The prediction of PM2.5 concentration is a canonical predictive challenge due to the distribution of PM2.5 appears serious spatiotemporal variability at multiple scales. Currently, using satellite-based remote sensing data to estimate ground-level PM2.5 is a promising method for providing spatiotemporal continuous information of PM2.5. In this letter, we proposed a deep neural network (DNN)-based PM2.5 prediction model to capture the spatiotemporal variability of ground-level PM2.5 using the remote sensing aerosol optical depth (AOD) data from the Himawari-8 satellite along with the conventional meteorological observation variables (denoted as PM25-DNN). The PM25-DNN model was trained and tested using the data from Beijing-Tianjin-Hebei region of China in 2017, and we compared the prediction performance between the PM25-DNN and the current state-of-the-art methods in this field. The results show that the PM25-DNN outperforms the other models with the cross-validated coefficient of determination (R²), root-mean-square error (RMSE), mean prediction error (MPE), and relative prediction error (RPE) were 0.84, 19.9 μg/m³, 11.89 μg/m³, and 41.21%, respectively. Then, the trained PM25-DNN model was applied to estimate the hourly gridded PM2.5 with 1-km spatial resolution. Our results indicate that the DNN architecture can capture the essential spatiotemporal distribution associated with PM2.5 only using AOD data and conventional meteorological observational variables without more handcrafted features. The proposed PM25-DNN model can greatly improve the accuracy of PM2.5 estimation, and it provides a new perspective for PM2.5 monitoring using end-to-end deep learning method.

Airborne eddy covariance (EC) is one of the most effective ways to directly measure turbulent flux at a regional scale. This study aims to find the optimum spatial window length for turbulent heat fluxes calculation from airborne eddy covariance measurements under near neutral to unstable atmospheric stability conditions, to reduce the negative influences from mesoscale turbulence, and to estimate local meaningful turbulent heat fluxes accurately. The airborne flux measurements collected in 2008 in the Netherlands were used in this study. Firstly, the raw data was preprocessed, including de-spike, segmentation, and stationarity test. The atmospheric stability conditions were classified as near neutral, moderately unstable, or very unstable; the stable condition was excluded. Secondly, Ogive analysis for turbulent heat fluxes from all available segmentations of the airborne measurements was used to determine the possible window length range. After that, the optimum window length for turbulent heat flux calculations was defined based on the analysis of all possible window lengths and their uncertainties. The results show that the choice of the optimum window length strongly depends on the atmospheric stability conditions. Under near neutral conditions, local turbulence is mixed insufficiently and vulnerable to heterogeneous turbulence. A relatively short window length is needed to exclude the influence of mesoscale turbulence, and we found the optimum window length ranges from 2000 m to 2500 m. Under moderately unstable conditions, the typical scale of local turbulence is relative large, and the influence of mesoscale turbulence is relatively small. We found the optimum window length ranges from 3900 m to 5000 m. Under very unstable conditions, large convective eddies dominate the transmission of energy so that the window length needs to cover the large eddies with large energy transmission. We found the optimum window length ranges from 4500 m to 5000 m. This study gives a comprehensive methodology to determine the optimizing window length in order to compromise a balance between the accuracy and the surface representativeness of turbulent heat fluxes from airborne EC measurements.

Airborne eddy covariance (EC) measurement is one of the most effective methods to directly measure the surface mass and energy fluxes at the regional scale. It offers the possibility to bridge the scale gap between local- and global-scale measurements by ground-based sites and remote-sensing instrumentations, and to validate the surface fluxes estimated by satellite products or process-based models. In this study, we developed an unmanned aerial vehicle (UAV)-based EC system that can be operated to measure the turbulent fluxes in carbon dioxides, momentum, latent and sensible heat, as well as net radiation and photosynthetically active radiation. Flight tests of the developed UAV-based EC system over land were conducted in October 2020 in Inner Mongolia, China. The in-flight calibration was firstly conducted to correct the mounting error. Then, three flight comparison tests were performed, and we compared the measurement with those from a ground tower. The results, along with power spectral comparison and consideration of the differing measurement strategies indicate that the system can resolve the turbulent fluxes in the encountered measurement condition. Lastly, the challenges of the UAV-based EC method were discussed, and potential improvements with further development were explored. The results of this paper reveal the considerable potential of the UAV-based EC method for land surface process studies.

High-resolution albedo has the advantage of a higher spatial scale from tens to hundreds of meters, which can fill the gaps of albedo applications from the global scale to the regional scale and can solve problems related to land use change and ecosystems. The Sentinel-2 satellite provides high-resolution observations in the visible-to-NIR bands, giving possibilities to generate a high-resolution surface albedo at 10 m. This study attempted to evaluate the performance of the four data-driven machine learning algorithms (i.e., random forest (RF), artificial neural network (ANN), k-nearest neighbor (KNN), and XGBoost (XGBT)) for the generation of a Sentinel-2 albedo over flat and rugged terrain. First, we used the RossThick-LiSparseR model and the 3D discrete anisotropic radiative transfer (DART) model to build the narrowband surface reflectance and broadband surface albedo, which acted as the training and testing datasets over flat and rugged terrain. Second, we used the training and testing datasets to drive the four machine learning models, and evaluated the performance of these machine learning models for the generation of Sentinel-2 albedo. Finally, we used the four machine learning models to generate a Sentinel-2 albedo and compared them with in situ albedos to show the models’ application potentials. The results show that these machine learning models have great performance in estimating Sentinel-2 albedos at a 10 m spatial scale. The comparison with in situ albedos shows that the random forest model outperformed the others in estimating a high-resolution surface albedo based on Sentinel-2 datasets over the flat and rugged terrain, with an RMSE smaller than 0.0308 and R² larger than 0.9472.

Accurate land cover (LC) datasets are the basis for global environmental and climate change studies. Recently, numerous open-source annual LC datasets have been created due to advances in remote sensing technology. However, the agreements and sources of error that affect the accuracy of current annual LC datasets are not well understood, which limits the widespread use of these datasets. We compared four annual LC datasets, namely the CLCD, MCD12Q1, CCI-LC, and GLASS-LC, in the Yellow River Basin (YRB) to identify their spatial and temporal agreement for nine LC classes and to analyze their sources of error. The Mann–Kendall test, Sen’s slope analysis, Taylor diagram, and error decomposition analysis were used in this study. Our results showed that the main LC classes in the four datasets were grassland and cropland (total area percentage > 80%), but their trends in area of change were different. For the main LC classes, the temporal agreement was the highest between the CCI-LC and CLCD (0.85), followed by the MCD12Q1 (0.21), while the lowest was between the GLASS-LC and CLCD (−0.11). The spatial distribution of area for the main LC classes was largely similar between the four datasets, but the spatial agreement in their trends in area of change varied considerably. The spatial variation in the trends in area of change for the cropland, forest, grassland, barren, and impervious LC classes were mainly located in the upstream area region (UA) and the midstream area region (MA) of the YRB, where the percentage of systematic error was high (>68.55%). This indicated that the spatial variation between the four datasets was mainly caused by systematic errors. Between the four datasets, the total error increased along with landscape heterogeneity. These results not only improve our understanding of the spatial and temporal agreement and sources of error between the various current annual LC datasets, but also provide support for land policy making in the YRB.