An assessment of design of support vector machines for snow cover mapping in Ilgaz Forest District region

This study aims to investigate the design and assessment of support vector machines (SVM) for fractional snow cover (FSC) mapping from moderate resolution imaging spectroradiometer (MODIS) data in Ilgaz Forest District region lying within the borders of Çankırı and Kastamonu provinces. For this purpose, thirteen MODIS - Landsat 7/8 image pairs acquired between March 2000 and April 2016 are used. SVM models are trained by using MODIS top-of-atmospheric reflectance values of bands 1-7, normalized difference snow index, normalized difference vegetation index and land cover class as predictor variables. Reference FSC maps are generated from higher spatial resolution Landsat binary snow cover maps. During the training and the testing, the effects of the training data size and the sampling type on the predictive performance of SVM models are investigated. An additional search is also conducted to reveal whether the choice of kernel function has a significant contribution to the FSC mapping performance. The results on the independent test scenes indicate that the developed SVM models with radial basis function (RBF), linear, 2nd order polynomial, 3rd order polynomial and 4th order polynomial kernels are in good agreement with reference FSC data with average values of R ≥ 0.91. In contrast, the standard MODIS snow fraction product, namely, MOD10 FSC, exhibits slightly poorer performance with average R = 0.77. The SVM models with RBF kernel is computationally more efficient in model building with average CPU times of 279, 2,300 and 8,457 seconds for small-, medium- and large-sized training data sets, respectively.