Remote-sensing Model Research Articles

Assessing Remote Sensing-Based Maize Crop Biophysical Characteristics and Evapotranspiration Estimations

The rapid growth in population, climate variability, and decreasing water resources necessitate innovative agricultural practices to ensure food security and resource conservation. This study investigates the effectiveness of various multispectral imagery from remote sensing (RS) platforms, Unmanned Aerial Systems (UAS), PlanetDove microsatellites, Sentinel-2, Landsat 8/9, and proximal MSR-5 in assessing crop biophysical characteristics (CBPCs) and actual crop evapotranspiration (ETa) for maize fields in northeastern Colorado. The research aims to evaluate the accuracy of vegetation indices (VIs) derived from these platforms in estimating key CBPCs, including leaf area index (LAI), crop height (Hc), and fractional vegetation cover (Fc), as well as ETa. Field experiments were conducted during 2022 at the USDA-ARS Limited Irrigation Research Farm in Greeley, Colorado, U.S.A., using different irrigation strategies. Surface reflectance data collected using a handled sensor and observed LAI, Hc, and Fc values, served as ground truth for validating RS estimates. The study applied various statistical analyses to compare the performance of different RS platforms and models. Results indicate that higher-resolution platforms, particularly UAS, provided higher accuracy in estimating VIs and CBPCs than satellite platforms. The study also highlights the influence of environmental conditions on the accuracy of RS models, with locally calibrated models outperforming those developed in dissimilar conditions. The findings underscore the potential of advanced RS technologies in enhancing precision agriculture practices and optimizing water resource management.

Spatiotemporal Variabilities in Evapotranspiration of Alfalfa: A Case Study Using Remote Sensing METRIC and SSEBop Models and Eddy Covariance

Prolonged drought exacerbated by climate change in the Mesilla Valley, one of the major agricultural areas of New Mexico, USA, is causing a shortage of surface water from the Rio Grande for irrigation. Farmers in the Valley are using groundwater for irrigation and complementing it with limited surface water from the river (Rio Grande). Managing irrigation water better is vital to sustaining agriculture in the Valley. Remote sensing (RS)-based crop evapotranspiration (ETa) models offer significant advantages over traditional methods. The ET maps generated by these RS models provide valuable information that can be used to manage irrigation water and crops in water-scarce areas. This study used METRIC and SSEBop RS models to map the ET of alfalfa on a private farm that is managed as commonly practiced in the Valley. The integrated ET values of the two models are compared to those of the ETa measured using the eddy covariance method. The comparison showed that 91.55% of the variability in SSEBop ETa estimates can be explained by the variability in the METRIC ETa estimates, and the variability in eddy covariance ETa can explain 93.07% of the variability in METRIC ETa and 86.01% in the SSEBop Eta estimates. Both METRIC and SSEBop reflected the ETa of alfalfa during full growth and harvesting periods. However, the absolute percent mean relative difference (MRD) of ET was higher for two out of three cuttings by SSEBop (>32%) compared to those for METRIC and eddy covariance. The spatiotemporal variabilities in crop ET estimates using METRIC and SSEBop showed a need to improve on-farm irrigation conveyance and on-the-field irrigation efficiency. Overall, RS models can provide spatiotemporal maps of ET that can be used for decision-making to manage irrigation water better and improve crop yield on a field, farm, and regional scale.

An Application of Machine-Learning Model for Analyzing the Impact of Land-Use Change on Surface Water Resources in Gauteng Province, South Africa

The change in land-use diversity is attributed to the anthropogenic factors sustaining life. The surface water bodies and other crucial natural resources in the study area are being depleted at an alarming rate. This study explored the implications of the changing land-use diversity on surface water resources by using a random forest (RF) classifier machine-learning algorithm and remote-sensing models in Gauteng Province, South Africa. Landsat datasets from 1993 to 2022 were used and processed in the Google Earth Engine (GEE) platform, using the RF classifier. The results indicate nine land-use diversity classes having increased and decreased tendencies, with high F-score values ranging from 72.3% to 100%. In GP, the spatial coverage of BL has shrunk by 100.4 km2 every year over the past three decades. Similarly, BuA exhibits an annual decreasing rate of 42.4 km2 due to the effect of dense vegetation coverage within the same land use type. Meanwhile, water bodies, marine quarries, arable lands, grasslands, shrublands, dense forests, and wetlands were expanded annually by 1.3, 2.3, 2.9, 5.6, 11.2, 29.6, and 89.5 km2, respectively. The surface water content level of the study area has been poor throughout the study years. The MNDWI and NDWI values have a stronger Pearson correlation at a radius of 5 km (r = 0.60, p = 0.000, n = 87,260) than at 10 and 15 km. This research is essential to improve current land-use planning and surface water management techniques to reduce the environmental impacts of land-use change.

The precision of satellite-based net irrigation quantification in the Indus and Ganges basins

Abstract. Even though irrigation is the largest direct anthropogenic interference in the natural terrestrial water cycle, limited knowledge of the amount of water applied for irrigation exists. Quantification of irrigation via evapotranspiration (ET) or soil moisture residuals between remote-sensing models and hydrological models, with the latter acting as baselines without the influence of irrigation, have successfully been applied in various regions. Here, we implement a novel ensemble methodology to estimate the precision of ET-based net irrigation quantification by combining different ET and precipitation products in the Indus and Ganges basins. A multi-model calibration of 15 models independently calibrated to simulate rainfed ET was conducted before the irrigation quantification. Based on the ensemble average, the 2003–2013 net irrigation amounts to 233 mm yr−1 (74 km3 yr−1) and 101 mm yr−1 (67 km3 yr−1) in the Indus and Ganges basins, respectively. Net irrigation in the Indus Basin is evenly split between dry and wet periods, whereas 70 % of net irrigation occurs during the dry period in the Ganges Basin. We found that, although annual ET from remote-sensing models varied by 91.5 mm yr−1, net irrigation precision was within 25 mm per season during the dry period for the entire study area, which emphasizes the robustness of the applied multi-model calibration approach. Net irrigation variance was found to decrease as ET uncertainty decreased, which is related to the climatic conditions, i.e., high uncertainty under arid conditions. A variance decomposition analysis showed that ET uncertainty accounted for 73 % of the overall net irrigation variance and that the influence of precipitation uncertainty was seasonally dependent, i.e., with an increase during the monsoon season. The results underline the robustness of the framework to support large-scale sustainable water resource management of irrigated land.

Fuzzy-based global water quality assessment and water quality cells identification using satellite data

The water environmental impact assessment and management programs increasingly rely on accurate and quantitative estimates of water quality parameters through remote sensing, owing to the limitation of the time-consuming field-based approaches. Numerous studies have utilised the remote-derived water-quality products and existing water quality index WQI models, but they are typically site-specific and yield significant errors for the accurate assessment and monitoring of coastal and inland water bodies. This study presents a generalized WQI model that incorporates a flexible number of parameters, simplifying them to produce comprehensive water quality index values with the fuzzy logic approach. To derive these index values, three major water quality parameters such as Chl, TSS and aCDOM443 were estimated using new remote-sensing models, and the corresponding indices Trophic State Index (TSI), Total Suspended Solids Index (TSSI) and CDOM Index (CI) were produced by a generalized index model. Finally, WQI products were derived based on the Mamdani-based Fuzzy Inference System (FIS) and individual contribution of the water quality parameters to WQI was analysed to establish ‘Water Quality Cells’ WQcells, which are represented by the dominant WQ parameter. The new models were tested on MODIS-Aqua and Sentinel-3 OLCI data in different regional and global oceanic waters. Further, a time series analysis was performed in regional coastal oceanic waters (along the Indian coast) to study the seasonal variations of individual water quality parameters and WQI over the period from 2011 to 2020. The results demonstrated that the FIS is efficient in handling the parameters with varying units and their relative importance. The water quality cells were identified in the bloom-dominated (Arabian Sea), TSS-dominated (Point Calimere, India and Yangtze River estuary, China) and CDOM-dominated (South Carolina coast, USA) regions. The time series analysis revealed that the water quality of the Indian coast exhibits cyclic seasonal variations due to the annual occurrence of the south-west and north-east monsoons. These results are critical for monitoring and assessing the quality of surface waters in coastal and inland environments and enabling water resources managers to formulate and implement management plans for a variety of water bodies cost-effectively.

LASSO (L1) Regularization for Development of Sparse Remote-Sensing Models with Applications in Optically Complex Waters Using GEE Tools

Remote-sensing data are used extensively to monitor water quality parameters such as clarity, temperature, and chlorophyll-a (chl-a) content. This is generally achieved by collecting in situ data coincident with satellite data collections and then creating empirical water quality models using approaches such as multi-linear regression or step-wise linear regression. These approaches, which require modelers to select model parameters, may not be well suited for optically complex waters, where interference from suspended solids, dissolved organic matter, or other constituents may act as “confusers”. For these waters, it may be useful to include non-standard terms, which might not be considered when using traditional methods. Recent machine-learning work has demonstrated an ability to explore large feature spaces and generate accurate empirical models that do not require parameter selection. However, these methods, because of the large number of included terms involved, result in models that are not explainable and cannot be analyzed. We explore the use of Least Absolute Shrinkage and Select Operator (LASSO), or L1, regularization to fit linear regression models and produce parsimonious models with limited terms to enable interpretation and explainability. We demonstrate this approach with a case study in which chl-a models are developed for Utah Lake, Utah, USA., an optically complex freshwater body, and compare the resulting model terms to model terms from the literature. We discuss trade-offs between interpretability and model performance while using L1 regularization as a tool. The resulting model terms are both similar to and distinct from those in the literature, thereby suggesting that this approach is useful for the development of models for optically complex water bodies where standard model terms may not be optimal. We investigate the effect of non-coincident data, that is, the length of time between satellite image collection and in situ sampling, on model performance. We find that, for Utah Lake (for which there are extensive data available), three days is the limit, but 12 h provides the best trade-off. This value is site-dependent, and researchers should use site-specific numbers. To document and explain our approach, we provide Colab notebooks for compiling near-coincident data pairs of remote-sensing and in situ data using Google Earth Engine (GEE) and a second notebook implementing L1 model creation using scikitlearn. The second notebook includes data-engineering routines with which to generate band ratios, logs, and other combinations. The notebooks can be easily modified to adapt them to other locations, sensors, or parameters.

Estimating Carrot Gross Primary Production Using UAV-Based Multispectral Imagery

Gross primary productivity (GPP) is an essential parameter to assess the efficiency of terrestrial ecosystems on carbon transfer. Although GPP is regularly measured with eddy covariance (EC) systems, these are restricted to the tower footprint area, and remote sensing (RS) products have estimated GPP using multispectral vegetation indexes (VIs) from farms to whole ecosystems. Indeed, nowadays, unmanned aerial vehicle (UAV)-based RS technology is becoming more accessible. Accordingly, we propose the estimation of GPP using VIs at high spatial resolutions using UAVs and multi-spectral cameras. A small typical farm in Colombia was cultivated with carrot as our base crop. An EC system was installed to estimate GPP and was used as a reference. A total of 24 VIs from UAV-based RS products were selected and compared with the GPP of the EC system. A cross-validation process was performed, and seven VIs obtained a high R2 score (0.76–0.78). The accumulated GPP, estimated with the best index (NIRv) was 520.3 g C m−2, while the GPP-EC estimate was 580.4 g C m−2 (10.3% error). This work showed that it is possible to estimate the GPP of carrot crops using UAV-based RS, VIs, and linear regression models, which can be used in further research on GPP using UAVs.

Net Primary Productivity Estimation Using a Modified MOD17A3 Model in the Three-River Headwaters Region

Remote sensing (RS) models can easily estimate the net primary productivity (NPP) on a large scale. The majority of RS models try to couple the effects of temperature, water, stand age, and CO2 concentration to attenuate the maximum light use efficiency (LUE) in the NPP models. The water effect is considered the most unpredictable, significant, and challenging. Because the stomata of alpine plants are less sensitive to limiting water vapor loss, the typically employed atmospheric moisture deficit or canopy water content may be less sensitive in signaling water stress on plant photosynthesis. This study introduces a soil moisture (SM) content index and an alpine vegetation photosynthesis model (AVPM) to quantify the RS NPP for the alpine ecosystem over the Three-River Headwaters (TRH) region. The SM content index was based on the minimum relative humidity and maximum vapor pressure deficit during the noon, and the AVPM model was based on the framework of a moderate resolution imaging spectroradiometer NPP (MOD17) model. A case study was conducted in the TRH region, covering an area of approximately 36.3 × 104 km2. The results demonstrated that the AVPM NPP greatly outperformed the MOD17 and had superior accuracy. Compared with the MOD17, the average bias of the AVPM was −9.8 gCm−2yr−1, which was reduced by 91.8%. The average mean absolute percent error was 57.0%, which was reduced by 68.2%. The average Pearson’s correlation coefficient was 0.4809, which was improved by 30.0%. The improvements in the NPP estimation were mainly attributed to the decreasing estimation of the water stress coefficient on the NPP, which was considered the higher constraint of water impact on plant photosynthesis. Therefore, the AVPM model is more accurate in estimating the NPP for the alpine ecosystem. This is of great significance for accurately assessing the vegetation growth of alpine ecosystems across the entire Qinghai–Tibet Plateau in the context of grassland degradation and black soil beach management.

RingMo: A Remote Sensing Foundation Model With Masked Image Modeling

Deep learning approaches have contributed to the rapid development of Remote Sensing (RS) image interpretation. The most widely used training paradigm is to utilize ImageNet pre-trained models to process RS data for specified tasks. However, there are issues such as domain gap between natural and RS scenes, and the poor generalization capacity of RS models. It makes sense to develop a foundation model with general RS feature representation. Since a large amount of unlabeled data is available, the self-supervised method has more development significance than the fully supervised method in remote sensing. However, most of the current self-supervised methods use contrastive learning, whose performance is sensitive to data augmentation, additional information, and selection of positive and negative pairs. In this paper, we leverage the benefits of generative self-supervised learning for RS images, and propose a Remote sensing foundation Model framework called RingMo, which consists of two parts. First, a large-scale dataset is constructed by collecting two million RS images from satellite and aerial platforms, covering multiple scenes and objects around the world. Second, we propose an RS foundation model training method designed for dense and small objects in complicated RS scenes. We show that the foundation model trained on our dataset with RingMo method achieves state-of-the-art on eight datasets across four downstream tasks, demonstrating the effectiveness of the proposed framework. Through in-depth exploration, we believe it is time for RS researchers to embrace generative self-supervised learning and leverage its general representation capabilities to speed up the development of RS applications.

Robust extension of the simple sea-surface irradiance model to handle cloudy conditions for the global ocean using satellite remote sensing data

Sea-surface solar radiation (abbreviated as photosynthetically available radiation, PAR) in the visible wavelength (400–700 nm) is an essential parameter to estimate marine primary productivity and understanding phytoplankton dynamics, upper ocean physics and biogeochemical processes. Although many remote-sensing models were developed to estimate daily PAR (DPAR) from ocean colour data, these models often produce biases in the DPAR products under cloudy-sky and complex atmospheric conditions due to the lack of parameterization to deal with the cloud cover conditions and insufficient in-situ DPAR data. This study presents an Extended Sea-surface Solar Irradiance Model (ESSIM) for estimating DPAR over the global ocean. The ESSIM uses the direct and diffuse components from the Simple sea-surface Solar Irradiance Model (SSIM) along with a new parameter to handle cloudy conditions. The ESSIM produced DPAR products with greater accuracy under both clear and cloudy conditions. Its performance was tested on the time-series MODIS-Aqua images and compared with the concurrent in-situ data and the results from two global models. Results showed that the DPAR values produced by ESSIM agree with in-situ data better than the global models for all-sky conditions (with a mean relative error of 11.267 %; a root mean square error of 5.563 Em−2day−1; and a mean net bias of 2.917 Em−2day−1). The ESSIM performed slightly better than the SSIM for clear conditions and the Frouin's Operational Algorithm (FOA) for all-sky conditions. As the new parameterization accounts for cloudy conditions, the ESSIM produced more accurate results for cloud cover conditions across latitudes (up to 60°). The time-series Level-3 binned MODIS-Aqua data (global gridded) also demonstrated that the ESSIM improved the accuracy of DPAR products and produced spatially and temporally consistent DPAR products over the global ocean regardless of the seasons and sky conditions.

Quantifying the Natural Climate Solution Potential of Agricultural Systems by Combining Eddy Covariance and Remote Sensing

AbstractLivestock agriculture accounts for ∼15% of global anthropogenic greenhouse gas (GHG) emissions. Recently, natural climate solutions (NCS) have been identified to mitigate farm‐scale GHG emissions. Nevertheless, their impacts are difficult to quantify due to farm spatial heterogeneity and effort required to measure changes in carbon stocks. Remote sensing (RS) models are difficult to parameterize for heterogeneous agricultural landscapes. Eddy covariance (EC) in combination with novel techniques that quantitatively match source area variations could help update such vegetation‐specific parameters while accounting for pronounced heterogeneity. We evaluate a plant physiological parameter, the maximum quantum yield (MQY), which is commonly used to calculate gross and net primary productivity in RS applications. RS models often rely on spatially invariable MQY, which leads to inconsistencies between RS and EC models. We evaluate if EC data improve RS models by updating crop specific MQYs to quantify agricultural GHG mitigation potentials. We assessed how farm harvest compared to annual sums of (a) RS without improvements, (b) EC results, and (c) EC‐RS models. We then estimated emissions to calculate the annual GHG balance, including mitigation through plant carbon uptake. Our results indicate that EC‐RS models significantly improved the prediction of crop yields. The EC model captures diurnal variations in carbon dynamics in contrast to RS models based on input limitations. A net zero GHG balance indicated that perennial vegetation mitigated over 60% of emissions while comprising 40% of the landscape. We conclude that the combination of RS and EC can improve the quantification of NCS in agroecosystems.

Optimization of Samples for Remote Sensing Estimation of Forest Aboveground Biomass at the Regional Scale

Accurately estimating forest aboveground biomass (AGB) based on remote sensing (RS) images at the regional level is challenging due to the uncertainty of the modeling sample size. In this study, a new optimizing method for the samples was suggested by integrating variance function in Geostatistics and value coefficient (VC) in Value Engineering. In order to evaluate the influence of the sample size for RS models, the random forest regression (RFR), nearest neighbor (K-NN) method, and partial least squares regression (PLSR) were conducted by combining Landsat8/OLI imagery in 2016 and 91 Pinus densata sample plots in Shangri-La City of China. The mean of the root mean square error (RMSE) of 200 random sampling tests was adopted as the accuracy evaluation index of the RS models and VC as a relative cost index of the modeling samples. The research results showed that: (1) the statistical values (mean, standard deviation, and coefficient of variation) for each group of samples based on 200 experiments were not significantly different from the sampling population (91 samples) by t-test (p = 0.01), and the sampling results were reliable for establishing RS models; (2) The reliable analysis on the RFR, K-NN, and PLSR models with sample groups showed that the VC decreases with increasing samples, and the decreasing trend of VC is consistent. The number of optimal samples for RFR, K-NN, and PLSR was 55, 54, and 56 based on the spherical model of variance function, respectively, and the optimal results were consistent. (3) Among the established models based on the optimal samples, the RFR model with the determination coefficient R2 = 0.8485, RMSE = 12.25 Mg/hm2, and the estimation accuracy P = 81.125% was better than K-NN and PLSR. Therefore, they could be used as models for estimating the aboveground biomass of Pinus densata in the study area. For the optimal sample size and sampling population, the RFR model of Pinus densata AGB was established, combining 26 variable factors in the study area. The total AGB with the optimal samples was 1.22 × 107 Mg, and the estimation result with the sampling population was 1.24 × 107 Mg based on Landsat8/OLI images. Respectively, the average AGB was 66.42 Mg/hm2 and 67.51 Mg/hm2, with a relative precision of 98.39%. The estimation results of the two sample groups were consistent.

Improved global evapotranspiration estimates using proportionality hypothesis-based water balance constraints

Accurate estimation of global evapotranspiration (ET) is critical to understand the water and energy cycles in the Earth system. Satellite-driven ET algorithms serve as an effective way to estimate the global ET. However, many algorithms have been designed independently of water balance constraints, which potentially limit their ability to estimate ET in water-limited and high interception regions. As ET remains one of the most uncertain terms in the global water budgets, incorporating water balance constraints into algorithms should improve the performance of ET estimates. In this study, we developed a general solution (denoted PEW) based on the proportionality hypothesis to incorporate available water control into the widely used Priestley Taylor-Jet Propulsion Laboratory (PT-JPL) ET algorithm. Simulated performances of the PEW model and PT-JPL algorithm were evaluated against 106 FLUXNET eddy covariance (EC) towers data at the site scale. Meanwhile, model results were compared at the global scale with the means of the widely used ET products. We found that the PEW model has smaller errors than the original PT-JPL algorithm, with the greatest improvements in water-limited regions and areas characterized by the high interception. Moreover, by incorporating the water balance constraints into the ET algorithm, the PEW model has the ability to distinguish variations in ET affected by El Nino-Southern Oscillation. In summary, our study offers a convincing evidence regarding the incorporation of water balance constraints into remote sensing algorithms for more accurately mapping global terrestrial ET with an enhanced understanding of ET variation under climate change. This model is the first of its kind among remote-sensing models to provide global land ET estimation with the proportionality hypothesis-based water balance constraints.

Combining Different Transformations of Ground Hyperspectral Data with Unmanned Aerial Vehicle (UAV) Images for Anthocyanin Estimation in Tree Peony Leaves

To explore rapid anthocyanin (Anth) detection technology based on remote sensing (RS) in tree peony leaves, we considered 30 species of tree peonies located in Shaanxi Province, China. We used an SVC HR~1024i portable ground object spectrometer and mini-unmanned aerial vehicle (UAV)-borne RS systems to obtain hyperspectral (HS) reflectance and images of canopy leaves. First, we performed principal component analysis (PCA), first-order differential (FD), and continuum removal (CR) transformations on the original ground-based spectra; commonly used spectral parameters were implemented to estimate Anth content using multiple stepwise regression (MSR), partial least squares (PLS), back-propagation neural network (BPNN), and random forest (RF) models. The spectral transformation highlighted the characteristics of spectral curves and improved the relationship between spectral reflectance and Anth, and the RF model based on the FD spectrum portrayed the best estimation accuracy (R2c = 0.91; R2v = 0.51). Then, the RGB (red-green-blue) gray vegetation index (VI) and the texture parameters were constructed using UAV images, and an Anth estimation model was constructed using UAV parameters. Finally, the UAV image was fused with the ground spectral data, and a multisource RS model of Anth estimation was constructed, based on PCA + UAV, FD + UAV, and CR + UAV, using MSR, PLS, BPNN, and RF methods. The RF model based on FD+UAV portrayed the best modeling and verification effect (R2c = 0.93; R2v = 0.76); compared with the FD-RF model, R2c increased only slightly, but R2v increased greatly from 0.51 to 0.76, indicating improved modeling and testing accuracy. The optimal spectral transformation for the Anth estimation of tree peony leaves was obtained, and a high-precision Anth multisource RS model was constructed. Our results can be used for the selection of ground-based HS transformation in future plant Anth estimation, and as a theoretical basis for plant growth monitoring based on ground and UAV multisource RS.

Regional Evapotranspiration Estimation by the Improved MOD16-sm Model and Its Application in Central China

Evapotranspiration (ET) is a key component of the hydrological cycle, but traditional monitoring approaches are always based on measurements, which cannot satisfy the requirements of research on a regional scale. Hence, ET estimation by remote sensing is essential. MOD16 is a remote-sensing model based on the P-M equation and has good applicability. However, it describes soil moisture indirectly by RH, etc., which may cause uncertainties in ET estimating, so this study attempts to utilize the NDWI as a supplement to soil moisture information and makes improvements on the MOD16 model (with the resultant new model being named MOD16-sm). Specific work includes two aspects: one is model verification through making comparisons between ET estimates and measurements, and the other is a model application effect test analyzing the spatiotemporal characteristics of ET and exploring how ET responds to climate and land-use changes. Model verification indicated that the accuracy of the improved MOD16-sm model increased, with a higher R2 of 0.71, a lower RMSE 0.9 mm, and a lower MAE 0.91 mm, and that the improved MOD16-sm model was convincing. The application effect test of the MOD16-sm model showed that the average relative change rate of annual ET was 1.7%, showing an upward trend, and areas with growth trends of ET also had high vegetation coverage. As for the impacts of climate and land-use changes on ET, ET was positively correlated with precipitation, whereas it had no relevant correlation with air temperature in most areas, and the ET of all land-use types displayed significant increasing trends resulting from climate change. The application effect test demonstrated that ET estimates by the improved MOD16-sm model were reasonable.

EMTCAL: Efficient Multiscale Transformer and Cross-Level Attention Learning for Remote Sensing Scene Classification

In recent years, convolutional neural network (CNN)-based methods have been widely used for remote sensing (RS) scene classification tasks and achieved excellent results. However, CNNs are not good at exploring contextual information, which is essential for fully understanding RS scenes. A new model named transformer attracts researchers’ attention to address this problem, which is skilled in mining the latent contextual information in RS scenes. Nevertheless, since the contents of RS scenes are diverse in type and various in scale, the performance of the original transformer in RS scene classification cannot reach what we expect. In addition, due to the specific self-attention mechanism, the time costs of the transformer are high, which hinders its practicability in the RS community. To overcome the above limitations, we propose a new model named efficient multi-scale transformer and cross-level attention learning (EMTCAL) for RS scene classification in this paper. EMTCAL combines the advantages of CNN and transformer to mine information within RS scenes fully. First, it uses a multi-layer feature extraction module (MFEM) to acquire global visual features and multi-level convolutional features from RS scenes. Second, a contextual information extraction module (CIEM) is proposed to capture rich contextual information from multi-level features. In CIEM, taking the characteristics of RS scenes and the computational complexity into account, we propose an efficient multi-scale transformer (EMST). EMST can mine the abundant knowledge with various scales hidden in RS scenes and model their inherent relations at small-time costs. Third, a cross-level attention module (CLAM) is developed to aggregate and explore correlations of multi-level features. Finally, a class score fusion module (CSFM) is designed to integrate the contributions of global and aggregated multi-level features for the discriminative scene representations. Extensive experiments are conducted on three public RS scene data sets. The positive results demonstrate that our EMTCAL can achieve superior classification performance and outperform many state-of-the-art methods.

A Lightweight Multi-Scale Crossmodal Text-Image Retrieval Method in Remote Sensing

Remote sensing (RS) crossmodal text-image retrieval has become a research hotspot in recent years for its application in semantic localization. However, since multiple inferences on slices are demanded in semantic localization, designing a crossmodal retrieval model with less computation but well performance becomes an emergent and challenging task. In this article, considering the characteristics of multi-scale and target redundancy in RS, a concise but effective crossmodal retrieval model (LW-MCR) is designed. The proposed model incorporates multi-scale information and dynamically filters out redundant features when encoding RS image, while text features are obtained via lightweight group convolution. To improve the retrieval performance of LW-MCR, we come up with a novel hidden supervised optimization method based on knowledge distillation. This method enables the proposed model to acquire dark knowledge of the multi-level layers and representation layers in the teacher network, which significantly improves the accuracy of our lightweight model. Finally, on the basis of contrast learning, we present a method employing unlabeled data to boost the performance of RS retrieval model further. The experiment results on four RS image-text datasets demonstrate the efficiency of LW-MCR in RS crossmodal retrieval (RSCR) tasks. We have released some codes of the semantic localization and made it open to access at <uri>https://github.com/xiaoyuan1996/retrievalSystem</uri>.

Combining Remote Sensing and Meteorological Data for Improved Rice Plant Potassium Content Estimation

Potassium (K) plays a significant role in the formation of crop quality and yield. Accurate estimation of plant potassium content using remote sensing (RS) techniques is therefore of great interest to better manage crop K nutrition. To improve RS of crop K, meteorological information might prove useful, as it is well established that weather conditions affect crop K uptake. We aimed to determine whether including meteorological data into RS-based models can improve K estimation accuracy in rice (Oryza sativa L.). We conducted field experiments throughout three growing seasons (2017–2019). During each year, different treatments (i.e., nitrogen, potassium levels and plant varieties) were applied and spectra were taken at different growth stages throughout the growing season. Firstly, we conducted a correlation analysis between rice plant potassium content and transformed spectra (reflectance spectra (R), first derivative spectra (FD) and reciprocal logarithm-transformed spectra (log [1/R])) to select correlation bands. Then, we performed the genetic algorithms partial least-squares and linear mixed effects model to select important bands (IBs) and important meteorological factors (IFs) from correlation bands and meteorological data (daily average temperature, humidity, etc.), respectively. Finally, we used the spectral index and machine learning methods (partial least-squares regression (PLSR) and random forest (RF)) to construct rice plant potassium content estimation models based on transformed spectra, transformed spectra + IFs and IBs, and IBs + IFs, respectively. Results showed that normalized difference spectral index (NDSI (R1210, R1105)) had a moderate estimation accuracy for rice plant potassium content (R2 = 0.51; RMSE = 0.49%) and PLSR (FD-IBs) (R2 = 0.69; RMSE = 0.37%) and RF (FD-IBs) (R2 = 0.71; RMSE = 0.40%) models based on FD could improve the prediction accuracy. Among the meteorological factors, daily average temperature contributed the most to estimating rice plant potassium content, followed by daily average humidity. The estimation accuracy of the optimal rice plant potassium content models was improved by adding meteorological factors into the three RS models, with model R2 increasing to 0.65, 0.74, and 0.76, and RMSEs decreasing to 0.42%, 0.35%, and 0.37%, respectively, suggesting that including meteorological data can improve our ability to remotely sense plant potassium content in rice.

Changes in Nutrient Concentrations in Shenzhen Bay Detected Using Landsat Imagery between 1988 and 2020

Shenzhen Bay (SZB), situated between Shenzhen and Hong Kong, is a typical bay system. The water quality of the bay is notably affected by domestic and industrial discharge. Rivers and various types of drainage outlets carry terrestrial pollutants into SZB, resulting in elevated concentrations of nitrogen and phosphorous as well as relatively poor water quality. For over 200 years, Hong Kong has practiced oyster farming within brackish estuarine waters. Oyster farming is a type of mariculture which includes oyster breeding in oyster rafts. Remote sensing is a monitoring technique characterized by large spatial coverage, high traceability, and low cost, making it advantageous over conventional point-based and ship-borne monitoring methods. In this study, remote-sensing models were established using machine-learning algorithms to retrieve key water-quality factors (dissolved inorganic nitrogen (DIN) and orthophosphate-phosphorous (PO4_P) concentrations, CDIN and CPO4_P, respectively) from long-term time-series data acquired by the Landsat satellites. (1) Spatially, the water quality in Inner SZB was worse than that in Outer SZB. (2) The water quality temporarily deteriorated between the end of the 20th century and the beginning of the 21st century; then it gradually improved in the late 2000s. (3) Monitoring the water quality in an oyster-farming area revealed that oyster farming did not adversely affect the water quality. (4) The result of monitoring the water quality in river estuaries in SZB shows that water quality was mainly affected by river input.

Detection of Xylella fastidiosa in almond orchards by synergic use of an epidemic spread model and remotely sensed plant traits

The early detection of Xylella fastidiosa (Xf) infections is critical to the management of this dangerous plan pathogen across the world. Recent studies with remote sensing (RS) sensors at different scales have shown that Xf-infected olive trees have distinct spectral features in the visible and infrared regions (VNIR). However, further work is needed to integrate remote sensing in the management of plant disease epidemics. Here, we research how the spectral changes picked up by different sets of RS plant traits (i.e., pigments, structural or leaf protein content), can help capture the spatial dynamics of Xf spread. We coupled a spatial spread model with the probability of Xf-infection predicted by a RS-driven support vector machine (RS-SVM) model. Furthermore, we analyzed which RS plant traits contribute most to the output of the prediction models. For that, in almond orchards affected by Xf (n = 1426 trees), we conducted a field campaign simultaneously with an airborne campaign to collect high-resolution thermal images and hyperspectral images in the visible-near-infrared (VNIR, 400–850 nm) and short-wave infrared regions (SWIR, 950–1700 nm). The best performing RS-SVM model (OA = 75%; kappa = 0.50) included as predictors leaf protein content, nitrogen indices (NIs), fluorescence and a thermal indicator (Tc), alongside pigments and structural parameters. Leaf protein content together with NIs contributed 28% to the explanatory power of the model, followed by chlorophyll (22%), structural parameters (LAI and LIDFa), and chlorophyll indicators of photosynthetic efficiency. Coupling the RS model with an epidemic spread model increased the accuracy (OA = 80%; kappa = 0.48). In the almond trees where the presence of Xf was assayed by qPCR (n = 318 trees), the combined RS-spread model yielded an OA of 71% and kappa = 0.33, which is higher than the RS-only model and visual inspections (both OA = 64–65% and kappa = 0.26–31). Our work demonstrates how combining spatial epidemiological models and remote sensing can lead to highly accurate predictions of plant disease spatial distribution.