful in-season wheat yield forecasts were possible with a 3-member-MME, combining data from the SCMs of the ECMWF, the National Aeronautics and Space Administration and the French national meteorological service. This MME forecasted wheat yield on national level at the beginning of November, one month before harvest, with a R² of 0.32, a RMSE of 9.9% and a ROC score of 0.7. This approach can be applied to other crops and regions.
... Study [106] used CNN and LSTM with Gaussian procedure to foresee the crop yield using 3D histograms produced from images acquired from remotely sensed SR, land cover, and LST. Study [107] extended the work by testing the ability of a trained model in one area to be transferred to another region. They initialized the LSTM model with NN parameters trained on the Argentina dataset and replaced the last dense layer with the untrained dense layer before training the model again on Brazil data. ...
The growing global requirement for food and the need for sustainable farming in an era of a changing climate and scarce resources have inspired substantial crop yield prediction research. Deep learning (DL) and machine learning (ML) models effectively deal with such challenges. This research paper comprehensively analyses recent advancements in crop yield prediction from January 2016 to March 2024. In addition, it analyses the effectiveness of various input parameters considered in crop yield prediction models. We conducted an in-depth search and gathered studies that employed crop modeling and AI-based methods to predict crop yield. The total number of articles reviewed for crop yield prediction using ML, meta-modeling (Crop models coupled with ML/DL), and DL-based prediction models and input parameter selection is 125. We conduct the research by setting up five objectives for this research and discussing them after analyzing the selected research papers. Each study is assessed based on the crop type, input parameters employed for prediction, the modeling techniques adopted, and the evaluation metrics used for estimating model performance. We also discuss the ethical and social impacts of AI on agriculture. However, various approaches presented in the scientific literature have delivered impressive predictions, they are complicated due to intricate, multifactorial influences on crop growth and the need for accurate data-driven models. Therefore, thorough research is required to deal with challenges in predicting agricultural output.
... Satellite imaging is a very useful technique for monitoring the natural phenomena and human activities on the surfaces of the Earth. Lots of applications rely on satellite images such as crop monitoring, weather forecasting, urban planning, wildfire management and so on [1][2][3][4][5][6][7][8]. However, the acquisition of satellite images can be very expensive and the spatial and temporal resolution (the frequency that a satellite image is captured) may be limited due to the physical constraints of sensors [1]. ...
During the acquisition of satellite images, there is generally a trade-off between spatial resolution and temporal resolution (acquisition frequency) due to the onboard sensors of satellite imaging systems. High-resolution satellite images are very important for land crop monitoring, urban planning, wildfire management and a variety of applications. It is a significant yet challenging task to achieve high spatial-temporal resolution in satellite imaging. With the advent of diffusion models, we can now learn strong generative priors to generate realistic satellite images with high resolution, which can be utilized to promote the super-resolution task as well. In this work, we propose a novel diffusion-based fusion algorithm called \textbf{SatDiffMoE} that can take an arbitrary number of sequential low-resolution satellite images at the same location as inputs, and fuse them into one high-resolution reconstructed image with more fine details, by leveraging and fusing the complementary information from different time points. Our algorithm is highly flexible and allows training and inference on arbitrary number of low-resolution images. Experimental results show that our proposed SatDiffMoE method not only achieves superior performance for the satellite image super-resolution tasks on a variety of datasets, but also gets an improved computational efficiency with reduced model parameters, compared with previous methods.
... Many scholars have proposed feasible and effective NN structures to achieve superior large-scale crop yield estimation performance. For example, Spiking Neural Network (SNN) [5], Deep Gaussian Process (DGP) [6], Deep Transfer Learning (DTL) [7], etc. In particular, [6] added DGP to convolutional neural networks (CNNs) and long and short-term memory (LSTM) networks and outperformed all competing methods. ...
Accurately identifying the growth stages of winter wheat is essential for effective crop management. The timing of each growth stage depends on various factors, including the sowing date, the type of winter wheat, and unpredictable factors such as precipitation, solar radiation, and temperature. Estimating the start of each stage can be challenging. However, deep learning models have proven to be effective in many agriculture-related applications, such as yield prediction, disease control, and nutrient recommendation.
This study proposes the use of a convolutional neural network (CNN) model that can predict the emergence and stem elongation time of winter wheat by identifying tiny color changes. The model is trained on satellite images captured over a 3-year period. The proposed CNN model can achieve a prediction accuracy of up to 93.75% for stem elongation stage time and up to 91.75% for winter wheat emergence stage time. The proposed model outperforms other machine learning methods, such as SVM and Random Forest used for similar tasks in previous studies. The grayscale index is used as the input of the machine learning models. It can achieve 70% to 74.5% accuracy in the prediction of emergence and 82.5% to 84.4% accuracy in the prediction of stem elongation.
... Thus, selecting the appropriate model for crop yield prediction modeling is challenging. People use the CNN method to convert time remote sensing images into a histogram to model the spectral time relationship [113]. Additionally, some researchers have proposed applying spatial statistical models [114] to crop yield prediction, such as geographically weighted regression. ...
With the rapid development of data acquisition and storage technology, spatio-temporal (ST) data in various fields are growing explosively, so many ST prediction methods have emerged. The review presented in this paper mainly studies the prediction of ST series. We propose a new taxonomy organized along three dimensions: ST series prediction methods (focusing on time feature learning, focusing on spatial feature learning, and focusing on spatial–temporal feature learning), techniques of ST series prediction (the RNN-, CNN-, and transformer-based models, as well as the CNN-based-composite model and GNN-based-composite models, and the miscellaneous model) and ST series prediction results (single target and multi-target). We first introduce and explain each dimension of the taxonomy in detail. After providing this three-dimensional view, we comprehensively review and compare the recent related ideas in the literature and analyze their advantages and limitations. Moreover, we summarize the key information of the existing literature and provide guidance for researchers to select suitable models. Second, we summarize the different applications of deep learning models in ST series prediction based on current literature and list relevant datasets and download links per application classifications. Lastly, we comprehensively analyze the current innovation and challenges and suggest future directions for researching ST series prediction after comparing and analyzing the computing performance of these forecasting models. In addition, each method or model solves one aspect of the challenge, which means that two or more methods should be combined to solve more challenges at the same time. We hope this article provides readers a broader and deeper understanding of the field of ST series research.
... This approach is advantageous, as it leverages existing models and data to quickly adapt to the characteristics of the target domain. In the context of deep neural networks, initial layers tend to capture general features, whereas later layers are more focused on specific tasks [81][82][83]. Therefore, in GNSS-R SM retrieval studies, researchers can freeze the weights of the initial layers of pre-trained DL architecture and fine-tune or retrain the last few layers. ...
Soil moisture (SM) is an important parameter in water cycle research. Rapid and accurate monitoring of SM is critical for hydrological and agricultural applications, such as flood detection and drought characterization. The Global Navigation Satellite System (GNSS) uses L-band microwave signals as carriers, which are particularly sensitive to SM and suitable for monitoring it. In recent years, with the development of Global Navigation Satellite System–Reflectometry (GNSS-R) technology and data analysis methods, many studies have been conducted on GNSS-R SM monitoring, which has further enriched the research content. However, current GNSS-R SM inversion methods mainly rely on auxiliary data to reduce the impact of non-target parameters on the accuracy of inversion results, which limits the practical application and widespread promotion of GNSS-R SM monitoring. In order to promote further development in GNSS-R SM inversion research, this paper aims to comprehensively review the current status and principles of GNSS-R SM inversion methods. It also aims to identify the problems and future research directions of existing research, providing a reference for researchers. Firstly, it introduces the characteristics, usage scenarios, and research status of different GNSS-R SM observation platforms. Then, it explains the mechanisms and modeling methods of various GNSS-R SM inversion research methods. Finally, it highlights the shortcomings of existing research and proposes future research directions, including the introduction of transfer learning (TL), construction of small models based on spatiotemporal analysis and spatial feature fusion, and further promoting downscaling research.
... The approach of learning by transfer is used in various applications, such as the prediction of the performance of numerous crops worldwide, where Wang et al. [19] used remote sensors with satellite images to estimate the outcome of soy crops through algorithms and deep learning, offering an inexpensive and efficient alternative in comparison to conventional techniques that are generally expensive and difficult to expand in regions with limited access to data. ...
The early and precise identification of the different phenological stages of the bean (Phaseolus vulgaris L.) allows for the determination of critical and timely moments for the implementation of certain agricultural activities that contribute in a significant manner to the output and quality of the harvest, as well as the necessary actions to prevent and control possible damage caused by plagues and diseases. Overall, the standard procedure for phenological identification is conducted by the farmer. This can lead to the possibility of overlooking important findings during the phenological development of the plant, which could result in the appearance of plagues and diseases. In recent years, deep learning (DL) methods have been used to analyze crop behavior and minimize risk in agricultural decision making. One of the most used DL methods in image processing is the convolutional neural network (CNN) due to its high capacity for learning relevant features and recognizing objects in images. In this article, a transfer learning approach and a data augmentation method were applied. A station equipped with RGB cameras was used to gather data from images during the complete phenological cycle of the bean. The information gathered was used to create a set of data to evaluate the performance of each of the four proposed network models: AlexNet, VGG19, SqueezeNet, and GoogleNet. The metrics used were accuracy, precision, sensitivity, specificity, and F1-Score. The results of the best architecture obtained in the validation were those of GoogleNet, which obtained 96.71% accuracy, 96.81% precision, 95.77% sensitivity, 98.73% specificity, and 96.25% F1-Score.
... Other papers relating DL to yield estimation models in general have been published in recent years, and they were designed mainly for corn, soybean, and wheat [94][95][96][97][98][99][100][101][102], among other crops, and hence, they were excluded from our review. These studies are still limited in number, for this is still an incipient area in the field of yield estimation and forecasting, a point also noted in the systematic reviews of [13,103,104]. ...
The sugarcane crop has great socioeconomic relevance because of its use in the production of sugar, bioelectricity, and ethanol. Mainly cultivated in tropical and subtropical countries, such as Brazil, India, and China, this crop presented a global harvested area of 17.4 million hectares (Mha) in 2021. Thus, decision making in this activity needs reliable information. Obtaining accurate sugarcane yield estimates is challenging, and in this sense, it is important to reduce uncertainties. Currently, it can be estimated by empirical or mechanistic approaches. However, the model’s peculiarities vary according to the availability of data and the spatial scale. Here, we present a systematic review to discuss state-of-the-art sugarcane yield estimation approaches using remote sensing and crop simulation models. We consulted 1398 papers, and we focused on 72 of them, published between January 2017 and June 2023 in the main scientific databases (e.g., AGORA-FAO, Google Scholar, Nature, MDPI, among others), using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology. We observed how the models vary in space and time, presenting the potential, challenges, limitations, and outlooks for enhancing decision making in the sugarcane crop supply chain. We concluded that remote sensing data assimilation both in mechanistic and empirical models is promising and will be enhanced in the coming years, due to the increasing availability of free Earth observation data.
... They used histogram representation of MODIS SR, LST, and Land Cover data. They also stated that the transfer learning (TL) approach improves the results [18]. ...
Predicting agricultural yields is imperative for effective planning to sustain the growing global population. Traditionally, regression-based, simulation-based, and hybrid methods were employed for yield prediction. In recent times, there has been a notable shift towards the adoption of Machine Learning (ML) methods, with Deep Learning (DL), particularly Convolutional Neural Networks (CNNs) and Long-Short Term Memory (LSTM) networks, emerging as popular choices for their enhanced predictive accuracy. This research introduces a cost-effective DL architecture tailored for corn yield prediction, considering computational efficiency in processing time, data size, and NN architecture complexity. The proposed architecture, named SEDLA (Simple and Efficient Deep Learning Architecture), leverages the spatial and temporal learning capabilities of CNNs and LSTMs, respectively, with a unique emphasis on exploring the impact of kernel size in CNNs. Simultaneously, the study aims to exclusively employ satellite and yield data, strategically minimizing input variables to enhance the model's simplicity and efficiency. Notably, the study demonstrates that employing larger kernel sizes in CNNs, especially when processing histogram-based Surface Reflectance (SR) and Land Surface Temperature (LST) data from Moderate Resolution Imaging Spectroradiometer (MODIS), allows for a reduction in the number of hidden layers. The efficacy of the architecture was evaluated through extensive testing on corn yield prediction across 13 states in the United States (U.S.) Corn Belt at county-level. The experimental results showcase the superiority of the proposed architecture, achieving a Mean Absolute Percentage Error (MAPE) of 6.71 and Root Mean Square Error (RMSE) of 14.34, utilizing a single-layer CNN with a 15x15 kernel in conjunction with LSTM. These outcomes surpass existing benchmarks in the literature, affirming the efficacy and potential of the suggested DL framework for accurate and efficient crop yield predictions.
... Khaki and Wang et al. designed a deep neural network model ( Fig. 1) for predicting corn yields at 2247 locations from 2008 to 2016 [45]. Wang et al. designed a deep learning framework to predict soybean crop yields in Argentina, and they also achieved satisfactory results by using transfer learning methods to predict soybean yields in Brazil with less data [46]. The key to a deep neural network model is that it does not require the specification of appropriate functions to fit the relationships between the data. ...
Background
Camellia oleifera, an essential woody oil tree in China, propagates through grafting. However, in production, it has been found that the interaction between rootstocks and scions may affect fruit characteristics. Therefore, it is necessary to predict fruit characteristics after grafting to identify suitable rootstock types.
Methods
This study used Deep Neural Network (DNN) methods to analyze the impact of 106 6-year-old grafting combinations on the characteristics of C.oleifera, including fruit and seed characteristics, and fatty acids. The prediction of characteristics changes after grafting was explored to provide technical support for the cultivation and screening of specialized rootstocks. After determining the unsaturated fat acids, palmitoleic acid C16:1, cis-11 eicosenoic acid C20:1, oleic acid C18:1, linoleic acid C18:2, linolenic acid C18:3, kernel oil content, fruit height, fruit diameter, fresh fruit weight, pericarp thickness, fresh seed weight, and the number of fresh seeds, the DNN method was used to calculate and analyze the model. The model was screened using the comprehensive evaluation index of Mean Absolute Error (MAPE), determinate correlation R² and and time consumption.
Results
When using 36 neurons in 3 hidden layers, the deep neural network model had a MAPE of less than or equal to 16.39% on the verification set and less than or equal to 13.40% on the test set. Compared with traditional machine learning methods such as support vector machines and random forests, the DNN method demonstrated more accurate predictions for fruit phenotypic characteristics, with MAPE improvement rates of 7.27 and 3.28 for the 12 characteristics on the test set and maximum R² improvement values of 0.19 and 0.33. In conclusion, the DNN method developed in this study can effectively predict the oil content and fruit phenotypic characteristics of C. oleifera, providing a valuable tool for predicting the impact of grafting combinations on the fruit of C. oleifera.
The rapid and accurate acquisition of soil moisture (SM) information is essential. Although Unmanned Aerial Vehicle (UAV) remote sensing technology has made significant advancements in SM monitoring, existing studies predominantly focus on developing models tailored to specific regions. The transferability of these models across different regions remains a considerable challenge. Therefore, this study proposes a transfer learning-based framework, using two representative small agricultural watersheds (Hongxing region and Woniutu region) in Northeast China as case studies. This framework involves pre-training a model on a source domain and fine-tuning it with a limited set of target domain samples to achieve high-precision SM inversion. This study evaluates the performance of three algorithms: Random Forest (RF), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) network. Results show that the fine-tuned model significantly mitigates the decline in prediction accuracy caused by regional differences. The fine-tuned LSTM model achieved the highest retrieval accuracy, with the following results: 10% samples (R = 0.615, RRMSE = 15.583%), 30% samples (R = 0.682, RRMSE = 13.97%), and 50% samples (R = 0.767, RRMSE = 16.321%). Among these models, the LSTM model exhibited the most significant performance improvement and the best transferability. This study underscores the potential of transfer learning for enhancing cross-regional SM monitoring and providing valuable insights for future UAV-based SM monitoring.
Remotely sensed geospatial data are critical for applications including precision agriculture, urban planning, disaster monitoring and response, and climate change research, among others. Deep learning methods are particularly promising for modeling many remote sensing tasks given the success of deep neural networks in similar computer vision tasks and the sheer volume of remotely sensed imagery available. However, the variance in data collection methods and handling of geospatial metadata make the application of deep learning methodology to remotely sensed data nontrivial. For example, satellite imagery often includes additional spectral bands beyond red, green, and blue and must be joined to other geospatial data sources that may have differing coordinate systems, bounds, and resolutions. To help realize the potential of deep learning for remote sensing applications, we introduce TorchGeo, a Python library for integrating geospatial data into the PyTorch deep learning ecosystem. TorchGeo provides data loaders for a variety of benchmark datasets, composable datasets for uncurated geospatial data sources, samplers for geospatial data, and transforms that work with multispectral imagery. TorchGeo is also the first library to provide pre-trained models for multispectral satellite imagery (e.g., models that use all bands from the Sentinel-2 satellites), allowing for advances in transfer learning on downstream remote sensing tasks with limited labeled data. We use TorchGeo to create reproducible benchmark results on existing datasets and benchmark our proposed method for preprocessing geospatial imagery on the fly. TorchGeo is open source and available on GitHub: https://github.com/microsoft/torchgeo.
Crop breeders often face challenges due to limited data availability when making crucial decisions, such as selecting top-performing varieties/hybrids for further experiments, registration, and commercialization. Evaluating all varieties/hybrids across all fields is impractical due to high experimental and time costs, as well as the limited number of locations for planting. This article aims to evaluate the performance of various maize hybrids in untested locations using historical data. The problem is approached through a matrix framework, where hybrids and fields correspond to rows and columns, respectively, with entries representing the yield of a specific hybrid at a given location. As this matrix is typically sparse, the task is to fill in missing data. Agronomists are primarily interested in the performance of top hybrids at specific locations for smart seed selection. To address this, we introduce a novel application of the Data Fusion by Matrix Factorization (DFMF) algorithm for predicting crop yields using maize data from the 2019 Syngenta Crop Challenge. The DFMF results are compared with the Random Forest (RF) algorithm as a benchmark, focusing on model performance for smart seed selection. Our analysis highlights the advantages of the DFMF approach over the traditional RF method in this context.
Spatio-temporal prediction tasks play a crucial role in facilitating informed decision-making through anticipatory insights. By accurately predicting future outcomes, the ability to strategize, preemptively address risks, and minimize their potential impact is enhanced. The precision in forecasting spatial and temporal patterns holds significant potential for optimizing resource allocation, land utilization, and infrastructure development. While existing review and survey papers predominantly focus on specific forecasting domains such as intelligent transportation, urban planning, pandemics, disease prediction, climate and weather forecasting, environmental data prediction, and agricultural yield projection, limited attention has been devoted to comprehensive surveys encompassing multiple objects concurrently. This paper addresses this gap by comprehensively analyzing techniques employed in traffic, pandemics, disease forecasting, climate and weather prediction, agricultural yield estimation, and environmental data prediction. Furthermore, it elucidates challenges inherent in spatio-temporal forecasting and outlines potential avenues for future research exploration.
Climate change poses a significant threat to global food security and livelihoods, weakening the resilience of global agriculture. Emerging fields like AI in agriculture enhance efficiency by automating tasks like crop monitoring and irrigation, leading to improved yields and resource optimization. In addition, AI-powered analytics enable data-driven decision-making, helping farmers respond proactively to environmental changes and optimize crop management strategies. This chapter delves into essential statistical modelling approaches crucial for climate-resilient agriculture, emphasizing the pivotal role of advanced data analytics in the agricultural sector. Through a series of case studies, the chapter illustrates the application of machine learning (ML) models, remote sensing, geographical information systems (GIS), and enviromics across diverse agro-ecological contexts. These applications encompass climate data analysis, crop predictions, spatial analysis of climate data, enviromics for breeding data, water resource management, the design of climate resilient farming systems, which in turn helps in drought-resistant crop breeding, disease diagnosis, soil health monitoring, and yield optimization. Moreover, the chapter provides an overview of contemporary state-of-the-art statistical models in the realm of climate-resilient agriculture.
Remote sensing technology has many applications in agriculture, including monitoring and analyzing local food systems in urban areas. In urban environments, space is often limited, and food is often produced on small plots of land; this makes it challenging to monitor and assess the condition of the crops using traditional methods. Traditional methods of predicting local food availability may not provide as much detail nor consider factors such as the specific weather conditions in a particular location or the health and growth of individual crops. Remote sensing technology can provide more detailed and specific information about these factors, allowing for more accurate predictions about local food availability. One of the critical applications of remote sensing for local food systems in urban areas is detecting and monitoring farmland and crops. These methods are an essential part of land use planning and policy recommendations. Crop yield prediction can be measured by applying satellite remote sensing to determine physiological conditions during the growing season. Machine learning has shown to be effective in a variety of data-driven applications. This study used Sentinel-2 images from March to October 2018 to map the normalized difference vegetation index (NDVI) and leaf area index (LAI) via satellite. Different growth stages were observed. The generated models were validated by linear regression and random forest algorithms. The results showed that the random forest method had good accuracy. NDVI had the highest accuracy (R2 = 0.90) in September 2018 and October 2018; however, the accuracy of LAI was very high (R2 = 0.91) as of July 2018. These results presented the late growth stage had good accuracy for estimating local crop production as crops reached the peak vegetative periods. Thus, machine learning provides reliable tracking of local food production at local and regional levels. The results show that vegetation indices can be used to calculate site-specific local crop management and predict yield. These integrated models can be used for logistics and decision-making related to local agricultural production.
Forecasting agricultural variables at national level is crucial for crafting policies aimed at ensuring food security and enhancing the stability of the entire agricultural food chain. However, this task is often daunting due to sparse data and the rarity of extensive records. This study seeks to assess the effectiveness of ten deep learning models in predicting monthly milk production in France, Germany, and Italy. It uses climatic and economic data from open datasets as input. The findings suggest that deep learning models offer a superior alternative to traditional statistical methods, yielding robust results without the need for complex model architectures. Additionally, the predominant autoregressive nature of milk production underscores the limitations of environmental variables in capturing external influences such as milk quotas. Despite this, the models demonstrate high accuracy, paving the way for potential applications in accurately forecasting national-scale agricultural variables. This could have implications for the development of dairy insurance products and risk management strategies.
The lack of reliable data in developing countries is a major obstacle towards
sustainable development, food security, and disaster relief. Poverty data, for
example, is typically scarce, sparse in coverage, and labor-intensive to
obtain. Remote sensing data such as high-resolution satellite imagery, on the
other hand, is becoming increasingly available and inexpensive. Unfortunately,
such data is highly unstructured and we lack techniques to automatically
extract useful insights to inform policy decisions and help direct humanitarian
efforts. We propose a novel machine learning approach to extract large-scale
socioeconomic indicators from high-resolution satellite imagery. The main
challenge is that training data is very scarce, making it difficult to apply
modern techniques such as Convolutional Neural Networks (CNN). We therefore
propose a transfer learning approach where nighttime light intensities are used
as a data-rich proxy. We train a fully convolutional CNN model to predict
nighttime lights from daytime imagery, simultaneously learning features that
are useful for poverty prediction. The model learns filters identifying
different terrains and man-made structures, including roads, buildings, and
farmlands, without any supervision beyond nighttime lights. We demonstrate that
these learned features are highly informative for poverty mapping, even
approaching the predictive performance of survey data collected in the field.
Agricultural monitoring, especially in developing countries, can help prevent famine and support humanitarian efforts. A central challenge is yield estimation, i.e., predicting crop yields before harvest. We introduce a scalable, accurate, and inexpensive method to predict crop yields using publicly available remote sensing data. Our approach improves existing techniques in three ways. First, we forego hand-crafted features traditionally used in the remote sensing community and propose an approach based on modern representation learning ideas. We also introduce a novel dimensionality reduction technique that allows us to train a Convolutional Neural Network or Long-short Term Memory network and automatically learn useful features even when labeled training data are scarce. Finally, we incorporate a Gaussian Process component to explicitly model the spatio-temporal structure of the data and further improve accuracy. We evaluate our approach on county-level soybean yield prediction in the U.S. and show that it outperforms competing techniques.
Measuring consumption and wealth remotely
Nighttime lighting is a rough proxy for economic wealth, and nighttime maps of the world show that many developing countries are sparsely illuminated. Jean et al. combined nighttime maps with high-resolution daytime satellite images (see the Perspective by Blumenstock). With a bit of machine-learning wizardry, the combined images can be converted into accurate estimates of household consumption and assets, both of which are hard to measure in poorer countries. Furthermore, the night- and day-time data are publicly available and nonproprietary.
Science , this issue p. 790 ; see also p. 753
New advances in satellite data acquisition and processing offer promise for monitoring agricultural lands globally. Using these data to estimate crop yields for individual fields would benefit both crop management and scientific research, especially for areas where reliable ground-based estimates are not currently made. Here we introduce a generalized approach for mapping crop yields with satellite data and test its predictions for yields across more than 17,000 maize fields and 11,000 soybean fields spanning multiple states and years in the Midwestern United States. The method, termed SCYM (a scalable satellite-based crop yield mapper), uses crop model simulations to train statistical models for different combinations of possible image acquisition dates, and these are then applied to Landsat and gridded weather data within the Google Earth Engine platform, where the Landsat is composited to find the “best” dates of observations on a pixel-by-pixel basis. SCYM estimates successfully captured a significant fraction of maize yield variation in all state-years, with a range of 14–58% and an average of 35% for this particular study region and crop. Similar results were observed for soybean, with an average of 32% of yield variation captured. The multi-year yield estimates were also used to examine the temporal persistence of yield advantages for the top yielding fields in different counties, which is one measure of how important factors such as farmer skill are in explaining yield gaps. The strength of the SCYM approach lies in its ability to leverage physiological knowledge embedded in crop models to interpret satellite observations in a scalable way, as it can be readily applied to new crops, regions, and types and timing of remote sensing observations without the need for ground calibration.
This paper investigates the potential of using vegetation index profiles from AVHRR data to monitor crop yield. A retrospective study and predictive study is presented, for 1986-1988 and 1989 respectively, for an area in Northern Greece. Results are encouraging for operational crop monitoring. Yield for wheat, cotton, rice and maize crops has been estimated to a high degree of accuracy using a simple linear relationship between NDVI and yield. However input from an agro-meteorological model is recommended to modify the model during the grain-filling period of the wheat crop. The estimates stabilize 50-100 days prior to harvest enabling an early assessment of crop yield to be made.
We used data from NASA's Moderate Resolution Imaging Spectroradiometer (MODIS) in association with county-level data from the United States Department of Agriculture (USDA) to develop empirical models predicting maize and soybean yield in the Central United States. As part of our analysis we also tested the ability of MODIS to capture inter-annual variability in yields. Our results show that the MODIS two-band Enhanced Vegetation Index (EVI2) provides a better basis for predicting maize yields relative to the widely used Normalized Difference Vegetation Index (NDVI). Inclusion of information related to crop phenology derived from MODIS significantly improved model performance within and across years. Surprisingly, using moderate spatial resolution data from the MODIS Land Cover Type product to identify agricultural areas did not degrade model results relative to using higher-spatial resolution crop-type maps developed by the USDA. Correlations between vegetation indices and yield were highest 65–75 days after greenup for maize and 80 days after greenup for soybeans. EVI2 was the best index for predicting maize yield in non-semi-arid counties (R2 = 0.67), but the Normalized Difference Water Index (NDWI) performed better in semi-arid counties (R2 = 0.69), probably because the NDWI is sensitive to irrigation in semi-arid areas with low-density agriculture. NDVI and EVI2 performed equally well predicting soybean yield (R2 = 0.69 and 0.70, respectively). In addition, EVI2 was best able to capture large negative anomalies in maize yield in 2005 (R2 = 0.73). Overall, our results show that using crop phenology and a combination of EVI2 and NDWI have significant benefit for remote sensing-based maize and soybean yield models.
NOAA: How the failed 2014-15 El Nino fueled the strong 2015-16 El Nino
- Michael Mcphaden
- Aaron Levine
Michael McPhaden and Aaron Levine. 2016. NOAA: How the failed 2014-15
MOD09A1 MODIS/Terra Surface Reflectance 8-Day L3 Global 500m SIN Grid V006
- Eric Vermote
Eric Vermote. 2015. MOD09A1 MODIS/Terra Surface Reflectance 8-Day L3
Global 500m SIN Grid V006. (2015). https://doi.org/10.5067/modis/mod09a1.
006
MYD11A2 MODIS/Aqua Land Surface Temperature/Emissivity 8-Day L3 Global 1km SIN Grid V006
- Zhengming Wan
- S Hook
Zhengming Wan and S. Hook. 2015. MYD11A2 MODIS/Aqua Land Surface
Temperature/Emissivity 8-Day L3 Global 1km SIN Grid V006. (2015). https:
//doi.org/10.5067/modis/myd11a2.006
Instituto Brasileiro de Geografia e Estatistica
- Brasil Sistema
- Ibge De Recuperacao Automatica
Brasil Sistema IBGE de Recuperacao Automatica, Instituto Brasileiro de Geografia e Estatistica. [n. d.].
Eric Vermote. 2015. MOD09A1 MODIS/Terra Surface Reflectance 8-Day L3 Global 500m SIN Grid V006
- Eric Vermote
Zhengming Wan and S. Hook. 2015. MYD11A2 MODIS/Aqua Land Surface Temperature/Emissivity 8-Day L3 Global 1km SIN Grid V006
- Zhengming Wan
- S Hook