Page Banner

United States Department of Agriculture

Agricultural Research Service

Title: Spatial Crop Yield and Soil Water Content: Measurement, Scaling and Topographic Analyses

Authors
item Green, Timothy
item Green, Timothy
item Erskine, Robert
item Erskine, Robert
item Martinez, A - COLORADO STATE UNIVERSITY
item Martinez, A - COLORADO STATE UNIVERSITY
item Murphy, Michael
item Murphy, Michael
item Ahuja, Lajpat
item Ahuja, Lajpat
item Salas, J - COLORADO STATE UNIVERSITY
item Salas, J - COLORADO STATE UNIVERSITY
item Ramirez, J - COLORADO STATE UNIVERSITY
item Ramirez, J - COLORADO STATE UNIVERSITY

Submitted to: International Congress on Modeling and Simulation Proceedings
Publication Type: Proceedings
Publication Acceptance Date: October 1, 2001
Publication Date: December 10, 2001
Citation: Green, T.R., Erskine, R.H., Martinez, A., Murphy, M.R., Ahuja, L.R., Salas, J.D., Ramirez, J.A. 2001. Spatial crop yield and soil water content: measurement, scaling and topographic analyses. International Congress on Modeling and Simulation Proceedings. December 10, 2001.

Interpretive Summary: The need to transfer information across a range of space-time scales (i.e., scaling) is coupled with the need to predict variables and processes of interest across landscapes. Agricultural landscapes offer a unique set of problems and space-time data availability with the onset of satellite-based positioning and crop yield monitoring. The present study addresses quantification of the spatial variability of rainfed crop yield and near-surface soil water over a farm field using three methods: 1) geostatistical and fractal analyses; 2) multiple linear regression (MLR) using topographic attributes for explanatory variables; and 3) nonparametric estimation by Spatial Artificial Neural Networks (SANN). Method 1 is useful for scaling the spatial moments of each variable to determine appropriate scales of measurement and management. Methods 2 and 3 take advantage of empirical and process knowledge of topographic controls on water movement and microenvironments, where topographic attributes estimated from a digital elevation model at some scale (10 m by 10 m here) help explain the observed spatial variability in crop yield. Soil water (top 30 cm) displays more random spatial variability, and its dynamic nature makes it difficult to predict in both space and time. Despite such variability, spatial structure is evident and can be approximated by simple fractals out to lag distances of about 400 m. The SANN technique is more flexible than point-to-point parametric correlations, including the use of spatial activation functions for interpolation within a field. Using topographic attributes as input, SANN provides a minimum prediction error for crop yield in 1997, which explains 68% of the spatial variance.

Technical Abstract: The need to transfer information across a range of space-time scales (i.e., scaling) is coupled with the need to predict variables and processes of interest across landscapes. Agricultural landscapes offer a unique set of problems and space-time data availability with the onset of satellite-based positioning and crop yield monitoring. The present study addresses quantification of the spatial variability of rainfed crop yield and near-surface soil water over a farm field using three methods: 1) geostatistical and fractal analyses; 2) multiple linear regression (MLR) using topographic attributes for explanatory variables; and 3) nonparametric estimation by Spatial Artificial Neural Networks (SANN). Method 1 is useful for scaling the spatial moments of each variable to determine appropriate scales of measurement and management. Methods 2 and 3 take advantage of empirical and process knowledge of topographic controls on water movement and microenvironments, where topographic attributes estimated from a digital elevation model at some scale (10 m by 10 m here) help explain the observed spatial variability in crop yield. Soil water (top 30 cm) displays more random spatial variability, and its dynamic nature makes it difficult to predict in both space and time. Despite such variability, spatial structure is evident and can be approximated by simple fractals out to lag distances of about 400 m. The SANN technique is more flexible than point-to-point parametric correlations, including the use of spatial activation functions for interpolation within a field. Using topographic attributes as input, SANN provides a minimum prediction error for crop yield in 1997, which explains 68% of the spatial variance.

Submitted to: International Congress on Modeling and Simulation Proceedings
Publication Type: Proceedings
Publication Acceptance Date: October 1, 2001
Publication Date: December 10, 2001
Citation: Green, T.R., Erskine, R.H., Martinez, A., Murphy, M.R., Ahuja, L.R., Salas, J.D., Ramirez, J.A. 2001. Spatial crop yield and soil water content: measurement, scaling and topographic analyses. International Congress on Modeling and Simulation Proceedings. December 10, 2001.

Interpretive Summary: The need to transfer information across a range of space-time scales (i.e., scaling) is coupled with the need to predict variables and processes of interest across landscapes. Agricultural landscapes offer a unique set of problems and space-time data availability with the onset of satellite-based positioning and crop yield monitoring. The present study addresses quantification of the spatial variability of rainfed crop yield and near-surface soil water over a farm field using three methods: 1) geostatistical and fractal analyses; 2) multiple linear regression (MLR) using topographic attributes for explanatory variables; and 3) nonparametric estimation by Spatial Artificial Neural Networks (SANN). Method 1 is useful for scaling the spatial moments of each variable to determine appropriate scales of measurement and management. Methods 2 and 3 take advantage of empirical and process knowledge of topographic controls on water movement and microenvironments, where topographic attributes estimated from a digital elevation model at some scale (10 m by 10 m here) help explain the observed spatial variability in crop yield. Soil water (top 30 cm) displays more random spatial variability, and its dynamic nature makes it difficult to predict in both space and time. Despite such variability, spatial structure is evident and can be approximated by simple fractals out to lag distances of about 400 m. The SANN technique is more flexible than point-to-point parametric correlations, including the use of spatial activation functions for interpolation within a field. Using topographic attributes as input, SANN provides a minimum prediction error for crop yield in 1997, which explains 68% of the spatial variance.

Technical Abstract: The need to transfer information across a range of space-time scales (i.e., scaling) is coupled with the need to predict variables and processes of interest across landscapes. Agricultural landscapes offer a unique set of problems and space-time data availability with the onset of satellite-based positioning and crop yield monitoring. The present study addresses quantification of the spatial variability of rainfed crop yield and near-surface soil water over a farm field using three methods: 1) geostatistical and fractal analyses; 2) multiple linear regression (MLR) using topographic attributes for explanatory variables; and 3) nonparametric estimation by Spatial Artificial Neural Networks (SANN). Method 1 is useful for scaling the spatial moments of each variable to determine appropriate scales of measurement and management. Methods 2 and 3 take advantage of empirical and process knowledge of topographic controls on water movement and microenvironments, where topographic attributes estimated from a digital elevation model at some scale (10 m by 10 m here) help explain the observed spatial variability in crop yield. Soil water (top 30 cm) displays more random spatial variability, and its dynamic nature makes it difficult to predict in both space and time. Despite such variability, spatial structure is evident and can be approximated by simple fractals out to lag distances of about 400 m. The SANN technique is more flexible than point-to-point parametric correlations, including the use of spatial activation functions for interpolation within a field. Using topographic attributes as input, SANN provides a minimum prediction error for crop yield in 1997, which explains 68% of the spatial variance.

Last Modified: 10/1/2014
Footer Content Back to Top of Page