Welcome to Zhanglong's Space

Introducing Stats Journey, an interactive learning platform for statistics and data science education built with modern web technologies.

Presentation at AAGI Annual Science Symposium on systematic design superiority in on-farm agricultural trials.

This is the presentation that I gave on 4th September at AASC 2024 Rottnest Island.

Case Studies in Advanced Analysis of Large Strip On-farm Experiments

Abstract: On-farm experiments (OFE) are gaining attention among farmers and agronomists for testing various research questions on real farms. The Analytics for the Australian Grains Industry (AAGI) has developed several techniques for analysing OFE data. Geographically Weighted Regression (GWR) and the multi-environment trial (MET) technique, which partitions paddocks into pseudo-environments (PEs), have proven effective. Additionally, we have explored the potential of the Generalised Additive Model (GAM) for handling temporal and spatial variability, given its flexibility in accommodating non-linear variables. In this presentation, we will demonstrate case studies using these techniques to analyse OFE data and compare the outcomes of different approaches.

PDF.

This is the 5-minute presentation that I gave to GRDC western panel and growers at Pawsey Centre..

Analytics Innovations in On-farm Experiments

PDF.

In the realm of precision farming, the application of Machine Learning (ML) and Deep Learning (DL) algorithms has proven to be invaluable for analyzing and modeling agricultural data that varies across both space and time. While most research has heavily leaned toward utilizing remote sensing data, particularly in image recognition, the advancements in deep learning methodologies for regression problems have been comparatively slow.

Regression techniques are crucial for large-scale agricultural trials, especially when the goal is to identify optimal management practices for specific locations within a field. To enhance predictive accuracy, it’s essential to incorporate various factors—topographical, environmental, and ground conditions—into the modeling process.

To address this gap, our recent work introduces an innovative framework that leverages spatio-temporal estimation and prediction of agricultural data through deep learning and Gaussian processes. By integrating a deep Gaussian Process component, we explicitly model the spatio-temporal structure of the data, while utilizing standard deep learning techniques to account for other influential factors.

We applied our framework to two real-world datasets, demonstrating its effectiveness and accuracy in yield prediction. This approach not only enhances our understanding of agricultural dynamics but also paves the way for more informed decision-making in precision farming.

Stay tuned for further insights as we delve deeper into the applications and implications of this research!