Submitted to: Journal of Biosystems Engineering
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/9/2014
Publication Date: 12/18/2014
Citation: Chung, S., Sudduth, K.A., Drummond, S.T., Kitchen, N.R. 2014. Spatial variability of soil properties using nested variograms at multiple scales. Journal of Biosystems Engineering. 39(4):377-388.
Interpretive Summary: Precision agriculture relies on the ability to efficiently and economically collect and interpret data describing the variability within cropped fields. Data analysis is often done using geostatistical techniques, which consider not only the value of a parameter (for example soil pH) but also its location in relation to other data points. The simplest and most often used geostatistical model is the variogram, a representation of how a parameter varies spatially. Often the simple variogram structure does not do a good job of representing the complexity of variability in the data, and nested variograms have been developed to address this issue. In this study we applied the nested variogram approach to better understand the variability present in two Missouri corn/soybean fields. Parameters studied were soil chemical properties and soil electrical conductivity. We found that the nested variogram approach was not successful with soil chemical property data, possibly due to the sparse nature of such data. However, the dense, sensor-based soil electrical conductivity data were well-represented by the nested variogram approach. This allowed us to make inferences about the spatial structure of the sensor data, which included both short-range and long-range components. We attributed the short-range variation to soil profile cation content affected by past erosion, and the long-range variation to soil texture differences related to landscape morphology. This research may benefit scientists and data analysts who wish to more thoroughly examine spatial datasets as a first step in developing management zones for site-specific field operations.
Technical Abstract: Determining the spatial structure of data is important in understanding within-field variability for site-specific crop management. The structure of variability determines the required spatial intensity of data collection and can be used for directing the delineation of management zones. Especially when site variables are correlated, or are a combined response to multiple causative factors, an understanding of the spatial structures present in the data may help to illuminate interrelationships that are important in subsequent explanatory analyses. In this study, correlation, principal component analysis, and single and nested variogram models were applied to soil electrical conductivity and chemical property data for two fields in central Missouri, USA. Some, but not all variables which were highly correlated, or which were strongly expressed in the same principal component, exhibited similar spatial ranges when fit with a single variogram model. However, single variogram results were dependent on the active lag distance used, with short distances (30 m) required to fit short-range variability. Longer active lag distances, as are more often used in practice, only revealed long-range spatial components. Nested models generally yielded a better fit than single models for sensor-based conductivity data, where multiple scales of spatial structure were apparent. Gaussian-spherical nested models fit well to the data at both short (30 m) and long (300 m) active lag distances, generally capturing both short-range and long-range spatial components. As soil conductivity relates strongly to profile texture, we hypothesize that the short-range components may relate to the scale of erosion processes, while the long-range components are indicative of the scale of landscape morphology. In this study, we investigated the effect of changing active lag distance on the calculation of the range parameter. Future work investigating scale effects on other variogram parameters, including nugget and sill variances, may lead to better model selection and interpretation. Once this is done, separation of nested spatial components by factorial kriging may help to better define the correlations present between spatial datasets.