Remote sensing provides rich sources of data for the monitoring of land surface dynamics. However, single-sensor systems are constrained from providing spatially high-resolution images with high revisit frequency due to the inherent sensor design limitation. To obtain images high in both spatial and temporal resolutions, a number of image fusion algorithms, such as spatial and temporal adaptive reflectance fusion model (STARFM) and enhanced STARFM (ESTARFM), have been recently developed. To capitalize on information available in a fusion process, we propose a Bayesian data fusion approach that incorporates the temporal correlation information in the image time series and casts the fusion problem as an estimation problem in which the fused image is obtained by the Maximum A Posterior (MAP) estimator. The proposed approach provides a formal framework for the fusion of remotely sensed images with a rigorous statistical basis; it imposes no requirements on the number of input image pairs; and it is suitable for heterogeneous landscapes. The approach is empirically tested with both simulated and real-life acquired Landsat and Moderate Resolution Imaging Spectroradiometer (MODIS) images. Experimental results demonstrate that the proposed method outperforms STARFM and ESTARFM, especially for heterogeneous landscapes. It produces surface reflectances highly correlated with those of the reference Landsat images. It gives spatio-temporal fusion of remotely sensed images a solid theoretical and empirical foundation that may be extended to solve more complicated image fusion problems.

Land surface temperature (LST) is a key diagnostic indicator of agricultural water use and crop stress. LST data retrieved from thermal infrared (TIR) band imagery, however, tend to have a coarser spatial resolution (e.g., 100 m for Landsat 8) than surface reflectance (SR) data collected from shortwave bands on the same instrument (e.g., 30 m for Landsat). Spatial sharpening of LST data using the higher resolution multi-band SR data provides an important path for improved agricultural monitoring at sub-field scales. A previously developed Data Mining Sharpener (DMS) approach has shown great potential in the sharpening of Landsat LST using Landsat SR data co-collected over various landscapes. This work evaluates DMS performance for sharpening ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS) LST (~70 m native resolution) and Visible Infrared Imaging Radiometer Suite (VIIRS) LST (375 m) data using Harmonized Landsat and Sentinel-2 (HLS) SR data, providing the basis for generating 30-m LST data at a higher temporal frequency than afforded by Landsat alone. To account for the misalignment between ECOSTRESS/VIIRS and Landsat/HLS caused by errors in registration and orthorectification, we propose a modified version of the DMS approach that employs a relaxed box size for energy conservation (EC). Sharpening experiments were conducted over three study sites in California, and results were evaluated visually and quantitatively against LST data from unmanned aerial vehicles (UAV) flights and from Landsat 8. Over the three sites, the modified DMS technique showed improved sharpening accuracy over the standard DMS for both ECOSTRESS and VIIRS, suggesting the effectiveness of relaxing EC box in relieving misalignment-induced errors. To achieve reasonable accuracy while minimizing loss of spatial detail due to the EC box size increase, an optimal EC box size of 180-270 m was identified for ECOSTRESS and about 780 m for VIIRS data based on experiments from the three sites. Results from this work will facilitate the development of a prototype system that generates high spatiotemporal resolution LST products for improved agricultural water use monitoring by synthesizing multi-source remote sensing data.

Robust information on consumptive water use (evapotranspiration, ET) derived from remote sensing can significantly benefit water decision-making in agriculture, informing irrigation schedules and water management plans over extended regions. To be of optimal utility for operational usage, these remote sensing ET data should be generated at the sub-field spatial resolution and daily-to-weekly timesteps commensurate with the scales of water management activities. However, current methods for field-scale ET retrieval based on thermal infrared (TIR) imaging, a valuable diagnostic of canopy stress and surface moisture status, are limited by the temporal revisit of available medium-resolution (100 m or finer) thermal satellite sensors. This study investigates the efficacy of a data fusion method for combining information from multiple medium-resolution sensors toward generating high spatiotemporal resolution ET products for water management. TIR data from Landsat and ECOSTRESS (both at ~ 100-m native resolution), and VIIRS (375-m native) are sharpened to a common 30-m grid using surface reflectance data from the Harmonized Landsat-Sentinel dataset. Periodic 30-m ET retrievals from these combined thermal data sources are fused with daily retrievals from unsharpened VIIRS to generate daily, 30-m ET image timeseries. The accuracy of this mapping method is tested over several irrigated cropping systems in the Central Valley of California in comparison with flux tower observations, including measurements over irrigated vineyards collected in the GRAPEX campaign. Results demonstrate the operational value added by the augmented TIR sensor suite compared to Landsat alone, in terms of capturing daily ET variability and reduced latency for real-time applications. The method also provides means for incorporating new sources of imaging from future planned thermal missions, further improving our ability to map rapid changes in crop water use at field scales.

In classification of satellite images acquired over smallholder agricultural landscape with complex spectral profiles of various crop types, exploring image spatial information is important. The deep convolutional neural network (CNN), originally designed for natural image recognition in the computer vision field, can automatically explore high level spatial information and thus is promising for such tasks. This study tried to evaluate different CNN structures for classification of four smallholder agricultural landscapes in Heilongjiang, China using pan-sharpened 2 m GaoFen-1 (meaning high resolution in Chinese) satellite images. CNN with three pooling strategies: without pooling, with max pooling and with average pooling, were evaluated and compared with random forest. Two different numbers (~70,000 and ~290,000) of CNN learnable parameters were examined for each pooling strategy. The training and testing samples were systematically sampled from reference land cover maps to ensure sample distribution proportional to the reference land cover occurrence and included 60,000–400,000 pixels to ensure effective training. Testing sample classification results in the four study areas showed that the best pooling strategy was the average pooling CNN and that the CNN significantly outperformed random forest (2.4–3.3% higher overall accuracy and 0.05–0.24 higher kappa coefficient). Visual examination of CNN classification maps showed that CNN can discriminate better the spectrally similar crop types by effectively exploring spatial information. CNN was still significantly outperformed random forest using training samples that were evenly distributed among classes. Furthermore, future research to improve CNN performance was discussed.

Accurate and frequent monitoring of evapotranspiration (ET) at sub-field scales can provide valuable information for agricultural water management, quantifying crop water use and stress toward the goal of increasing crop water use efficiency and production. Using land-surface temperature (LST) data retrieved from Landsat thermal infrared (TIR) imagery, along with surface reflectance data describing albedo and vegetation cover fraction, surface energy balance models can generate ET maps down to a 30 m spatial resolution. However, the temporal sampling by such maps can be limited by the relatively infrequent revisit period of Landsat data (8 days for combined Landsats 7 and 8), especially in cloudy areas experiencing rapid changes in moisture status. The Sentinel-2 (S2) satellites, as a good complement to the Landsat system, provide surface reflectance data at 10–20 m spatial resolution and 5 day revisit period but do not have a thermal sensor. On the other hand, the Visible Infrared Imaging Radiometer Suite (VIIRS) provides TIR data on a near-daily basis with 375 m resolution, which can be refined through thermal sharpening using S2 reflectances. This study assesses the utility of augmenting the Harmonized Landsat and Sentinel-2 (HLS) dataset with S2-sharpened VIIRS as a thermal proxy source on S2 overpass days, enabling 30 m ET mapping at a potential combined frequency of 2–3 days (including Landsat). The value added by including VIIRS-S2 is assessed both retrospectively and operationally in comparison with flux tower observations collected from several U.S. agricultural sites covering a range of crop types. In particular, we evaluate the performance of VIIRS-S2 ET estimates as a function of VIIRS view angle and cloud masking approach. VIIRS-S2 ET retrievals (MAE of 0.49 mm d⁻¹ against observations) generally show comparable accuracy to Landsat ET (0.45 mm d⁻¹) on days of commensurate overpass, but with decreasing performance at large VIIRS view angles. Low-quality VIIRS-S2 ET retrievals linked to imperfect VIIRS/S2 cloud masking are also discussed, and caution is required when applying such data for generating ET timeseries. Fused daily ET time series benefited during the peak growing season from the improved multi-source temporal sampling afforded by VIIRS-S2, particularly in cloudy regions and over surfaces with rapidly changing vegetation conditions, and value added for real-time monitoring applications is discussed. This work demonstrates the utility and feasibility of augmenting the HLS dataset with sharpened VIIRS TIR imagery on S2 overpass dates for generating high spatiotemporal resolution ET products.