Predicted Soil Water Content Research Articles

Quantification of soil water content by machine learning using enhanced high-resolution ERT

The accurate acquisition of soil water content is a fundamental cornerstone of research into hydrological processes and agricultural engineering. Electrical Resistivity Tomography (ERT) has been validated for hydrological studies and soil monitoring. The establishment of a quantitative relationship between ERT resistivity data and soil water content is usually based on rock physics models. However, the applicability of such models in complex environments and the acquisition of the relevant parameters pose a certain challenge. In addition, the spatial resolution of ERT limits its application in soil moisture assessment. Therefore, a machine learning-based approach is proposed in this study to determine the quantitative relationship between resistivity and soil water content. We investigate the integration of three machine learning models (KNN, RF, XGBOOST) with ERT to predict the water content of clay soils. The results show that the RF model achieves an R2 of 0.92 with an RMSE of 0.41. To improve the ERT resolution, a new data collection method (MRU) is introduced in this study by increasing the density of ERT data collection. A comparative analysis is conducted between traditional ERT data collection methods and the MRU approach in terms of soil water content prediction accuracy. The results show that the MRU method of data collection improves the accuracy of soil water content prediction by an average of 57% compared to traditional methods. This study confirms the feasibility of using machine learning models to establish mappings between resistance and water content and shows that the MRU data collection method for ERT effectively improves the accuracy of predicting soil water content. These results provide a new perspective for hydrological process research and agricultural monitoring technology.

A Study on Hyperspectral Soil Moisture Content Prediction by Incorporating a Hybrid Neural Network into Stacking Ensemble Learning

The accurate prediction of soil moisture content helps to evaluate the quality of farmland. Taking the black soil in the Nanguan District of Changchun City as the research object, this paper proposes a stacking ensemble learning model integrating hybrid neural networks to address the issue that it is difficult to improve the accuracy of inversion soil moisture content by a single model. First, raw hyperspectral data are processed by removing edge noise and standardization. Then, the gray wolf optimization (GWO) algorithm is adopted to optimize a convolutional neural network (CNN), and a gated recurrent unit (GRU) and an attention mechanism are added to construct a hybrid neural network model (GWO–CNN–GRU–Attention). To estimate soil water content, the hybrid neural network model is integrated into the stacking model along with Bagging and Boosting algorithms and the feedforward neural network. Experimental results demonstrate that the GWO–CNN–GRU–Attention model proposed in this paper can better predict soil water content; the stacking method of integrating hybrid neural networks overcomes the limitations of a single model’s instability and inferior accuracy. The relative prediction deviation (RPD), root mean square error (RMSE), and coefficient of determination (R2) on the test set are 4.577, 0.227, and 0.952, respectively. The average R2 and RPD increased by 0.056 and 1.418 in comparison to the base learner algorithm. The study results lay a foundation for the fast detection of soil moisture content in black soil areas and provide a data source for intelligent irrigation in agriculture.

Modeling the response of sesame (Sesamum indicum L.) to different soil fertility levels under rain-fed conditions in the semi-arid areas of western Tigray, Ethiopia

Sesame, a crucial oilseed crop in Ethiopia, ranks second only to coffee in its importance as an exported agricultural commodity. However, inadequate soil fertility management has hampered its productivity despite its substantial international market demand. Hence, this study was conducted to model the response of sesame to different nitrogen fertilizer levels using the AquaCrop model, and to assess the capability of the model as a decision-support tool for optimizing soil fertility management strategies in the study area. The experiment was laid out in a randomized complete block design, consisting of four nitrogen fertilizer rates (0, 23, 46, and 69 kg/ha nitrogen) and three distinct sesame varieties (Setit-1, Setit-2, and Humera-1). Over the course of the three cropping seasons, data on soil physical and chemical properties, crop growth, yield and yield components were collected for each treatment. Evaluation of model performance relied upon established metrics of coefficient of determination (R2), root mean square error (RMSE), normalized root mean square error (N-RMSE), model efficiency (E), and degree of agreement (D). Analysis of results revealed the AquaCrop model appropriately calibrated for simulation of soil water content, showing R2 values ranging from 0.92 to 0.98, RMSE values varying from 6.5 to 13.9 mm, E values from 0.78 to 0.94, and D values from 0.95 to 0.99. Similarly, simulation outputs for aboveground biomass (AB) demonstrated good accuracy of the model, with R2 values varying from 0.92 to 0.98, RMSE values ranging from 0.33 to 0.54 tons/ha, and D values from 0.9 to 0.98. Notable accuracy was also observed in the simulation of canopy cover (CC), revealing R2 values between 0.95 and 0.99, and RMSE values ranging from 5.3 to 8.6 %. In conclusion, this study substantiates the successful calibration and validation of the AquaCrop model for predicting sesame response to diverse nitrogen fertilizer levels. The performance of the model in predicting soil water content, CC, AB, and yield highlights its potential as a valuable tool for optimizing soil fertility management and enhancing sesame cultivation practices in Ethiopia.

An Improved Pedotransfer Function for Soil Hydrological Properties in New Zealand

This paper describes a new pedotransfer function (PTF) for the soil water content of New Zealand soils at seven specific tensions (0, −5, −10, −20, −40, −100, −1500 kPa) using explanatory variables derived from the S-map soil mapping system. The model produces unbiased and physically plausible estimates of the response at each tension, as well as unbiased and physically plausible estimates of the response differences that define derived properties (e.g., macroporosity and total available water content). The PTF is a development of an earlier model using approximately double the number of sites compared with the earlier study, a change in fitting methodology to a semi-parametric GAM Beta response, and the inclusion of sample depth. The results show that the new model has resulted in significant improvements for the soil water content estimates and derived quantities using standard goodness-of-fit measures, based on validation data. A comparison with an international PTF using explanatory variables compatible with variables available from S-map (EUPTF2) suggests that the model is better for prediction of soil water content using the limited information available from the S-map system.

Estimating soil water content from thermal images with an artificial neural network

Data of soil water content measured at regular time intervals could be used to set an irrigation schedule for farmland. Changes in soil water content could also provide an early warning of potential slope slippage. An artificial neural network is proposed for detecting soil water content from five variables: soil and water temperature changes after sunlight exposure, plant coverage rates, fine-grained soil content, and effective soil particle size. Temperature changes were detected using thermal images taken using an unmanned aerial vehicle equipped with a thermal camera in the morning and at noon. Plant coverage was estimated using visible light photograph, and the fine-grained soil content, effective particle size, and soil water content were determined from laboratory tests of soil samples. An optimized neural network was established by analyzing the performance of various neural network parameters. The mean square error of the optimized neural network reached a minimum value of 0.10061 in the validation data set at the 12th epoch, the training was stopped to avoid overfitting. The correlation coefficients for the training, validation, testing, and overall data sets were 0.880, 0.772, 0.810, and 0.847 respectively. The target values and the predicted values of the neural network are highly correlated. Thus, the proposed neural network is effective for predicting soil water content. A relative importance analysis revealed that soil and water temperature changes after sunlight exposure had the greatest influence on the neural network’s results.

A Novel Three-Segment Model to Describe the Entire Soil–Water Characteristic Curve

This study aims to accurately describe the soil–water characteristic curve (SWCC) across the full range from saturation to oven dryness. We propose a smooth, continuous three-segmented SWCC model that divides the saturation range into wet, air-dried, and oven-dried segments. The two model junction points are anchored at matric suctions of 104.5 and 106.5 cm, respectively. The soil water content at 104.5 cm represents the maximum soil hygroscopy, reflecting the maximum water content in air-dried soil, while the soil water content at 106.5 cm characterizes the minimum soil water content. This imbues the junction points with specific physical significance regarding soil moisture content and matric potential. The model was tested with the water retention data of nine soils across the SWCC and compared with three existing SWCC models based on the adjusted coefficient of determination (adjR2) and root mean square error (RMSE). The results indicated that the proposed model accurately described the entire SWCC. The three-segmented model yielded an adjR2 of >0.99 and an RMSE of ≤0.022 cm3 cm−3, outperforming other models. We also introduce a new method for predicting soil water data in air-dried and oven-dried segments. The results showed that the predicted soil water content values were accurate.

Regional soil water content monitoring based on time-frequency spectrogram of low-frequency swept acoustic signal

Acoustic waves offer a non-destructive, safe, and cost-effective means of monitoring the environment, with a potential application in soil water content monitoring. However, extracting soil water information from acoustic signals is still challenging. To tackle this issue, we have developed a low-frequency swept acoustic signal detection device and system. We conducted soil penetration testing using low-frequency swept acoustic signals. The swept-frequency acoustic signals passing through the soil were transformed into time–frequency spectrogram. Using the Swin-Transformer model, we established a regression model between the time–frequency spectrogram of the swept frequencies and the soil water content. Predictions were made both on a laboratory test dataset and through field trials using the calibrated model. The results indicate that the RMSE, MAE, and R2 values between the observed and the model's outputs of water content (%) for the test laboratory dataset are 0.191, 0.081, and 0.999, respectively, using the Swin-Transformer model. In the case of the field trials, the RMSE, MAE, and R2 values between the predicted and observed values are 6.715 %, 1.829 %, and 0.711, respectively. These studies demonstrate that this method is highly effective in predicting soil water content, with the best results achieved at a resolution of 20 PPI (Pixels Per Inch) and within the frequency range of 260–360 Hz. It provides an efficient approach for acoustic soil water content detection, effectively resolves the difficulty in building models caused by the single-parameter limitation in traditional acoustic model.

Assessing the Potential of UAV-Based Multispectral and Thermal Data to Estimate Soil Water Content Using Geophysical Methods

Knowledge of the soil water content (SWC) is important for many aspects of agriculture and must be monitored to maximize crop yield, efficiently use limited supplies of irrigation water, and ensure optimal nutrient management with minimal environmental impact. Single-location sensors are often used to monitor SWC, but a limited number of point measurements is insufficient to measure SWC across most fields since SWC is typically very heterogeneous. To overcome this difficulty, several researchers have used data acquired from unmanned aerial vehicles (UAVs) to predict the SWC by using machine learning on a limited number of point measurements acquired across a field. While useful, these methods are limited by the relatively small number of SWC measurements that can be acquired with conventional measurement techniques. This study uses UAV-based data and thousands of SWC measurements acquired using geophysical methods at two different depths and before and after precipitation to predict the SWC using the random forest method across a vineyard in the central United States. Both multispectral data (five reflectance bands and eleven vegetation indices calculated from these bands) and thermal UAV-based data were acquired, and the importance of different reflectance data and vegetation indices in the prediction of SWC was analyzed. Results showed that when both thermal and multispectral data were used to estimate SWC, the thermal data contributed the most to prediction accuracy, although multispectral data were also important. Reflectance data contributed as much or more to prediction accuracy than most vegetation indices. SWC measurements that had a larger sample size and greater penetration depth (~30 cm sampling depth) were more accurately predicted than smaller and shallower SWC estimates (~18 cm sampling depth). The timing of SWC estimation was also important; higher accuracy predictions were achieved in wetter soils than in drier soils, and a light precipitation event also improved prediction accuracy.

Temperature-dependent relationship between soil water content and electrical conductivity

Water content is a fundamental parameter for analysing soil composition and engineering behaviour. However, conceptual water content sensors cannot be used in high-temperature and high-pressure environments. To assess the suitability of using electrical conductivity (EC) as a predictor of soil water content at high temperatures, a series of EC and water content tests of clay and sand is conducted. Subsequently, the relationships between EC and water content at different temperatures are analysed. On this basis, the feasibility of eight EC models for predicting soil water content is discussed. The results show that temperature has a vital impact on the EC test of soil and EC is more sensitive to temperature than to dry density and water content. The temperature effect of soil EC is more obvious with an increase in dry density, and EC does not increase monotonically with an increase in temperature. The predicted effect of some EC models tends to be better with an increase in dry density or temperature. This research provides a reference for the EC test and model establishment of high-temperature soils.

Combination of effective color information and machine learning for rapid prediction of soil water content

Soil water content (SWC) is one of the critical indicators in various fields such as geotechnical engineering and agriculture. To avoid the time-consuming, destructive, and laborious drawbacks of conventional SWC measurements, the image-based SWC prediction is considered based on recent advances in quantitative soil color analysis. In this study, a promising method based on the Gaussian-fitting gray histogram is proposed for extracting characteristic parameters by analyzing soil images, aiming to alleviate the interference of complex surface conditions with color information extraction. In addition, an identity matrix consisting of 32 characteristic parameters from eight color spaces is constituted to describe the multi-dimensional information of the soil images. Meanwhile, a subset of 10 parameters is identified through three variable analytical methods. Then, four machine learning models for SWC prediction based on partial least squares regression (PLSR), random forest (RF), support vector machines regression (SVMR), and Gaussian process regression (GPR), are established using 32 and 10 characteristic parameters, and their performance is compared. The results show that the characteristic parameters obtained by Gaussian-fitting can effectively reduce the interference from soil surface conditions. The RGB, CIEXYZ, and CIELCH color spaces and lightness parameters, as the inputs, are more suitable for the SWC prediction models. Furthermore, it is found that 10 parameters could also serve as optimal and generalizable predictors without considerably reducing prediction accuracy, and the GPR model has the best prediction performance (R2 ≥ 0.95, RMSE ≤ 2.01%, RPD ≥ 4.95, and RPIQ ≥ 6.37). The proposed image-based SWC predictive models combined with effective color information and machine learning can achieve a transient and highly precise SWC prediction, providing valuable insights for mapping soil moisture fields.

Estimation of soil water content using electromagnetic induction sensors under different land uses

The complex nature of podzolic soils makes investigating their subsurface challenging. Near-surface geophysical techniques, like electromagnetic induction (EMI), offer significant assistance in studying podzolic soils. Multi-coil (MC-EMI) and multi-frequency (MF-EMI) sensors were selected to maximize soil water content (SWC) prediction in this study. The objectives were to (i) compare apparent electrical conductivity (ECa) measurements from the MC and MF-EMI sensors under different land use conditions, (ii) investigate the spatial variation of ECa, SWC, texture, soil organic matter (SOM), and bulk density (BD) under different land use conditions, and (iii) use statistical and geostatistical analysis to evaluate the effectiveness of ECa measurements in characterizing SWC under different land use conditions, considering the texture, SOM, and BD contents. The study found that MC-EMI had statistically significant relations (p-value < 0.05) with SWC relative to the MF-EMI. Multiple linear regression (MLR) models were also shown to be more effective in representing SWC variations (higher coefficient of determination and lower root mean square error) than simple linear regression models. MC-EMI sensor provided better SWC predictions compared to the MF-EMI sensor, possibly due to larger sampling depths differences between time domain reflectometry measured SWC (SWCTDR) and MF-EMI sensor than those between SWCTDR and MC-EMI sensor. Lastly, cokriging of measured SWC was found to offer more accurate maps than cokriging of predicted SWC obtained from MLR across different land use conditions. The study has shown that EMI may not be highly effective for shallow depths, and ECa can be affected by various soil properties, making it difficult to extrapolate other parameters. However, EMI still shows promise as a reliable method for predicting SWC in boreal podzolic soils. Research into EMI’s usefulness for this purpose has yielded promising results, as indicated in this study. Further investigation is needed to fully harness the potential of this promising technique.

Prediction of Soil Water Content Based on Hyperspectral Reflectance Combined with Competitive Adaptive Reweighted Sampling and Random Frog Feature Extraction and the Back-Propagation Artificial Neural Network Method

The soil water content (SWC) is a critical factor in agricultural production. To achieve real-time and nondestructive monitoring of the SWC, an experiment was conducted to measure the hyperspectral reflectance of soil samples with varying levels of water content. The soil samples were divided into two parts, SWC higher than field capacity (super-θf) and SWC lower than field capacity (sub-θf), and the outliers were detected by Monte Carlo cross-validation (MCCV). The raw spectra were processed using Savitzky–Golay (SG) smoothing and then the spectral feature variable of SWC was extracted by using a combination of competitive adaptive reweighted sampling (CARS) and random frog (Rfrog). Based on the extracted feature variables, an extreme learning machine (ELM), a back-propagation artificial neural network (BPANN), and a support vector machine (SVM) were used to establish the prediction model. The results showed that the accuracy of retrieving the SWC using the same model was poor, under two conditions, i.e., SWC above and below θf, mainly due to the influence of the lower accuracy of the super-θf part. The number of feature variables extracted by the sub-θf and super-θf datasets were 25 and 18, respectively, accounting for 1.85% and 1.33% of the raw spectra, and the variables were widely distributed in the NIR range. Among the models, the best results were achieved by the BPANN model for both the sub-θf and the super-θf datasets; the R2p, RMSEp, and RRMSE of the sub-θf samples were 0.941, 1.570%, and 6.685%, respectively. The R2p, RMSEp, and RRMSE of the super-θf samples were 0.764, 1.479%, and 4.205%, respectively. This study demonstrates that the CARS–Rfrog–BPANN method was reliable for the prediction of SWC.

Variability of in situ soil water retention curves under different tillage systems and growing seasons

Soil pore space can change over time due to tillage, frost, wetting and drying cycles, plant roots, and compaction. However, a soil water retention curve (SWRC) is usually considered to be static and is determinedin a laboratory on a limited number of small soil samples. Field SWRCs were determined at a depth of 20 cm on Mollic Gleysol. Measurements were made in two growing seasons at different spatial locations on plots with different tillage systems: no-tillage (NT) and conventional tillage (CT). The unimodal van Genuchten retention function (VG) was fitted to each SWRC and model parameters were estimated. We evaluated the variability of the curves within growing seasons, predicted soilwater contents at field capacity, and evaluated the differences in the estimated parameters of the VG model between different growing seasons and tillage systems considering intraseasonal and spatial variability. The shape of the curves depended more on the environmental conditions during the growing season than on the tillage system. In the wetter and cooler growing season of 2020, when soybeans were grown, the soil retained water in a narrower range. In the drier growing season of 2021, when maize was grown, intraseasonal variability in SWRCs was greater than in 2020. Regardless of the growing season, estimated α values were more intraseasonally variable under the CT than under the NT. Based on the SWRCs, predicted water contents at field capacity varied intraseasonally, seasonally, and spatially. Accounting for spatial and intraseasonal variability, the estimated VG parameter θr differed between years, regardless of tillage system. In 2020, the parameter α was higher under the NT system than under the CT system. Under the CT, α differed between years, while under NT it did not. For the parameter n, there were no significant differences between tillage systems or growing seasons. Seasonal, intraseasonal, and spatial variability in soil hydraulic properties should be considered in soil water modelling studies.

Spatial Prediction and Mapping of Soil Water Content by TPE-GBDT Model in Chinese Coastal Delta Farmland with Sentinel-2 Remote Sensing Data

Soil water content is an important indicator used to maintain the ecological balance of farmland. The efficient spatial prediction of soil water content is crucial for ensuring crop growth and food production. To this end, 104 farmland soil samples were collected in the Yellow River Delta (YRD) in China, and the soil water content was determined using the drying method. A gradient boosting decision tree (GBDT) model based on a tree-structured Parzen estimator (TPE) hyperparametric optimization was developed, and then the soil water content was predicted and mapped based on the soil texture and vegetation index from Sentinel-2 remote sensing images. The results of statistical analysis showed that the soil water content had a high coefficient of variation (55.30%), a non-normal distribution, and complex spatial variability. Compared with other models, the TPE-GBDT model had the highest prediction accuracy (RMSE = 6.02% and R2 = 0.71), and its mapping results showed that the areas with high soil water content were distributed on both sides of the river and near the estuary. Furthermore, the results of Shapley additive explanation (SHAP) analysis showed that the soil texture (PC2 and PC5), modified normalized difference vegetation index (MNDVI), and Sentinel-2 red edge position (S2REP) index provided important contributions to the spatial prediction of soil water content. We found that the hydraulic physical properties of soil texture and the vegetation characteristics (such as vegetation coverage, root action, and transpiration) are the key factors affecting the spatial migration and heterogeneity of the soil water content in the study area. The above results show that the TPE algorithm can quickly capture the hyperparameters that are most suitable for the GBDT model, so that the GBDT model can ensure prediction accuracy, reduce the loss function with less training data, and accurately learn of the nonlinear relationship between soil water content and environmental factors. This paper proposes a machine learning method for hyperparameter optimization that shows considerable potential to predict the spatial heterogeneity of soil water content, which can effectively support regional farmland soil and water conservation and high-quality agricultural development.

Research on soil moisture content combination prediction model based on ARIMA and BP neural networks

AbstractPredicting soil moisture accurately is the precondition of realizing accurate irrigation and improving the utilization rate of water resource and the necessary step of developing water‐saving agriculture, which can alleviate the water shortage in our agricultural effectively. In order to further improve the accuracy of soil water content prediction, a combined soil water content prediction model based on Autoregressive moving average model (ARIMA model) and back propagation neural network (BP neural network) neural network is proposed. The model considers the linear and nonlinear characteristics of soil water content data, combines them according to the characteristics of the model itself, gives full play to the advantages of ARIMA model and BP neural network. At the same time, two data smoothing methods were used to establish the ARIMA model, and the adaptive moment estimation algorithm (Adam algorithm) and mind evolutionary algorithm (MEA) optimization BP neural network model were used to propose an improved combined prediction model to predict soil water content data. The experimental results show that the average relative error of the improved combinatorial prediction model is 1.51%, which is 4.18%, 0.95% and 3.1% lower than the combinatorial prediction model, BP neural network model and ARIMA model, respectively, and the overall prediction effect is better, which can be used to save agricultural water and provide a strong basis for the development of water‐saving agriculture in China. At the same time, it can also ensure that crop production is increased and the purpose of national food security is guaranteed.

MODELLING OF POTATO CROP WATER PRODUCTIVITY AND YIELD RESPONSE UNDER SUBSURFACE DRIP IRRIGATION SYSTEM

The need for increased food production by applying low water quantity is pausing more challenges to farmers. The production of Irish Potatoes in Kenya in arid and semi-arid areas is challenging for most farmers due to low yields. The purpose of the study was to simulate potato crop water productivity. A field experiment was conducted at Egerton University during the 2021-2022 dry season. The effect of deficit subsurface drip irrigation allied with superabsorbent polymer soil amendment on growth parameters and yield of potatoes was evaluated. Twelve treatments composed of three water application levels 70%IL, 85%IL,100%IL and four super absorbent polymer (SAP) application rates No SAP, 6 kg/ha, 10 kg/ha,and 14kg/ha laid in factorial experimental design were applied in December 2021 to May 2022 season. Irrigation was undertaken when the allowable water depletion reached 50%. The AquaCrop model was utilised to predict soil water content, biomass, canopy cover, dry yield, and water productivity (WP). The 100% water application level with 0kg/ha SAP was used in model calibration, whereas the remaining treatments were used to validate the model. The model simulated results compared with observed canopy cover, biomass and WP produced a good performance fit with R2 range of 0.86 to 0.98, root mean square error (RMSE) of 10.6%, NSE of 0.9, and Wilmot's degree of agreement of 0.98. However, the model underestimated the water content due to the application of SAP in the soil. Because the SAP improved soil characteristics relevant for water holding capacity without changing the physical texture and structure of the soil. The findings indicated that the combined effect of SAP and subsurface drip resulted in increased yield in deficit water application. The calibrated model was successfully applied in simulating WP and yield. The calibrated AquaCrop model can be applied in planning, operation and management of potato-irrigated farms that will lead to improved water productivity and yield in water constrained areas like semi-arid Kenya.

Soil-Surface-Image-Feature-Based Rapid Prediction of Soil Water Content and Bulk Density Using a Deep Neural Network

This study aimed to develop a deep neural network model for predicting the soil water content and bulk density of soil based on features extracted from in situ soil surface images. Soil surface images were acquired using a Canon EOS 100d camera. The camera was installed in the vertical direction above the soil surface layer. To maintain uniform illumination conditions, a dark room and LED lighting were utilized. Following the acquisition of soil surface images, soil samples were collected using a metal cylinder to obtain measurements of soil water content and bulk density. Various features were extracted from the images, including color, texture, and shape features, and used as inputs for both a multiple regression analysis and a deep neural network model. The results show that the deep neural network regression model can predict soil water content and bulk density with root mean squared error of 1.52% and 0.78 kN/m3. The deep neural network model outperformed the multiple regression analysis, achieving a high accuracy for predicting both soil water content and bulk density. These findings suggest that in situ soil surface images, combined with deep learning techniques, can provide a fast and reliable method for predicting important soil properties.

A Stacked Machine Learning Algorithm for Multi-Step Ahead Prediction of Soil Moisture

A trustworthy assessment of soil moisture content plays a significant role in irrigation planning and in controlling various natural disasters such as floods, landslides, and droughts. Various machine learning models (MLMs) have been used to increase the accuracy of soil moisture content prediction. The present investigation aims to apply MLMs with novel structures for the estimation of daily volumetric soil water content, based on the stacking of the multilayer perceptron (MLP), random forest (RF), and support vector regression (SVR). Two groups of input variables were considered: the first (Model A) consisted of various meteorological variables (i.e., daily precipitation, air temperature, humidity, and wind speed), and the second (Model B) included only daily precipitation. The stacked model (SM) had the best performance (R2 = 0.962) in the prediction of daily volumetric soil water content for both categories of input variables when compared with the MLP (R2 = 0.957), RF (R2 = 0.956) and SVR (R2 = 0.951) models. Overall, the SM, which, in general, allows the weaknesses of the individual basic algorithms to be overcome while still maintaining a limited number of parameters and short calculation times, can lead to more accurate predictions of soil water content than those provided by more commonly employed MLMs.

Prediction of deep soil water content (0–5 m) with in-situ and remote sensing data

Accurate and timely evaluation of soil water content (SWC) in water-limited arid and semiarid regions will provide an essential reference for scientifically based water management strategies and vegetation restoration practices. Theoretical and empirical models based on remote sensing data have been successfully applied to predict large-scale SWC. However, most existing models are limited to the surface soil layer, while the prediction of SWC in the deep soil layer (1–5 m) has been overlooked because of the shortage of long-term and large-scale filed-observed data. In this study, a south-north transect (ca. 860 km) was selected across the Chinese Loess Plateau (CLP). Based on both long-term remote sensing and in-situ data (2013–2016), pedo-transfer functions (PTFs) for the SWC in the 0–5 m profile were developed using multiple regression, random forest, and artificial neural network. The results showed that evapotranspiration, potential evapotranspiration, soil surface temperature, total shortwave broadband albedo, and normalized difference vegetation index were important remote sensing parameters, whereas soil texture and bulk density were important in-situ parameters for SWC PTFs development. Among the three PTF development methods, machine learning (i.e. random forest and artificial neural network) obtained a higher accuracy than multiple regression. For different combinations of input parameters, the introduction of in-situ factors significantly improved the accuracy of PTFs compared with PTFs based on remote sensing data only. The artificial neural network developed a PTF with only five input variables that predicted SWC with reasonable accuracy (root mean square error = 0.039, R2 = 0.697, mean absolute percentage = 24.8) and is thus useful for many applications on the Loess Plateau of China. In the future, more attention should be given to the role of in-situ parameters when developing PTFs for deep SWC prediction.

End‐user‐oriented pedotransfer functions to estimate soil bulk density and available water capacity at horizon and profile scales

AbstractSoil available water capacity (AWC) and bulk density (BD) are key properties for understanding water flows in soils, land‐use planning and irrigation management. As measuring these properties is costly and time‐consuming, pedotransfer functions (PTFs) are commonly used to predict BD and soil moisture at a specific matric potential and then estimate AWC. Currently, operational tools to estimate AWC at the soil profile scale from data easily available are lacking. In this study, new PTFs based on the regression‐tree method Cubist were developed at a regional scale to predict soil water contents at −10 kPa (field capacity) and −1585 kPa (permanent wilting point), and BD. A first PTF (PC model) required commonly measured soil properties (sand, silt, clay, organic carbon contents), while the second (PM model) required four additional predictors: qualitative information deriving from description of the soil profile. The models were validated with an independent dataset. Both models outperformed existing PTFs. AWC was then estimated at the horizon scale, soil‐profile scale and in the upper 30 cm of soil. The PM model performed better than the PC model with the training dataset at the profile scale (Nash‐Sutcliffe efficiency = 0.87 and 0.60; RMSE = 0.186 and 0.326 mm cm−1, respectively). Independent validation of AWC estimates at the profile scale by the PM and PC models yielded RMSE of 0.192–0.204 and 0.229–0.244 mm cm−1, respectively. The sensitivity of AWC to estimates of soil depth and coarse‐fragment content was tested. Results confirmed the importance of these variables resulting in soil observation to estimate AWC accurately.