Simultaneous corn and soybean yield prediction from remote sensing data using deep transfer learning

AbstractLarge-scale crop yield estimation is, in part, made possible due to the availability of remote sensing data allowing for the continuous monitoring of crops throughout their growth cycle. Having this information allows stakeholders the ability to make real-time decisions to maximize yield potential. Although various models exist that predict yield from remote sensing data, there currently does not exist an approach that can estimate yield for multiple crops simultaneously, and thus leads to more accurate predictions. A model that predicts the yield of multiple crops and concurrently considers the interaction between multiple crop yields. We propose a new convolutional neural network model called YieldNet which utilizes a novel deep learning framework that uses transfer learning between corn and soybean yield predictions by sharing the weights of the backbone feature extractor. Additionally, to consider the multi-target response variable, we propose a new loss function. We conduct our experiment using data from 1132 counties for corn and 1076 counties for soybean across the United States. Numerical results demonstrate that our proposed method accurately predicts corn and soybean yield from one to four months before the harvest with an MAE being 8.74% and 8.70% of the average yield, respectively, and is competitive to other state-of-the-art approaches.

Download Full-text

YieldNet: A Convolutional Neural Network for Simultaneous Corn and Soybean Yield Prediction Based on Remote Sensing Data

10.1101/2020.12.05.413203 ◽

2020 ◽

Author(s):

Saeed Khaki ◽

Hieu Pham ◽

Lizhi Wang

Keyword(s):

Remote Sensing ◽

Continuous Monitoring ◽

Large Scale ◽

State Of The Art ◽

Remote Sensing Data ◽

Yield Potential ◽

Soybean Yield ◽

Growth State ◽

Sensing Data ◽

Learning Framework

AbstractLarge scale crop yield estimation is, in part, made possible due to the availability of remote sensing data allowing for the continuous monitoring of crops throughout its growth state. Having this information allows stakeholders the ability to make real-time decisions to maximize yield potential. Although various models exist that predict yield from remote sensing data, there currently does not exist an approach that can estimate yield for multiple crops simultaneously, and thus leads to more accurate predictions. A model that predicts yield of multiple crops and concurrently considers the interaction between multiple crop’s yield. We propose a new model called YieldNet which utilizes a novel deep learning framework that uses transfer learning between corn and soybean yield predictions by sharing the weights of the backbone feature extractor. Additionally, to consider the multi-target response variable, we propose a new loss function. Numerical results demonstrate that our proposed method accurately predicts yield from one to four months before the harvest, and is competitive to other state-of-the-art approaches.

Download Full-text

Forecasting of crop yields based on Earth remote sensing data (using soybeans as an example)

Вычислительные технологии ◽

10.25743/ict.2019.24.6.015. ◽

2019 ◽

Author(s):

А.С. Степанов

Keyword(s):

Remote Sensing ◽

Crop Yields ◽

Gaussian Function ◽

Remote Sensing Data ◽

Earth Remote Sensing ◽

Science System ◽

Soybean Yield ◽

Sensing Data ◽

Independent Variable ◽

Simulation Results

Описан подход к прогнозированию урожайности сельскохозяйственных культур с использованием данных дистанционного зондирования Земли. В качестве основного параметра прогностической регрессионной модели использовались значения вегетационного индекса NDVI. В статье приведена оценка возможности раннего прогнозирования до достижения индексом NDVI максимальных значений с применением гауссианы в качестве аппроксимирующей функции, соответствующей еженедельным композитам NDVI. Для пахотных земель Тамбовского р-на Амурской области рассчитана ошибка определения максимума NDVI в зависимости от календарной недели прогнозирования. Построенная модель использована для предварительной оценки урожайности сои в регионе в 2018 г. Purpose. Develop and describe a general approach to forecasting crop yields (using soybeans as an example). Methodology. Crop yields were estimated using regression models. Values of the vegetative index (NDVI) were considered with Vega-Science system. The normalized NDVI values were approximated by the Gauss function using the LevenbergMarquardt algorithm to enable early prediction with Python language. Findings. Values of the normalized index were determined by the preceding fiveyears period. For normalized values, approximating Gaussians were constructed and the parameters of the Gaussian function were calculated. The maximum was predicted for the NDVI values at various calendar weeks of the simulated year. The maximum values of NDVI composites in 20092018 were accounted for 3032 calendar weeks. According to the simulation results, it was found that the average absolute error in predicting the maximum NDVI for 10 years at the weeks 2932 did not exceed 3, for weeks 2728 4 and for the weeks 2126 7. At the next stage, a regression model was built to predict yield, where the calculated NDVI maximum was used as an independent variable, and soybean yield calculated according to the statistics of Rosstat on sown areas and gross soybean harvest in the region acted as an independent variable. Analysis of the error in predicting soybean yield for 2018 was obtained according to the simulation results of 20092017. It was shown that the absolute forecast error when using the data of 2232 calendar weeks of 2018 did not exceed 9.1. Originality/Value. The proposed approach to determining crop yields demonstrate high accuracy, while the method provides the possibility of early forecasting. The use of Earth remote sensing data and developed software modules of Python contribute to the operational formation of the forecast and, accordingly, the possibility of adjusting the agricultural plans.

Download Full-text

The use of remote sensing data to extract information from agricultural land with emphasis on soil salinity

Soil Research ◽

10.1071/sr03033 ◽

2003 ◽

Vol 41 (7) ◽

pp. 1243 ◽

Cited By ~ 45

Author(s):

F. M. Howari

Keyword(s):

Remote Sensing ◽

Soil Salinity ◽

Large Scale ◽

Agricultural Land ◽

Rio Grande Valley ◽

Remote Sensing Data ◽

Rio Grande ◽

The United States ◽

Sensing Data ◽

Matching Techniques

The rapid growth of information technologies has provided exciting new sources of data, interpretation tools, and modelling techniques to soil research and education communities at all levels. This paper presents some examples of the capability of remote sensing data such as Landsat ETM+, airborne visible/infrared imaging spectrometer (AVIRIS), colour infrared aerial photos (CIR), and high-resolution field spectroradiometer (GER 3700) to extract surface information about soil salinity. The study used image processing techniques such as supervised classification, spectral extraction, and matching techniques to investigate types and occurrences of salts in the Rio Grande Valley on the United States–Mexico border. Soil salinity groups were established using soil physico-chemical properties and image elements (absorption-reflectivity profiles, band combinations, grey tones of the investigated images, and textures of soil and vegetation covers as they appear in images). The lack of vegetation or scattered vegetation on salt-affected soil (SAS) surfaces makes it possible to detect salt in several locations of the investigated area. The presented remote sensing datasets reveal the presence of gypsum and halite as the dominant salt crusts in the Rio Grande Valley. This information can help agricultural scientists and engineers to produce large-scale maps of salt-affected lands, which will help improve salinity management in watersheds and ecosystems.

Download Full-text