Modeling spatial data for precision agriculture and remote sensing
Conference Paper
Overview
Identity
Additional Document Info
View All
Overview
abstract
The rapid advancement of sensor and information technologies enables the collection of voluminous spatial data in precision agriculture. On the other hand, the modeling and analysis of spatial data for better management decision making is somewhat lagging behind, in particular at the large-scale production level. In this paper, two spatial datasets were analyzed and compared using three modeling approaches, namely, ordinary least squares (OLS), residual maximum likelihood (REML) geostatistics, and spatial simultaneous autoregressive model (SSAR). The two datasets were collected from a research farm in Texas, USA and a commercial field in Mississippi, USA. Procedures to preprocess and merge different data layers into a common spatial resolution and support were described in detail. The results showed that model parameters estimated by REML and SSAR agreed well with each other, both deviating substantially from OLS estimates. Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) suggested that REML and SSAR had a better overall fit of data than OLS. Leave-one-out cross validation showed that REML predicted soil properties more accurately than OLS. For magnesium, the R2 value increased from 0.46 to 0.80; and for clay, the R2 value increased from 0.22 to 0.34. As more and more spatial data will be collected for precision agriculture, spatial data analysis should be employed by researchers and farmers as a primary means for data modeling, prediction, and decision-making.
