Progress 08/01/18 to 07/31/23
Outputs
Target Audience:The target audiences were: (1) the postdocs, graduate students, and undergraduate students who were directly supported by the project and participated in the research, (2) researchers at USDA-NRCS National Soil Survey Center, (3) professional societies (American Society of Agricultural & Biological Engineers andSoil Science Society of America), (4) several industry partners (Yard Stick PBC, General Mills), and (5) farmers and the general public who are interested in precision agriculture, drone technologies, and soil science. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?The project trained Nuwan Wijewardane as a PhD student and then as a postdoc (UNL-BSE).The project trained Omar Murad as a Postdoc (UNL-BSE). The project trained 4MSstudents (UNL-BSE), 1 PhDstudent in UNL-Statistics, and 1 PhD student in UNL-Computer Science. The projecttrained 1 MS student in TAMU-Soil Science. The project also trained two undergraduate students at UNL in using the instruments such as VisNIR, MIR and pXRF for soil analysis, andthe programming language R for data analysis. Several postdocs andgraduate students working on the project joined the NRCS Earth Team volunteers and received training by the NRCS scientists for soil survey and soil analysis. How have the results been disseminated to communities of interest?The project led to four peer-reviewed journal articles and two referred conference proceeding papers. The team attended the following professional society meetings and made technical presentations to disseminated the results: American Society of Agricultural and Biological Engineers (2019-2022), ASA-CSSA-SSSA (2019, 2021 and2022), Proximal Soil Sensing Symposium (2019). The team also gave one invited talkat NE Society of Soil Scientist meeting (2019). PI Ge worked with UN-FAO on a training manual titled "A primer on soil analysis using visible and near-infrared (vis-NIR)and mid-infrared (MIR) spectroscopy: Soil spectroscopy training manual# 1"for the developing counties. Thisproject partially supported the development of this manual, with the goal of capacity building for developing countries for precision agriculture. This manual is being translated to multiple languages. What do you plan to do during the next reporting period to accomplish the goals?
Nothing Reported
Impacts
What was accomplished under these goals?
Obj. 1. A UAV airframe intergrating a hyperspectral camera and a LiDAR scanner was developed. Protocols ofimage acquisition and processing to obtain high-quality, well calibrated hyperspectral images were developed. This UAV airframe and an existingcontinuous-depth soil VisNIR penetrometer were deployed in six agricultural fields in Nebraska and Kansas to collect the sensor data. More than 1000 soil samples at multiple depths were collected from these six agricultural fields; and these soil samples were processed and analyzed in the lab for soil properties such as organic matter, total nitrogen, pH, cation exchange capacity, and macro- and micro-nutrient contents. On the last year of the project, we had additional resource to develop a robotic system that could measure the optical reflectance fromsurface soil. Obj. 2. We developed calibration models to estimate soil properties from the image and sensor data acquired by the UAV and the soil VisNIR penetrometer. Because the image and sensor data were measured on field soils with varying moisture content and aggregrate conditions, an intermediate step was investigated to calibrate the models based on the lab-measured VisNIR spectral data of dried-ground soil samples. New chemometricapproaches (such as External Parameter Orthogonalization and Direct Standardization) were studied to correct forthe effectof moistureand aggregation on the model estimates. Through proper model calibration and validation, we showed that soil properties including organic matter, total nitrogen, and certainmacronutrients (Ca, Mg) couldbe estimated with R2 > 0.70. Models performance for organic matter and total nitrogen were quite consistent across all six fields, whereas models for Ca and Mg were field-dependent.With the estimated soil data from the model, we were able to conduct layer-by-layer krigging to create and visualize 3D soil maps of organic matter andtotal nitrogen for the experimental fields.Toward the last two years of the project, we also investigated the use of portable XRF (X-ray fluorescence) data (and its fusion with VisNIR reflectance data) to improve the estimation of soil macro- and micronutrients. Obj. 3. We created and developed VSSAL (Virtual Soil SpectrumAnalytical Laboratory). VSSAL was an online soil property estimation and visualization system. It accepted real-time, field-measured soil VisNIR spectral data or manually uploaded data to estimate soil properties. The models for soil property estimation were pre-trained and loaded into VSSAL, but the system also had the capability to train a model on-the-fly. Oneinnovative part of VSSAL was a new data storage and retrieval component fueled by deep-learning (and accelerated by GPUs), which greatly increased the speed of data retrieval for near real-time model building.
Publications
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2019
Citation:
Wijewardane, N.K., Wang, L., Zhan, Y., Franz, T., Yu, H., Zhou, Y., Shi, Y., Ge, Y., 2019. Mapping infield variability of soil properties to support precision agriculture using UAV, multi-depth EC, and aerial hyperspectral imagery, In Proceedings of 5th Global Workshop on Proximal Soil Sensing. Columbia, MO, USA.
- Type:
Journal Articles
Status:
Published
Year Published:
2020
Citation:
Wijewardane, N.K., Hetrick, S., Ackerson, J., Morgan, C.L.S., 2020. VisNIR integrated multi-sensing penetrometer for in situ high-resolution vertical soil sensing. Soil and Tillage Research 199. 104604. https://doi.org/10.1016/j.still.2020.104604
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2019
Citation:
Sun, J., Wu, C., Ge, Y., Li, Y., Yu, H., 2019. Spatial-temporal scientific data clustering via deep convolutional neural network. 2019 IEEE International Conference on Big Data (Big Data). DOI:10.1109/BigData47090.2019.9006507
- Type:
Theses/Dissertations
Status:
Published
Year Published:
2020
Citation:
Thapa, S., 2020. Virtual soil spectrum analysis laboratory (VSSAL). MS Thesis. Department of Computer Science. University of Nebraska-Lincoln.
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
Wijewardane, N.K., Ge, Y., Sanderman, J., Ferguson, R., 2021. Fine grinding is needed to maintain the high accuracy of MIR spectroscopy for soil property estimation. Soil Science Society of America Journal 85, 263-272. https://doi.org/10.1002/saj2.20194
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Sheng, W., Guan, K., Zhang, C., Lee, D.K., Margenot, A.J., Ge, Y., Peng, J., Zhou, W., Zhou, Q., Huang, Y., 2021. Using soil library hyperspectral reflectance and machine learning to predict soil organic carbon: assessing potential of airborne and spaceborne optical soil sensing. Remote Sensing of Environment 271, 112914. https://doi.org/10.1016/j.rse.2022.112914
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Murad, O.F., Jones, E.J., Minasny, B., McBratney, A.B., Wijewardane, N., Ge, Y. 2021. A VisNIR penetrometer system for predicting soil carbon under Australian conditions. Biosystems Engineering 224, 197-212. https://doi.org/10.1016/j.biosystemseng.2022.10.011
- Type:
Theses/Dissertations
Status:
Published
Year Published:
2023
Citation:
Ghimire, B. 2023. Spectroscopic sensor data fusion to improve the prediction of soil nutrient contents (Master's Thesis). Biological Systems Engineering-Dissertations, Theses, and Student Research. Department of Biological Systems Engineering, University of Nebraska, Lincoln.
- Type:
Theses/Dissertations
Status:
Published
Year Published:
2023
Citation:
Harun, H. 2023. A robotic system for in-situ measurement of soil total carbon and nitrogen (Master's Thesis). Biological Systems Engineering-Dissertations, Theses, and Student Research. Department of Biological Systems Engineering, University of Nebraska, Lincoln.
|
Progress 08/01/22 to 07/31/23
Outputs
Target Audience:NRCS scientists. The team continued to work closely with USDA-NRCS' Kellogg Soil Survey Lab (KSSL). Team members Omar Murad, Sadia Mitu, Bidhan Ghimire, and Husein Harun continued as NRCS' Earth Team Volunteers and contributed a few hours every month at KSSL for soil analysis and spectral scanning, and interacted with NRCS soil scientists. PD Yufeng Ge continued to serve as a member of FAO (Food and Agriculture Organization of the United Nations) GLOSLAN-spec steering committee. Additional research collaborations formed with Innovative Solution for Digital Agriculture,Woodwell Climate Research Center, and Soil Health Institute. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?The project provided training for one postdoc, three graduate students, and oneundergraduate student, all in the department of Biological Systems Engineering of UNL. The graduate students received training in VisNIR, MIR, and XRF instruments for soil analysis and programming languages (R, python) for machine learning/deep learning and spectroscopic modeling. The undergraduate student received training in VisNIR and MIR soil analysis. The postdoc attended the SSSA and ASABE meeting. The graduate student attended the ASABE meeting. As the NRCS Earth Team Volunteers, the postdoc and all three graduate students received training at NRCS Kellogg Soil Survey Laboratory for lab-based soil analysis. How have the results been disseminated to communities of interest?The results were disseminated through peer reviewed journal articles (one published, one is under review). The team attended the ASA-CSSA-SSSA annual meeting (1 oral presentation) and the ASABE annual meeting (3 oral presentations, 1 poster, July 2023). What do you plan to do during the next reporting period to accomplish the goals?
Nothing Reported
Impacts
What was accomplished under these goals?
Objective 1: Aground robotic platform that carried two compact lightweight spectrometerswas developed. A robotic end-effector that integrated an optical sensing head (for reflectance measurement from the soil surface) and a hoe (to prepare a smooth surface for the measurement) was developed. The robotic system was tested in a 5-acre field for its performance in autonomous navigation, soil sensing, and mapping. Around 160 soil samples were collected from the predetermined GPS waypoints (where the robotic system navigated and took measurements) to validate the system. It was shown that soil properties such as total carbon and total nitrogen could be estimated with R2 of 0.7 using the measurements by the robotic system. Objective 2: We continued to explore the sensor fusion approach to combine the VisNIR, MIR and pXRF data to better estimate soil macronutrient contents (nitrogen, phosphorus, potassium, calcium, and magnesium). We used ~720 samples collected from the previous foursampling campaigns. Fusion algorithms including low-level, mid-level, and high-level fusions were tested. The results showed that mid-level and high-level fusions gave better estimations than low-levelor no fusion. The best models gave R2 of ~0.8 for Ca and Mg, 0.7 for P, and >0.5 for N and K. Objective 3 was completed and therefore no accomplishments in this reporting period.
Publications
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Murad, M.O.F., Jones, E.J., Minasny, B., McBratney, A.B., Wijewardane, N., Ge, Y., 2022. Assessing a VisNIR penetrometer system for in-situ estimation of soil organic carbon under variable soil moisture conditions. Biosystems Engineering 224, 197-212. https://doi.org/10.1016/j.biosystemseng.2022.10.011
- Type:
Theses/Dissertations
Status:
Published
Year Published:
2023
Citation:
Ghimire, B. 2023. Spectroscopic sensor data fusion to improve the prediction of soil nutrient contents (Master's Thesis). Biological Systems Engineering-Dissertations, Theses, and Student Research. Department of Biological Systems Engineering, University of Nebraska, Lincoln.
- Type:
Theses/Dissertations
Status:
Published
Year Published:
2023
Citation:
Harun, H. 2023. A robotic system for in-situ measurement of soil total carbon and nitrogen (Master's Thesis). Biological Systems Engineering-Dissertations, Theses, and Student Research. Department of Biological Systems Engineering, University of Nebraska, Lincoln.
- Type:
Journal Articles
Status:
Submitted
Year Published:
2023
Citation:
Murad, M.O.F., Ackerson, J., Tolles, C., Meissner, K., Morgan, C.L.S., Ge, Y., 2023. Estimating soil organic carbon content at variable moisture content using a low-cost spectrometer. Geoderma. Under Review.
|
Progress 08/01/21 to 07/31/22
Outputs
Target Audience:NRCS scientists. The team continued to workclosely with USDA-NRCS' Kellogg Soil Survey Lab (KSSL). Team members Omar Murad, Sadia Mitu, Bidhan Ghimire, and Husein Harun joined NRCS' Earth Team Volunteer and contributed a few hours every month at KSSL for soil analysis. Food and Agriculture Organization of the United Nations (FAO). PI Ge was elected to be a member of FAO's GLOSLAN-spec steering committee. Ge also led the wrting of a training manual titled "A primer on soil analysis using visible and near-infrared (vis-NIR) and mid-infrared (MIR) spectroscopy" (https://www.fao.org/documents/card/en/c/cb9005en/). This manual will betranslated into several different languages, and then distributed to many research institutions and soil labs around the world. Graduate students and soil science faculty at University of Nebraska-Lincoln. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?The project provided training for one postdoc and three graduate students, all in the department of Biological Systems Engineering of UNL. The postdoc and graduate students received training in VisNIR, MIR, and XRF instruments for soil analysis and programming languages (R, python) for machine learning/deep learning and spectroscopic modeling. The postdoc and two graduate students attended the ASABE annual meeting and ASA-SSSA-CSSA meeting and made oral presentations. How have the results been disseminated to communities of interest?The results were published through peer reviewed journal articles (one published, one has gone through 2 rounds of review). The team attendedthe ASA-CSSA-SSSA annual meeting (Nov 2021, 1 oral presentation), and theASABE annual meeting (3oral presentations, July 2022). What do you plan to do during the next reporting period to accomplish the goals?Objective 1: We have completed this objective of the UAV and soil depth-sensing physical system. It was tested in 6 agricultural fields of varying sizes, and over 1100 soil samples were collected and analyzed in the lab for sensor calibration and validation. No further work will be done for this objective. Objective 2: We will continue to investigate methods to improve the models for estimating soil properties from various sensor data. In particular we will focus on the estimation ofN, P, K, and other nutrient data. We will research different deep learning neural network architectures and model fine-tuning to improve estimation. Furthermore, combine spatial modeling (e.g., semi-variogram analysis and kriging) into machine-learning models will be attempted. Objective 3: We will continue to improve and test the already built 3D soil visualization and VSSAL system, with the data generated from this project, and some other data available to the team (such as NRCS' national-scale soil database). The goal is to improve the robustness and usability of thissoil mapping cyber system. In addition, we will continue to develop the autonomous ground platform for measurement of surface soil properties.
Impacts
What was accomplished under these goals?
The goal of this project is to create and demonstrate a scalable cyber-physical system that enables rapid and cost-effective mapping of soil properties in 3D. We will combine the newest technologies of unmanned aerial vehicle(UAV), continuous depth proximal soil sensing, innovative 3D statistical modeling and real-time data-guided sampling to achieve this goal. Our working hypothesis is that soil and environmental covariates, measured densely at surface by UAV and sparsely at subsurface by the continuous-depth sensing, will allow accurate modeling and reconstruction of soil properties in 3D. The specific research objectives are as follows. Objective 1: Develop and deploy a new instrument platform that combines UAV and a continuous-depth soil sensor to collect high resolution soil data in lateral and vertical dimensions Objective 1. In previous reporting periods, we completed the development of the two physical systems: the continuous-depth soil sensing system and a UAV that integrates a hyperspectral camera and a 3DLiDAR scanner. In the period, we continued to test the system in one additional field at NE. The field was 10 ha. in size and with drylandcorn-soy rotation. UAV was flownin May/2022 (before planting whensoil was visible) to take the hyperspectral images and 3D point clouds of the field. Soil samplingwasconducted within one week of UAV flying;and 125 soil samples were collected from a regular grid pattern of 25 x 25 m. The UAV hyperspectral image was calibrated, geo-rectified, and overlaid to represent the entire field. Soil samples were analyzed in the lab for properties including moisture content, organic carbon, total nitrogen, pH, cation exchange capacity (CEC), and nutrient contents. Data from this field will be combined with those from other fields (5 fields in NE and KS) for spectral modeling andsoil property mapping. Objective 2: Investigate the novel spatial statistical approach to model the distribution of soil properties in 3D and data-guided sampling We will put forward a novel statistical-based and process-based approach to model the distribution of soil properties in 3D, by coupling covariance in both lateral and vertical dimensions. Objective 2. We continued to explore the useof advanced modeling (machine learning and deep learning) andsensor fusion to estimate soil properties at low cost. Sensors we have used in the project including lab-based X-ray fluorescence, VisNIR, and MIR; and field-based VisNIR (from UAV and depth-sensing penetrometer), elevation, soil electrical conductivity, and gamma-ray counts. These analyses have been done on observations from individual fields, as well as combined observations from all the fields.Specifically, we investigated more traditional machine learning approaches such as Random Forest and Partial Least Squares Regression, but also newer deep learning approaches. Our result showed that deep learning didnot improve modeling performance over the traditional machine learning. The reason might be that our sample size (even with n > 1100) is still small to harness the benefit of deep learning. Soil properties like organic carbon, total N, CEC can be estimated from the sensor data more successfully; other properties such as phosphorus,potassium, and some macro/micro nutrients are more challenging to estimate. In terms of sensor fusion, we attempted approaches to fuse the information from various sensors at the raw data level and at the feature level. More research is being conducted in this regard. Objective 3: Create a scalable cyber system for interactive soil mapping and visualization in 3D In this objective, we will address the computational challenge associated with 3D modeling and map visualization so that the system can be put into practical use. Objective 3. We continued to develop VSSAL - Virtual Soil Spectrum Analysis Laboratory, the cyber component of the 3D soil mapping system. Specifically, we worked on thereception of real-time, field-measured soil VisNIR spectral data through wireless transmission, deployment of VisNIR models in VSSALfor soil property estimation, and return of the estimations to the field in near real-time. One new feature added to the VSSAL is deep-learning fueled data retrieval, which we expect to greatly increase the accuracy of data/sample retrieval and enhance model calibration on-the-fly for real-time soil property determination. The VSSAL is also integrated into a more general Agricultural Data Management and Analysis Platform, which includes not only soil data, but also other agricultural data such as crop data, remote sensing imagery, etc. In addition to the objectives above, the team is also developing an autonomous platform that can navigate in the agricultural fields and measure thepropertiesof surface soil. This physical system is an extension of our continuous-depth soil sensing system.
Publications
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Wang, S., Guan, K., Zhang C., Lee, D., Margenot, A.J., Ge, Y., Peng, J., Zhou, W., Zhou, Q., Huang, Y., 2022. Using soil library hyperspectral reflectance and machine learning to predict soil organic carbon: Assessing potential of airborne and spaceborne optical soil sensing. Remote Sensing of Environment 271, 112914. https://doi.org/10.1016/j.rse.2022.112914.
- Type:
Journal Articles
Status:
Under Review
Year Published:
2022
Citation:
Murad, O.F., Jones, E.J., Minasny, B., McBratney, A.B., Wijewardane, N., Ge, Y. 2022. A VisNIR penetrometer system for predicting soil carbon under Australian conditions. Biosystems Engineering. Under Review
|
Progress 08/01/20 to 07/31/21
Outputs
Target Audience:The team has worked closely with USDA-NRCS' Kellogg Soil Survey Lab (KSSL) and the staff members at KSSL in our project. KSSL provides a national-scale soil vis-NIR and MIR spectral library (with more than 150,000 archived soil samples and measured soil properties) to support the UAV hyperspectral soil mapping system and the vis-NIR soil penetrometer system. The team has worked with FAO (Food and Agriculture Organization of the United Nations) to develop an introductory document onthe use of vis-NIR and MIR soil spectroscopy as a rapid and low-cost tool for soil analysis in precision agriculture and soil-based site specific management. Thisdocument is intended as a training material for developing countries. Our team has worked with a group of scientists to establish a "global soil spectral calibation library and estimation service" under the umbrella of the Global Soil Laboratory Network of the Global Soil Partnership of the FAO. Changes/Problems:Our team has faced challenges with covid-19. For most of the time in this reporting period, our university has strict policies for social distancing and travel restrictions. We were not allowed to share a vehiclewhen driving to the field sites, which hamperedour progressin field work and soil sample collection. We also reduced the speed ofsample processing and analysis due to the need for social distancing among project personnel working in the lab. Travel restrictionlimited our capacityto attend professional society meetings and disseminate project results to the community. We had planned to attend the 2021 ASABE meeting and to make a presentation there. But the organizer decided to move the meeting completely online and we decided to cancel our talk. In fact, these same problemsexisted in the previous reporting cycle and just continued to impact us into this reporting cycle. We hope as the trend with the Pandemic has been getting better, we will be able to be back to normalcy and improve our productivity with the project. What opportunities for training and professional development has the project provided?In this reporting period, three of the project participants Dr. Omar Murad, Sadia Mitu, and Bidhan Ghimire joined the Earth Team Volunteers atUSDA-NRCS. This opportunity was possible through our collaborative work with NRCS in the scope of this project. Each of them volunteered a couple of hours every week and worked in the NRCS' Kellogg Soil Survey Lab (KSSL)to learn labtechniques for soil analysis, soil scanning using vis-NIR and MIR instruments,and the soil information system maintained at the KSSL. This professional development opportunity is beneficial for the students, especially if they want to pursue a career in a federal agency like NRCS. The project trained an undergraduate student (Luke Freyhof)from the Biological Systems Engineering Department in the use of vis-NIR and MIR instruments for soil scanning, and multivariate statistical modeling using R. The previous postdoc working on the project (Dr. Nuwan Wijewardane) is now a tenure trackassistant professor at the Agricultural & Biological Engineering Department of Mississippi State University (since Mar 2021). How have the results been disseminated to communities of interest?In this reporting period, two manuscripts out of the collaborative work of this project weresubmitted for peer review. The team attended 2021 virtual "Application of Proximal and Remote Sensing Technologies for Soil Investigations" Symposium and gave apresentation titled "In-situ VisNIR Penetrometer System for Predicting Soil Organic Carbon". What do you plan to do during the next reporting period to accomplish the goals?Obj. 1. In the next reporting period, we plan to locatetwo agricultural fields with sufficient spatial variations to test the combined UAV and continuous-depth soil sensing physical system. Importantly, we will need to identify the fields and take measurements when bare soil is exposed (not in the growing season when there isvegetation or with high crop residues). The brief steps for this field work in each field are: (1) flying the UAV to acquire high-resolution hyperspectral imagery of bare soil of the field,(2) conducting grid sampling (n1 = 75) to collect surface soil samples (0-15 cm), (3) at a subset of sampling locations (n2 = 20) using the continuous-depth soil sensing probe to collect high resolution depth measurements, (4) collecting soil core samples from these 20 locations,(5) lab processing and analyzing the surface and core soil samples, and (6) data analysis to correlate the image data / depth-sensing data with the lab-based soil property data. Obj. 2. With the data we already collected from the four fields, we will continue to investigate more advanced machine-learning approaches to improve the accuracy of estimating soil properties from proximally-sensed soil data. Particularly, we will investigateto fusedata from different sensing modalities (vis-NIR, MIR, X-ray fluorescence, electrical conductivity, elevation, gamma-ray) and incorporatethe spatial modeling (e.g., semi-variogram analysis and kriging) into machine-learning modeling to improve prediction. After the completion of Obj. 1 above, theseapproaches will be assessedusing the new data from the two new fields. With the already developed 3D soil visualization tool and VSSAL (the real-time soil property prediction tool), we will attempt to put everything together to demonstrate near real-time 3Dmapping (both at surface and at depth) of important soil properties (e.g., organic C, total nitrogen, clay content) from the UAV and continue-depth soil sensing physical platform.
Impacts
What was accomplished under these goals?
Obj. 1. In the previous reporting period, we have completed the development of the two physical systems:the continuous-depth soil sensing system and a UAV system that integrates a hyperspectral imager and a 3D scanning LiDAR. The continuous-depth soil sensing system (vis-NIR integrated soil penetrometer) has been tested in numerous fields before. In this reporting period, we continued to test this penetrometer in a number of production-scale agricultural fields in Nebraska and Kansas. We also flied and tested the UAV system in a few fieldswith corn and wheat at various growth stages. The result showed that the hyperspectral signals from the UAV can effectively distinguish the water and nitrogenstresses of the crops, therefore demonstrating its usefulness. Ourgoal though,is to use this UAV system to capture hyperspectral images and LiDAR point clouds of the bare soil. Obj. 2. In this reporting period, our research focused on the accurate modeling and estimating of soil properties from multiple sources of proximally-sensed soil data. The soil properties of interest were organic matter, total nitrogen, pH, clay,and cation exchange capacity. The proximally-sensed data were collected in the previous reporting periods including the soil reflectance spectra in the vis-NIR and MIR regions, X-ray fluorescence spectra, soil bulk electrical conductivityand elevation, gamma-ray counts, and hyperspectral imagery from the UAV. Different machine-learning approaches such as Partial Least Squares Regression, Random Forest, Support Vector Regression, and Neural Networks were investigated for this purpose. For each field (we had data from four fields with a total of over 1000 samples), we split the data into a 70% training set and 30% test set for model development and assessment, respectively. Theresults showed that different soil properties could be estimated fromthe proximally-sensed data with different levels of success. For example, organic matter, total nitrogen,and clay can be estimated satisfactorily (with R2 > 0.75), whereas extractable phosphorus and potassium were estimated poorly (with R2 < 0.5). We also found that leveraging the spatial correlation among the field samples generally improved the accuracy of soil property modeling. Obj. 3. In this reporting period, we developed VSSAL: Virtual Soil Spectrum Analysis Laboratory. VSSAL is a cyber system that (1) allows users to upload vis-NIR or MIR reflectance spectra of a batch of soil samples,(2) estimate the major soil physical and chemical properties on-the-fly (such as organic matter, pH, textures, and cation exchange capacity), and (3) return the estimated soil properties to the users. The VSSAL cyber system is consisted of the following components (1) a front-end graphic user interface developed by JavaScript and Python (to process user requests), (2) a MongoBD database and Hierarchical Data Format files to store the vis-NIR and MIR soil spectral data, (3) an SQL Lite database to store the users and sessions information. The VSSAL employs advanced machine learning algorithms such as artificial neural network and support vector regression to model and estimatethe properties of the submitted soil samples. In addition, the VSSAL can adaptively search for the training samples in the database to build the most appropriate models for the submitted samples. The VSSAL is a significant cyber system thatleads to near real-time estimate and visualization of soil properties from the remotely-sensedsoil reflectance spectra (for example, either from remote-sensing/UAV platformor optical sensors mounted on a ground vehicle). Currently we are thoroughly testing the VSSAL using a wide range of soil samples available to our research team.
Publications
- Type:
Journal Articles
Status:
Under Review
Year Published:
2021
Citation:
Murad, O.F., Jones, E.J., Minasny, B., McBratney, A.B., Wijewardane, N., Ge, Y. 2021. A VisNIR penetrometer system for predicting soil carbon under Australian conditions. Biosystems Engineering. Unver Review.
- Type:
Journal Articles
Status:
Under Review
Year Published:
2021
Citation:
Sheng, W., Guan, K., Zhang, C., Lee, D.K., Margenot, A.J., Ge, Y., Peng, J., Zhou, W., Zhou, Q., Huang, Y., 2021. Using soil library hyperspectral reflectance and machine learning to predict soil organic carbon: assessing potential of airborne and spaceborne optical soil sensing. Remote Sensing of Environment. Under Revision.
|
Progress 08/01/19 to 07/31/20
Outputs
Target Audience:The team has worked closely with USDA-NRCS in the development of the 3D soil mapping system because NRCS field scientists will be the primary users of the system for soil survey and mapping. NRCS soil characterization lab provided our team their national VisNIR soil spectral libraries that support both the UAV hypersppectral soil mapping system and the VisNIR soil penetrometer. The team has worked with the industry partner (General Mills) and Applied Ecololgical Services on the project. They tested our VisNIR soil penetrometer at 6 production fields in Kansas with 62 soil cores down to 1 m. The main focus of this collaboration is for in-situ, low-cost soil organic matter measurements and carbon market. Changes/Problems:Covid-19 has exerted a few challenges and impacts on our project. First, because of the policies like the need to do social distancing and driving to field sites on separate vehicles, the efficiency of field data collection and soil sample analyses were lowered. The delayed experimental data acquisition has further delayed some work in Objective 2 and Objective 3. In general, we are still on track of the progress, with four out of six fields being surveyed already. We plan to resume field surveys to collect data from the remaining two fields as soon as the Covid-19 situation becomes better. Another impact of Covid-19 on our project is the dissemination of the project and the results to professional society meetings. For example, we had planned to attend the 2020 ASABE meeting and to make a presentation there. However, because the meeting was moved completely online and additional work for presenters was needed to pre-record the presentation, we decided to cancel our talk. Again we will attend more professional meetings for project result dissemination as soon as the Covid-19 situation becomes better. What opportunities for training and professional development has the project provided?The project trained one postdoc in the Department Biological Systems Engineering, one graduate student in the Department of Statistics, and two graduate students in the Department of Computer Science, all at the University of Nebraska-Lincoln. One of the graduate students in Computer Science was writing his thesis and expected to graduate with an MS degree in Dec 2020. How have the results been disseminated to communities of interest?We published two journal articles in "Soil Tillage and Research" and "Soil Science Society of America Journal". We also published a peer-reviewed conference paper in "2019 IEEE International Conference on Big Data". What do you plan to do during the next reporting period to accomplish the goals?In the first two years of the project, we intensively collected surface and depth-wise soil samples from 4 fields in Nebraska. Over one thousand soil samples were collected, among which ~850 samples were lab analyzed for important soil properties include organic carbon, pH, total nitrogen, cation exchange capacity, etc. The remaining 250 samples were stored in the lab and will also be analyzed in the next reporting period. In addition to these physical soil samples, we also used a variety of remote and proximal soil sensors to generate spatially-dense soil maps of these fields. These maps include the soil ECa (apparent electrical conductivity) maps, the Gamma-Ray maps, and the elevation maps. Two fields (at the time when there was no vegetation or crop residues) were imaged by a hyperspectral imaging system onboard a manned aircraft. In the next reporting period, we will continue our field campaign to collect surface and depth-wise soil samples from two additional fields, as well as using the ECa sensor, Gamma-ray sensor, and the hyperspectral imaging system on UAV to survey the fields. We will use the data collected from the fields to develop high resolution 3D soil properties maps. We will focus on organic carbon and total nitrogen first, and then move to other agronomically important soil properties such as pH and soil nutrients. We will use the lab-analyzed soil sample data and the readings from proximal sensors to develop estimation models for the soil properties at un-sampled locations and depths. We will investigate machine learning techniques such as support vector regression or artificial neural network for this purpose, considering fusing the measurements from different sensors. Because soil properties at the field scale usually exhibit spatial dependence, we will also incorporate spatial modeling (for example, regression kriging) to improve the estimation of soil properties using the ECa, elevation, and Gamma-Ray maps as co-variates. 3D kriging will be employed to develop soil property maps by considering spatial dependence of soil at both lateral and vertical directions. An interactive 3D soil mapping and visualization tool will be developed using the experimental data collected from these 6 fields. This tool will be hosted at UNL's Holland Computing Center for wider accessibility. We plan to draft two manuscripts for publication. The first manuscript will focus on the developed UAV hyperspectral system for soil property mapping. The second manuscript will focus on 3D soil property maps and their accuracy compared with the lab-analyzed data.
Impacts
What was accomplished under these goals?
The goal of this project is to create and demonstrate a scalable cyber-physical system that enables rapid and cost-effective mapping of soil properties in 3D. Objective 1.Develop and deploy a new platform that combines UAV and a continuous-depth soil sensor to collect high resolution soil data in lateral and vertical dimensions We purchased a Nano Push-broom hyperspectral system, Headwall Photonics. The system has a spectral range from 400 -1000 nm, with 270 spectral bands. The hyperspectral imager, along with3D scanning LiDAR, IMU, GPS antennas, and a data storage drive was integrated on a DJI Matric Pro drone. Once the UAV-based hyperspectral imaging system was developed, we tested its performance. Because at the time of testing we could not identify any field that was fallow, we decided to conduct the flying and image acquisition at a turf field at UNL. This field had 48 plots subjected to different water and nitrogen applications, presumably creating a large difference in the spectral signatures from these plots. On the same day of drone data collection, two ground measurements were made: the visual rating of turfgrass quality and the clipping dry biomass. The visual rating ranged from 1 to 9, where 1 represents bare soil and 9 indicates perfect turfgrass quality. We explored the preprocessing and modeling of hyperspectral images. The workflow included (1) converting raw digital numbers into reflectance values using a standard reflectance calibration tarp, (2) georectifying the hyperspectral data cubes using GPS and IMU sensors, and (3) mosaicking individual data cubes into the whole field map. After data preprocessing, the plot-level spectrum was extracted from the ortho-mosaicked map by averaging all pixels within one plot. Partial least squares regression (PLSR) modeling was adopted. Ten-fold cross-validation was used to calibrate and test the model. The results showed that biomass was estimated with R2=0.72 and RMSE=1.01 g/m2, and the visual rating was estimated with R2=0.80 and RMSE=0.44. After completing this test flight and data modeling, we concluded that the developed UAV-based hyperspectral imaging system performed satisfactorily. We were also confident about the experimental setup and data processing workflow to convert the raw measurements to reflectance spectra of the land surface. We will use this system for surface mapping of soil properties in the next reporting period. We did the 3D soil mapping and soil sample collection/analysis in two fields, with the primary focus of using various proximal soil sensors to infer soil properties rapidly. The first field was on the Havelock farm (UNL's research farm). This field was approximately 4 acres in size. The field campaign included ECa (apparent electric conductivity), elevation, and gamma-ray surveys in fall 2019. Physical soil sample collection in this field was conducted in spring 2020. 192 samples were obtained from 64 randomly selected locations: 3 samples at different depths (0-10, 10-25, and 25-40cm). The samples were air-dried, ground, and sieved. The samples were scanned by a portable X-ray Fluorescence (XRF) Analyzer for the elemental analysis. The same samples were scanned with a LabSpec® spectrometer to obtained VisNIR reflectance spectra from 350 to 2500 nm. A subsample of each sample was sent to a commercial lab (Ward Laboratories, Kearny, NE) and measured for different soil properties: TC, TN, pH, CEC, etc. The second field for the 3D soil mapping activities was on Rogers Memorial Farm (also UNL's research farm). This field was managed as a term-long corn/soybean/wheat rotation experiment field with no-till and cover crops after wheat harvest. There were 42 subplots receiving various combinations of soil tillage levels and cove crop types. We expect large variations in soil properties such as organic matter and nitrogen among the subplots. ECa and Gamma-ray mapping were conducted in this field in Nov 2019 using the same instruments and the survey protocols as for the first field. In fall 2020, three soil samples were taken from each treatment subplot (at equal distances along the centerline of the subplot). At each location, samples were taken from two depths: 0-10 and 10-20 cm, totaling ~250 samples. Samples were transported back to the lab and stored at 0°C. These samples were sent to a commercial laboratory for measuring soil properties. Objective 2.Investigate the novel spatial statistical approach to model the distribution of soil properties in 3D and data-guided sampling Analyses were done with the data collected from the first field to evaluate the performance of proximally sensed measurements to estimate soil properties. The dataset (192 samples) was first randomly split into the training (80%) and test (20%) sets to calibrate and validate models. The VisNIR spectra of the training set were used to calibrate models for XRF derived element data using four modeling techniques: partial least squares regression (PLS), artificial neural networks (ANN), random forests (RF), and support vector regression (SVR). The models were fine-tuned using ten-fold cross-validation. The calibrated models on the calibration set were used to predict for the test set. The prediction statistics R2, RMSE were calculated to evaluate the model performance. The prediction statistics showed that PLS performed well compared to other modeling techniques and R2> 0.5 was obtained for five different elements: Fe, Zn, Rb, As, Zr. Further analysis with lab measured soil properties is underway. Objective 3.Create a scalable cyber system for interactive soil mapping and visualization in 3D It is challenging to visualize scientific datasets with higher spatial-temporal dimensions. A feasible way is to cluster such a dataset and characterize its essential information into different groups for detailed examinations. We developed a new visual analytics approach for multivariate time-varying data in this direction. We explored the usage of autoencoder with a deep convolutional neural network to cluster the time steps of a scientific dataset. Our preliminary results showed that feature descriptors can be learned for individual variables and their combinations for real-world simulation data. Each time step could be represented as a set of vectors in the feature space. Using this representation, we could quantify the distance between the time steps and cluster them into different groups, where each group had similar patterns. Our visualization results qualitatively revealed the difference and the similarity among different classes. This approach made it easy for domain scientists to explore and select time steps from their large-scale spatial-temporal scientific datasets, as well as helped researchers design new algorithms for data compression and reduction. We have further extended this idea to tackle 3D hyperspectral soil datasets. In such a dataset, each voxel contained the spectral information in a waveform of hundreds or thousands of bands. No existing visualization techniques can be applied on this type of data.We developed a hybrid classification model with an autoencoder. For a given dataset, we trained a deep convolutional autoencoder and extracted a 3D feature space or latent space by inferencing the encoding part of the network. Each voxel was transformed into a feature descriptor (or a vector representation) in the feature space. Then, we clustered all feature descriptors using a clustering algorithm (e.g., MeanShift), and generated the classification of the voxels according to the feature descriptors of their spectral information. The classification results could be visualized using the conventional 3D volume rendering technique. A prototype example of interactive visualization for the classification result can be found at http://vis.unl.edu/~yu/homepage/resources/hyperspectral_classification/. We are working on a publication and developing visualization tools.
Publications
- Type:
Journal Articles
Status:
Published
Year Published:
2020
Citation:
Wijewardane, N.K., Hetrick, S., Ackerson, J., Morgan, C.L.S., 2020. VisNIR integrated multi-sensing penetrometer for in situ high-resolution vertical soil sensing. Soil and Tillage Research 199. 104604. https://doi.org/10.1016/j.still.2020.104604
- Type:
Journal Articles
Status:
Accepted
Year Published:
2020
Citation:
Wijewardane, N.K., Ge, Y., Sanderman, J., Ferguson, R., 2020. Fine grinding is needed to maintain the high accuracy of MIR spectroscopy for soil property estimation. Soil Science Society of America Journal. https://doi.org/10.1002/saj2.20194
- Type:
Theses/Dissertations
Status:
Published
Year Published:
2019
Citation:
Thapa, S., 2020. Virtual soil spectrum analysis laboratory (VSSAL). MS Thesis. Department of Computer Science. University of Nebraska-Lincoln.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2019
Citation:
Sun, J., Wu, C., Ge, Y., Li, Y., Yu, H., 2019. Spatial-temporal scientific data clustering via deep convolutional neural network. 2019 IEEE International Conference on Big Data (Big Data). DOI:10.1109/BigData47090.2019.9006507
10.1109/BigData47090.2019.9006507
|
Progress 08/01/18 to 07/31/19
Outputs
Target Audience:The research team has worked closely with soil scientists at USDA-NRCS (especially the National Soil Survey Center and the Kellogg Soil Survey Laboratory at Lincoln Nebraska) to use their national-scale soil VisNIR and MIR spectral libraries in our project. USDA-NRCS will be one major target user of the 3D soil information system we will develop in our project. The research team attended the following professional meetings to disseminate the project findings: (1) 2019 ASABE Annual Meeting (International); (2) 2019 Soil Science Society of America Meeting (International); (3) 2019 Nebraska Society of Professional Soil Scientists Meeting (Regional); and (4)5th Global Proximal Soil Sensing Workshop (International). ?The research team has also worked with a group of scientists from University of Sidney in Australia, Purdue University, and Woods Hole Research Center to advance field application of soil spectroscopy. Changes/Problems:Co-PI Dr. Cristine Morgan left Texas A&M University at the beginning of 2019 and she is now Chief Scientific Officer of Soil Health Institute. Upon her departure, Dr. Peyton Smith at Texas A&M has become the Co-PI of the project and the transition was smooth. Dr. Morgan is still working closely with the team on this project, and she is committed to participate throughout the life of the project. Her new role actually plays quite a positive role on the project as she brings a lot of new insights on soil health and how the technologies we develop in the project can enhance soil health. Therefore impact of this change of personnel on the project is minimal. What opportunities for training and professional development has the project provided?The project trained one postdoc (Dr. Nuwan Wijewardane) and three graduate students at UNL (Lin Wang from Biological Systems Engineering, Jianxin Sun from Computers Science Engineering, and Yinglun Zhan from Statistics). Lin Wang obtained her remote pilot certificate by the Department of Transportation, Federal Aviation Administration (FAA) so that she can operate the UAVs for data collection for the project. How have the results been disseminated to communities of interest?The dissemination of the results is mainly through peer-review publication and attending professional society meetings. One manuscript was submitted to "Soil & Tillage Research" and now is under revision. The team attended four professional meetings at the international and regional levels and made presentations about the project: 2019 SSSA meeting, 2019 ASABE meeting, 5th Global Workshop on Proximal Soil Sensing, and 2019 NE Society of Soil Scientist meeting. What do you plan to do during the next reporting period to accomplish the goals?Obj. 1 - We plan to continue the field sampling and data collection using the UAV and the depth sensing soil penetrometer platform. In addition, other sensors such as gamma ray sensor has become available to the research team through newly established collaboration and will be leveraged to the project to acquire additional data layers. We have identified two new fields in Nebraska and one new field in Texas. After spatial data collection, soil cores will be collected from these fields. Soil samples will be processed and analyzed for agronomically important properties such as particle sizes, organic carbon, bulk density, total nitrogen, pH, etc. We expect to process and analyze ~400 soil samples under this objective in the next reporting period. Together with the spatial data collected by the UAV, penetrometer, and other sensors, these soil data will be used for modeling in Obj. 2 and interactive visualization in Obj. 3. Obj. 2 - We will conduct 3D statistical modeling and analysis of data collected from the five fields (two fields in Year 1 and three fields in Year 2). Data analysis will be focused on the following two aspects. Firstly, we will investigate to improve the modeling accuracy of soil properties (such as clay, sand, organic carbon, total nitrogen, pH, etc.) from the in-situ VisNIR spectral data and other spatial data collected from soil surface. Our first year result only showed limited success with certain soil properties. We hope that by including measurements from more fields will increase the overall variability of the data and improve the model performance. Secondly, we will start 3D spatial modeling and interpolation (such as kriging) to produce 3D soil property maps. The quality of the 3D soil maps will be evaluated against the actual measurements using cross validation. Obj. 3 - We will continue to develop the Hyperspectral Volume Rendering (HVR) framework. Field data collected from the new fields in Obj. 1 and 3D soil property maps developed in Obj. 2 will be included in the framework. New interactive visualization methods will be developed to allow users view the spatial distribution of soil properties in 3D.
Impacts
What was accomplished under these goals?
For Objective 1: Develop and deploy a new instrument platform that combines UAV and a continuous-depth soil sensor to collect high resolution soil data in lateral and vertical dimensions We assembled a UAV system (DJI Matrice 600 Pro) with a multispectral and RGB camera (Micasense RedEdge and Zenmuse X5R) to acquire images of bare soils in the two agricultural fields. The first field was at Eastern Nebraska Research and Extension Center (Latitude 41°10'45"N and Longitude 96°26'16"W) Mead, NE. The field was approximately 65.4 ha in size and currently managed under the Carbon Sequestration Program (CSP) of University of Nebraska-Lincoln. A sampling grid of 7*8 was established and used to identify the sampling locations. The center of each sampling cell was used as the sampling point. A total of 58 sampling points (including two additional sampling points near the in-field weather stations) was identified for data collection. A soil penetrometer consisted of an optical sensing probe coupled with the ASD LabSpec spectrometer to measure VisNIR reflectance spectra along the depth, was used for depth soil sensing at each sampling point. This penetrometer was attached to a Giddings's probe and pushed at each sampling point up to a depth of ~1m to obtain reflectance spectra and penetrating force at different depths. A second push at each sampling point was conducted next to the first push to obtain replicated measurements. A validation soil core of 2" in diameter was extracted from each sampling point approximately 6" from the penetrometer pushes. The soil cores were obtained in plastic sleeves and capped to preserve soil properties. Upon transport to the lab, the soil cores were segmented into five segments: 0-15, 15-30, 30-60, 60-90, and 90-120 cm, and measured for different soil properties as moisture content, bulk density, pH, total carbon, and total nitrogen. These soil samples were then air-dried, ground, and sieved (passing 2-mm screen) to be scanned in the lab using the same ASD spectrometer coupled with a MugLite. A spectral library of ~1600 dry-ground soil samples from NE was extracted from USDA-NRCS' Kellogg Soil Survey Lab to calibrate VisNIR models. This spectral library was used to build support vector regression (SVR) models for pH, TC and TN with 10 random-segment cross-validation. A total of 285 samples was available for data analysis. This sample set was divided into two random segments as calibration transfer (30%) and test (70%) set. The SVR models were used to predict the dry ground spectra (DG) and penetrometer field spectra (VNIRP) of the test set. The calibration transfer set was then used to implement different calibration transfer techniques: direct standardization (DS), external parameter orthogonalization (EPO), and extra-weighted spiking (SPK) and to predict for the VNIRP spectra. Bulk Density of the samples was also modeled as a function of moisture content, penetrating force, and depth. As for the results, TC showed the highest model accuracy with R2 of 0.9 and RMSE of 0.2% when predicted with the DG spectra. The model failed to directly predict for VNIRP spectra with R2 of 0.08 and RMSE of 6.5%. However, EPO and SPK were able to rectify this by improving prediction R2 to 0.25 and 0.84, and RMSE to 0.64 and 0.22 respectively. Bulk density model showed a R2 of 0.89 with RMSE of 0.193 g/cm3. The second field was at Havelock Farm of UNL (40o51'43''N, 96o36'47''W). The field was 12.5 ha (340×405 m) in size. The major soil types were Crete Silty Clay Loam and Crete Silt Loam. Hyperspectral images of the field were obtained using an imagine spectrometer (AISA Eagle) carried by an airplane (UNL CALMIT airborne research program). The spectrometer has a spectral range of 400-970 nm, 356 spectral bands, 1024 spatial pixels, and 12-bit depth per image. The airplane was flown at an altitude of approximately 1000 m to obtain a ground resolution of 1 m2. A LiDAR digital elevation map (2 m resolution) of the field was obtained from Nebraska Department of Natural Resources. 143 surface (0-10 cm) soil samples were obtained from the field as the validation data using grid sampling. At each sampling location, a 1x1 m square area was identified and soil samples were collected from four corners and the center. These samples were mixed and composited as the representative sample for the sampling point. This sampling procedure was employed to account for the soil property variation within the 1 m2 area to match with the spatial resolution of the hyperspectral image. The collected soil samples were immediately put in sealed bags and sent to the lab for the measurement of six different soil properties: gravimetric moisture content, total carbon, total nitrogen, nitrate-N, phosphorus, and CEC. The modeling was conducted using either the point data or image data extracted from spatial data. The point data were extracted using the area weight method; and this procedure was implemented to extract data from ECa map, and all bands of UAV and hyperspectral images. The resulting data matrix consisted of six responses (i.e. soil properties measured), and 364 predictors (2 depths from ECa + LiDAR elevation + 5 bands from multispectral + 356 bands from hyperspectral) for 143 samples. The dataset was randomly split into two subsets: a calibration set (70%) for modeling and a test set (30%) to validate the models. Four modeling techniques were attempted: partial least squares regression (PLS), artificial neural networks (ANN), random forests (RF), and support vector regression (SVR). The results of this work showed high correlations between P and LiDAR DEM, and negative correlations between TC/TN with UAV bands. The highest model performance was observed for P with random forests model using all the data sources (i.e. UAV+LiDAR+ECa+Hyperspectral), which showed a R2 of 0.67 and RMSE of 9.94 ppm. None of the other models performed satisfactorily regardless of the modeling technique or data source used. We believe that including data from fields under different geographic regions, moisture regimes, texture, and including fields with more in-field variability, may be able to increase the variability within the dataset as well as the number of samples. For Objective3, we proposed a framework named Hyperspectral Volume Rendering (HVR) to overcome the difficulties of visualizing 3D soil hyperspectral data, including missing 3D spatial information, visual clutter and occlusion, and lack of user defined area of interest (AOI). The HVR framework provided a pipeline of unsupervised learning to enable understanding of 3D hyperspectral data. Firstly, several unsupervised embedding techniques were implemented to transform the raw high dimensional hyperspectral data into embedding space with low dimensionality. Secondly, several clustering techniques were used to separate the embedding space samples into groups based on the similarity. The resulting categorical clusters were then visualized in the HVR visualization interface with user interaction to efficiently find the distribution of clusters in 3D space. This framework utilized 3D volume rendering techniques to effectively reveal the spatial intensity pattern for each hyperspectral band. Users can specify AOI on hyperspectral intensity through the user interface to detect key patterns in 3D space. The HVR framework was versatile for its light-weighted web user interface, cross-platform accessibility and capability of handling both continuous and categorical data types.
Publications
- Type:
Journal Articles
Status:
Under Review
Year Published:
2019
Citation:
Wijewardane, N.K., Sarah Hetrick, Ackerson, J., Morgan, C.L.S., Ge, Y., 2019. VisNIR integrated multi-sensing penetrometer for in situ high-resolution vertical soil sensing. Soil & Tillage Research. Under Revision.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2019
Citation:
Wijewardane, N.K., Ge, Y., 2019. VisNIR spectroscopy for rapid, quantitative soil analysis. Nebraska Society of Professional Soil Scientists. Invited Presentation. Mar/8/2019. Lincoln, NE.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2019
Citation:
Wijewardane, N.K., Wang, L., Zhan, Y., Franz, T., Yu, H., Zhou, Y., Shi, Y., Ge, Y., 2019. Mapping Infield Variability of Soil Properties to Support Precision Agriculture Using UAV, Multi-Depth EC, and Aerial Hyperspectral Imagery, In Proceedings of 5th Global Workshop on Proximal Soil Sensing. Columbia, MO, USA.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2019
Citation:
Wijewardane, N.K., Wang, L., Zhan, Y., Franz, T., Yu, H., Zhou, Y., Shi, Y., Ge, Y., 2019. Mapping infield variability of soil properties using different spatial data: UAV, multi-depth EC, and aerial hyperspectral imagery, In: 2019 ASABE Annual International Meeting. Boston, MA, USA.
|
|