Remotely sensed vegetation index and LAI for parameter determination of the CSM-CROPGRO-Soybean model when in situ data are not available

Highlights

• Remotely sensed derived light interception and leaf area index can be used to calibrate a crop model.

• For yield, pod weight, and biomass prediction, the use of remote sensing data produce results as good as in situ data.

• Remote sensing data is an alternative when in situ data are not available for model calibration.

Abstract

An agricultural system is a complex combination of many different components that require different types of data for analysis and modeling. Remote sensing information is an alternative source of data for areas that only have a small amount of ground truth data. The goal of this study was to evaluate whether remotely sensed data can be used for calibration of genetic specific parameters (GSPs) with the ultimate goal of yield estimation. This study used the Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) from Moderate Resolution Imaging Spectroradiometer (MODIS) with measured Leaf Area Index (LAI) for soybean fields in Paraná, Brazil and Iowa, USA, to calibrate the cultivar parameters of the CSM-CROPGRO-Soybean model. Three calibration methods were performed including field-measured LAI, remotely sensed derived LAI, and remotely sensed derived Light Interception. The cultivar parameters sensitive to LAI and LI were calibrated for yield with a mean error of -4.5 kg/ha (0.1%) and with a R² of 0.89 for Parana. The availability of crop growth measurements for Iowa resulted in an average RMSE of 895 kg/ha (average nRMSE of 6%), and Willmott agreement index of 0.98 for time-series biomass, and an average RMSE of 941 kg/ha (average nRMSE of 15%) for pod weight. This study showed that remotely sensed LAI and LI from NDVI data can be used for calibration of GSPs with the ultimate goal of improving yield predictions based on local dynamic temporal and spatial variability.

Introduction

Soybean is a major agricultural commodity for Brazil and US in terms of acreage, production, and export. Understanding and estimating how this crop responds to different environmental conditions and crop management practices requires a complex system with combinations of various components that contain a number of interacting biological, physical, and chemical processes (Jones et al., 2017; Tsuji et al., 1998). The Decision Support System for Agrotechnology Transfer (DSSAT) is a suite of software models that simulate such complexity, as it encompasses models for different crops including the CROPGRO model for soybean and other grain legumes (Boote et al., 1998b; Jones et al., 2003; Hoogenboom et al., 2015). For the accurate use of the CROPGRO model for yield prediction, determining the cultivar parameters is essential. Nevertheless, calibration of individual parameters of a crop model may require many experimental datasets and other resources (Fukui et al., 2015; Hoogenboom et al., 2015; Jégo et al., 2010; Jones et al., 2003; Kajumula Mourice et al., 2014; Boote et al., 1998b; Tsuji et al., 1998). Obtaining region- and site-specific in situ data that can be used for regional yield estimation is normally challenging because the observed information usually not available, especially for large areas.

Indirect data collection can be conducted to fill possible in situ gaps. Because remote sensing data possess a significant potential for monitoring vegetation dynamics (Kasampalis et al., 2018) due to their synoptic coverage and frequent temporal sampling (Atzberger, 2013), such data can be used for modeling purposes. Li et al. (2015) stated that a combination of remote sensing and crop growth models can be an effective tool for grain yield estimation. For example, Chakrabarti et al. (2014) assimilated downscaled remote sensing soil moisture from the Soil Moisture and Ocean Salinity (SMOS) mission into the DSSAT-CROPGRO model using an Ensemble Kalman filter-based augmented state-vector technique that estimates states and parameters simultaneously. This framework was implemented in La Plata basin in Brazil for two years and the root mean square error (RMSE) between the assimilated and observed soybean yield were 16.8% during the first growing season and 4.4% during the second season. Fang et al. (2011) integrated the Cropping System Model‐CERES‐Maize with the MODIS LAI products for estimating corn yield in the state of Indiana, USA, and concluded that the inversion of a crop simulation model facilitates several different user input modes and outputs a series of agronomic and biophysical variables, including crop yield. Campos et al. (2018) developed an operational remote sensing approach to assist crop growth models in reproducing actual processes in the field by relating satellite-based remote sensing data and key canopy biophysical parameters. They proposed a relation-based crop coefficient using field data obtained from 11 years of irrigated and rainfed soybean and maize grown in eastern Nebraska. They concluded that the relationship between biomass production and the reflectance is strong, indicating that the use of remote sensing data for a quantitative analysis of crop biomass production and yield is reliable. Therefore, integrating biophysical data from remotely sensed data could improve the yield prediction of crop simulation models (Doraiswamy et al., 2005). The main advantages from incorporating remote sensing data into crop models are the representation of the missing spatial information and the more accurate description of the crop’s actual conditions during various stages of the growing season (Kasampalis et al., 2018). However, remote sensing-based techniques to calibrate crop growth models such as CROPGRO are not fully understood, especially for GSPs, that are critical for predicting crop development and growth and ultimately yield correctly. Using remote sensing to calibrate the model can be an asset for applying the model for gridded applications for large areas and making the use of the model viable when limited observational data are available.

LAI and LI are strong links between remote sensing and the CROPGRO model (Boote et al., 1998a; Hoogenboom et al., 2015). Both can be derived from remote sensing data. As described by Boote et al. (1998a), LAI and LI are state variables because the model simulates LAI over time and uses the LI and subsequent prediction of leaf-to-canopy assimilation, which affects biomass and final yield. Because both LAI and LI are outputs of the model, one can use observed data with model inversion techniques to solve for model parameters that affect LAI and LI. The goal of the study was to understand the utility of remote sensing observations to improve yield predictions from crop growth models. Specific objectives were (1) to test three methods to calibrate the GSPs using field-measured LAI, remote-sensing-derived LAI, and remote-sensing-derived Light Interception (LI), and (2) to compare results with field data to evaluate the feasibility and accuracy of these methods.