Spatial Data Analysis in Digital Agriculture Application on Crops Growth Monitoring

Digital agriculture has emerged as a cornerstone of modern agricultural practices. It leverages advanced technologies to collect extensive data on farming activities, which encompasses a wide range of sources, from historical records and traditional farming methods to high-resolution data acquired through modern machinery, drones, and satellite systems. As a key application of Big Data Analytics, digital agriculture provides the foundation for innovations in data-driven farming, and my research is positioned at the intersection of this rapidly evolving field and data science. My primary research focus is monitoring crop fields during the growing season, with the goal of identifying and predicting anomalies in crop development. Early detection of these anomalies allows farmers to implement timely interventions, minimizing potential adverse impacts. Specifically, my study concentrates on analyzing satellite data and developing advanced data analytics techniques to achieve the following key objectives: predicting the growth cycle of winter wheat, identifying its growth stages, and forecasting its yield. Auxiliary datasets, such as weather data, play a critical role in constructing robust machine-learning models. This research also collected weather data corresponding to each analyzed field to enhance the prediction accuracy of yield models. Furthermore, it explores the relationship between weather data and satellite data, integrating weather variables into the yield prediction models to improve their overall performance. The research methodology employed in this study follows the traditional data mining workflow, comprising three primary steps: (1) data collection, (2) data preprocessing, and (3) data analysis. Beyond merely acquiring relevant datasets, a thorough exploration of the data was undertaken to extract additional insights and construct a robust data warehouse. For datasets requiring intricate preprocessing—such as satellite data, which is often affected by significant noise—a comprehensive preprocessing pipeline was developed. This pipeline included the design of a dynamic polynomial model to address missing values, correct outliers, and improve overall data quality. We compared the existing machine learning models, like KNN, Naive Bayes, Gradient Boosting, Logistic Regression, and Random Forest, which were conducted to predict winter wheat growth stages and yield. Additionally, we developed specialized neural network architectures tailored to specific tasks and data types, such as Deep Neural Networks (DNN), Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and hybrid LSTM-CNN models. We aim to enable the identification of the most suitable model for each research objective, ensuring optimal performance across different tasks. At this stage of the project, I have successfully completed the collection and preprocessing of all necessary datasets and finalized the data analysis. This work aims to assess the impact of the data and models on agricultural decision-making processes and to explore their integration into the broader data analytics framework of CONSUS, contributing to the advancement of digital agriculture.