Utilization of Multi-Temporal Microwave Remote Sensing Data within a Machine Learning Approach for the Derivation of Soil Texture

Autor: Meyer, Swen, Rühmann, Jörg; Bönecke, Eric; Kramer, Eckart; Marzahn, Philip;
In:

Conference Schedule

Autor: ESA iving Planet Symposium 2022
Jahr: 2022

Einordung:
Institut: Professur Geodäsie und Geoinformatik

Abstract:
Precision Agriculture (PA) applied on a widespread basis can be a building block for reduced ecosystem degradation without compromising food security. PA is a management strategy that gathers, processes and analyzes temporal, spatial and individual data and combines it with other information to support management decisions according to estimated variability for improved resource use efficiency, productivity, quality, profitability and sustainability of agricultural production (ispa.org 2021). One problem of farmers in the implementation of PA applications is the lack of high spatial resolution soil information. Consequently, an agricultural management that is not adapted to site specific variable conditions, like soil properties can lead to harmful emissions into the surrounding ecosystems, while at the same time not gaining a maximum yield.
The EU-funded research project ‘pH-BB: Precision liming in Brandenburg’ aims at developing innovative nutrient management strategies based on proximal soil sensing data. In the project, an 800-hectare farm in Brandenburg close to Frankfurt (Oder) was intensively monitored with geo-physical on the go sensors, like the “Geophilus-System”. The system measured the soil´s apparent electrical resistivity and soil’s natural gamma activity. Together with 344 reference soil samples that were taken in a depth of 0-30 cm, high resolution soil texture maps with a pixel resolution of 2*2 mts were produced.
For this study, the pH-BB project provided us the data of the 344 soil samples, which were analyzed in the laboratory for the texture fractions clay, silt, and sand according to DIN ISO 11277:2002-08. We used the Google Earth Engine (GEE) to process 1474 Sentinel-1 (S1) SAR scenes available at the study site during the period 2016-03-01 - 2021-11-09. To derive long-term persistent characteristics of backscatter patterns, independent of vegetation properties, the S1 data collection was used to calculate two gridded data sets. The two calculated data sets included the coefficient of variation “vv_CV” and the maximum backscatter “vv_max” along the temporal domain. The reference samples were randomly split in to a training data set (N=241; eg. 70%) and a validation data set (N=103, eg 30%. At sampling locations of the training data set values of the “vv_CV” and “vv_max” were extracted.
We applied a random forest machine learning algorithm to the training dataset to train 3 single models for the target variables clay, silt, and sand using "vv_CV" and "vv_max" as co-variables. The developed models had been further applied to the gridded datasets to calculate final maps of the target variables clay, silt, and sand.
Comparison of the prediction results with the validation data set showed that the spatial distribution of the clay and silt fraction could be predicted with a root mean square error (rmse) of 7 mass % and that of the sand fraction with a rmse of 12 mass-%. A classification of the residual errors according to the German KA5 scheme showed that especially in the sand dominated soil classes the prediction errors were lower, whereas they increased in the loamy soil classes (dependent on the clay content).
With a performance 7-12 mass-%, the approach shows good potential for surface soil texture assessments even at high resolution or for global applications or as a first guess for high resolution soil monitoring, with devices such as the Geophilus-System.

Ansprechpartner

Weitere Information im WWW

Zur Publikation: Mitarbeiter, Projekte
Weitere Publikationen im Forschungsteilschwerpunkt