Progress 06/01/07 to 07/31/08
Outputs
OUTPUTS: In Phase I, we collected a diverse training set of 60 soil samples. Field data collection was performed at four locations throughout the state of Wisconsin. These include the University of Wisconsin Research Stations in Arlington, Marshfield, and Platteville and one private landowner, Opitz Farms near West Bend. At these four locations, five fields were used at the UW - Arlington Station, and one field was used at each of the other locations. Field sites were chosen based on several factors including being able to utilize relationships established during prior research and development projects, soil and crop types and their accessibility during the summer months of a Midwest growing season. Data collection began on June 28th, 2007 and was completed on September 6th, 2007. Field work included the collection of soil images with the SIP (Soil Imaging Penetrometer) along with other in situ probe based soil sensor data and physical soil samples collected adjacent to probe measurements. The soil samples were sent to a laboratory and images from the SIP were analyzed using a number of image processing and computer vision techniques. The laboratory and penetrometer data was used to train and validate an artificial neural network (ANN). Two other methods (physical approach, fractal approach) were also tested. The results of our findings have been disclosed within a Phase I final report submitted to the USDA as well as in a Phase II proposal. SIS, Inc has met with cooperators at the USGS Wisconsin Water Science Center in Middleton, WI and presented our results of Phase I research within a PowerPoint presentation. We then proposed a cooperative effort in the near term for testing our SIP derived SWRC using their CPT platform at a test site of their choice in northern Wisconsin. PARTICIPANTS: Daniel Rooney (Soil Physicist, Principle Investigator) was responsible for overall project direction and coordination and provided scientific input for soil property measurements with geophysical sensors; Woody Wallace (Geomorphologist) contributed to project management, soil imaging system design, integration, and evaluation; Nick Guries (Geomaticist, Programmer) contributed to image processing, and statistical analysis. Nick also worked to set up the Artificial Neural Network; John Samuelson (Soil Scientist, Field Applications Specialist) contributed to project management included budget tracking and assisted in laboratory management. Mark Kuehn (Soil Information Specialist, Geographer) contributed to project management including coordination of field work activities and sample management. Partner organizations that helped contribute their facilities to the use of this project include three University of Wisconsin Research Stations, located in Arlington, Marshfield, and Platteville. One private cooperator, Jeff Opitz allowed us to perform data collection on his farm near West Bend, WI. TARGET AUDIENCES: Our initial target audience is focused on the grape industry of California. We believe that managers and consultants who are currently using WRC data for working with water management issues in vineyards will see this technology as most valuable. Vineyards are chosen because they are a high value specialty crop and a first adopter of new technologies, including the SIS. As the business grows we will be able to expand the use of the technology to other agricultural markets (practically any irrigated or drained crop) and non-agricultural markets, such as construction, storm water infiltration system design, and environmental remediation. PROJECT MODIFICATIONS: Nothing significant to report during this reporting period.
Impacts
Of the three methods tested, training the artificial neural network (ANN) by using laboratory, Soil Image Penetrometer (SIP) data, and Soil Information System (SIS) penetrometer sensor data showed the most promising results with the lowest RSME (e.g. 1.18 % v/v at 0.33 bar) values and the highest coefficient determinations (e.g. r2 = 0.83 @ 0.33 bar). Two other methods for utilizing information from SIP imagery alone did not produce results due to limitations of the resolution of the camera. Therefore, using the SIS penetrometer sensor data+SIP ANN method holds the most promise for rapidly and accurately estimating the soil WRC and has the potential to significantly outperform laboratory-based pedo transfer functions (PTFs), while being much cheaper and less time consuming to collect and process. The combination of SIS+SIP slightly outperformed the SIS penetrometer alone and significantly out performed the use of the SIP alone. This result is not surprising given the range of pore size that is visible by the current sensor (pores corresponding to pressures less than 0.1 bar). Based on the information learned from this research, in the future we will use ANN to derive soil WRC measurements using only SIP and SIS data as inputs. Although we recognize that more testing over a wider geographic area would be required in order to demonstrate full commercial application, we feel that the Phase I results indicate a high degree of technical feasibility. That combined with the cost effectiveness of our approach compared to conventional methods puts us well positioned for successful commercialization. Our ability to integrate the new technology with the SIS technology that is already commercialized provides a clear path to commercialization of the new penetrometer and data that it produces.
Publications
- No publications reported this period
|