A comparison between machine learning and functional geostatistics approaches for data-driven analyses of solid transport in a pre-Alpine stream

Keywords

Statistical learning
Code:
66/2020
Title:
A comparison between machine learning and functional geostatistics approaches for data-driven analyses of solid transport in a pre-Alpine stream
Date:
Thursday 29th October 2020
Author(s):
Didkovsky, O.; Ivanov, V.; Papini, M.; Longoni, L.; Menafoglio, A.
Download link:
Abstract:
We address the problem of providing data-driven models for sediment transport in a pre-Alpine stream in Italy. We base our study on a large set of measurements collected from real pebbles, traced along the stream through Radio Frequency IDentificator (RFID) tags after precipitation events. We propose and evaluate two classes of data-driven models -- based on machine learning and functional geostatistics approaches respectively -- to predict the probability of movement of single pebbles within the stream. The first class is built upon gradient boosting decision trees, and allows one to estimate the probability of movement of a pebble, based on the pebbles’ geometrical features, river flow rate, locations, and subdomain types. The second class is built upon functional kriging, a recent geostatistical technique which allows one to predict a functional profile --i.e., the movement probability of a pebble, as a function of the pebbles geometrical features or of the stream's flow rate-- at unsampled locations in the study area. Although grounded on different perspectives, both these models aim to account for two main sources of uncertainty, namely (i) the complexity of river’s morphological structure, and (ii) the highly-nonlinear dependence between probability of movement, pebble’s size and shape, and the stream’s flow rate. We extensively compare the performances of the two methods in terms of classification accuracy, and show that, although these techniques are grounded on different perspectives, an overall consistency appears between the methods suggesting that both approaches may provide valuable modeling frameworks for the problem at hand. We finally discuss on the use of the developed models in a bottom-up strategy, which starts with the prediction/classification of a single pebble and then integrates the results into a forecast of the grain-size distribution of mobilized sediments.