Скачать или смотреть GeoPython 2024: GPU-Accelerated Yield Data Interpolation with ML Algorithms in Python

Lucas Johannsen, Sebastian Ramm
GPU-Accelerated Yield Data Interpolation with ML Algorithms in Python

This project pioneers GPU-accelerated interpolation of yield data using powerful ML algorithms in Python. With a focus on precision and efficiency, the implementation of Random Forest variants like RFsp offers a great solution for accurate and scalable interpolation tasks.

In response to the EU's Common Agricultural Policy and the introduction of eco-schemes in 2023, farmers face the challenge of selecting eco-schemes without incurring financial drawbacks. This project addresses the issue by developing a web-based decision support system for farmers, utilizing sensor data from agricultural machinery. Focusing on geo-referenced combine harvester yield data, the research compares traditional methods like Ordinary Kriging with ML-based approaches, specifically Random Forest and Deep Kriging, to interpolate yield maps. Implemented in Python, we leverage the RAPIDS GPU library from NVIDIA to reduce computation time when dealing with high-dimensional data. The study explores the influence of spatial features, the impact of covariates such as Digital Terrain Models, and the effects of varying data densities. Results demonstrate that ML-based interpolation, particularly with Random Forest variants like RFsp, offers precision comparable to or exceeding Ordinary Kriging, with added advantages of automation and an optimal balance between computation time and accuracy. This contributes valuable insights for implementing a robust decision support system for farmers navigating the adoption of eco-schemes.