Aarhus University Seal / Aarhus Universitets segl

Amélie Marie Beucher

Predicting artificailly drained areas by means of selective model ensemble

Publikation: KonferencebidragKonferenceabstrakt til konferenceForskning

Subsurface drain pipes in agricultural fields have a large impact on crop yields, the hydrological cycle, the leaching of nutrients and pesticides and numerous soil properties. However, the location of artificially drained areas is often unknown. In Denmark, drainage activities have been carried out since the mid-19th century, and it has been estimated that half of the cultivated area is artificially drained (Olesen, 2009).
A number of machine learning approaches can be used to predict artificially drained areas in geographic space. However, instead of choosing the most accurate model produced with one of these approaches, a better prediction can be achieved by combining the predictions of several models (Caruana et al., 2004, Sollich and Krogh, 1996).
As more approaches become available, the importance of the method used for selecting the models for use in the ensemble increases.
The study aims firstly to train a large number of models to predict the extent of artificially drained areas using various machine learning approaches. Secondly, the study will develop a method for selecting the models, which give a good prediction of artificially drained areas, when used in conjunction.
The approaches employed include decision trees, discriminant analysis, regression models, neural networks and support vector machines amongst others. Several models are trained with each method, using variously the original soil covariates and principal components of the covariates.
With a large ensemble, the time necessary for producing a map can easily exceed the practical limits set for the task. For this reason, the study aims not only to produce the ensemble which gives the best prediction of artificially drained areas, but also to take into account the time necessary for producing a map with each method, when selecting the models.
In this way, the developed method should be able to produce a highly accurate and robust map of artificially drained areas within a limited span of time.
StatusUdgivet - 2017
BegivenhedPedometrics Conference - Wageningen, Holland
Varighed: 26 jun. 20171 jul. 2017


KonferencePedometrics Conference

Se relationer på Aarhus Universitet Citationsformater

ID: 114481606