Progress 09/01/23 to 08/31/24
Outputs
Target Audience:The primary target audience of this NIFA project is the USDA National Agricultural Statistics Service (NASS). NASShas a federal mandate to report the crop progress and condition, including soil moisture conditions nationwide. Currently, NASS conducts a weekly survey of crop progress and condition and soil moisture condition for U.S. cropland and publishes state-level soil moisture conditions in the NASS Crop Progress and Condition Report. The survey has approximately 4,000 field respondents reporting the soil moisture condition qualitatively each week based on their field observations, which is a large-scale, extremely labor-intensive data collection effort. The USDA and agricultural-related communities rely on these state-level soil moisture reports for many decision-making and policy formations. The low/coarsespatial resolutionof the product prevented accurate statistics assessment, drought event monitoring, and flooded/extremely wet field identification, which might further affect disaster relief or mitigation efforts. Our project will address the low-resolution of the existing USDANASS soil moisture maps (Crop-CASMA, available from here: https://nassgeo.csiss.gmu.edu/CropCASMA/) by developing apartnership between US and Canadian institutes to produce novel Machine Learning High-Resolution Soil Moisture (ML-HRSM 2.0), which are an improved surface and rootzone soil moisture data productin Support of NASS Crop Monitoring and Management across the conterminous US (CONUS). Other target audiences of this NIFA project include researchers, students, extension specialists, governmental agencies, private sectors, and stakeholders such as farmers and ranchers in the US and selected regions in Canada. These audiences will use the model code developed from this project by our team and/or access the public-access final soil moisture data products (ML-HRSM 2.0) once they are validated (in Years 2 and 3 of the project). Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?The project has funded two PhD students (Yijia Xu, majoring in Biological Systems Engineering) and (Shuohao Cai, majoring in Soil Science) in Year 1. The projects have also funded one undergraduate student (Jake Jackan) for a summer training mini-project to work on satellite data analysis. The project has provided funding for PDs (Jingyi Huang, Zhou Zhang, Zhengwei Yang) to travel to conferences and the NIFA PD meeting (in Kansas in the summer of 2024) to present the project outcomes. How have the results been disseminated to communities of interest?The PDs and the graduate students have disseminated the project outcomes at national conferences, local meetings, and the Internet newsletter. 1. American Geophysical Union 2023 Fall Meeting, San Francisco, CA. The audience of this conference comprises scientists, researchers, students, policymakers, private industry representatives, stakeholders, and the media. 2.7th Satellite Soil Moisture Validation and Application Workshop, East Lansing, MI. The audience of this conference consists ofscientists, researchers, and students. The PhD student Shuohao Cai attended this conference and presented the project outcome and built a professional network with the group. 3. Wisconsin Agricultural Field Days at Hancock Ag Research Station, Wisconsin, organized by the Division of Extension of University of Wisconsin-Madison. The audience of this local field day are soil and water conservationists, consultants, and local farmers. The PD (Jingyi Huang) attended this meeting. 4. University of Wisconsin Alumni newsletter/magazine "Grow".How Wet is America's Soil? Nobody Really Knows, But AI Can Help CALS scientists are leveraging artificial intelligence to map soil moisture across the country. It's a possible game changer for farmers and a boon for climate models and wildfire forecasts. The audience are University of Wisconsin-Madison alumni and their families and friends. Available from here:https://grow.cals.wisc.edu/departments/features/how-wet-is-americas-soil-nobody-really-knows-but-ai-can-help What do you plan to do during the next reporting period to accomplish the goals?We plan to publish the two new manuscripts mentioned above in peer-reviewed journals in Year 2 of the project. We will also start working on Objectives 3 and 4 in Year 2. Particularly, we plan to travel to Canada in the summer of Year 2 to collaborate with the co-PD from Canada and collect additional in situ soil moisture datasets for evaluating the performance of our machine learning and data assimilation-based models outside the USA.
Impacts
What was accomplished under these goals?
Over the firstyear of the project (09/2023-08/2024), we have accomplished Objectives 1 and 2. In addition to the conference papers,the journal article, and the model codewe published (listed in "Products" and "Other products"), we have produced two new manuscripts for Objectives 1 and 2. 1) Use Unsupervised Domain Adaptation (UDA) to improve the performance of the ML models in data-limited regions in the CONUS; One manuscript has been submitted to a journal for peer review. The manuscript information is provided below. Journal: Remote Sensing of Environment Title: Integrating Snapshot and Time Series Data for Enhancing SMAP Soil Moisture Downscaling Corresponding Author: Ms. Yijia Xu (PhD student) Co-Authors: Shuohao Cai (PhD student); Jingyi Huang (PD); Jiangui Liu (collaborator from Canada); Jiali Shang (Co-PD from Canada); Zhengwei Yang (Co-PD); Zhou Zhang (co-PD) Abstract:Understanding soil moisture (SM) dynamics is crucial for various environmental and agricultural applications. While satellite-based SM products provide extensive coverage, their coarse spatial resolution often fails to capture local SM variability. Recent advancements in deep learning (DL) present new opportunities to integrate diverse data sources for enhancing SM estimates in both resolution and accuracy. In this study, we developed a multi-modal network (MMNet) that integrates data from different sensors and sources, incorporating both snapshot and time series data, to downscale Soil Moisture Active Passive (SMAP) Level 4 SM data. Specifically, the model takes multiple publicly available datasets as input: snapshot data includes optical imagery, Synthetic Aperture Radar (SAR) imagery, terrain attributes, landcover information, and soil properties, and time series data incorporates weather and land surface temperature (LST) collected over the preceding days. In-situ surface (0-5 cm) SM measurements from the Soil Climate Analysis Network (SCAN) and the United States Climate Reference Network (USCRN) served as ground truth. We evaluated MMNet under three scenarios: on-site, off-site, and cross-region, and compared its performance with models that utilize either snapshot or time series data. The results showed that 1) MMNet trained with on-site data provided accurate SM estimates over time in withheld years; 2) MMNet effectively captured SM dynamics in regions with sparse or no in-situ measurements; and 3) the integration of snapshot and time series data was crucial for maintaining the model's accuracy and generalizability across all scenarios. These findings highlight the effectiveness of MMNet across different settings, demonstrating its potential for producing high-resolution temporally and spatially continuous SM estimates, which could further support a broad range of environmental and agricultural applications. 2) Combine ML models with a physically based process model via a data assimilation framework to predict field-level soil water content (SWC) and plant available water storage (PAWS) across the CONUS; One manuscript will be submitted to a journal for peer review by December 2024. The manuscript has been submitted and accepted as an abstract for the American Geophysical Union fall meeting in 2024 (to be held in Washington, D.C.). The details of the accepted presentation are provided below: Title:Machine-Learning High-resolution soil moisture (ML-HRSM 2.0) mapping: performance in rainfed and irrigation crop fields Presenting Author: Shuohao Cai (PhD student) Co-Authors: Yijia Xu (PhD student); Zhengwei Yang (Co-PD);Jiali Shang (Co-PD from Canada);Jiangui Liu (collaborator from Canada); Zhou Zhang (co-PD); Jingyi Huang (PD). Abstract:Accurate and timely monitoring of soil moisture is important for water resource management, drought and flood forecasting, and nutrient transport estimation over cropland. In recent years, various soil moisture products have been developed using remote sensing algorithms, process-based or land surface models, and hybrid methods. However, few products capture field-level irrigation information due to limitations in spatial and temporal resolutions. This study presents a new framework that integrates machine learning (ML) with process-based models to produce high-resolution soil moisture products (100 m, daily to sub-daily). Initially, a process-based model was developed based on a 1-dimensional field water balance, simulating hourly soil moisture across five soil layers (0-5, 5-15, 15-30, 30-60, 60-100 cm). Then, three machine learning models (using Sentinel-1 and SMAP, Sentinel-2 and SMAP, or Landsat and SMAP) were independently developed based on satellite overpass time to produce 100-m soil moisture estimates for two soil layers (0-5 cm and 0-100 cm), where 3 ML combination has a temporal resolution of 2-6 days. Finally, the machine learning outputs were integrated into the process-based model using a Kalman filter data assimilation scheme. A total of 72 crop sites across the continental US from SCAN and USCRN soil moisture monitoring networks were used to train and validate the models. The final model is also applied over USDA ARS crop sites for the independent validation. This framework is designed to produce high spatiotemporal resolution soil moisture products (ML-HRSM 2.0) that can help farmers optimize water resource management.
Publications
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2024
Citation:
Yang, Z., Wang, X., Huang, J., & Zhang, Z. (2024, July). County Level Crop Yield Prediction Using Smap Derived Data Products and Deep Learning Model. In IGARSS 2024-2024 IEEE International Geoscience and Remote Sensing Symposium (pp. 3222-3225). IEEE.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Yang, Z., Huang, J., & Zhang, Z. (2023, July). Toward Field Level Drought and Irrigation Monitoring Using Machine Learning Based High-Resolution Soil Moisture (ML-HRSM) Data. In IGARSS 2023-2023 IEEE International Geoscience and Remote Sensing Symposium (pp. 3570-3573). IEEE.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Yang, Z., Willis, P., Huang, J., & Zhang, Z. (2023, December). Identifying Early Season Prevented Planting Fields Using Machine Learning Based High Resolution Soil Moisture (ML-HRSM). In AGU Fall Meeting Abstracts (Vol. 2023, pp. B23A-07).
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Hu, J., & Huang, J. (2023, December). Improved Soil Hydraulic Configuration in CLM5 Reduces Transpiration Uncertainty at Selected NEON Sites. In AGU Fall Meeting Abstracts (Vol. 2023, No. 2651, pp. B43H-2651).
- Type:
Peer Reviewed Journal Articles
Status:
Published
Year Published:
2024
Citation:
Peng, Y., Yang, Z., Zhang, Z., & Huang, J. (2024). A Machine Learning-Based High-Resolution Soil Moisture Mapping and SpatialTemporal Analysis: The mlhrsm Package. Agronomy, 14(3), 421.
|