This paper proposes the estimation of growth stage in cabbage farmland using multi-temporal satellite images. Although, the growth stages in the farming are defined as the establishing period, the growing period, the heading period, and harvest period, they are not applicable to satellite image analysis from the investigation of spectral features obtained on the ground. We show the spectral growth stages based on vegetable cover ratio, which are the first period, the middle period and the latter period, and adaptable to satellite image analysis. We propose a technique to identify the spectral growth stages, by using change of multi-temporal images, which were misclassified by a single satellite image analysis. The accuracy of the growth stage estimation is 76% using a single satellite image and 89% using multi-temporal ones. We demonstrate that the technique using multi-temporal images improves the accuracy of estimation.
In order to make the map of grain protein content of winter wheat in large area, we investigated the relationship between grain protein content and spectral response of visible and near infrared bands of satellite data. SPOT, QuickBird and Terra/ASTER satellites data were used, and they were mainly observed from June to July in the year from 2002 to 2004, selecting 4 study areas in central and eastern part of Hokkaido. Target wheat variety was "Hokushin". The results showed that we could estimate the grain protein content using the normalized deference vegetation index (NDVI) with more than 0.6 of the coefficient of determination (R2) between three weeks before maturing stage and the stage. This observation period almost corresponds to the period from late June to mid-July in Hokkaido and we evaluated the possibility to get satellite data using Terra/MODIS and Aqua/MODIS satellites data for study area. As the results, we would have enough chance to get clear satellite data in this period using the SPOT sensors that has pointing ability. The distribution map of the grain protein content can be obtained by applying the regression formula between NDVI and the protein content to each pixel of wheat field extracted from the satellite data, and makes clear the fluctuation of protein content not only in an area level but also in a field level. In addition, this map can provide effective information for crop management such as nitrogen fertilizer rate in successive years.