Estimating Crop Yield from Multi-temporal Satellite Data Using Multivariate Regression and Neural Network Techniques

Abstract

Accurate, objective, reliable, and timely predictions of crop yield over large areas are critical to helping ensure the adequacy of a nation’s food supply and aiding policy makers on import/export plans and prices. Development of objective mathematical models of crop yield prediction using remote sensing is highly desirable. In this study, we develop a new methodology using an artificial neural network (ANN) to estimate and predict corn and soybean yields on a county-by-county basis, in the “corn belt” area in the Midwestern and Great Plains regions of the United States. The historical yield data and long time-series NDVI derived from AVHRR and MODIS are used to develop the models. A new procedure is developed to train the ANN model using the SCE-UA optimization algorithm. The performance of ANN models is compared with multivariate linear regression (MLR) models and validation is made on the model’s stability and forecasting ability. The new algorithms can effectively train ANN models, and the prediction accuracy can be as high as 85 percent. Introduction Crop yield is a key element for rural development and an indicator for national food security. Accurate, objective, reliable, and timely predictions of crop yield over large areas are critical for national food security through policy making on import/export plans and prices. In recent years, a variety of mathematical models relating to crop yield have been proposed (Dan, 1998; Landan et al., 2000; Wheeler et al., 2000; Hansen et al., 2004). Remote sensing techniques have the potential to provide quantitative and timely information on agricultural crops over large areas, and many Estimating Crop Yield from Multi-temporal Satellite Data Using Multivariate Regression and Neural Network Techniques Ainong Li, Shunlin Liang, Angsheng Wang, and Jun Qin different methods have been developed to estimate crop yields (Guérif and Duke, 2000; Liang et al., 2004; Walthall et al., 2004; Doraiswamy et al., 2004; Wu, 2004; Xiao et al., 2005; Tao et al., 2005). One practical approach using satellite data is the development of empirical relationships between the integrated Normalized Difference Vegetation Index (NDVI) and crop yield. NDVI responds to changes in the amount of green biomass, chlorophyll content, and canopy water stress. It is simple and easy to implement, and can be effective in predicting surface properties when the vegetation canopy is not too dense or too sparse (Liang, 2004). The relationship between NDVI and production has been confirmed by various field experiments (Prince and Justice, 1991). Rasmussen (1992) showed that yield could be estimated directly from the regression with NDVI. However, the general drawback of most methods using statistical relationships between NDVI and crop yield is that they have a strong empirical character and that the correlation coefficients are moderate to low (e.g., Groten, 1993; Sharma et al., 1993). Therefore, although many studies have been conducted to estimate and predict crop yield using remote sensing data, the operational systems are mainly based on the anomalies of vegetation indices in a subjective fashion. Development of objective mathematical models using remote sensing is still highly desirable. In this study, we develop a methodology using Artificial Neural Networks (ANN) to simulate and predict corn and soybean yields on a county-by-county basis. NDVI values derived from multi-temporal remote sensing image (such as AVHRR and MODIS) within the crop growth season are used to characterize the whole growing process instead of simply extracting some specified values or using the integrated value. The performance of the ANN model is compared with the multivariate linear regression (MLR) model, and validation is made on the model’s stability. The model’s predictive power for yield as well as its management and update on time and space are discussed in the context of evaluating the feasibility of developing a yield forecasting system. Crop production estimation and forecasts have two components: acres to be harvested and expected yield per acre. We mainly focus on estimating the crop yield per area in this paper. PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING Oc t obe r 2007 1149 Ainong Li is with the Institute of Mountain Hazards and Environment, Chinese Academy of Sciences, Chengdu 610041, China; the Department of Geography, University of Maryland, College Park, MD 20742; and The Center for Disaster Reduction, Chinese Academy of Sciences, Beijing 100029, China (ainong1974@yahoo.com.cn). Shunlin Liang is with the Department of Geography, University of Maryland, College Park, MD 20742. Angsheng Wang is with The Center for Disaster Reduction, Chinese Academy of Sciences, Beijing 100029, China. Jun Qin is with School of Geography, Beijing Normal University, Beijing 100875, China, and the School of Geography, Beijing Normal University, Beijing 100875, China. Photogrammetric Engineering & Remote Sensing Vol. 73, No. 10, October 2007, pp. 1149–1157. 0099-1112/07/7310–1149/$3.00/0 © 2007 American Society for Photogrammetry and Remote Sensing PMSRS-04.qxd 9/14/07 11:20 PM Page 1149

4 Figures and Tables

Cite this paper

@inproceedings{Li2007EstimatingCY, title={Estimating Crop Yield from Multi-temporal Satellite Data Using Multivariate Regression and Neural Network Techniques}, author={Ainong Li and Shunlin Liang and Angsheng Wang and Jun Qin}, year={2007} }