A Geostatistical Approach for Modeling Soybean Crop Area and Yield Based on Census and Remote Sensing Data

Michel Chaves,Marcelo De Oliveira,Marcelo De Carvalho Alves,Thelma Sáfadi

doi:10.3390/rs10050680

Abstract

Advances in satellite imagery and remote sensing have enabled the acquisition of spatial data at several different resolutions. Geographic information systems (GIS) and geostatistics can be used to link geographic data from different sources. This article discusses the need to improve soybean crop detection and yield prediction by linking census data, GIS, remote sensing, and geostatistics. The proposed approach combines Brazilian Institute of Geography and Statistics (IBGE) census data with an eight-day enhanced vegetation index (EVI) time series derived from Moderate Resolution Imaging Spectroradiometer (MODIS) data to monitor soybean areas and yields in Mato Grosso State, Brazil. In situ data from farms were used to validate the obtained results. Binomial areal kriging was used to generate maps of soybean occurrence over the years, and Gaussian areal kriging was used to predict soybean crop yield census data inside detected soybean areas, which had a downscaling effect on the results. The global accuracy and the Kappa index for the soybean crop detection were 92.1% and 0.84%, respectively. The yield prediction presented 95.09% accuracy considering the standard deviation and probable error. Soybean crop detection and yield monitoring can be improved by this approach.

Highlights

Advances in satellite imagery and remote sensing enable scientists to acquire spatial data at several different resolutions [1]
This paper presents an uncertainty approach to the geostatistical mapping of harvested soybean areas and yields in the State of Mato Grosso (MT), Brazil, combining geographic sets of area-based data with remotely sensed data provided by an eight-day Moderate Resolution Imaging Spectroradiometer (MODIS) enhanced vegetation index (EVI) time series
Soybean crop detection and yield monitoring can be improved by this geostatistical approach

Summary

Introduction

Advances in satellite imagery and remote sensing enable scientists to acquire spatial data at several different resolutions [1]. An increase in spatial resolution is known as downscaling. In the context of remote sensing, downscaling refers to a decrease in the pixel size of remotely sensed images [2]. Data can be downscaled for area-to-point prediction or for areal interpolation, a kriging-based disaggregation technique [3]. Area-to-point prediction uses a downscaling process to predict the same continuous variable at a finer spatial resolution than the input. Once a prediction surface has been created, predictions can be aggregated back to a new set of polygons [2,3]

Methods

Results

Discussion

Conclusion