By incorporating temporal effects into the geographically weighted regression (GWR) model, an extended GWR model, geographically and temporally weighted regression (GTWR), has been developed to deal with both spatial and temporal nonstationarity simultaneously in real estate market data. Unlike the standard GWR model, GTWR integrates both temporal and spatial information in the weighting matrices to capture spatial and temporal heterogeneity. The GTWR design embodies a local weighting scheme wherein GWR and temporally weighted regression (TWR) become special cases of GTWR. In order to test its improved performance, GTWR was compared with global ordinary least squares, TWR, and GWR in terms of goodness-of-fit and other statistical measures using a case study of residential housing sales in the city of Calgary, Canada, from 2002 to 2004. The results showed that there were substantial benefits in modeling both spatial and temporal nonstationarity simultaneously. In the test sample, the TWR, GWR, and GTWR models, respectively, reduced absolute errors by 3.5%, 31.5%, and 46.4% relative to a global ordinary least squares model. More impressively, the GTWR model demonstrated a better goodness-of-fit (0.9282) than the TWR model (0.7794) and the GWR model (0.8897). McNamara's test supported the hypothesis that the improvements made by GTWR over the TWR and GWR models are statistically significant for the sample data.

This paper presents a novel model for blending remote sensing data of high spatial resolution (HSR), taken at infrequent intervals, with those available frequently but at low spatial resolution (LSR) in the context of monitoring and predicting changes in land usage and phenology. Named “SParse-representation-based SpatioTemporal reflectance Fusion Model” (SPSTFM), the model has been developed for predicting HSR surface reflectances through data blending with LSR scenes. Remarkably, this model forms a unified framework for fusing remote sensing images with temporal reflectance changes, phenology change (e.g., seasonal change of vegetation), or type change (e.g., conversion of farmland to built-up area), by establishing correspondences between structures within HSR images of given areas and their corresponding LSR images. Such corresponding relationship is achieved by means of the sparse representation, specifically by jointly training two dictionaries generated from HSR and LSR difference image patches and sparse coding at the reconstruction stage. SPSTFM was tested using both a simulated data set and an actual data set of Landsat Enhanced Thematic Mapper Plus-Moderate Resolution Imaging Spectroradiometer acquisitions. It was also compared with other related algorithms on two types of data: images primarily with phenology change and images primarily with land-cover type change. Experimental results demonstrate the superiority of SPSTFM in capturing surface reflectance changes on both categories of images.

We propose a novel spatiotemporal fusion method based on deep convolutional neural networks (CNNs) under the application background of massive remote sensing data. In the training stage, we build two five-layer CNNs to deal with the problems of complicated correspondence and large spatial resolution gaps between MODIS and Landsat images. Specifically, we first learn a nonlinear mapping CNN between MODIS and low-spatial-resolution (LSR) Landsat images and then learn a super-resolution CNN between LSR Landsat and original Landsat images. In the prediction stage, instead of directly taking the outputs of CNNs as the fusion result, we design a fusion model consisting of high-pass modulation and a weighting strategy to make full use of the information in prior images. Specifically, we first map the input MODIS images to transitional images via the learned nonlinear mapping CNN and further improve the transitional images to LSR Landsat images via the fusion model; then, via the learned SR CNN, the LSR Landsat images are supersolved to transitional images, which are further improved to Landsat images via the fusion model. Compared with the previous learning-based fusion methods, mainly referring to the sparse-representation-based methods, our CNNs-based spatiotemporal method has the following advantages: 1) automatically extracting effective image features; 2) learning an end-to-end mapping between MODIS and LSR Landsat images; and 3) generating more favorable fusion results. To examine the performance of the proposed fusion method, we conduct experiments on two representative Landsat-MODIS datasets by comparing with the sparse-representation-based spatiotemporal fusion model. The quantitative evaluations on all possible prediction dates and the comparison of fusion results on one key date in both visual effect and quantitative evaluations demonstrate that the proposed method can generate more accurate fusion results.