Spectral Prediction App

Try our app for predicting soil properties from spectral measurements, using the Bayesian Additive Regression Trees algorithm. After uploading your spectral measurements you can retrieve predictions of specified soil properties together with associated uncertainty estimates.

Launch Spectral Prediction App

Linux Installation

Download and install:

> sudo apt-get install gdebi-core wget
> wget http://spectpred.qed.ai/media/spectpred_desktop/SpectralPrediction/origin/tags/v1.2.5/linux/Release/SpectralPrediction-1.2.5-Linux.deb
> sudo gdebi SpectralPrediction-1.2.5-Linux.deb

Run

> /usr/bin/SpectralPrediction

System Description

The diagram below shows the workflow of the Spectral Prediction App (SPA). This system has two main parts:

  1. the spectral prediction engine, which is based on the Bayesian Additive Regression Trees algorithm, and is hosted in local field laboratories, and
  2. a centralized database of spectral and webchemistry data, hosted on a central server, whose contents are uploaded by collaborating laboratories.

During operation, these two parts have a symbiotic relationship that requires constant communication: local field labs will regularly retrieve the most updated prediction models from the central server, and the central server regularly receives new uploaded data from the local field labs in order to train better prediction models.

prediction workflow

BART Algorithm

The BART algorithm, devised by Robert McCulloch (University of Chicago) and his collaborators, has been shown to produce state-of-the-art prediction results in many practical applications, while also providing natural statistical uncertainty estimates through a Bayesian framework. These properties can be compared with random forests and neural networks, which, while often exhibiting good performance in practice, do not provide rigorous uncertainty estimates. From a high-level perspective, the BART algorithm uses Markov Chain Monte Carlo algorithm to calculate the posterior distributions for each regression tree. The disadvantage of BART is the intense computation required for MCMC convergence. However, the centralized server relieves local field labs of this computational burden by executing the training (on advanced hardware) and caching the results for other parties to use.

For more details about the BART algorithm, see the following documents:

Training Dataset Prediction Results

Using a training dataset of ~2000 samples acquired from AfSIS Phase I, we compared the predictions for eight kinds of soil properties (C,N,Al,Ca,K,Mg,Na,pH) across three kinds of spectral machines (MIR, NIR, Alpha) with their true measured values. Clicking the button below will display a grid of interactive scatterplots that illustrate this comparison.

View Training Dataset Prediction Results