Abstract

The approach of the Statistical Reporting Service (SRS) for using LANDSAT remote sensor data is to use it as an auxiliary variable with existing operational ground surveys. SRS objectives have been to investigate the use of LANDSAT data to improve crop-acreage estimates at several levels for which acreage statistics are needed; such as counties, groups of counties such as Crop Reporting Districts (CRD1s), and entire states.

To determine the feasibility of these objectives, the Illinois crop-acreage experiment was established in 1975. The experiment employs LANDSAT data for the state of Illinois and data from SRS's June Enumerative Survey (JES) for Illinois. The JES data was collected and edited by the Illinois Cooperative Crop Reporting Service. In addition the JES data was supplemented by monthly-updates conducted throughout the growing season and by low-altitude color-infrared photography for 202 of the 300 JES segments in Illinois.

This paper describes:

1. The statistical methodology for the auxiliary use of LANDSAT data to estimate crop acreages,

2. The procedure for designing the pixel classifier which is required by this methodology, and

3. Results obtained by applying this methodology for three LANDSAT frames in western Illinois.

Software systems have been developed jointly by SRS and the Center for Advanced Computation of the University of Illinois which implement the estimation methodology.

The use of LANDSAT data as an auxiliary variable developed from a realization that using LANDSAT data as a survey variable produces biased estimates. The two major types of bias in using LANDSAT data as a survey variable are:

1. Mensuration biases due to the large pixel size of the LANDSAT data (57m x 79m), and

2. Classifier-related procedural biases due to different discrimination functions (linear or quadratic), training sets, prior probabilities, and classification categories used in the design of the classifier.

Date of this Version

1977

Share

COinS