Multispectral image time-series have been promising for some years; yet, the substantial advance of the technology involved, with unprecedented combinations of spatial, temporal, and spectral capabilities for remote sensing applications, raises new challenges, in particular, the need for methodologies that can process the different dimensions of satellite information. Considering that the multi-collinearity problem is present in remote sensing time-series, regression models are widespread tools to model multi-way data. This paper presents the results of the analysis of a high order data of Sentinel-2-time series, conducted in the framework of extreme weather event. A feature extraction method for multi-way data, N-CovSel was used to identify the most relevant features explaining the loss of yield in Mediterranean vineyards during the 2019 heatwave. Different regression models (uni-way and multi-way) from features extracted from the N-CovSel algorithm were calibrated based on available heat wave impact data for 107 vineyard blocks in the Languedoc-Roussillon region and multispectral time-series predictor data for the period May to August. The performance of the models was evaluated by the r 2 and the root mean square of error (RMSE) as follows: for the temporal N-PLS model ( r 2 = 0.62—RMSE = 11%), for the spatial N-PLS model ( r 2 = 0.61—RMSE = 12%) and the temporal-spectral PLS model ( r 2 = 0.63—RMSE = 11%). The results validated the effectiveness of the proposed N-CovSel algorithm in order to reduce the number of total variables and restricting it to the most significant ones. The N-CovSel algorithm seems to be a suitable choice to interpret complex multispectral imagery by temporally discriminating the most appropriate spectral information.