Progress 01/15/24 to 01/14/25
Outputs
Target Audience:In our study, we designed and implemented innovative models and tools aimed at delivering immediate, in-season crop yield predictions for two primary agricultural commodities: corn and soybeans. These predictions, tailored for county-level accuracy across the Continental United States (CONUS), utilize cutting-edge remote sensing and advanced deep learning techniques. The yield data will be made accessible to the public at no cost via an online cyber-platform. This initiative holds significant value for a diverse range of users, including US farmers who rely on it for critical operational decisions, insurance firms for claims processing and compliance evaluation, agricultural finance entities for market analysis and business support, and agribusinesses for optimizing production, distribution, processing, and marketing strategies. Additionally, the methodologies, algorithms, and yield predictions developed in this study will be shared with the USDA National Agricultural Statistics Service (NASS), enhancing NASS crop yield estimation operations. The tool we propose has the potential to supplement the existing NASS county crop yield reports, elevating operational efficiency and data accuracy while bolstering USDA NASS's capabilities in crop yield prediction. Consequently, this project promises to enrich NASS's current crop yield estimation efforts and exert a profound impact on a wide array of stakeholders. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?In PI Zhang's group: One PhD student has been working on remote sensing data processing and machine learning model development for corn yield prediction. One manuscript is under review with the student as the first leading author. Inco-PI Huang's group: APh.D. graduate student at the department of Geography is being trained to work on the project. Shehas been working on real-time data acquisition and storage using Google Cloud services, and real-time crop prediction and modeling with Google Vertex AI platform. She has also been working on machine learning model development for high resolution remote sensing image analytics for change detections. She was able to present her research in the 7th ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery. Songxi has published one journal paper and one conference paper with two manuscripts under review as the first leading author. Another M.S. Studentat the Department of Geography has been working on developing innovative methods for change detection with remote sensing imagery. In addition, an undergraduate studentat the department of Compute Science and Geography (double major) has been working on the spatial web portal (client) development. She has co-authored on one manuscript under review. How have the results been disseminated to communities of interest?The project results have been disseminated through journal publications, invited research seminars and international conferences, such as the American Geophysical Union (AGU) fall meeting, and ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery. Some of the research results have also been incorporated into the class of PIs Zhang and Huang for teaching and student training activities. Through Co-PI Yang, the project progress has also been reported to stakeholder USDA NASS. The results were also cross disseminated to another NASA funded project team for collaboration. The developed Bayesian neural network (BNN) based crop yield prediction model will be coupled with NASA Land Information System (LIS) for crop yield prediction. The final implementation will be open source and available to general public along with NASA project deliverables. The implemented algorithm and models will be disseminated at national and international conferences, proceedings, and journals. All software implementations will be open-sourced and available at GitHub. What do you plan to do during the next reporting period to accomplish the goals?1. Develop novel yield prediction framework to integrate process-based model and machine learning technique to further enhance prediction performance 2. Apply strategies such as partial domain adaptation and multiple instance regression to further improve the model's learning abilities 3. Further assessment of the developed algorithms by comparing the prediction results with NASS remote sensing model results 4. Deploy the spatial web port and web services to a server supervised by Co-PD Huang, and hosted in the department of Geography. 5. Integrate the cloud enabled data processing, modeling, and model prediction component into USDA NASS's crop yield estimation and dissemination system for operational test. 6. Test the model deployed at USDA NASS for production and further refine the model for final production operation.
Impacts
What was accomplished under these goals?
Objective 1: We proposed a Knowledge-Guided Machine Learning with Soil Moisture (KGML-SM) framework for county-level corn yield prediction. First, to address the limitations of previous Knowledge- Guided Machine Learning (KGML) methods, we integrated soil moisture as an intermediary variable, emphasizing its critical role in plant growth and drought response. This approach allowed us to incorporate both remote sensing data and domain knowledge, leveraging soil moisture to improve model performance, particularly in counties with limited reported yield data. Furthermore, we explored the relationships between drought, soil moisture, and corn yield, assessing the importance of various features and analyzing the impact of soil moisture on yield predictions across different regions and time periods. The developed KGML-SM model demonstrated superior performance compared to traditional machine learning models, providing complementary insights to existing yield prediction frameworks and offering robust predictions even in drought-affected areas. Objective 2:We have already accomplished this objective in previous years. Objective 3: We have completed the development of the prototype of the cyber-agricultural system for automating near real-time crop yield prediction and dissemination. The system is designed to provide an end-to-end solution for data-driven agricultural analysis with three integrated components (tasks): (T1) Real-time data acquisition and storage, responsible for archiving and retrieving datasets. An automatic system is developed to preprocess, integrate, and download unstructured, heterogeneous data from GEE using Google Cloud services; (T2) Real-time crop prediction and modeling with Google Vertex AI platform providing components to customize various machine learning models; and (T3) Spatial web portal , providing the key functions to support data search, analysis, visualization, and animation service requested from end users through geovisualization or animation with interactive tool.In particular, the spatial web portal (T3), developed using Leaflet and D3 libraries in JavaScript, is deployed on Github and accessible through https://scdmlab.github.io/cyber-agricultural-system/. In addition, we also implemented web services for enhancing the data and model output access through the third party.
Publications
- Type:
Peer Reviewed Journal Articles
Status:
Published
Year Published:
2024
Citation:
Xu, Y., Zhou, J., & Zhang, Z. (2024). A new Bayesian semi-supervised active learning framework for large-scale crop mapping using Sentinel-2 imagery. ISPRS Journal of Photogrammetry and Remote Sensing, 209, 17-34.
- Type:
Peer Reviewed Journal Articles
Status:
Published
Year Published:
2024
Citation:
Xu, Y., Ebrahimy, H., & Zhang, Z. (2024). Bayesian Joint Adaptation Network for Crop Mapping in the Absence of Mapping Year Ground-Truth Samples. IEEE Transactions on Geoscience and Remote Sensing.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2024
Citation:
Wang, X., Ebrahimy, H., Yang, Z., Zhang, Z., Bindlish, R., Huang, Y., Feng, G., Kumar, S., & Ruane, A. C. (2024). Developing a Novel Knowledge-Guided Deep Learning Algorithm for County Level Crop Yield Prediction in the Face of Climate Change in the US Midwest. AGU Fall Meeting Abstracts, 2024.
|
Progress 01/15/23 to 01/14/24
Outputs
Target Audience:In this project, we developed models and tools to provide near real-time in-season crop yield predictions for two mainly commodity crops (i.e., corn and soybeans) at the county-level for the CONUS, based on state-of-the-art remote sensing and deep learning technologies. The predicted yield information will be disseminated through a web-based cyber-platform for free public use, which is critical for a variety of stakeholders, such as for US producers to make operational decisions, for insurance companies to help with claim and compliance assessment, for agricultural finance companies to analyze the market and support businesses decision makings, and for agricultural business to better plan production, distribution, processing, and marketing. The developed methods, algorithms, and predicted yield information will also be delivered to USDA National Agricultural Statistics Service (NASS) for improving NASS crop yield estimation operation. The proposed tool could potentially provide supplemental information to the current NASS county crop yield estimation report, improve the operational efficiency and data quality, and enhance USDA NASS crop yield prediction capability. Therefore, the proposed project will add value to NASS's existing crop yield estimation program and have broad impacts on many stakeholders. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?In PI Zhang's group: One PhD was supported by this project and has graduated. He is currently working as a Postdoc scholar at Stanford University. Another PhD student is currently being trained under this project. One manuscript is under review with the student as the first leading author. In co-PI Huang's group: A female Ph.D. graduate student (Meiliu) at the department of Geography was trained to work on the project. Meiliu has been working on street view data processing and machine learning model development for high resolution land use mapping. She was able to present her research in multiple conferences, including the 2023 30th International Conference on Geoinformatics, and the 2023 Association of American Geographer (AAG) Annual Meeting. One journal paper and one conference paper have been published with Meiliu as the first leading author.Another female Ph.D. graduate student (Songxi Yang) is currently being trained and working the data processing and downloading automation, and modeling running through the cloud computing techniques. One manuscript on using cyberinfrastructure to enable automated crop yield prediction and dissemination is under preparation with Songxi as the first leading author. In addition, one undergraduate (Haiyue Liu) is also currently being trained under this project on web portal interface development. How have the results been disseminated to communities of interest?The project results have been disseminated through journal publications, invited research seminars and international conferences, such as the American Geophysical Union (AGU) fall meeting, 2023 Spatial Data Science Symposium, and AAG annual meeting. Some of the research results have also been incorporated into the class of PIs Zhang and Huang for teaching and student training activities. Through Co-PI Yang, the project progress has also been reported to stakeholder USDA NASS. The results were also cross disseminated to another NASA funded project team for collaboration. The developed Bayesian neural network (BNN) based crop yield prediction model will be coupled with NASA Land Information System (LIS) for crop yield prediction. The final implementation will be open source and available to general public along with NASA project deliverables. The implemented algorithm and models will be disseminated at national and international conferences, proceedings, and journals. All software implementations will be open-sourced and available at GibHut. What do you plan to do during the next reporting period to accomplish the goals?1. Develop novel yield prediction framework to integrate process-based model and machine learning technique to further enhance prediction performance 2. Apply strategies such as partial domain adaptation and multiple instance regression to further improve the model's learning abilities 3. Further assess the developed algorithms by comparing the prediction results with NASS remote sensing model results 4. Improve the web portal prototype by adding more user functions, enhancing its user-friendliness and make the portal publicly accessible. 5. Enable the yield prediction models to run on the Google Vertex AI platform (https://cloud.google.com/vertex-ai?hl=en) or other cloud solutions.
Impacts
What was accomplished under these goals?
Objecitive 1: (1) Satellite image processing and feature extraction. We gathered MODIS MCD43A4 satellite imagery with 500m resolution, MODIS MYD11A2 land surface temperatures at 1km resolution, and Parameter elevation Regressions on Independent Slopes Model (PRISM) weather data at 4km resolution. Soil properties were obtained from the Soil Survey Geographic database (SSURGO) dataset at 30m resolution. Compared to previous work, we have incorporated a lot of new data. Our datasets covered 2008 to 2022, with an emphasis on newly added data from 2019 onwards. Additionally, we utilized the Soil Moisture Active Passive (SMAP) soil moisture data, offering global soil moisture details at 9km resolution, including surface and rootzone moisture levels, as well as daily global 1km resolution surface soil moisture from SMAP's L-band radiometer. This soil moisture data spanned from 2015 to the present. Cropland Data Layer (CDL) mask was used to identify the corn field lands with 30m spatial resolution. On the Google Earth Engine (GEE) platform, we spatially aggregated all the remote sensing and agricultural features, then grouped them temporally to 16-day intervals from mid-May to early-October to cover the corn-growing period. The 1km soil moisture data were masked by 2016 US county shape file and aggregated into county level. We obtained reasonable results with the Bayesian neural network (BNN) on the new dataset. Subsequently, we will conduct a more in-depth analysis of the model in both drought-prone and normal areas to uncover the impact of soil moisture on yield prediction. In this year, the Visible Infrared Imaging Radiometer Suite (VIIRS) satellite imagery will be adapted to replace MODIS data products for modeling due to degradation of the MODIS sensors. (2) Development and evaluation of Attention-weighted Multiple Instance Learning We proposed an Attention-weighted Multiple Instance Learning (Att-MIL) model for county-level corn yield prediction. First, to avoid the information loss problem in previous work, we examined each county at the pixel level and applied Multiple Instance Learning (MIL) to leverage detailed information within a county. To collect the features in pixel-level, we have changed some of the data processing details. When the resolution of a feature dataset was smaller than that of the CDL, the same feature value was assigned to all CDL pixels within one feature dataset pixel. After the data preprocessing, each CDL pixel had a corresponding feature variable containing all the above features. Furthermore, we integrated the year to grasp specific time-related features and calculated 5-year historical average yield to provide a baseline for model robustness. In addition, our method addressed the "mixed pixel" issue caused by the inconsistent resolution between feature datasets and crop mask, which may introduce noise into the model and therefore hinder accurate yield prediction. Specifically, the attention mechanism is employed to automatically assign weights to different pixels, which can mitigate the influence of mixed pixels. The developed model outperformed four other machine learning models over the past five years in the U.S. corn belt. We also demonstrated the advantages of our approach from both spatial and temporal perspectives. Finally, through an in-depth study of the relationship between mixed pixels and attention, it is verified that our approach can capture critical feature information while filtering out noise from mixed pixels. Objective 2: We worked on objective 2 and developed a Bayesian-based joint distribution adaptation approach, named Bayesian Joint distribution Adaptation Network (BJAN), for county-level corn yield prediction based on time-series remote sensing and weather variables. By solving a minimax problem on the uncertainty of Bayesian predictor, BJAN was trained to align source and target domains by considering crop yield response in the target domain based on the task-specific regression predictor. Also, the concept of joint distribution adaptation was adopted in BJAN to align the joint distributions of input variables and output crop yield across heterogeneous regions. Experiments on cross-year crop mapping scenarios in the CONUS have been conducted to evaluate the model's effectiveness. It was observed that BJAN effectively mitigates inter-annual domain shifts, enabling accurate crop mapping in years without crop type labels. It also demonstrated superior performance over representative marginal-distribution, conditional-distribution, and joint-distribution unsupervised domain adaptation methods. This approach will be applied to cross-region corn yield prediction to improve the yield prediction accuracy in low-agriculture regions with limited yield records. Objective 3: (1) web portal prototype development- We improved the web portal prototype developed in year #1 based on the libraries of Leaflet and D3 in JavaScript for users to interact with our prediction results. This version provides more functions and is more convenient and user-friendly. A few key functions developed in the last year include: 1.Producing statistical charts 2.Editing the map, e.g., adding features (point/line/polygon) and annotation to the map; 3.Saving the map with the selected information by the user into different formats, including pdf, png, and jpg; 4.Downloading the open-sourced map data selected by the user into different formats, including cvs, shapefile, and geojson; In this year, we will continue adding more useful functions. In particular, the model running interface and interaction function will be implemented. This function will enable users to interactively identify the model input information through the web interface, e.g., selecting a region of interests, crop type and year, then trigger the model running process, and finally use our portal to visualize and analyze the crop yield modeling outputs. In addition, we plan to publish it as a website or integrate it into NASS Crop production visualization, mapping, and report system, enabling it to be publicly accessible. (2) Cyberinfrastructure enabled model running- In addition to the web interface development, we will explore different solutions to enable public users to run the crop yield prediction models through our platform. In particular, we have developed a module to integrate various Google Cloud services for completing Google Earth Engine (GEE) tasks, including data processing and downloading automation in a cloud environment. With this module, all required datasets will be automatically downloaded and processed, ready to be fed for our models developed through our Objective 1 and Objective 2 tasks. In this year, we plan to enable the model to run on the Google Vertex AI platform (https://cloud.google.com/vertex-ai?hl=en ) or other cloud solutions.
Publications
- Type:
Journal Articles
Status:
Accepted
Year Published:
2024
Citation:
Xu, Y., Ma, Y., & Zhang, Z. (2024). Self-supervised pre-training for large-scale crop mapping using Sentinel-2 time series. ISPRS Journal of Photogrammetry and Remote Sensing, 207, 312-325.
- Type:
Journal Articles
Status:
Published
Year Published:
2023
Citation:
Wu M., Huang Q.*, Gao S., Zhang Z., 2023. Mixed Land Use Measurement and Mapping with Street View Images and Spatial Context-Aware Prompts via Zero-shot Multimodal Learning. International Journal of Applied Earth Observation and Geoinformation, 125 (2023): 103591. DOI: doi.org/10.1016/j.jag.2023.103591
- Type:
Journal Articles
Status:
Published
Year Published:
2023
Citation:
Vongkusolkit J., Peng B., Wu M., Huang Q.*, Andresen C. G., 2023. Near Real-Time Flood Mapping with Weakly Supervised Machine Learning. Remote Sensing, 15(13): 2363. DOI: 10.3390/rs15133263.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Wu M., Huang Q.*, Gao S., 2023. Measuring access inequality in a hybrid physical-virtual world: A case study of racial disparity of healthcare access during CoVID-19. In Proceedings of 2023 30th International Conference on Geoinformatics, Jul 19-21, 2023 London, UK, pg.1-10. DOI: 10.1109/Geoinformatics60313.2023.10247690.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Xu, Y. & Zhang, Z. (2023). Unsupervised Domain Adaptation for Crop Mapping without Current Year Ground-truth. AGU Fall Meeting Abstracts, 2023.
|
Progress 01/15/22 to 01/14/23
Outputs
Target Audience:In this project, we developed models and tools to provide near real-time in-season crop yield predictions for two mainly commodity crops (i.e., corn and soybeans) at the county-level for the CONUS, based on state-of-the-art remote sensing and deep learning technologies. The predicted yield information will be disseminated through a web-based cyber-platform for free public use, which is critical for a variety of stakeholders, such as for US producers to make operational decisions, for insurance companies to help with claim and compliance assessment, for agricultural finance companies to analyze the market and support businesses decision makings, and for agricultural business to better plan production, distribution, processing, and marketing. The developed methods and algorithms, and predicted yield information will also be delivered to USDA National Agricultural Statistics Service (NASS) for improving NASS crop yield estimation operation. The proposed tool could potentially provide supplemental information to the current NASS county crop yield estimation report, improve the operational efficiency and data quality, and enhance USDA NASS crop yield prediction capability. Therefore, the proposed project will add value to NASS's existing crop yield estimation program and have broad impacts on many stakeholders. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?Two PhD students are beingtrained under this project. One student has been working on remote sensing data processing and machine learning model development for corn yield prediction. Three journal articles have been published and one manuscript is under reviewwith the student as the first leading author. Another student has been working on developing a cyber-platform for yield prediction results from dissemination. Besides, she has been working on developing machine learning models for high-resolution land use mapping.She was able to present her research in multiple conferences, including 5th ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery, and the 2022 AGU Fall Meeting. She also has one journal manuscript under review as the first leading author. How have the results been disseminated to communities of interest?The project results have been disseminated through invited research seminars and international conferences, such as the American Geophysical Union (AGU) fall meeting, the Association for the Advancement of Artificial Intelligence (AAAI) conference, and ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery. Some of the research results have also been incorporated into the class of PIs Zhang Huang for teaching and student training activities. Through Co-PI Yang, the project progress has also been reported to stakeholder USDA NASS. The results were also cross disseminated to another NASA funded project team for potential collaboration. What do you plan to do during the next reporting period to accomplish the goals? Develop novel machine learning models to integrate remote sensing images and environmental factors to further enhance prediction performance Apply strategies such as partial domain adaptation and multiple instance regression to further improve the model's learning abilities Further assess the developed algorithms by comparing the prediction results with NASS remote sensing model results Continue developing and improving the web portal for crop yield modeling, and visualization.
Impacts
What was accomplished under these goals?
Objecitive1: (1) Satellite image processing and feature extraction Informative predictors have been collected from multiple sources of datasets, including satellite images from the daily MODIS MCD43A4 product, land surface temperatures from the MODIS MYD11A2 product, weather variables from the Parameter elevation Regressions on Independent Slopes Model (PRISM) dataset, and soil properties from Soil Survey Geographic database (SSURGO). Google Earth Engine (GEE) platform was leveraged to preprocess the data. Data collection in the U.S. corn belt started from 2001 to 2019 while data collection in Argentina started from 2006 to 2019. Specifically, spatial filtering was first conducted by using the MODIS Land Cover Type product (MCD12Q1 v6) at 500-m spatial resolution and NASS Cropland Data Layer (CDL) as the crop masks to mask out observations on non-cultivated croplands in each county. Then, sequential observations including VIs and climate observations were aggregated into a 16-day interval to cover the complete planting and growing season for corn. Finally, the input variables were paired with county-level yield records and used for model development. (2)Development and evaluation of Bayesian neural network We worked on objective 1 and proposed a Bayesian neural network (BNN) model for corn yield prediction and uncertainty analysis. We evaluated the performance of the BNN model for the end-of-season prediction on Oct 4th as well as the in-season prediction during the growing season. It was observed that the BNN model outperformed several other widely used machine learning and deep learning models by a large margin. For the in-season prediction, the proposed BNN model was able to achieve optimal accuracy in the middle of August, which is two months before the harvest. We also evaluated the predictive uncertainty from the BNN model and found that the observation noise due to the crop mask, prolonged exposure to extreme heat, and severe water would potentially increase the predictive uncertainty. County-level corn yield and predictive uncertainty maps have been developed from the model estimates to visualize the spatial variations, which will be incorporated into the web platform for public access. Objecitive 2: (1) Development and evaluation of unsupervised domain adaptation models We worked on objective 2 and developed two adversarial domain adaptation models, including the adaptive domain adversarial domain adaptation neural network (ADANN) and the Bayesian domain adversarial domain adaptation neural network (BDANN), for county-level corn yield prediction based on time-series remote sensing and weather variables. The ADANN model was designed to adaptively adjust the weighting parameter between the prediction loss and the domain loss. Based on ADANN, we further proposed BDANN by applying Bayesian learning to the model training. Both ADANN and BDANN were evaluated in two ecoregions of the U.S. corn belt and compared with other widely used machine learning models including the original DANN model with a fixed weighting parameter. Evaluation results across two ecoregions in the U.S. corn belt demonstrated that the proposed ADANN and BDANN had better spatial transferability with more stable performance against RF, DNN, and DANN in four testing years 2016-2019. County-level corn yield maps have been developed from the model estimates to visualize the spatial variations, which will be incorporated into the web platform for public access. (2) Development and evaluation of multi-source domain adaptation models We continued to work on objective 2 based on the previous works. A multi-source domain adaptation method named multi-source maximum predictor discrepancy (MMPD) was proposd for county-level corn yield prediction based on time-series remote sensing and weather variables. By using the maximum predictor discrepancy, MMPD was trained to align source and target domains by considering crop yield response in the target domain based on task-specific regression models. Also, the multi-source domain strategy was adopted in MMPD to avoid negative interference among source samples from heterogeneous regions. Experiments on three domain adaptation scenarios in the U.S. corn belt and Argentina have been conducted to evaluate the model performance. It was observed that MMPD outperformed representative single-source and multi-source unsupervised domain adaptation methods. County-level corn yield maps have been developed from the model estimates to visualize the spatial variations, which will be incorporated into the web platform for public access. Objecitive 3: We have developed a web portal prototype based on the libraries of Leaflet and D3 in Javascript for users to interact with our prediction results. Currently, the web map portal supports the following main functions: Presenting the general information of NIFA Corn Yield Prediction Project at the menu, e.g., introduction, publications, documentation, profiles of the research group, and the web map user guideline; Displaying the studied areas by census tract, with crop yield prediction results rendered as a choropleth map, along with web map interaction basics such as zooming in/out; Hovering on a census tract of interest, and its information of crop type, yield, and prediction value, date and error will be displayed in a panel at the top right corner; Selecting a location, crop type, year, prediction date, and an attribute (i.e., crop yield value, prediction value, and prediction error), and the filtered information will be displayed on the map. Other functions will also be developed in the next step as below (but not limited to): Editing the map, e.g., adding features (point/line/polygon) and annotation to the map; Saving the map with the selected information by the user; Downloading the open-sourced map data selected by the user; Contacting the technician for help; Linking to the web map Github portal for any further information.
Publications
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Ma, Y. and Zhang, Z., 2022. A Bayesian Domain Adversarial Neural Network for Corn Yield Prediction. IEEE Geoscience and Remote Sensing Letters, 19, pp 1-5.
- Type:
Journal Articles
Status:
Under Review
Year Published:
2022
Citation:
Ma, Y., Yang, Z., Zhang, Z., Maximum Predictor Discrepancy for Multi-source Unsupervised Domain Adaptation on Corn Yield Prediction. Submitted to IEEE Transactions on Geoscience and Remote Sensing. Manuscript under review
- Type:
Journal Articles
Status:
Under Review
Year Published:
2022
Citation:
Cao, Z., Ma, Y., Zhang, Z., Corn Yield Prediction based on Remotely Sensed Variables Using Variational Autoencoder and Multiple Instance Regression. Submitted to IEEE Geoscience and Remote Sensing Letters. Manuscript under review
- Type:
Journal Articles
Status:
Under Review
Year Published:
2023
Citation:
Wu M., Huang Q., Gao S., Zhang Z., 2022. Vision-Language Multimodal Learning for Mixed Land Use Measurement and Mapping with Street View Images. Submitted to ISPRS Photogrammetry and Remote Sensing. Manuscript under reivew
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
Ma, Y., Yang, Z, and Zhang, Z., 2022. Corn Yield Prediction Using Remote Sensing Observations and Multi-source Unsupervised Domain Adaptation, AGU Fall Meeting Abstracts 2022.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
Ma, Y., Zhang, Z, 2022. Multi-source Unsupervised Domain Adaptation on Corn Yield Prediction. AAAI-22 AI for Agriculture and Food Systems (AIAFS) Workshop.
|
|