Estimating Fractional Shrub Cover Using Simulated EnMAP Data: A Comparison of Three Machine Learning Regression Techniques

Abstract

Anthropogenic interventions in natural and semi-natural ecosystems often lead to substantial changes in their functioning and may ultimately threaten ecosystem service provision. It is, therefore, necessary to monitor these changes in order to understand their impacts and to support management decisions that help ensuring sustainability. Remote sensing has proven to be a valuable tool for these purposes, and especially hyperspectral sensors are expected to provide valuable data for quantitative characterization of land change processes. In this study, simulated EnMAP data were used for mapping shrub cover fractions along a gradient of shrub encroachment, in a study region in southern Portugal. We compared three machine learning regression techniques: Support Vector Regression (SVR); Random Forest Regression (RF); and Partial Least Squares Regression (PLSR). Additionally, we compared the influence of training sample size on the prediction performance. All techniques showed reasonably good results when trained with large samples, while SVR always outperformed the other algorithms. The best model was applied to produce a fractional shrub cover map for the whole study area. The predicted patterns revealed a gradient of shrub cover between regions affected by special agricultural management schemes for nature protection and areas without land use incentives. Our results highlight the value of EnMAP data in combination with machine learning regression techniques for monitoring gradual land change processes. OPEN ACCESSRemote Sens. 2014, 6 3428

Application of Abundance Map Reference Data for Spectral Unmixing

“…Several studies plotted unmixed fraction versus RD fraction and computed a linear regression of the result. Using this technique, perfect unmixing would result in an R 2 and slope equal to unity, with the intercept equal to zero [10,13].…”

“…These studies also implemented methods to compare AMRD with spectral unmixing results, with the most common strategy being calculation of mean absolute error (MAE) and/or root-mean square error (RMSE) [11][12][13]. Several studies plotted unmixed fraction versus RD fraction and computed a linear regression of the result.…”

“…Specifically, RSRD performs standard classification or unmixing on fine scale imagery (e.g., 1-m GSD National Ecological Observatory Network (NEON) IS), then aggregates fine scale results to co-registered coarse scale pixels (e.g., 15-m GSD Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) IS), thereby creating AMRD for the coarse scale imagery. RSRD methodology is similar to previous studies which have created sub-pixel RD using high-resolution RGB videography [10], RGB imagery [11], multi-spectral imagery [12], and IS imagery [13][14][15].…”

Monitoring Natural Ecosystem and Ecological Gradients: Perspectives with EnMAP

“…In this paper, we present some of the preparatory research activities of the EnMAP Scientific Advisory Group (EnSAG) on natural ecosystems and ecosystem transitions (EnSAG-Ecosystems), alongside the work previously published [37,43] and that published in the current issue [53]. Our aim is to demonstrate the benefit from high quality spaceborne IS data, like EnMAP, for characterizing and quantifying gradual ecosystem transitions and natural ecosystems.…”

“…Few studies have so far used simulated EnMAP data [42] to monitor gradual ecosystem changes [28,37,43]. The high temporal and systematic coverage of the EnMAP system will allow for continuous monitoring and the study of ecosystem processes and properties [44], as demonstrated by some pilot studies using data from experimental spaceborne systems [45,46].…”