Accurate and continuous PM2.5 data is essential for effective prevention of PM2.5 pollution. Despite the achievements of deep learning methods in estimating PM2.5 concentrations, existing neural network models have relied too much on the self-learning capability and have ignored geographic patterns of PM2.5. Few have taken a geographic perspective when modeling PM2.5, resulting in lower model interpretability. In this paper, rather than inputting spatiotemporal information directly into the networks, we propose an improved geographic pattern based residual neural network (IGeop-ResNet) for estimating PM2.5 concentrations in the Beijing-Tianjin-Hebei region (BTH) of China considering spatial heterogeneity and spatial autocorrelation by introducing spatial eigenvector and attention mechanism, as well as the encoding and embedding methods for temporal categorical variables. A DEM-weighted loss function was introduced to enhance the spatial predictive ability, particularly in high-altitude regions. The results show that the IGeop-ResNet model achieves excellent spatial predictive abilities (R2 of 0.925 in terms of station-based cross-validation) and offers a certain level of interpretability compared to the Ori-STResNet (ordinary directly inputs temporal and spatial information in the ResNet model) and the Geop-ResNet model (without the DEM-weighted loss function). Continuous maps derived from the IGeop-ResNet model suggest the PM2.5 concentrations in the BTH region exhibited a downward trend from 2015 to 2018 and experienced a sharp drop in 2017. The results indicate that NO2 is the Granger cause of PM2.5, while the relationship between SO2 and PM2.5 is insignificant.
Spectral reconstruction technology extracts rich detail information from limited spectral bands, thereby enhancing both of the image quality and the resolution capabilities. It finds application in non-destructive testing, elevating the precision and robustness of detection. Current studies primarily focus on improving the local information perception of convolutional neural networks or modeling long-distance dependencies with Transformer. However, such approaches fail to effectively integrate global–local modeling information, resulting in poor accuracy in image reconstruction. This paper introduces a Progressive CNN-Transformer Alternating Reconstruction Network (PCTARN) to alternately utilize robust convolutional attention and transpose Transformer self-attention. A Dual-Path CNN-Transformer Alternating Reconstruction Module (DPCTARM) is proposed to dynamically introduce global–local dynamic priors at various levels to facilitate extracting high- and low-frequency features. This enhancement effectively strengthens PCTARN’s capability to discern valuable signals. To verify the proposed method, a spectral dataset based on seven selected red tide algae is collected. And a peak signal-to-noise ratio (PSNR) metric of 34.58 dB is achieved, which is at least 0.44 dB higher than the methods such as MAUN and MST++. While the Params and FLOPS are reduced by over 41.9 % and 38.4 %, respectively. Since the performance of the proposed PCTARN depends not only on image quality but also on spectral fidelity, an application of spectral detection on red tide are conducted for this purpose. Four feature bands are selected from multispectral images and reconstructed into 20-band hyperspectral images by using PCTARN. Species identification and cell concentration detection are conducted based on the reconstructed images. The results demonstrate that PCTARN can enhance the spatial signal and spectral peak differences of red tide samples, achieving an identification accuracy of 94.21 % and a coefficient of determination (R2) of 0.9660 in species identification and cell concentration detection, which are respectively improved by 11.55 % and 11.59 % compared to those of 4-band multispectral detection.
The European Space Agency (ESA) under the Climate Change Initiative (CCI) has developed a multi-satellite global, daily Soil Moisture (SM) dataset that has paved the ways for agricultural drought studies. To evaluate the performance of this ESACCI SM, two SM-based indices i.e. parametric distribution-based Standardized Soil Moisture Index (SSMI) and non-parametric distribution-based Empirical Standardized Soil Moisture Index (ESSMI) are computed to characterize agricultural drought in the Southern Plateau and Hills (SPH) in India from 1991 to 2020. SSMI and ESSMI are then compared with the Standardized Precipitation Index (SPI) and Standardized Precipitation Evapotranspiration Index (SPEI). The yearly temporal analysis revealed a consistent pattern among all the four indices with 2003 and 2020 marked as the driest and wettest years, respectively. On the other hand, monthly temporal analysis indicated SSMI and ESSMI lagged behind SPI and ESSMI suggesting a delayed response of SM to precipitation. Spatial distributions of indices showed that the SM-based indices effectively capture temporal variations of dryness or wetness across seasons. The near normal and mild to moderate droughts predominated (both spatially and temporally) the SPH and SSMI better captured the extreme drought areas compared to ESSMI. Further, Dynamic Threshold Run Theory (DTRT) is introduced to identify and characterize drought events based on their duration, frequency, intensity and peak. The findings revealed a resemblance in spatial distribution between the duration and frequency. The drought peak and intensity revealed a moderate nature of drought conditions. Overall, this study highlights the effectiveness of ESACCI SM product to characterize the agricultural droughts.
To attain sustainable development goals and understand urban growth patterns, continuous and precise monitoring of built-up area heights is essential. This helps reveal how urban form evolution impacts the thermal environment. Previous research often used isolated images, ignoring the temporal dimension of thermal infrared and reflectance data from Landsat sensors. Additionally, cost-effective and efficient methods for reconstructing time-series built height are lacking. To fill this knowledge gap, we utilized Landsat time-series data to reconstruct the yearly trends in urban form in Beijing, China, spanning from 1990 to 2020. Continuous Change Detection and Classification (CCDC) time series analysis method was used to identify urban growth and renewal years. Employing a reference height for 2020 and logical reasoning method, we reconstructed the annual dynamics of built-up heights, pinpointing years of significant change. Finally, we analyzed the alterations in urban form over the past three decades and their impact on surface temperature changes. Our change detection method achieved an overall accuracy of 86 %, demonstrating its effectiveness in determining the year of change. When compared with data from Lianjia and LiDAR point cloud, our height reconstruction method showed impressive accuracy, with R2 values of 0.9773 and 0.9526, respectively. Analysis of summer and winter LST values revealed distinct temperature patterns across different building heights, with mid-rise buildings exhibiting the highest LST in summer and low-rise buildings registering the highest LST in winter. During periods of urban growth, both mean and amplitude values of LST increased, while during urban renewal (demolition), they decreased. The date of annual temperature peaks advanced during urban growth but delayed during urban renewal (demolition). Our time series analysis framework offers a new method for understanding the yearly dynamics of urban form and its influence on surface temperature, with potential applications in carbon emission and urban climate modeling studies.
The process of crop type mapping generates land use maps, which serve as critical tools for efficient evaluation of production factors and impacts of agricultural practice. Yet, despite the necessity for comprehensive solutions in space and time, the state of research still exhibits significant limitations in these two dimensions: (1) From a temporal perspective, the primary focus of past research in crop type mapping has been on the economically most meaningful, main-season crops, thereby largely neglecting the explicit study of off-season vegetation despite its pivotal roles in year-round management cycles. (2) Viewed spatially, study areas in crop type mapping show distinct limitations from a multi- and transnational standpoint, despite intense cross-regional and international interrelations of agricultural production and an increasing number of countries publishing crop reference data. With a focus on Europe, this research aims to tackle the two described shortcomings (a) by investigating to what extent a selection of major off-season, winter vegetation types in continental Europe can be classified and (b) by analyzing the transnational applicability of the Hierarchical Crop and Agriculture Taxonomy (HCAT) for remote sensing-based crop type mapping across the European Union (EU). This study uses ESA’s Sentinel-2 satellite data, EU’s administrative farming declarations, and HCAT labels to analyze off-season farming measures, based on a study period from late summer to spring, in Austria, France, Germany, and Slovenia. We demonstrate that deep learning models effectively identify major productive and agroecogically significant winter vegetation in continental Europe. HCAT proves thereby valuable for transnational crop classification, excelling in mixed-country experiments and showing potential for transfer learning. This study’s findings provide a solid foundation for advancing transnational as well as winter and all-year crop type mapping, thereby serving as contribution towards temporally and spatially holistic research on agricultural practices’ sociocultural, economic, and environmental impacts.
Understanding spatial interactions in urban environments has become critical in the context of spatio-temporal big data. However, Spatial–temporal big data often exhibit non-uniformity, necessitating the imputation of spatial interaction relationships derived from the analysis of such data. Previous studies often used simplified grid-based or TAZ approaches that ignore the complex interactions for spatial interaction imputation, leading to limitations in accuracy. In this paper, we proposed a two-layer spatial interaction imputation framework (SIF) for accurate multi-scale spatial interaction imputation. To our knowledge, this is the first time that we impute spatial interactions in multi-scale urban areas. In the first layer, it utilised a hierarchical spatial units division algorithm inspired by Shannon’s information entropy to hierarchically classify study area using point of interest (POI) data; In the second layer, it integrates the classified areas and travel flow data into a spatial interaction graph convolutional network (SI-GCN) for spatial interaction imputation. Two case studies were conducted in Beijing, China and New York City, USA, using over eight million taxi data and one million bike-sharing data. The results showed the superior performance of SIF compared to baseline models. The results also analysed the travel behaviours in both Cities, as well as the impact of social, economic and environmental factors on passengers’ spatial choices when travelling.
Well-facilitated farmland (WFF) construction is greatly responsible for agricultural sustainable development. How to quantitatively plan the WFF construction distribution and schedule is still challenging. This study thus introduced a simple but robust method, and took the typical dryland Yulin city to spatially identify its potential WFF construction areas and temporally determine construction priorities based on public data. By integrating satellite-observed croplands with survey-based statistical data, this study firstly obtained density maps of constant croplands. We found that constant cropland densities decreased from west to east in Yulin city. Jingbian and Dingbian counties of the west gave relatively dense distributions. Secondly, by overlaying evaluation indictors of WFF construction, we found over 96% of constant croplands had WFF construction potentials. Slope and fractional vegetation coverage (FVC) showed evidently spatial differences, which comprehensively reflected the potentials and difficulties of WFF construction. Therefore, an index SF, by considering normalized slope and FVC, was subsequently introduced to rank potential WFF construction priorities. According to the completion ratio and the assumption that giving priorities to develop better basic condition regions, batches of WFF construction areas were identified under the equal proportion planning scenario for each county (S1). Besides, a scenario of city-wide unified planning (S2) was also discussed. WFF construction areas in S2 were further concentrated in northwestern counties compared to those in S1. Both scenarios recommended that construction priorities were given to northwestern counties. This study could provide valuable references for arranging distributions and schedules of WFF construction.
The accurate estimation of Above Ground Biomass (AGB) is the basis for plantation forest carbon trading. This study focused on Picea crassifolia artificial plantations, extracting individual tree crown diameters and heights using Unmanned Aerial Vehicles (UAV) data and calculating the individual tree biomass using allometric growth equations. These results were then used to train a satellite image AGB prediction model. In additional, satellite images were resampled to different resolutions to assess the impact of satellite image resolution on model the accuracy. Finally, the model with the highest accuracy among the deep learning algorithms was selected to predicts the AGB within the P. crassifolia plantation forest. The results indicated that the accuracy of single tree crown diameters extracted from P. crassifolia point clouds significantly surpassed those extracted from general point clouds and Crown Height Model (CHM), while the accuracy of the heights extracted from all three sources was similar; RepLKNet outperformed GoogLeNet and ResNet in identifying plantation forest; random forest slightly outperformed XGBoost in the capability of AGB prediction, while the accuracy of the AGB prediction models initially increasd and then decreasd with satellite image resolution, reaching the highest accuracy at a resolution of 50 m. This indicates that the optimal satellite image resolution for estimating the AGB in the study area was affected by scale effects of 50 m. Compared with the combination of satellite data and manual field measurements, the concurrent use of UAVs and satellites offers significant advantages in terms of efficiency and accuracy. UAVs can replace manual sampling for carbon sequestration transactions for plantations.
Mapping water bodies from remotely sensed imagery is crucial for understanding hydrological and biogeochemical processes. The identification of water extent is mainly dependent on optical and synthetic aperture radar (SAR) images. However, the use of remote sensing for water body mapping is often undermined by the mixed pixel dilemma inherent to traditional hard classification approaches. At the same time, the presence of clouds in optical imagery and speckle noise in SAR imagery, coupled with the difficulty in differentiating between water-like surfaces and actual water bodies, significantly compromise the accuracy of water body identification. This paper proposes a DEEP feature collaborative convolutional neural network (CNN) for Water Super-Resolution Mapping based on Optical and SAR images (DeepOSWSRM), which collaboratively leverages Sentinel-1 and Sentinel-2 imagery to address the challenges of missing data and mixed pixels. The Sentinel-1 image provides complementary water distribution information for the cloudy areas of the Sentinel-2 image, while the Sentinel-2 image enhances the perception capabilities for small water bodies in the Sentinel-1 image. Using PlanetScope imagery as the true reference data, the effectiveness of the proposed method was assessed through two experimental scenarios: one utilizing synthetic coarse-resolution imagery degraded from Sentinel-1 and Sentinel-2 data and another using actual Sentinel-1 and Sentinel-2 data, encompassing both simulated and real cloud conditions. A comparative analysis was conducted against three state-of-the-art CNN-based water mapping methods and two CNN SRM methods. The findings demonstrate that the proposed DeepOSWSRM method successfully produces accurate, fine-resolution water body maps, with its performance mainly benefiting from the fusion of SAR and optical images.