Soil moisture (SM) is an important biophysical parameter by which to evaluate water resource potential, especially for agricultural activities under the pressure of global warming. The recent advancements in different types of satellite imagery coupled with deep learning-based frameworks have opened the door for large-scale SM estimation. In this research, high spatial resolution Sentinel-1 (S1) backscatter data and high temporal resolution soil moisture active passive (SMAP) SM data were combined to create short-term SM predictions that can accommodate agricultural activities in the field scale. We created a deep learning model to forecast the daily SM values by using time series of climate and radar satellite data along with the soil type and topographic data. The model was trained with static and dynamic features that influence SM retrieval. Although the topography and soil texture data were taken as stationary, SMAP SM data and Sentinel-1 (S1) backscatter coefficients, including their ratios, and climate data were fed to the model as dynamic features. As a target data to train the model, we used in situ measurements acquired from the International Soil Moisture Network (ISMN). We employed a deep learning framework based on long short-term memory (LSTM) architecture with two hidden layers that have 32 unit sizes and a fully connected layer. The accuracy of the optimized LSTM model was found to be effective for SM prediction with the coefficient of determination () of 0.87, root mean square error (RMSE) of 0.046, unbiased root mean square error (ubRMSE) of 0.045, and mean absolute error (MAE) of 0.033. The model’s performance was also evaluated concerning above-ground biomass, land cover classes, soil texture variations, and climate classes. The model prediction ability was lower in areas with high normalized difference vegetation index (NDVI) values. Moreover, the model can better predict in dry climate areas, such as arid and semi-arid climates, where precipitation is relatively low. The daily prediction of SM values based on microwave remote sensing data and geophysical features was successfully achieved by using an LSTM framework to assist various studies, such as hydrology and agriculture.
This paper evaluates the potential of spaceborne bistatic interferometric synthetic aperture radar images for the monitoring of biophysical variables in wetlands, with a special interest on paddy rice. The assessment is made during the rice cultivation period, from transplanting to harvesting time (May to October) for fields around Gala lake (Turkey), one of the largest and most productive paddy rice planting area in the country. Detailed ground truth measurements describing biophysical parameters are collected in a dedicated campaign. A stack of 16 dual-pol TanDEM-X images is used for the generation of 32 digital elevation models (DEMs) over the studied area. The quality of the data allows the use of the interferometric phase as a state variable capable to estimate crop heights for almost all the growing stages. The early vegetative rice stage, which is characterized by flooded fields, cannot be represented by the interferometric phase due to a low signal-to-noise ratio but can be easily detected by amplitude and interferometric coherence thresholding. A study on the impact of the polarization in the signal backscatter is also performed. An analysis of the differences between HH and VV DEMs shows the varying signal penetration for the two polarizations at different growing stages. The validation with reference data demonstrates the capability to establish a direct relationship between interferometric phase and rice growth. The very high coherence of TanDEM-X data yields elevation estimates with root-mean-square error in a decimetric level, supporting temporal change analysis on a field-by-field basis.
19 May 2016 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Crop monitoring and phenology estimation based on the satellite systems have become an important research area due to high demand on crops. Satellites with synthetic aperture radar (SAR) sensor are highly preferred on such studies because of not only their day/night and all weather acquisition capabilities but also their ability to detect small morphological changes in monitored target, regarding the wavelength of signals. Besides, thanks to the high temporal resolution of new generation space-based sensors, it has been possible to monitor growth cycle of crops by classification algorithms. This paper focused on building a feasible phenology classification schema for paddy-rice using multitemporal co-polar TerraSAR-X images. Phenology classification was conducted with support vector machines (SVM) with linear and nonlinear kernel, k-nearest neighbors (kNN), and decision trees (DT). The key implementation challenges such as the number of classes, the identification of the boundaries of the classes, and the selection of textural and polarimetric features were deeply analyzed. According to all the evaluations conducted, the classification schema was finalized to be used for obtaining thematic maps for two independent rice-cultivated agricultural areas located in Spain and Turkey. The results of these experiments enable one to draw a conclusion about feasibility of machine learning (ML) algorithms in operational phenology monitoring.
Fast soil moisture content (SMC) mapping is necessary to support water resource management and to understand crop growth, quality, and yield. Therefore, earth observation (EO) plays a key role due to its ability of almost real-time monitoring of large areas at a low cost. This study aimed to explore the possibility of taking advantage of freely available Sentinel-1 (S1) and Sentinel-2 (S2) EO data for the simultaneous prediction of SMC with cycle-consistent adversarial network (CycleGAN) for time-series gap filling. The proposed methodology, first, learns latent low-dimensional representation of the satellite images, then learns a simple machine learning (ML) model on top of these representations. To evaluate the methodology, a series of vineyards, located in South Australia ’s Eden valley are chosen. Specifically, we presented an efficient framework for extracting latent features from S1 and S2 imagery. We showed how one could use S1 to S2 feature translation based on CycleGAN using S1 and S2 time series when there are missing images acquired over an area of interest. The resulting data in our study is then used to fill gaps in time-series data. We used the resulting latent representations to predict SMC with various ML tools. In the experiments, CycleGAN and the autoencoders were trained with data randomly chosen around the site of interest, so we could augment the existing dataset. The best performance was demonstrated with random forest (RF) algorithm, whereas linear regression model demonstrated significant overfitting. The experiments demonstrate that the proposed methodology outperforms the compared state-of-the-art methods if there are missing optical and synthetic-aperture radar (SAR) images.