Forecast of wheat yield throughout the agricultural season using optical and radar satellite images |
| |
| Institution: | 1. United States Department of Agriculture—Agricultural Research Service, Animal and Natural Resources Institute, Crop Systems and Global Change Laboratory, Bldg 001, Rm 342, Barc-West, 10300 Baltimore Avenue, Beltsville, MD 20705, USA;2. Norwegian University of Life Sciences, PO Box 5003, NO-1432 As, Norway;1. Technische Universität München, LS Genetik, 85354 Freising, Germany;2. Institute of Biochemical Plant Pathology, Helmholtz Zentrum München, 85764 Neuherberg, Germany;1. Department of Biophysics, Biology Faculty, M.V. Lomonosov Moscow State University, 119992 Moscow, Russia;2. Technical University Berlin, Institute of Chemistry, Max-Volmer-Laboratory of Biophysical Chemistry, Straβe des 17. Juni 135, D-10623 Berlin, Germany;1. China University of Geosciences, School of Information Engineering, 29 Xueyuan Road, Beijing 100083, China;2. Peking University, Institute of Remote Sensing and GIS, 5 Yiheyuan Road, Beijing 100871, China;1. INRA, UMR791 Modélisation systémique appliquée aux ruminants, 16 rue Claude Bernard, 75231 Paris cedex 05, France;2. AgroParisTech, UMR791 Modélisation systémique appliquée aux ruminants, 16 rue Claude Bernard, 75231 Paris cedex 05, France |
| |
| Abstract: | The aim of this study is to estimate the capabilities of forecasting the yield of wheat using an artificial neural network combined with multi-temporal satellite data acquired at high spatial resolution throughout the agricultural season in the optical and/or microwave domains. Reflectance (acquired by Formosat-2, and Spot 4–5 in the green, red, and near infrared wavelength) and multi-configuration backscattering coefficients (acquired by TerraSAR-X and Radarsat-2 in the X- and C-bands, at co- (abbreviated HH and VV) and cross-polarization states (abbreviated HV and VH)) constitute the input variable of the artificial neural networks, which are trained and validated on the successively acquired images, providing yield forecast in near real-time conditions. The study is based on data collected over 32 fields of wheat distributed over a study area located in southwestern France, near Toulouse. Among the tested sensor configurations, several satellite data appear useful for the yield forecasting throughout the agricultural season (showing coefficient of determination (R2) larger than 0.60 and a root mean square error (RMSE) lower than 9.1 quintals by hectare (q ha?1)): CVH, CHV, or the combined used of XHH and CHH, CHH and CHV, or green reflectance and CHH. Nevertheless, the best accurate forecast (R2 = 0.76 and RMSE = 7.0 q ha?1) is obtained longtime before the harvest (on day 98, during the elongation of stems) using the combination of co- and cross-polarized backscattering coefficients acquired in the C-band (CVV and CVH). These results highlight the high interest of using synthetic aperture radar (SAR) data instead of optical ones to early forecast the yield before the harvest of wheat. |
| |
| Keywords: | Wheat Yield forecast Optical Microwave TerraSAR-X Radarsat-2 Formosat-2 Spot-4/5 Artificial neural networks |
| 本文献已被 ScienceDirect 等数据库收录！ |
|
