Tom Hengl (OpenGeoHub)
Sessions
Diffuse reflectance infrared spectroscopy has become an indispensable tool for rapid estimation of numerous soil health indicators and soil properties as an noninvasive alternative to the wet chemistry. With a hands-on approach, the workshop addresses topics of near infra-red fundamentals, chemometrics, sample preparation, instrumentation techniques and calibrating models predicting the basic soil health indicators. The target group of the workshop are non experts in the topic, i. e. soil managers, advisors or soil scientists. Therefore the workshop will be focused on the rapid in-situ measurement with handheld near-infrared spectrometer (example with NeoSpectra instrument) explaining what soil indicators are detectable with acceptable accuracy, good practice in spectral measurement of the soil samples in situ or in lab, i. e. user-friendly protocols developed in the AI4SoilHealth project. The core of the workshop will be practical training on how to build predictive models in R (and/or Python) using available machine learning tools and open soil spectral libraries (Open Soil Spectra Library, LUCAS etc). The lecturers will assist during the workshop to guide the participants through premade online computational notebooks. Aimed at advancing soil property estimation through fast, accurate, and cost-effective methods, this session underscores spectroscopy as a transformative tool for soil health monitoring, and environmental sustainability in general, positioning participants to integrate these methods into diverse soil-related research areas.
This workshop is an extension to the submitted oral talk “Probability based stratified sampling for both mapping and estimating the population parameters of the soil health indicators at field scale”. In the beginning of the workshop, the participants will become familiar with the appropriate (fit-for-purpose) sampling designs for monitoring the soil health indicators at the field based on the presence/absence of the legacy sampling data and environmental covariates. The brief introduction will be followed by practical coding secession in R using premade computational notebooks. The participants will compare the novel probability based balanced stratified sampling algorithm with 1) simple random sampling, and 2) feature space coverage sampling algorithm using legacy data. In the first experiment, the aim will be to optimize the number of sampling locations of the classic grid sampling performed in a real-world field using the available auxiliary Earth observation environmental layers from the Soil Health Data Cube (https://shdc.ai4soilhealth.eu/). In the second experiment, we will use the available soil maps from the datacube to design a new optimized sampling network for spatiotemporal predictive modeling. The target group of the workshop are non experts in the topic, i. e. soil managers, advisors, young researchers or soil scientists.