Search

Study Region: In the Yangtze River basin of China. Study focus: The emerging Explainable Artificial Intelligence (XAI) methods provide us an opportunity to understand the nonlinear relationship that the Deep Learning(DL) model learned inside. The construction of the Three Gorges Dam (TGD) has successfully minimized the likelihood of flooding in the Yangtze River basin. The XAI methods can help us to know the nonlinear relationship behind it. We apply the Long Short Term Memory (LSTM) network, in conjunction with two XAI methods, SHapley Additive exPlanation (SHAP) and Expected Gradient (EG), to do our work.In our DL model, we use YiChang (YC) station runoff,Precipitation (Pre) and vapour pressure deficit (VPD) data from the middle and lower river basin as input, while the output of the model generates runoff data at the DaTong (DT) station, XAI methods enable us to calculate the significance of each input feature is for generating the output feature in a DL model. In this study, we examine the difference in importance scores between the Before Three Gorges Dam (BTGD) period and the After Three Gorges Dam (ATGD) period. New Hydrological Insights for the Region: In the BTGD period, YC runoff was the primary contributor to flooding at the DT station. However, in the ATGD period, the largest contribution to flooding in the middle and lower Yangtze River basin has shifted from YC runoff to the the middle and lower reaches of precipitation. Our results suggest that the XAI can show the nonlinear relationship between the TGD and downstream flood clearly and the TGD can effectively mitigate flooding in the middle and lower river basins by regulating runoff from the upper river basin. The work shows the potential of XAI to explain the nonlinear relationship in the hydrology field.

Study region: In the Yangtze River basin of China. Study focus: We applied a recently popular deep learning (DL) algorithm, Transformer (TSF), and two commonly used DL methods, Long-Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), to evaluate the performance of TSF in predicting runoff in the Yangtze River basin. We also add the main structure of TSF, Self-Attention (SA), to the LSTM and GRU models, namely LSTM-SA and GRU-SA, to investigate whether the inclusion of the SA mechanism can improve the prediction capability. Seven climatic observations (mean temperature, maximum temperature, precipitation, etc.) are the input data in our study. The whole dataset was divided into training, validation and test datasets. In addition, we investigated the relationship between model performance and input time steps. New hydrological insights for the region: Our experimental results show that the GRU has the best performance with the fewest parameters while the TSF has the worst performance due to the lack of sufficient data. GRU and the LSTM models are better than TSF for runoff prediction when the training samples are limited (such as the model parameters being ten times larger than the samples). Furthermore, the SA mechanism improves the prediction accuracy when added to the LSTM and the GRU structures. Different input time steps (5 d, 10 d, 15 d, 20 d, 25 d and 30 d) are used to train the DL models with different prediction lengths to understand their relationship with model performance, showing that an appropriate input time step can significantly improve the model performance.

Drought is a natural disaster that occurs globally and can damage the environment, disrupt agricultural production and cause large economic losses. The accurate prediction of drought can effectively reduce the impacts of droughts. Deep learning methods have shown promise in drought prediction, with convolutional neural networks (CNNs) being particularly effective in handling spatial information. In this study, we employed a deep learning approach to predict drought in the Fenhe River (FHR) basin, taking into account the meteorological conditions of surrounding regions. We used the daily SAPEI (Standardized Antecedent Precipitation Evapotranspiration Index) as the drought evaluation index. Our results demonstrate the effectiveness of the CNN model in predicting drought events 1~10 days in advance. We evaluated the predictions made by the model; the average Nash–Sutcliffe efficiency (NSE) between the predicted and true values for the next 10 days was 0.71. While the prediction accuracy slightly decreased with longer prediction lengths, the model remained stable and effective in predicting heavy drought events that are typically difficult to predict. Additionally, key meteorological variables for drought predictions were identified, and we found that training the CNN model with these key variables led to higher prediction accuracy than training it with all variables. This study approves an effective deep learning approach for daily drought prediction, particularly when considering the meteorological conditions of surrounding regions.

Soil moisture is an essential component for the planetary balance between land surface water and energy. Obtaining long-term global soil moisture data is important for understanding the water cycle changes in the warming climate. To date several satellite soil moisture products are being developed with varying retrieval algorithms, however with considerable missing values. To resolve the data gaps, here we have constructed two global satellite soil moisture products, i.e., the CCI (Climate Change Initiative soil moisture, 1989–2021; CCIori hereafter) and the CM (Correlation Merging soil moisture, 2006–2019; CMori hereafter) products separately using a Convolutional Neural Network (CNN) with autoencoding approach, which considers soil moisture variability in both time and space. The reconstructed datasets, namely CCIrec and CMrec, are cross-evaluated with artificial missing values, and further againt in-situ observations from 12 networks including 485 stations globally, with multiple error metrics of correlation coefficients (R), bias, root mean square errors (RMSE) and unbiased root mean square error (ubRMSE) respectively. The cross-validation results show that the reconstructed missing values have high R (0.987 and 0.974, respectively) and low RMSE (0.015 and 0.032 m3/m3, respectively) with the original ones. The in-situ validation shows that the global mean R between CCIrec (CCIori) and in-situ observations is 0.590 (0.581), RMSE is 0.093 (0.093) m3/m3, ubRMSE is 0.059 (0.058) m3/m3, bias is 0.032 (0.037) m3/m3 respectively; CMrec (CMori) shows quite similar results. The added value of this study is to provide long-term gap-free satellite soil moisture products globally, which helps studies in the fields of hydrology, meteorology, ecology and climate sciences.

The North China Plain (NCP) represents a significant agricultural production region in China, with winter wheat serving as one of its main grain crops. Accurate identification of winter wheat through remote sensing technology holds significant importance in ensuring food security in the NCP. In this study, we have utilized Landsat 8 and Landsat 9 imagery to identify winter wheat in the NCP. Multiple convolutional neural networks (CNNs) and transformer networks, including ResNet, HRNet, MobileNet, Xception, Swin Transformer and SegFormer, are used in order to understand their uncertainties in identifying winter wheat. At the same time, these deep learning (DL) methods are also compared to the traditional random forest (RF) method. The results indicated that SegFormer outperformed all methods, of which the accuracy is 0.9252, the mean intersection over union (mIoU) is 0.8194 and the F1 score (F1) is 0.8459. These DL methods were then applied to monitor the winter wheat planting areas in the NCP from 2013 to 2022, and the results showed a decreasing trend.

Soil moisture over the Tibetan Plateau (TP) can affect hydrological cycles on local and remote scales through land–atmosphere interactions. However, TP long-term surface soil moisture characteristics and their response to climate change are still unclear. In this study, we firstly evaluate two satellite-based products—SSM/I (the Special Sensor Microwave Imagers) and ECV COMBINED (the Essential Climate Variable combined)—and three reanalysis products—ERA5-Land (the fifth generation of the land component of the European Centre for Medium-Range Weather Forecasts atmospheric reanalysis), MERRA2 (the second version of Modern-Era Retrospective Analysis for Research and Applications), and GLDAS Noah (the Noah land surface model driven by Global Land Data Assimilation System)—against two in situ observation networks. SSM/I and GLDAS Noah outperform the other soil moisture products, followed by MERRA2 and ECV COMBINED, and ERA5-Land has a certain degree of uncertainty in evaluating TP surface soil moisture. Analysis of long-term soil moisture characteristics during 1988–2008 shows that annual and seasonal mean soil moisture have similar spatial distributions of soil moisture decreasing from southeast to northwest. Additionally, a significant increasing trend of soil moisture is found in most of the TP region. With a non-linear machine learning method, we quantify the contribution of each climatic variable to warm-season soil moisture. It indicates that precipitation dominates soil moisture changes rather than air temperature. Pixel-wise partial correlation coefficients further show that there are significant positive correlations between precipitation and soil moisture over most of the TP region. The results of this study will help to understand the role of TP soil moisture in land–atmosphere coupling and hydrological cycles under climate change.

Accurate assessment of the extent of crop distribution and mapping different crop types are essential for monitoring and managing modern agriculture. Medium and high spatial resolution remote sensing (RS) for Earth observation and deep learning (DL) constitute one of the most major and effective tools for crop mapping. In this study, we used high-resolution Sentinel-2 imagery from Google Earth Engine (GEE) to map paddy rice and winter wheat in the Bengbu city of Anhui Province, China. We compared the performance of different popular DL backbone networks with the traditional machine learning (ML) methods, including HRNet, MobileNet, Xception, and Swin Transformer, within the improved DeepLabv3+ architecture, Segformer and random forest (RF). The results showed that the Segformer based on the combination of the Transformer architecture encoder and the lightweight multilayer perceptron (MLP) decoder achieved an overall accuracy (OA) value of 91.06%, a mean F1 Score (mF1) value of 89.26% and a mean Intersection over Union (mIoU) value of 80.70%. The Segformer outperformed other DL methods by combining the results of multiple evaluation metrics. Except for Swin Transformer, which was slightly lower than RF in OA, all DL methods significantly outperformed RF methods in accuracy for the main mapping objects, with mIoU improving by about 13.5~26%. The predicted images of paddy rice and winter wheat from the Segformer were characterized by high mapping accuracy, clear field edges, distinct detail features and a low false classification rate. Consequently, DL is an efficient option for fast and accurate mapping of paddy rice and winter wheat based on RS imagery.

Soil moisture (SM) is an important variable in mediating the land-atmosphere interactions. Earth System Models (ESMs) are the key tools for predicting the response of SM to future climate change. Many ESMs provide outputs for SM; however, the estimated SM accuracy from different ESMs varies geographically as each ESM has its advantages and limitations. This study aimed to develop a merged SM product with improved accuracy and spatial resolution in China for 2015-2100 through data fusion of 25 ESMs with a deep-learning (DL) method. A DL model that can simultaneously perform data fusion and spatial downscaling was used to analyze SM’s future trend in China. Through the model, monthly SM data in four future scenarios (SSP1-2.6, SSP2-4.5, SSP3-7.0, SSP5-8.5) from 2015 to 2100, with a high resolution at 0.25°, was obtained. The evaluation metrics include mean absolute error (MAE), root mean square difference (RMSD), unbiased root mean square difference (ubRMSD), and coefficient of correlation (r). The evaluation results showed that our merged SM product is significantly better than each of the ESMs and the ensemble mean of all ESMs in terms of accuracy and spatial distribution. In the temporal dimension, the merged product is equivalent to the original data after deviation correction and equivalent to reconstructing the fluctuation of the whole series in a high error area. By further analyzing the spatiotemporal patterns of SM with the merged product in China, we found that northeast China will become wetter whereas South China will become drier. Northwest China and the Qinghai-Tibet Plateau would change from wetting to drying under a medium emission scenario. From the temporal scale of the results, the rate of SM variations is accelerated with time in the future under different scenarios. This study demonstrates the feasibility and effectiveness of the proposed procedure for simultaneous data fusion and spatial downscaling to generate improved SM data. The merged data have great practical and scientific implications.

Precise monitoring of floods is significant in disaster management and loss reduction; however, remote sensing data resource and methods can largely affect the monitoring accuracy of flooded areas. In this study, we use cloud-free Sentinel-1 Synthetic Aperture Radar (SAR) imagery, preferable to the optical imagery. We have used 5 convolutional neural networks (CNNs), including HRNet, DenseNet, SegNet, ResNet and DeepLab v3 + for flood monitoring in the Poyang Lake area, and compared their performances with the traditional methods — the bimodal threshold segmentation (BTS) and the OSTU method. The HRNet has superior performance in water body identification with the highest precision and efficiency, based on a parallel structure to not only extract rich semantic information but also maintain high-resolution features in the whole process. Besides, speckle noise reduction by deep convolutional neural networks in SAR imagery is better compared with the Refined Lee filter. The CNNs are then used to monitor the temporal evolution of summer flooding (May-Nov.) in 2020. Results show the smallest water coverage of Poyang Lake in late May; it gradually increases to the maximum in mid-July, and then shows a downward trend until November.

Land evaporation (LET) is an important variable in climate change, water cycle and water resources management. Mastering the projected changes in LET is significant for crop water requirements and the energy cycle. The global climate model (GCM) is a vital tool for future climate change research. However, the GCMs have low spatial resolution and relatively high errors. We use a deep learning (DL)-based model to deal with this problem. The DL approach can downscale the model data and merge simultaneously. We applied the DL approach to a suit of models from the Coupled Model Intercomparison Project 6th edition (CMIP6) LET data. From the result of all the evaluation metrics, the DL merged data greatly improved in both spatial and time dimensions. The mean RMSE is 5.85 mm and the correlation is 0.95 between the DL merged data and reference data (historical reliable evaporation data). The future LET evidently increases in four scenarios (SSP1–2.6, SSP2–4.5, SSP3–7.0, and SSP5–8.5), and the upward intensity rises from the low to high emission scenarios. The highest increasing regions are in the Tibet Plateau and the south of China and the trend is larger than 10 mm/decade in the high scenarios. From the seasonal point of view, the increasing trend in spring and summer is far larger than for autumn and winter. The Tibet Plateau and the northeast of China have the largest upward trend in the spring of SSP5–8.5, higher than 1.6 mm/decade.

Remote sensing imageries processed through empirical and deterministic approaches help predict multiple agronomic traits throughout the growing season. Accurate identification of cotton crop from remotely sensed imageries is a significant task in precision agriculture. This study aims to utilize a deep learning-based framework for cotton crop field identification with Gaofen-1 (GF-1) high-resolution (16 m) imageries in Wei-Ku region, China. An optimized model for the pixel-wise multidimensional densely connected convolutional neural network (DenseNet) was used. Four widely-used classic convolutional neural networks (CNNs), including ResNet, VGG, SegNet, and DeepLab v3+, were also used for accuracy assessment. The results infer that DenseNet can identify cotton crop features within a relatively shorter time about 5 h for training convergence. The model performance was examined by multiple indicators (P, F1, R, and mIou) produced through the confusion matrix, and the derived cotton fields were then visualized. The DenseNet model has illustrated considerable improvements in comparison with the preceding mainstream models. The results showed that the retrieval precision was 0.948, F1 score was 0.953, and mIou was 0.911. Furthermore, its performance is relatively better in discriminating cotton crop fields’ fine structures when clouds, mountain shadows, and urban built up.

Fire is a common circumstance in the world. It causes direct casualties and economic losses, and also brings severe negative influences on the atmospheric environment. In the background of climate warming and rising population, it is important to understand the fire responses regarding the spatio-temporal changes. Thus, a long-term change analysis of fires is needed in China. We use the remote sensed MOD14A1/MYD14A1 fire products to analyze the seasonal variations and long-term trends, based on five main land cover types (forest, cropland, grassland, savannas and urban areas). The fires are found to have clear seasonal variations; there are more fires in spring and autumn in vegetated lands, which are related to the amount of dry biomass and temperature. The fire numbers have significantly increased during the study period, especially from spring to autumn, and those have decreased in winter. The long-term fire trends are different when delineated into different land cover types. There are significant increasing fire trends in grasslands and croplands in North, East and Northeast China during the study period. The urban fires also show increasing trends. On the contrary, there are significant decreasing fire trends in forests and savannas in South China where it is most densely vegetated. This study provides an overall analysis of the spatio-temporal fire changes from satellite products, and it may help to understand the fire risk in the changing climate for a better risk management.

The accurate acquisition of water information from remote sensing images has become important in water resources monitoring and protections, and flooding disaster assessment. However, there are significant limitations in the traditionally used index for water body identification. In this study, we have proposed a deep convolutional neural network (CNN), based on the multidimensional densely connected convolutional neural network (DenseNet), for identifying water in the Poyang Lake area. The results from DenseNet were compared with the classical convolutional neural networks (CNNs): ResNet, VGG, SegNet and DeepLab v3+, and also compared with the Normalized Difference Water Index (NDWI). Results have indicated that CNNs are superior to the water index method. Among the five CNNs, the proposed DenseNet requires the shortest training time for model convergence, besides DeepLab v3+. The identification accuracies are evaluated through several error metrics. It is shown that the DenseNet performs much better than the other CNNs and the NDWI method considering the precision of identification results; among those, the NDWI performance is by far the poorest. It is suggested that the DenseNet is much better in distinguishing water from clouds and mountain shadows than other CNNs.