Reliability Evaluation and Migration of Wetland Samples

Abstract

As one of the most important steps in classification and mapping research, sample acquisition and updating cost a lot of time and energy of researchers. The high temporal and spatial dynamic characteristics of wetland makes the reliable wetland sample selection more challenging. It is especially important for time series wetland classification to address the problem of how to apply sample sets to images with different periods. This paper evaluated the reliability of the historical samples by using optical and SAR remote sensing data from the perspective of water inundation frequency, and three sample migration methods based on rule set, reclassification and spectral similarity were proposed to carry out wetland classification experiment in the Tibetan Plateau. In addition, the relationship between the number of samples and classification accuracy is analyzed. The migration and reuse of sample sets can quickly obtain accurate wetland sample sets, which lays the foundation for the study of wetland mapping in time series.

Integration of Sentinel-1 and Sentinel-2 Data for Ground Truth Sample Migration for Multi-Temporal Land Cover Mapping

“…Previous studies have indicated the suitability of these parameters for pixel similarity comparison between satellite images at different times [15,40,49]. Spectral vectors are composed of distance and direction for land surface bodies [39]. The ED measures the spectral distance (length) between two spectral vectors [45].…”

“…SAD measures the angle between two spectral vectors to represent the similarity between the mentioned vectors [50] and is not sensitive to illumination or shade and can therefore emphasize target spectra-shape characteristics [15]. SAD equals 1 in the case of complete similarity between two training sample pixels [39]. In Formula (2), θ is the spectral angle, X i(t 1 ) is the reference training sample pixel at time t 1 , Y i(t 2 ) is the target training sample pixel at time t 2 , and i shows the spectral band index from 1 to N (number of bands).…”

“…Qualified and updated training samples are prerequisites for the majority of change analyses using multi-temporal satellite images [33]. Additionally, training sample selection is a subjective task dependent on the interpreter [39]. Therefore, training sample migration can assist in addressing the shortage of top-quality training samples required for supervised classification models [15,32].…”

“…Therefore, training sample migration can assist in addressing the shortage of top-quality training samples required for supervised classification models [15,32]. This technique can be employed to produce sufficient training samples for different LCs [15,39,40] and facilitate the updating of LC maps [30,31]. This study aims to investigate ground truth sample migration based on Sentinel-1 (S1) and Sentinel-2 (S2) data.…”

Studying How Machine Learning Maps Mangroves in Moderate-Resolution Satellite Images

“…In the existing mangrove-related research, reference data was collected manually through field surveys and interpretation of very high-resolution images. Studies [39] and [40] explored automated methods for collecting training samples but required existing products as references, making them inapplicable to land cover types lacking specific available products. Murray et al (2022) developed a global reference dataset from seven coastal ecosystems (coastTrain: https://www.coasttrain.org), including mangroves, to support current and future remote sensing classification models [41].…”