Search

Winter catch crops are promoted in the European Union under the Common Agricultural Policy to improve soil health and reduce nitrate leaching from agricultural fields. Currently, Member States often monitor farmers' adoption through on-site inspections for a limited subset of parcels. Because of its potential for region-wide coverage, this study investigates the potential of Sentinel-2 satellite time series to classify catch crops at the field level in Flanders (Belgium). The first objective was to classify catch crops and identify the optimal model and time series input for this task. The second objective was to apply these findings in a real-world scenario, aiming to provide reliable early-season predictions in a separate target year, testing early-season performance and temporal transferability. The following three models were compared: Random Forest (RF), Time Series Forest (TSF), and a One-Dimensional Convolutional Neural Network (1D-CNN). The results showed that, with a limited field-based training dataset, RF produced the most robust results across different time series inputs, achieving a median F1-score of >88% on the best dataset. Additionally, the early-season performance of the models was delayed in the target year, reaching the F1-score threshold of 85% at least one month later in the season compared to the training years, with large timing differences between the models. [ABSTRACT FROM AUTHOR]

Detailed information on forest tree species is crucial to inform management and policy and support environmental and ecological research. Sentinel-2 imagery is useful for obtaining spatially explicit and frequent information on forest tree species due to its suitable spatial, spectral, and temporal resolutions. However, classification workflows often do not generalise well to time periods that are not seen by the model during the calibration phase. This study investigates the temporal transferability of dominant tree species classification. To this end, the Random Forest, Support Vector Machine, and Multilayer Perceptron algorithms were used to classify five tree species in Flanders (Belgium) with regularly spaced Sentinel-2 time series from 2018 to 2022. Cross-year single-year input scenarios were compared with same-year single-year input scenarios to quantify the temporal transferability of the five evaluated years. This resulted in a decrease in overall accuracy between 2.30 and 14.92 percentage points depending on the algorithm and evaluated year. Moreover, our results indicate that the cross-year classification performance could be improved by using multi-year training data, reducing the drop in overall accuracy. In some cases, gains in overall accuracy were even observed. This study highlights the importance of including interannual spectral variability during the training stage of tree species classification models to improve their ability to generalise in time. [ABSTRACT FROM AUTHOR]

Until now, few research has addressed the use of machine learning methods for classification at the sub-pixel level. To close this knowledge gap, in this article, six machine learning methods were compared for the specific task of sub-pixel land-cover extraction in the spatially heterogeneous region of Flanders (Belgium). In addition to the classification accuracy at the pixel and the municipality level, three evaluation criteria reflecting the methods’ ease-of-use were added to the comparison: the time needed for training, the number of meta-parameters, and the minimum training set size. Robustness to changing training data was also included as the sixth evaluation criterion. Based on their scores for these six criteria, the machine learning methods were ranked according to three multi-criteria ranking scenarios. These ranking scenarios correspond to different decision-making scenarios that differ in their weighting of the criteria. In general, no overall winner could be designated: no method performs best for all evaluation scenarios. However, when both time available for preprocessing and the magnitude of the training data set are unconstrained, Support Vector Machines (SVMs) clearly outperform the other methods. [ABSTRACT FROM PUBLISHER]