|Zhang, Huihui - Texas A&M University|
|Lacey, Ron - Texas A&M University|
|Martin, Daniel - Dan|
|Bora, Ganesa - North Dakota State University|
Submitted to: International Journal of Agricultural and Biological Engineering
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/16/2009
Publication Date: 12/30/2009
Citation: Zhang, H., Lan, Y., Lacey, R., Huang, Y., Hoffmann, W.C., Martin, D.E., Bora, G. 2009. Analysis of variograms with various sample sizes from a multispectral image. International Journal of Agricultural and Biological Engineering. 2:62-69.
Interpretive Summary: Advanced remote sensing (imaging) technologies possess a capability to instantaneously acquire vegetative reflectance data but new analytical methods are needed to characterize the spatial variability of plant conditions within a field. A four-wavelength multispectral camera was used to acquire airborne images for the spatial analysis of weed infestations. A study was undertaken to analyze different sample sizes of multispectral images. Spatial analyses were used to generate a Normalized Difference Vegetation Index (NDVI) image for a subset of the larger multispectral image. By fitting with numerical models, different sample sizes were identified for specific imaging wavelengths. The information will be particularly useful as a guide for field sampling. Remote sensing technologies and spatial analysis techniques evaluated here will increase the speed of sampling and enhance the use of aerially-acquired images for precision application in pest management.
Technical Abstract: Variograms play a crucial role in remote sensing application and geostatistics. In this study, the analysis of variograms with various sample sizes of remotely sensed data was conducted. A 100 X 100 pixel subset was chosen from an aerial multispectral image which contained three wavebands, green, red, and near infra-red (NIR). An NDVI image also was derived from the subset image. These four image datasets were imported into R software for spatial analysis. Variograms were examined on these four full image datasets and sub-samples using a design-based simple random sampling method. For NIR and red band data, the variograms and the parameters of fitted models became more stable and consistent when sample size was larger than half of the full dataset. Variograms computed on green band and NDVI data did not change with sample size.