Search research reports:
Precision Conservation With Geospatial Technologies
T.G. Mueller, S. A. Shearer
Department of Plant and Soil Sciences
The NRCS provides substantial payments through the conservation reserve program (CRP) to land owners who have conservation buffers installed and maintained. This is because these structures have been shown to prevent ephemeral gully erosion which accounts for a substantial amount of the total water erosion that occurs in agricultural fields.
To determine whether CRP-eligible grassed waterways can be established in agricultural fields, an NRCS conservationist must first make an on-farm site assessment. This involves walking across fields in order to identify areas where there is evidence of erosion resulting from concentrated water flow. This is a slow and time-consuming process and eroded channels scattered across large fields can easily be missed. Given contracting budgets and growing responsibilities of NRCS conservationists, they have less time to make these field visits. Tools are needed that will help them rapidly and accurately identify areas that are eligible to receive CRP payments for these conservation structures.
We are developing neural network and regression models to predict the erosion from concentrated flow. Terrain attributes are used as predictor variables such as LS (length slope factor), WET (topographic wetness index), and PLAN (Plan curvature) are terrain attributes derived from precision GPS measurements. Cost effective mapping procedures will be developed in order to help NRCS conservationists more rapidly identify and prioritize potential locations for grassed waterways and buffer strips for enrollment in CRP. These maps will be similar to the one presented in Fig.1. , but in addition to estimating the potential erosion associated with each feature, we will also estimate the potential for delivering sediment beyond the edge-of-field and will estimate the potential for sequestration of carbon.
Furthermore, we will we will evaluate the accuracy of prediction maps created using elevation data sources that differ in spatial resolution: USGS digital elevation models (DEMs) and light detecting and ranging (LIDAR). LIDAR data is much less expensive than surveys created from precision GPS measurements on a per hectare basis. The USGS DEMs are freely available on the internet. However, the adequacy with which prediction maps can be created from these datasets is unknown but preliminary results suggests that USGS data will be adequate in some cases. We will determine how these models should be parameterized across the different physiographic regions in Kentucky.
We expect this work to lead to the development of regional models that predict where waterways will be needed. We hope that these maps of erosion potential will be on the internet and available to government personnel, farmers, and citizens.
2009 Project Description
Our findings were disseminated in an Agronomy Journal Publication (Pike et al., 2009). This was the journal cover story for that issue. It was also featured in a press release by the tri-societies and by the College of Agriculture, University of Kentucky. These press releases were featured on numerous web sites (e.g., Science Daily, GPS Daily, Southeast Farm Press, Science Blog, Phys.Org, Geology Times) and discussed on a Pod Cast (Discovery News). This work was also presented to the American Society of Agronomy in 2009.
We developed an innovative approach to predict where concentrated-flow water erosion will likely occur and presented a prediction model for five fields in Central Kentucky (Pike et al., 2009). The input data for this analysis were GPS-derived terrain attributes. The models were fit with field observations of erosion. We considered both logistic regression and neural network models. This work was validated with a leave-one-field-out procedure. We have also tested this model with terrain attributes derived from USGS data. We are also considering terrain attributes derived from LIDAR information. Additionally, we are also testing the model developed by Pike et al., (2009) on fields outside the area where the model was generated.
Pike, A.C., T.G. Mueller, A. Schorgendorfer, S.A. Shearer, and A.D. Karathanasis. 2009. Erosion indices derived from terrain attributes using Logistic Regression and Neural Networks Agron J 101:1068-1079.