Identifying and characterizing yield limiting soil factors with the aid of remote sensing and data mining techniques

摘要

Soil provides crop with nutrients, water and root support. But, soils vary a great deal in terms of origin, appearance, characteristics and production capacity. Better understanding of the causality between yield and yield-limiting soil factor(s) is essential for site-specific crop management. The objectives of this study were deriving a spatiotemporal yield trend map of a 144 km(2) paddy rice growing region located at an alluvial plain in southwestern Taiwan from satellite images and exploring the potential yield-limiting soil factor(s) in conjunction with general soil survey data. Due to the complexity of data sets, classification and regression trees analysis (CART) was used to relate soil characteristics to yield classes in the spatiotemporal yield trend map, and followed by comparisons of soil characteristics between those consistently-high and -low yielding areas to explore the interactions between yields and soil properties. Through the above data mining analysis, high soil pH, severe leaching loss of applied nitrogen fertilizers, and excessive reductive root environment were suspected to be the major soil related low-yielding mechanisms spread within studied region. Soil characteristics that induced these low-yielding mechanisms were identified and mapped. Error analysis indicated that 61.8 % of the consistently low-yield areas could be correctly identified by just a few soil characteristics. Improvements of management practices to alleviate the negative effects on yields were also proposed based on the identified low yielding mechanisms. Our study highlighted the pressing need and possible methodologies to adjust management strategies for narrowing yield variability and increasing crop production.