Quantifying the effect of environments on crop emergence, development and yield using sensing and deep learning techniques
Abstract
The world population is estimated to increase by 2 billion in the next 30 years, and global crop production needs to double by 2050 to meet the projected demands from rising population, diet shifts, and increasing biofuels consumption. Improving the production of the major crops has become an increasing concern for the global research community. However, crop development and yield are complex and determined by many factors, such as crop genotypes (varieties), growing environments (e.g., weather, soil, microclimate and location), and agronomic management strategies (e.g., seed treatment and placement, planting, fertilizer and pest management). To develop next-generation and high-efficiency agriculture production systems, we will have to solve the complex equation consisting of the interactions of genotype, environment and management (GxExM) using emerging technologies. Precision agriculture is a promising agriculture practice to increase profitability and reduce environmental impact using site-specifccurate measurement of crop, soil and environment. The success of precision agriculture technology heavily relies on access to accurate and high-resolution spatiotemporal data and reliable prediction models of crop development and yield. Soil texture and weather conditions are important factors related to crop growth and yield. The percentages of sand, clay and silt in the soil affect the movement of air and water, as well as the water holding capacity. Weather conditions, including temperature, wind, humidity and solar irradiance, are determining factors for crop evapotranspiration and water requirements. Compared to crop yield, which is easy to measure and quantify, crop development effects due to the soil texture and weather conditions within a season can be challenging to measure and quantify. Evaluation of crop development by visual observation at field scale is time-consuming and subjective. In recent years, sensor-based methods have provided a promising way to measure and quantify crop development. Unmanned aerial vehicles (UAVs) equipped with visual sensors, multispectral sensors and/or hyperspectral sensors have been used as a high-throughput data collection tool by many researchers to monitor crop development efficiently at the desired time and at field-scale. In this study, UAV-based remote sensing technologies combining with soil texture and weather conditions were used to study the crop emergence, crop development and yield under the effects of varying soil texture and weather conditions in a cotton research field. Soil texture, i.e., sand and clay content, calculated using apparent soil electrical conductivity (EC [subscript a]) based on a model from a previous study, was used to estimate soil characteristics, including field capacity, wilting point and total available water. Weather data were obtained from a weather station 400 m from the field. UAV imagery data were collected using a high-resolution RGB camera, a multispectral camera and a thermal camera from the crop emergence to before harvesting on a monthly basis. An automatic method to count emerged crop seedlings based on image technologies and a deep learning model was developed for near real-time cotton emergence evaluation. The soil and elevation effects on the stand count and seedling size were explored. The effects of soil texture and weather conditions on cotton growth variation were examined using multispectral images and thermal images during the crop development growth stages. The cotton yield variations due to soil texture and weather conditions were estimated using multiple-year UAV imagery data, soil texture, weather conditions and deep learning techniques. The results showed that field elevation had a high impact on cotton emergence (stand count and seedling size) and clay content had a negative impact on cotton emergence in this study. Monthly growth variations of cotton under different soil textures during crop development growth stages were significant in both 2018 and 2019. Soil clay content in shallow layers (0-40 cm) affected crop development in early growth stages (June and July) while clay content in deep layers (40-70 cm) affected the mid-season growth stages (August and September). Thermal images were more efficient in identifying regions of water stress compared to the water stress coefficient Ks calculated using data of soil texture and weather conditions. Results showed that cotton yield for each one of the three years (2017-2019) could be predicted using the model trained with data of the other two years with prediction errors of MAE = 247 (8.9 %) to 384 kg ha [superscript -1] (13.7 %), which showed that quantifying yield variability for a future year based on soil texture, weather conditions and UAV imagery was feasible. Results from this research indicated that the integration of soil and weather information and UAV-based image data is a promising way to understand the effects of soil and weather on crop emergence, crop development and yield.
Degree
Ph. D.