Search

This study uses National Centers for Environmental Prediction (NCEP) Stage IV (Stage IV) precipitation data over the state of Alaska to assess and cross compare precipitation estimates from the most recent versions of multiple precipitation products, including satellite-based passive microwave (PMW) [Special Sensor Microwave Imager/Sounder (SSMIS)–F17, Microwave Humidity Sounder (MHS)–MetOp-B, MHS–NOAA-19, Advanced Microwave Scanning Radiometer 2 (AMSR2), Advanced Technology Microwave Sounder (ATMS), and Global Precipitation Measurement Microwave Imager (GMI) in V05 and V07], active microwave [AMW or radar; Global Precipitation Measurement (GPM) dual-frequency precipitation radar (DPR) in V06 and V07], combined active and passive microwave (DPRGMI in V06 and V07), infrared [Atmospheric Infrared Sounder (AIRS)], reanalysis [fifth major global reanalysis produced by ECMWF (ERA5) and Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2), and satellite–gauge [Global Precipitation Climatology Project (GPCP) V1.3 and GPCP V3.2] products. PMW estimates are generally improved in V07 compared to V05 in terms of overall bias, pattern, and capturing precipitation extremes. DPR and DPRGMI show low skill in capturing different precipitation features. ERA5 and MERRA-2 show the highest agreement with Stage IV for all precipitation rate metrics. AIRS and GPCP capture the overall precipitation pattern and magnitude fairly well, performing better than the radar and comparable to the PMW V07 products, although the geographical maps suggest that they provide a relatively smoothed spatial distribution of mean precipitation rates. The outcomes of this study shed light on the performance of various precipitation products over Alaska (partly representing high-latitude regions) and can be useful to guide the development of multisensor products. [ABSTRACT FROM AUTHOR]

Performing remote sensing scene classification (RSSC) directly on satellites can alleviate data downlink burdens and reduce latency. Compared to convolutional neural networks (CNNs), the all-adder neural network (A2NN) is a novel basic neural network that is more suitable for onboard RSSC, enabling lower computational overhead by eliminating multiplication operations in convolutional layers. However, the extensive floating-point data and operations in A2NNs still lead to significant storage overhead and power consumption during hardware deployment. In this article, a shared scaling factor-based de-biasing quantization (SSDQ) method tailored for the quantization of A2NNs is proposed to address this issue, including a powers-of-two (POT)-based shared scaling factor quantization scheme and a multi-dimensional de-biasing (MDD) quantization strategy. Specifically, the POT-based shared scaling factor quantization scheme converts the adder filters in A2NNs to quantized adder filters with hardware-friendly integer input activations, weights, and operations. Thus, quantized A2NNs (Q-A2NNs) composed of quantized adder filters have lower computational and memory overheads than A2NNs, increasing their utility in hardware deployment. Although low-bit-width Q-A2NNs exhibit significantly reduced RSSC accuracy compared to A2NNs, this issue can be alleviated by employing the proposed MDD quantization strategy, which combines a weight-debiasing (WD) strategy, which reduces performance degradation due to deviations in the quantized weights, with a feature-debiasing (FD) strategy, which enhances the classification performance of Q-A2NNs through minimizing deviations among the output features of each layer. Extensive experiments and analyses demonstrate that the proposed SSDQ method can efficiently quantize A2NNs to obtain Q-A2NNs with low computational and memory overheads while maintaining comparable performance to A2NNs, thus having high potential for onboard RSSC. [ABSTRACT FROM AUTHOR]

Unsupervised domain adaptation (UDA) methods have become a research hotspot in remote sensing scene classification to reduce dependence on labeled samples. However, most current methods focus on extracting domain invariant features, ignoring the problem of large intraclass differences and the imbalanced sample numbers between categories in remote sensing images. To address these issues, we propose a remote sensing scene domain adaptive method based on middle-layer feature extraction and nuclear norm maximization (MFE-NM). In the MFE module, the middle-layer features of the feature extractor are randomly extracted and processed. Since the receptive field of the middle-layer features is smaller and the resolution is higher, the effective use of the middle-layer features can reduce the impact of image feature heterogeneity caused by large intraclass differences in remote sensing images. In addition, it can be concluded that the constrained nuclear norm can simultaneously improve the prediction diversity and discriminability of the model through theoretical derivation. Therefore, the NM module is proposed to solve the problem of reduced prediction diversity caused by entropy minimization methods when dealing with scene classification problems with imbalanced sample numbers between categories. Extensive experiments and analyses on three public remote sensing datasets demonstrate the effectiveness and competitiveness of our proposed method. [ABSTRACT FROM AUTHOR]

Modern smart agriculture initiative presents more requests for soil moisture (SM) monitoring over large agricultural areas. Remote sensing techniques facilitate high-resolution surface SM (SSM) estimation at a large scale but lack root zone SM (RZSM) information. Establishing the deduction method of RZSM from the SSM has long been the focus of most attention. Data assimilation methods are promising techniques for RZSM estimation, developing numerous assimilated reanalysis datasets, e.g., ERA5 and the latest Soil Moisture Active and Passive (SMAP) L4 SM product. However, data latency and large computation during data collecting and processing often inhibits further applications. This work proposes a rapid estimation scheme for estimating RZSM with short latency and small computations, based on the Exponential Filter (EF) method. The EF model with single parameter T was firstly calibrated and validated using the SSM and RZSM of ERA5 reanalysis dataset, obtaining the optimum parameter T map for each grid. Then, the fast-updating SMAP L3 SSM product together with the scale-matched optimum T were adopted as inputs into the EF model to retrieve RZSM estimation of each grid. Specifically, such estimation scheme was tested over the central and eastern agricultural areas of China, using a dense monitoring network of 796 SM observation sites, which contains various land uses, as well as meteorological and hydrological conditions. The calibrated optimum parameter T presented an increasing trend with good physical explanations. Furthermore, all the estimated RZSMs were found to have good performances on capturing the temporal-spatial variations of RZSM and well reflecting seasonal RZSM changes. Overall, such an estimation scheme was proven to be a desirable alternative for estimating RZSM over large agricultural areas. [ABSTRACT FROM AUTHOR]

Droughts have caused many damages in many countries and might be aggravated around the world. Therefore, it is urgent to predict and monitor drought accurately. Soil moisture and its corresponding drought index (e.g., soil water deficit index, SWDI) are the key variables to define drought. However, in situ soil moisture observations are inaccessible in many areas. This study applies support vector machine (SVM) by using a new set of inputs to investigate the performance of in situ and remote sensing products (CMORPH-CRT, IMERG V05 and TRMM 3B42V7) for soil moisture and SWDI forecast over the Xiang River Basin. This study also assesses whether the addition of remote sensing soil moisture as input can improve the performance of SWDI prediction. The results are as follows: (1) the new set of inputs is suitable for drought prediction based on SVM; (2) using in situ precipitation as input to SVM shows the best performance for soil moisture prediction, which followed by TRMM 3B42V7, IMERG V05 and CMORPH-CRT; (3) in situ precipitation and IMERG V05 as input are more suitable for indirect SWDI prediction, while CMORPH-CRT and TRMM 3B42V7 are more suitable for direct SWDI prediction; (4) the addition of soil moisture with in situ precipitation or CMORPH-CRT both can improve the performance of direct SWDI prediction; (5) the lead time for drought prediction with SVM over the Xiang River Basin is about 2 weeks. [ABSTRACT FROM AUTHOR]

The remote sensing of chlorophyll a concentration from ocean color satellites has been an essential variable quantifying phytoplankton in the past decades, yet estimation of accessory pigments from ocean color remote sensing data has remained largely elusive. In this study, we validated the concentrations of multiple pigments (Cpigs) retrieved from in situ and MEdium Resolution Imaging Spectrometer (MERIS) measured remote sensing reflectance (Rrs(λ)) in the global oceans. A multi-pigment inversion model (MuPI) was used to semi-analytically retrieve Cpigs from Rrs(λ). With a set of globally optimized parameters, the accuracy of the retrievals obtained with MuPI is quite promising. Compared with High-Performance Liquid Chromatography (HPLC) measurements near Bermuda, the concentrations of chlorophyll a, b, c ([Chl-a], [Chl-b], [Chl-c]), photoprotective carotenoids ([PPC]), and photosynthetic carotenoids ([PSC]) can be retrieved from MERIS data with a mean unbiased absolute percentage difference of 38%, 78%, 65%, 36%, and 47%, respectively. The advantage of the MuPI approach is the simultaneous retrievals of [Chl-a] and the accessory pigments [Chl-b], [Chl-c], [PPC], [PSC] from MERIS Rrs(λ) based on a closure between the input and output Rrs(λ) spectra. These results can greatly expand scientific studies of ocean biology and biogeochemistry of the global oceans that are not possible when the only available information is [Chl-a]. [ABSTRACT FROM AUTHOR]

This study focuses on tracing the error sources of the latest global satellite mapping of precipitation for global precipitation measurement (GPM-GSMaP) over the Tibetan Plateau by the corresponding satellite information flag provided by the retrieval system. We investigated the characteristics of GPM-GSMaP estimates from 11 passive microwave sensors (including 7 imagers and 4 sounders) and 1 geo-infrared data source integrating the morphing technique (i.e., morph). Assessment results show that imagers are generally superior to sounders and morph. Among the seven types of imagers, the Tropical Rainfall Measuring Mission Microwave Imager, which rarely underestimates heavy rain events, exhibits the best performance with total bias of −8.8%. In contrast, the worst performance is found in the special sensor microwave imager/sounder on defense meteorological satellite program (DMSP)-F19 with the largest bias of approximately 39.8%. The GPM Microwave Imager shows acceptable accuracy in detecting heavy rain, but it tends to overestimate light and moderate rain. Compared to the imagers, all the sounders produced larger biases (>120%), although they show lower probability of missing rainfall events. Particularly, the AMSU-A/MHS (i.e., Advanced Microwave Sounding Unit-A; Microwave Humidity Sounder) on national oceanic and atmospheric administration (NOAA) satellites (i.e., NOAA-18 and −19) display higher biases than those on meteorological operational satellite (MetOp) satellites (i.e., MetOp-A and −B), with dramatic overestimation up to 168.4%. Additionally, the error components of morph show a similar pattern to those of sounders, except for substantial underestimation of heavy rain events. The approach of tracing error reported here can enable both GPM-GSMaP users and developers to better understand the error sources of precipitation retrievals. [ABSTRACT FROM AUTHOR]

Multi-source soil moisture (SM) products provide a vigorous tool for the estimation of soil moisture on a large scale, but it is crucial to carry out the evaluation of those products before further application. In the present work, an evaluation framework on multi-source SM datasets over central and eastern agricultural areas of China was firstly proposed, based on a dense in situ SM monitoring network of 838 stations from 11 July 2012 to 31 December 2017. Each station adopted the most accurate gravimetric method for measuring the actual soil moisture. The effects of land use types and wet–dry conditions on the performances of multi-source SM products were further analyzed. Most satellite/reanalysis SM products could capture the spatial–temporal changes in soil moisture, especially for ERA5 products that matched the closest to the station-measured SM; by contrast, those satellite products showed poor spatial–temporal performances. Such phenomenon was also quantitatively demonstrated by the four statistical metrics correlation coefficient (CC), p-value, bias and root mean squared error (RMSE) between the satellite/reanalysis SM products and the ground-observed SM series. Further, most satellite/reanalysis SM products had poor performances in Forestland and Grassland areas, with a lower CC and a larger positive bias and RMSE. Such overestimation on soil moisture is possibly influenced by the inestimable parameter vegetation geometry and the vegetation water content in the radiative transfer models. The arid areas showed the worst CC between the station-observed SM data and different satellite/reanalysis SM products; meanwhile, the humid and semi-arid areas presented larger SM estimation errors than the other areas, especially for the satellite products. The fairly dry surface soil (arid area) and open water surface contamination (humid area) are suggested to hinder the reading of microwave-based retrieval systems. Additionally, the reanalysis SM products outperformed the satellite SM products in the evaluated areas, with better spatial–temporal performances, seasonality reflection and higher accuracy on SM estimation (higher CC, and lower bias and RMSE). This is because the reanalysis datasets assimilated various sources of datasets, especially the ground-observed data, with high quality. The evaluated results could provide guidance for fusing different satellite/reanalysis products, as a new feasible alternative to monitoring SM information in the future. [ABSTRACT FROM AUTHOR]

Object detection in optical remote sensing images (RSIs) is challenging due to cluttered backgrounds, complicated weather conditions as well as variation in the types, scales, and orientations of the objects in question. Numerous methods have been proposed to handle different scenarios in object detection in remote sensing images, but under severe imaging conditions such as hazy scenes and turbulent sensors these methods almost universally fail to perform satisfactorily. In this paper, we propose a new detection framework, called Knowledge-guided Joint Restoration and Detection Network (KJRD-Net), to facilitate the robust detection of tiny objects in hazy optical RSIs. This network is the first method specifically developed to address tiny object detection in hazy RSIs. The network consists of 4 parts: the pre-trained auto-encoder, the parallel image dehazing network and super-resolution network, the image-level coordination block and the object detector. The image is first fed through the auto-encoder to extract comprehensive semantic knowledge. Then, haze removal and super-resolution is performed on the image by the parallel dehazing and super-resolution network. After that, the semantic information and the image restoration cues extracted by the first two components is fed through the image-level coordination block to acquire the final restored image. Finally, the processed images, incorporated with knowledge extracted by prior network structures, are fed into an object detection network. The network is trained end-to-end. In addition, we construct the DOTA-H dataset, a novel dataset focusing on tiny object detection in hazy RSIs. Extensive experiments show that our framework is effective at detecting tiny objects from RSIs of low quality, improving the mAP of the Faster R-CNN baseline on our DOTA-H dataset from 41.2 to 48.6. • We present a network for tiny object detection on hazy remote sensing images. • The network achieves high performance by integrating low-level and high-level tasks. • A dataset is proposed to facilitate further research in the subject. [ABSTRACT FROM AUTHOR]

The demand for accurate long-term precipitation data is increasing, especially in the Lancang-Mekong River Basin (LMRB), where ground-based data are mostly unavailable and inaccessible in a timely manner. Remote sensing and reanalysis quantitative precipitation products provide unprecedented observations to support water-related research, but these products are inevitably subject to errors. In this study, we propose a novel error correction framework that combines products from various institutions. The NASA Modern-Era Retrospective Analysis for Research and Applications (AgMERRA), the Asian Precipitation Highly-Resolved Observational Data Integration Towards Evaluation of Water Resources (APHRODITE), the Climate Hazards group InfraRed Precipitation with Stations (CHIRPS), the Multi-Source Weighted-Ensemble Precipitation Version 1.0 (MSWEP), and the Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks-Climate Data Records (PERSIANN) were used. Ground-based precipitation data from 1998 to 2007 were used to select precipitation products for correction, and the remaining 1979–1997 and 2008–2014 observe data were used for validation. The resulting precipitation products MSWEP-QM derived from quantile mapping (QM) and MSWEP-LS derived from linear scaling (LS) are evaluated by statistical indicators and hydrological simulation across the LMRB. Results show that the MSWEP-QM and MSWEP-LS can better capture major annual precipitation centers, have excellent simulation results, and reduce the mean BIAS and mean absolute BIAS at most gauges across the LMRB. The two corrected products presented in this study constitute improved climatological precipitation data sources, both time and space, outperforming the five raw gridded precipitation products. Among the two corrected products, in terms of mean BIAS, MSWEP-LS was slightly better than MSWEP-QM at grid-scale, point scale, and regional scale, and it also had better simulation results at all stations except Strung Treng. During the validation period, the average absolute value BIAS of MSWEP-LS and MSWEP-QM decreased by 3.51% and 3.4%, respectively. Therefore, we recommend that MSWEP-LS be used for water-related scientific research in the LMRB. [ABSTRACT FROM AUTHOR]

• An enhancement strategy is proposed to calibrate snow parameters and capture snowmelt runoff. • The improvement in runoff simulation accuracy primarily stems from the precise capture of snowmelt runoff. • The joint reconstruction of parameter schemes and physical processes enhances the modeling capability. Hydrological models stand as a pivotal instrument for runoff simulation, while encountering notable uncertainties in output due to intricate terrain conditions and limited ground-based observations, especially in high-elevation basins. Leveraging satellite-based images presents a promising avenue for deciphering the hydrological model's state variables. In pursuit of refining runoff simulation, this study developed a two-step enhancement strategy, which involved (1) calibrating snow-related parameters in the hydrological model using remotely sensed snow cover area (SCA) and snow water equivalent (SWE) and (2) capturing snowmelt runoff through the hydrological model and image-based products. Coupled with the Variable Infiltration Capacity (VIC) model, we adopted this strategy as a case study in the Dadu River Basin, China. The results indicated (1) the daily Nash-Sutcliffe Efficiency (NSE) of runoff simulation reached 0.84 by the enhancement strategy, markedly surpassing the strategy reliant on soil parameters with a single calibrated reference (daily NSE of 0.66), and exhibited comparability to the strategy incorporating snow parameters but calibrated solely based on observed discharge (daily NSE of 0.83). (2) the enhancement strategy demonstrated hydrological consistency with snowmelt information derived from imagery. Specifically, multi-year average contributions of model's and image-based snowmelt calculations were 31.8% and 33.4%, respectively. Additionally, simulated dates of snow accumulation and ablation appeared an average deviation of approximately one week compared to the imagery results. This study elucidates a potential methodological approach for offering valuable insights into hydrological processes within analogous high-elevation basins globally. [ABSTRACT FROM AUTHOR]

The optical neural network system based on the 4f system (4f-ONN) is a feasible solution for on-orbit real-time target detection and recognition on remote sensing images, as it can directly modulate and encode the two-dimensional image information. However, traditional 4f systems based on spatial light modulators (SLMs) often encounter misalignment errors during assembly due to the reflective optical path used in SLMs. Additionally, implementing electronic SLM in space-based applications introduces problems such as particle number reversal, significantly reducing the reliability of the system. To address these issues, this paper proposes the adoption of diffractive optical elements (DOEs) to construct the 4f-ONN system. The DOE-based design offers a more compact structure, enhanced reliability, and reduced energy consumption, making it highly suitable for on-orbit image processing applications. Due to the expensive and time-consuming nature of the DOE manufacturing process, a design approach for a 4f system based on DOE was pursued through software simulation in this paper. The simulation phase involved the utilization of electronic neural networks to acquire the physical parameters of the DOE mask, while incorporating the array theorem and Fraunhofer diffraction theorem to accurately calculate the physical dimensions of the DOE. The effectiveness of the adopted DOE-based 4f system was initially validated through experiments on a simple pattern dataset. Subsequently, simulation experiments were conducted on three public datasets, namely Mnist, Fashion-mnist, and QuickDraw16, to confirm the efficacy of the DOE-based 4f system in classification tasks. Lastly, target recognition experiments were performed on the GF-2 dataset, and a corresponding hardware system was developed to demonstrate the potential of the DOE-based 4f system in on-orbit image processing. • A structurally more compact and stable DOE-based 4f system is proposed. • The effects of quantization and manufacturing errors on the performance of the DOE-based 4f system are analyzed. • Extensive experiments demonstrate the efficiency of the proposed DOE-based 4f system in ship recognition within remote sensing images. [ABSTRACT FROM AUTHOR]

• The drought indices SDCI, VHI and IDCI were constructed by four weighting methods. • Determination of weights was validated by comparing against scPDSI, SMCI and CY. • New classification schemes were executed by cumulative frequency curve of scPDSI. • Reliability of schemes was verified by drought affected area and frequencies. With the continuous development of remote sensing technology, an increasing number of drought indices are proposed to assess the drought process in different regions. However, most of the studies used the empirical coefficient method and standard drought classification scheme, without considering regional applicability, which may overestimate or underestimate drought intensity. Therefore, the GLDAS soil moisture dataset was firstly verified in the study area, Wei River basin, China, and then the response of precipitation accumulation to NDVI and the lagged effect were analyzed. Secondly, the Scaled Drought Condition Index (SDCI), Vegetation Health Index (VHI), and Integrated Drought Condition Index (IDCI) were constructed and the optimal weights of the drought indicators were determined against the scPDSI, SMCI and CY based on four weighting methods. Finally, the new classification schemes of the three indicators were executed based on the cumulative frequency curve of the self-calibrating Palmer Drought Severity Index (scPDSI) from June to September during the period 2000–2019, and the reliability of the new schemes was verified by comparing with drought frequency and drought affected area. Furthermore, the variabilities of drought frequency of the new classification scheme based on the three indicators were analyzed. The results showed that the correlation coefficients of 2-month accumulation of precipitation and NDVI produced the best value and no lagged effect was detected. The SDCI and VHI determined by the AHP weighting method and the IDCI determined by the entropy weighting method showed better performance. The new classified schemes of the three drought indicators demonstrated better performance than the original classified scheme with more similar frequency distributions to scPDSI and higher correlation coefficients against the drought affected area. Additionally, the three indicators based on the revised classification scheme all indicated a decreasing trend in the frequency of drought occurrence for all the different drought levels; the SDCI and VHI were better at capturing drought events than IDCI. Overall, the results of this study proved that the optimal weighting methods and new classification schemes could effectively improve the precision of drought assessment and may provide helpful information for agricultural drought disaster prevention. [ABSTRACT FROM AUTHOR]

Abstract In this study, we propose to assess climate impact on forest cover (represented by EVI) at multiple scales in areas undergoing substantial land cover change, using Landsat imagery with human-induced land cover change effect excluded. Taking the Qingliu River catchment located in a subtropical humid monsoon area in China as a case study, the results indicate that EVI increases significantly (p < 0.01) during 1989–2014 with a magnitude of 0.026/decade. Spatial distribution of EVI is distinct in summer and growing season. Temperature and precipitation show high partial correlations with EVI, with better partial correlation found between EVI and temperature. Their partial correlations with EVI on monthly scale are higher than those on annual scale. Besides, precipitation and pan evaporation show accumulative lag effects (4 months) on forest EVI, while temperature has no lag effect. Finally, an empirical formula is established to quantify the relationship among EVI and its main driving factors (temperature and precipitation) by considering the precipitation threshold (200 mm). The findings should provide scientific supports for local forest management and ecosystem services, and should also support the hydrological effect assessment of vegetation cover change under climate change for the study area. Graphical abstract Unlabelled Image Highlights • The framework allows for assessing climate impact on forest cover more precisely. • Human-induced land cover change was excluded from assessing climate impact on forest cover. • Climate-induced spatiotemporal variations of forest EVI on different scales were analyzed. • Minimum temperature and precipitation are two key control factors for forest EVI. • Quantitative relationship among forest EVI and climatic factors is established. [ABSTRACT FROM AUTHOR]

• A deep learning model (Multi-Level Transformer Fusion model) for QPE improvement. • The model takes inner interactions of multi-source data into account. • The proposed model performed well in the SRYRB (a sparsely-gauged basin) • The proposed model outperformed the selected comparative models in the SRYRB. • A daily precipitation dataset (1999–2019, 0.05°) with QPE improved is produced for SRYRB. Improving quantitative precipitation estimation (QPE) in sparsely-gauged basins via merging remote sensing precipitation data and rain gauge data still remains a challenge since most existing merging models degrade when the rain gauge data become limited. To address the challenge, we propose an attention-mechanism-based deep learning model, Multi-Level Transformer Fusion (MLTF) model, which allows to capture the inner interactions among the multi-source input data (TRMM 3B42 V7 data, GridSat-B1 data, and DEM) in the sparsely-gauged basin. Taking the source region of the Yellow River basin (SRYRB) as a representative case study, we demonstrate the performance of the proposed model and compare it with conventional methods (e.g., Multiplicative Bias Correction, Additive Bias Removal, linear regression, and Kriging) and deep-learning-based models (CNN, CNN-LSTM). Results indicate that the merged precipitation in SRYRB produced by the MLTF model exhibits a RMSE reduction of 27.1 %, MAE decrease of 11.2 %, and CC increase of 19.2 % in comparison to the original TRMM data, outperforming all the selected comparative methods. Finally, an improved daily precipitation dataset during 1999–2019 with a spatial resolution of 0.05° is produced for the study area. This study proposes a new method for QPE improvement in a sparsely-gauged basin, which would provide valuable data support for regional hydrological study and water resources management. [ABSTRACT FROM AUTHOR]

[Display omitted] • Remote sensing dataset construction scheme to support real-time forecast is provided. • Combined the corrected WT to develop hybrid models valid for real-world application. • Hybrid model can overcome the influence of complex environments on standalone model. • The WT-LSTM model exhibits the best forecasting performance, followed by WT-XGB. • The shapely additive explanations method enhances the forecasting interpretability. Groundwater level forecasting is significantly crucial for the sustainable management of water resources, especially for arid and semi-arid regions where groundwater resources are highly dependent on. However, the complex groundwater dynamic systems in these regions are strongly influenced by climate change and human activities, which poses severe challenges to the development of accurate groundwater level forecasting models. This study first explored the selection and processing scheme of remote sensing products. Based on this, a new strategy to support real-time groundwater level forecasting with the hybrid data-driven model as the core algorithm is proposed. We applied a Deep Learning algorithm and two Ensemble Machine Learning algorithms, combined with the corrected Wavelet Transformation (WT) to develop hybrid models (WT-LSTM, WT-RF and WT-XGB) valid for real-world applications. The SHapley Additive exPlanations (SHAP) method was used to enhance the interpretability of the forecasting strategy. Real-world applications in the Xi'an and Yinchuan regions of Northwest China have shown that WT-LSTM is the hybrid model with the best overall performance with 0.843, 0.749 and 0.712 mean NSE at 1-, 2- and 3-month forecasting lead time, followed by WT-XGB with 0.763, 0.642 and 0.590 mean NSE, respectively. The accuracy of the WT-based hybrid models is significantly improved compared to the standalone model. Further analysis demonstrates that the performance of the standalone models is influenced by the local climate, especially human activities, while the corrected WT method can overcome such drawbacks. The LSTM-based models have a stronger capability than RF-based model to capture the hydrological signal affecting the local groundwater level from dataset based on remote sensing products. The SHAP method also validates the above findings and the reliability of the forecasting models. We conclude that the groundwater level forecasting strategy proposed in this study improves accuracy, interpretability and generalizability, and provides new insights and a reliable scientific basic for real-time groundwater level forecasting. [ABSTRACT FROM AUTHOR]

Drought is considered to be one of the most serious natural disasters in China, which can result in enormous damage to nature and socio-economy. Compared to traditional ground-based monitoring techniques, remote sensing can effectively compensate for spatial discontinuities at ground stations. The use of remote sensing technology for drought monitoring has irreplaceable advantages. The applicability of the TRMM3B43 dataset for precipitation was firstly verified in the Wei River basin, and the spatiotemporal characteristics of precipitation were analyzed. Based on the TRMM3B43, MODIS NDVI, and MODIS LST datasets, the spatiotemporal variations of drought were secondly investigated by calculating the Precipitation Condition Index (PCI), Vegetation Condition Index (VCI), and Temporal Condition Index (TCI). Crop yield was employed as the reference of drought impact for evaluating the applicability of the Scaled Drought Condition Index (SDCI) based on the combination of the PCI, VCI, and VCI by four kinds of weight determination methods, i.e. Analytic Hierarchy Process (AHP), Entropy method, Criteria Importance Through Intercriteria Correlation (CRITIC), and Fuzzy Comprehensive Evaluation (FCE). Finally, the agricultural drought calculated by the SDCI was evaluated against drought area, disaster area, and crop failure area to verify the applicability of the SDCI for agricultural drought disaster assessment in the Wei River basin. The results showed that the SDCI determined by FCE has better correlations with crop yield (R2 =0.45) than the other methods. The SDCI values exhibited a "W" shape from 2003 to 2010 during the growing seasons and agricultural drought showed an increasing trend after 2013. The drought-prone areas shifted from north to south, with the degree of drought firstly decreasing and then increasing. In addition, the SDCI has better correlations with the disaster area (R2 =0.35) than the drought area (R2 = 0.16). At the municipal level, the SDCI could well assess agricultural drought. The results demonstrated that the SDCI can effectively monitor and assess drought impacts on agriculture and may provide helpful information for agricultural drought disaster prevention. • Drought condition was jointly assessed by VCI, TCI and PCI. • SDCI was computed by integrating VCI, TCI and PCI based on multi-source data. • Weights of SDCI were determined by crop yield based on four methods. • SDCI can well assess agricultural drought by comparing with drought disaster data. [ABSTRACT FROM AUTHOR]