Soil Moisture Prediction Research Articles

Soil Moisture Forecast Using Transfer Learning: An Application in the High Tropical Andes

Soil moisture is a critical variable in the hydrological cycle and the climate system, significantly impacting water resources, ecosystem functioning, and the occurrence of extreme events. However, soil moisture data are often scarce, and soil water dynamics are not fully understood in mountainous regions such as the tropical Andes of Ecuador. This study aims to model and predict soil moisture dynamics using in situ-collected hydrometeorological data for training and data-driven machine-learning techniques. Our results highlight the fundamental role of vegetation in controlling soil moisture dynamics and significant differences in soil water balance related to vegetation types and topography. A baseline model was developed to predict soil moisture dynamics using neural network techniques. Subsequently, by employing transfer-learning techniques, this model was effectively applied to different soil horizons and profiles, demonstrating its generalization capacity and adaptability. The use of neural network schemes and knowledge transfer techniques allowed us to develop predictive models for soil moisture trained on in situ-collected hydrometeorological data. The transfer-learning technique, which leveraged the knowledge from a pre-trained model to a model with a similar domain, yielded results with errors on the order of 1×10−6<ϵ<1×10−3. For the training data, the forecast of the base network demonstrated excellent results, with the lowest magnitude error metric RMSE equal to 4.77×10−6, and NSE and KGE both equal to 0.97. These models show promising potential to accurately predict short-term soil moisture dynamics with potential applications for natural hazard monitoring in mountainous regions.

A temporal gap filling approach for GNSS-IR soil moisture inversion based on generalized extension

Soil moisture monitoring is a research hotspot in the application of navigation satellite reflected signals. In order to retrieve soil moisture in any period with high precision and solve the problem that soil moisture cannot be retrieved when satellite observation data is partially missing, the principle of generalized extension in engineering science is applied to GNSS-IR technology in the paper. Firstly, the generalized extension approximation method is used to interpolate the precise ephemeris to obtain the altitude angle information of each satellite in each epoch. Then, the generalized extendedinterpolation and extrapolation models were used to supplement the initial phase of the reflected signal corresponding to the missing data as the input of the inversion model to predict soil moisture. The soil moisture inversion experiment was completed according to the observation data provided by PBO. The results show that the generalized extension method can cope with the missing data of different lengths and distributions in GNSS-IR soil moisture retrieval, When the data were confirmed for 15 days, the correlation coefficient of inversion results was 0.916, RSME was 0.03358 cm3/cm3, and MAE was 0.02912 cm3/cm3, and the correlation coefficient of more than 0.85 can be maintained when the amount of missing data reaches 60 % of the test set.

Seasonal forecasting of local-scale soil moisture droughts with Global BROOK90: a case study of the European drought of 2018

Abstract. Prolonged deficit of soil moisture can result in significant ecosystem and economical losses. General slowdown of vegetation growth and development, withering of foliage cover, reduction of carbon, nutrients and water cycling, increase of fire and insect outbreaks are just a few examples of soil moisture drought impacts. Thus, an early and timely warning via monitoring and forecast could help to prepare for a drought and manage its consequences. In this study, a new version of Global BROOK90, an automated framework to simulate water balance at any location, is presented. The new framework integrates seasonal meteorological forecasts (SEAS5 forecasting system) from European Centre for Medium-Range Weather Forecasts (ECMWF). Here we studied how well the framework can predict the soil moisture drought on a local scale. Twelve small European catchments (from 7 to 115 km2) characterized by various geographical conditions were chosen to reconstruct the 2018–2019 period, when a large-scale prolonged drought was observed in Europe. Setting the ERA5-forced soil moisture simulations as a reference, we analysed how the lead time of the SEAS5 hindcasts influences the quality of the soil moisture predictions under drought and non-drought conditions. It was found that the hindcasted soil moisture fits well with the reference model runs only within the first (in some cases until the second and third) month of lead time. Afterwards, significant deviations up to 50 % of soil water volume were found. Furthermore, within the drought period the SEAS5 hindcast forcings resulted in overestimation of the soil moisture for most of the catchment, indicating an earlier end of a drought period. Finally, it was shown that application of the probabilistic forecast using the ensembles' quantiles to account for the uncertainty of the meteorological input is reasonable only for a lead time of up to 3 months.

Data Mining Approach to Predicting Soil Moisture Based on Meteorological Factors and Flow Rates

Data Mining Approach to Predicting Soil Moisture Based on Meteorological Factors and Flow Rates

A comprehensive study of deep learning for soil moisture prediction

Abstract. Soil moisture plays a crucial role in the hydrological cycle, but accurately predicting soil moisture presents challenges due to the nonlinearity of soil water transport and the variability of boundary conditions. Deep learning has emerged as a promising approach for simulating soil moisture dynamics. In this study, we explore 10 different network structures to uncover their data utilization mechanisms and to maximize the potential of deep learning for soil moisture prediction, including three basic feature extractors and seven diverse hybrid structures, six of which are applied to soil moisture prediction for the first time. We compare the predictive abilities and computational costs of the models across different soil textures and depths systematically. Furthermore, we exploit the interpretability of the models to gain insights into their workings and attempt to advance our understanding of deep learning in soil moisture dynamics. For soil moisture forecasting, our results demonstrate that the temporal modeling capability of long short-term memory (LSTM) is well suited. Furthermore, the improved accuracy achieved by feature attention LSTM (FA-LSTM) and the generative-adversarial-network-based LSTM (GAN-LSTM), along with the Shapley (SHAP) additive explanations analysis, help us discover the effectiveness of attention mechanisms and the benefits of adversarial training in feature extraction. These findings provide effective network design principles. The Shapley values also reveal varying data leveraging approaches among different models. The t-distributed stochastic neighbor embedding (t-SNE) visualization illustrates differences in encoded features across models. In summary, our comprehensive study provides insights into soil moisture prediction and highlights the importance of the appropriate model design for specific soil moisture prediction tasks. We also hope this work serves as a reference for deep learning studies in other hydrology problems. The codes of 3 machine learning and 10 deep learning models are open source.

GRU–Transformer: A Novel Hybrid Model for Predicting Soil Moisture Content in Root Zones

The accurate measurement of soil moisture content emerges as a critical parameter within the ambit of agricultural irrigation management, wherein the precise prediction of this variable plays an instrumental role in enhancing the efficiency and conservation of agricultural water resources. This study introduces an innovative, cutting-edge hybrid model that ingeniously integrates Gated Recirculation Unit (GRU) and Transformer technologies, meticulously crafted to amplify the precision and reliability of soil moisture content forecasts. Leveraging meteorological and soil moisture datasets amassed from eight monitoring stations in Hebei Province, China, over the period from 2011 to 2018, this investigation thoroughly assesses the model’s efficacy against a diverse array of input variables and forecast durations. This assessment is concurrently contrasted with a range of conventional machine learning and deep learning frameworks. The results demonstrate that (1) the GRU–Transformer model exhibits remarkable superiority across various aspects, particularly in short-term projections (1- to 2-day latency). The model’s mean square error (MSE) for a 1-day forecast is notably low at 5.22%, reducing further to a significant 2.71%, while the mean coefficient of determination (R2) reaches a high of 89.92%. Despite a gradual increase in predictive error over extended forecast periods, the model consistently maintains robust performance. Moreover, the model shows exceptional versatility in managing different soil depths, notably excelling in predicting moisture levels at greater depths, thereby surpassing its performance in shallower soils. (2) The model’s predictive error inversely correlates with the reduction in parameters. Remarkably, with a streamlined set of just six soil moisture content parameters, the model predicts an average MSE of 0.59% and an R2 of 98.86% for a three-day forecast, highlighting its resilience to varied parameter configurations. (3) In juxtaposition with prevalent models such as Support Vector Regression (SVR), K-Nearest Neighbors (KNN), Gradient Boosting Decision Tree (GBDT), XGBoost, Random Forest, and deep learning models like Deep Neural Network (DNN), Convolutional Neural Network (CNN), and standalone GRU-branch and Transformer-branch models, the GRU–Transformer framework demonstrates a significant advantage in predicting soil moisture content with enhanced precision for a five-day forecast. This underscores its exceptional capacity to navigate the intricacies of soil moisture data. This research not only provides a potent decision-support tool for agricultural irrigation planning but also makes a substantial contribution to the field of water resource conservation and optimization in agriculture, while concurrently imparting novel insights into the application of deep learning techniques in the spheres of agricultural and environmental sciences.

Inversion of large-scale citrus soil moisture using multi-temporal Sentinel-1 and Landsat-8 data

Soil moisture is a significant variable in agricultural study and precision irrigation decision-making. It determines the soil water availability for plants, directly influencing plant growth, yield and quality. Owing to the variations in regional microclimate, landform difference, soil type and vegetation coverage, the soil moisture has strong spatial-temporal heterogeneity on a large regional scale. Micro-wave remote sensing can be used to invert soil moisture based on the dielectric constant under different weather conditions, while optical remote sensing utilizes spectral characteristics to estimate the physiological and ecological information of vegetation. In this study, two new hybrid models (ACO-RF and SSA-RF) were structured by optimizing the standalone random forest (RF) based on the ant colony optimization algorithm (ACO) and sparrow search algorithm (SSA), and six input combinations based on the multi-temporal Sentinel-1 and Landsat-8 remote sensing data from different sensors (optical, thermal and radar sensors) were used. The standalone RF, ACO-RF, and SSA-RF models with different combinations of inputs were employed to predict the soil moisture at different depths (5 cm, 10 cm, 20 cm, 40 cm) in a large-scale drip-irrigated citrus orchard. The results showed that the ACO-RF and SSA-RF outperformed the standalone RF model in terms of prediction accuracy at a depth of 0–40 cm, with R2 of 0.800–0.921 and 0.504–0.798, RRMSE of 7.214–16.284% and 11.124–22.214%, respectively. In the hybrid model, the ACO-RF model had better prediction accuracy than the SSA-RF model, with R2 of 0.805–0.921 and 0.800–0.911, RRMSE of 7.214–13.244% and 8.274–16.284%, respectively. At depths of 5 cm, 10 cm and 20 cm, the inversion accuracy of the model with microwave inputs was higher than that with multispectral inputs, with R2 of 0.556–0.888 and 0.541–0.886, RRMSE of 9.015–19.544% and 9.124–22.214%, respectively. However, at a depth of 40 cm, the inversion accuracy of the model with multispectral inputs was higher than that with microwave inputs, with R2 of 0.532–0.841 and 0.508–0.831, RRMSE of 9.124–21.021% and 9.142–21.214%, respectively. The model with multispectral, thermal, and microwave inputs exhibited the highest accuracy in predicting soil moisture, with R2 of 0.635–0.921, RRMSE of 7.214−18.564%, respectively. Therefore, the ACO-RF with multisource remote sensing data is recommended to predict the soil moisture in the drip-irrigated citrus orchard. This approach can provide data support for making intelligent irrigation decisions on a large-scale grid land lots.

Predicting Soil Moisture Content Based on Laser-Induced Breakdown Spectroscopy-Informed Machine Learning

Predicting Soil Moisture Content Based on Laser-Induced Breakdown Spectroscopy-Informed Machine Learning

Prediction of soil moisture using machine learning techniques: A case study of an IoT‐based irrigation system in a naturally ventilated polyhouse

AbstractThe agricultural sector faces a massive challenge in enhancing food production for the growing population with limited water resources. For effective and optimum utilization of fresh water, developing smart irrigation systems based on the internet of things (IoT) is essential for scheduling irrigation based on crop water requirements. In this study, an IoT‐based irrigation system was developed and evaluated inside a greenhouse located in the experimental fields of Indian Council of Agricultural Research‐Central Institute of Agricultural Engineering (ICAR‐CIAE), Bhopal, India. Data on microenvironmental parameters such as temperature, relative humidity, light intensity, soil temperature and soil moisture were collected from the sensors developed inside the greenhouse. Soil moisture was predicted based on the field data collected via different machine learning techniques, such as the decision tree (DT), random forest (RF), multiple linear regression (MLR), extreme gradient boosting (XGB), K‐nearest neighbour (KNN) and artificial neural network (ANN) methods, with three input combinations. The ANN (coefficient of determination [R2] = 0.942, 0.939) models performed well but were found to be less effective than the RF (R2 = 0.991, 0.951) and XGB (R2 = 0.997, 0.941) models in the training and testing phases, respectively. The RF and XGB models outperformed the other models, while the MLR (R2 = 0.955, 0.875) technique underperformed. With respect to both the testing and training datasets, the models trained with all four inputs outperformed the models trained with two or three inputs.

Enhancing Deep Learning Soil Moisture Forecasting Models by Integrating Physics-based Models

Enhancing Deep Learning Soil Moisture Forecasting Models by Integrating Physics-based Models

Soil Moisture Prediction Model Based on Improved GRU Recurrent Neural Network

Soil moisture plays a crucial role in land water and energy cycles, and has a certain impact on weather and climate change. In agricultural production, crop moisture status can be determined based on soil moisture, and timely and effective irrigation strategies can be formulated to ensure grain yield while saving water resources, maximizing the value of agricultural water resource utilization, and achieving sustainable development. Therefore, the accuracy of soil moisture prediction has important research value for many fields such as agriculture and climate. In this paper, the super parameters of GRU Recurrent neural network are optimized by intelligent seagull optimization algorithm using a small number of influencing factors, namely, atmospheric temperature, atmospheric humidity, rainfall and soil moisture data, and a soil moisture prediction model is established. The model was used to predict soil moisture for the next 12 hours, 24 hours, 36 hours, and 48 hours, respectively. The final experiment showed that the model in this paper had better predictive effect on soil moisture, with the best predictive evaluation index data being MAPE (12h) = 4.4120%, R2 (12h) = 0.94605, and RMSE (12h) = 1.9998; By comparing the prediction results of multiple time steps vertically, it was found that the prediction accuracy of the model in this paper decreased more smoothly, meeting the requirements of soil moisture prediction.

Harnessing the power of transformers and data fusion in smart irrigation

In recent years, IoT sensors have enabled smart agriculture to grow rapidly with many compelling real-world applications. One such application is the case of smart irrigation. A particular interest exists in forecasting soil water potential to allow for the establishment of more efficient irrigation systems. Nonetheless, forecasting soil moisture is a complex task and depends on various information sources. Most existing work relies on local approaches, which are less effective at leveraging shared information across different data sources. Therefore, this paper presents a robust global approach for soil water potential forecasting, combining various environmental factors through the cutting-edge Temporal Fusion Transformer. Our proposed approach outperforms established baselines for forecasting soil water potential. As such, this work contributes to the growing body of research on data fusion in real-world applications.

Development of inter-grid-cell lateral unsaturated and saturated flow model in the E3SM Land Model (v2.0)

Abstract. The lateral transport of water in the subsurface is important in modulating terrestrial water energy distribution. Although a few land surface models have recently included lateral saturated flow within and across grid cells, it is not a default configuration in the Climate Model Intercomparison Project version 6 experiments. In this work, we developed the lateral subsurface flow model within both unsaturated and saturated zones in the Energy Exascale Earth System Model (E3SM) Land Model version 2 (ELMv2.0). The new model, called ELMlat, was benchmarked against PFLOTRAN, a 3D subsurface flow and transport model, for three idealized hillslopes that included a convergent hillslope, divergent hillslope, and tilted V-shaped hillslope with variably saturated initial conditions. ELMlat showed comparable performance against PFLOTRAN in terms of capturing the dynamics of soil moisture and groundwater table for the three benchmark hillslope problems. Specifically, the mean absolute errors (MAEs) of the soil moisture in the top 10 layers between ELMlat and PFLOTRAN were within 1 %±3 %, and the MAEs of water table depth were within ±0.2 m. Next, ELMlat was applied to the Little Washita experimental watershed to assess its prediction of groundwater table, soil moisture, and soil temperature. The spatial pattern of simulated groundwater table depth agreed well with the global groundwater table benchmark dataset generated from a global model calibrated with long-term observations. The effects of lateral groundwater flow on the energy flux partitioning were more prominent in lowland areas with shallower groundwater tables, where the difference in simulated annual surface soil temperature could reach 0.3–0.4 ∘C between ELMv2.0 and ELMlat. Incorporating lateral subsurface flow in ELM improves the representation of the subsurface hydrology, which will provide a good basis for future large-scale applications.

Estimation of spatial distribution of soil moisture on steep hillslopes by state-space approach (SSA)

Soil moisture is a critical variable that quantifies precipitation, floods, droughts, irrigation, and other factors with regard to decision-making and risk evaluation. An accurate prediction of soil moisture dynamics is important for soil and environmental management. However, the complex topographic condition and land use in hilly and mountainous areas make it a challenge to monitor and predict soil moisture dynamics in these areas. In this study, the determinants of soil moisture variability were determined by structural equation modeling, and then an attempt was made to estimate the spatial distribution of soil moisture content on steep hillslope using the state-space method. Herein, soil moisture at different depths (0–10, 10–20, and 20–30 cm) was monitored by portable time-domain reflectometer (TDR) along this hillslope (100 m × 180 m). It showed that the spatial variability of soil moisture decreased with increasing soil wetness, primarily in the topsoil (0–10 cm). Soil moisture was correlated with elevation (r = 0.28, 0.50, and 0.28), capillary porosity (r = 0.06, 0.37, and 0.28), soil texture (r for Clay: 0.20, 0.24, and 0.16; r for Sand: −0.25, −0.18, and −0.28), organic carbon (r = −0.31, −0.08, and 0.10) and land use (r = −0.01, 0.28, and 0.24) under different conditions (dry, moderate, and wet). Among these determinants, elevation made direct contributions to soil moisture variation, especially under moderate conditions, while land use made its impacts by altering soil texture. It is encouraging that the state-space approach yielded precise and cost-effective predictions of soil moisture dynamics along this steep hillslope since it gives the minimum root-mean-square error (RMSE) and Akaike information criterion (AIC). Moreover, soil organic carbon (AIC = −4.497, RMSE = 0.104, R2 = 0.899), rock fragment contents (AIC = −4.366, RMSE = 0.111, R2 = 0.878), and elevation (AIC = −3.693, RMSE = 0.156, R2 = 0.629) effectively anticipated the spatial distribution of soil moisture under dry, moderate, and wet conditions, respectively. This study confirms the efficacy of the state-space approach as a valuable tool for soil moisture prediction in areas characterized by complex and spatially heterogeneous conditions.

Comparing modern and traditional modeling methods for predicting soil moisture in IoT-based irrigation systems

Proper water management is crucial for agriculture, as its irrigation systems can waste approximately 60% of water. Using IoT for irrigation based on soil moisture can help reduce waste. This study used multidimensional time series modeling to predict soil moisture using two years of data on a US farm, including factors such as soil moisture, temperature, and weather. The models assessed were machine learning algorithms like Extreme Gradient Boosting and Random Forests, Deep Learning's Spectral Temporal Graph Neural Network (StemGNN), and Vector Autoregression as a reference model. Random Forest was the most efficient and stable model. Furthermore, the spectral temporal graph neural network could not outperform the reference model. Comparing Random Forest with univariate algorithms like Naive, Exponential Smoothing, and Autoregressive Integrated Moving Average (ARIMA) showed univariate's better performance in 9 of 10 datasets, with ARIMA being best in 8. A substantial disparity is observed when considering the Mean Absolute Percentage Error (MAPE) metric. ARIMA yields an MAPE of 0.052, while StemGNN reports 0.127 as the major difference. ARIMA achieves a MAPE of 0.046 for the minor difference, whereas StemGNN obtains 0.038. In the remaining eight datasets, ARIMA consistently outperforms StemGNN by at least 30%. In summary, modern methods could not beat the older methods in the study data sets.

Relationships between Soil Moisture and Visible–NIR Soil Reflectance: A Review Presenting New Analyses and Data to Fill the Gaps

The ability to precisely monitor soil moisture is highly valuable in industries including agriculture and civil engineering. As soil moisture is a spatially erratic and temporally dynamic variable, rapid, cost-effective, widely applicable, and practical techniques are required for monitoring soil moisture at all scales. If a consistent numerical relationship between soil moisture content and soil reflectance can be identified, then soil spectroscopic models may be used to efficiently predict soil moisture content from proximal soil reflectance and/or remotely sensed data. Previous studies have identified a general decrease in visible–NIR soil reflectance as soil moisture content increases, however, the strength, best wavelengths for modelling, and domain of the relationship remain unclear from the current literature. After reviewing the relevant literature and the molecular interactions between water and light in the visible–NIR (400–2500 nm) range, this review presents new analyses and interprets new 1 nm resolution soil reflectance data, collected at >20 moisture levels for ten soil samples. These data are compared to the results of other published studies, extending these as required for further interpretation. Analyses of this new high-resolution dataset demonstrate that linear models are sufficient to characterise the relationship between soil moisture and reflectance in many cases, but relationships are typically exponential. Equations generalising the relationship between soil MC and reflectance are presented for a number of wavelength ranges and combinations. Guidance for the adjustment of these equations to suit other soil types is also provided, to allow others to apply the solutions presented here and to predict soil moisture content in a much wider range of soils.

Forecasting of soil moisture using machine learning in smart agriculture systems

Growing crops in modern conditions is a complex task and practically combines the practices of experience and the latest methods, including information technology, which has become part of the concept of "smart farming". An important factor in the stable predicted yield is the level of soil moisture, which is the result of changes in climatic factors such as air temperature, soil temperature, intensity of solar radiation, rainfall, wind speed, etc. A methodology for processing real historical indicators of climate change in a certain geographical area with subsequent training and application of machine learning models to predict soil moisture is proposed. To build a machine learning model, the following algorithms were selected and studied: the algorithm of regression trees, random forest, linear regression, M5P algorithms and the K* algorithm. The data source for training the models is the open information resource International Soil Moisture Network (ISMN) from ismn.earth/en. , which provides data on soil moisture and temperature, air temperature, and rainfall. Other data was used from the Open Meteo information service, which provides a free API and allows you to get historical data and weather forecast in specified coordinates during specified days. A data structure was developed to train the model for further prediction of soil moisture. An architecture has been developed and a software system for predicting soil moisture based on machine learning algorithms has been created using the Spring Framework, the WEKA library and Java FX with the ability to select and study the appropriate algorithms. Experiments have been carried out and the results of the duration of model training have been presented, while the algorithms of regression trees and linear regression require the least training time. A comparison of algorithms is made according to the following criteria: learning speed, cross-testing speed, prediction speed, testing performance indicators for real historical data. Based on the results of the study, conclusions are drawn about individual algorithms, the feasibility of using them to predict soil moisture based on climatic indicators. The obtained results will make it possible to evaluate and select the best models of machine learning in the design of the information and analytical system "smart agriculture" for forecasting soil moisture.

Soil Moisture Content Sensor: Utilizing Bacterial Cells as Living Sensors

Soil moisture affects water and heat exchange between soil and air, weather and climate, and plant growth. Therefore, controlling and predicting soil moisture is vital for irrigation, nutrient supply, absorption, and crop productivity. Soil microbial fuel cell (SMFC) is a recent technology that generates electricity from soil microorganisms and chemicals. SMFC can be used to sense soil moisture and treat polluted soil. In this study, we designed a low-cost, portable, and easy-to-install SMFC for soil moisture sensing, and it was tested at four moisture levels: 40%, 60%, 80%, and 100% soil water holding capacity (SWHC). The SMFC worked best at 60 - 80% SWHC, which is suitable moisture for many plants. The results suggest that the proposed SMFC can be a potential soil moisture sensor.

Assessment of the TEROS 10 and TEROS 12 sensors in soil moisture measurement

This study focuses on investigating the accuracy of the empirical equations proposed by the manufacturing company for the widely used TEROS 10 and TEROS 12 sensors. Specifically, the study examines whether the raw - unprocessed - measurements provided by the two sensors lead to reliable predictions of soil moisture and electrical conductivity values when these particular empirical equations are utilized. For this purpose, a prototype setup was developed, incorporating a lowcost microcontroller along with auxiliary power supply and data acquisition circuits for the sensors. The results highlight the need for specific calibration of the sensors for each soil type and salinity level. Additionally, the behaviour of the sensors operating at low frequency (70 MHz) under increasing salinity conditions is contrary to expectations.

Enforcing Water Balance in Multitask Deep Learning Models for Hydrological Forecasting

Abstract Accurate prediction of hydrological variables (HVs) is critical for understanding hydrological processes. Deep learning (DL) models have shown excellent forecasting abilities for different HVs. However, most DL models typically predicted HVs independently, without satisfying the principle of water balance. This missed the interactions between different HVs in the hydrological system and the underlying physical rules. In this study, we developed a DL model based on multitask learning and hybrid physically constrained schemes to simultaneously forecast soil moisture, evapotranspiration, and runoff. The models were trained using ERA5-Land data, which have water budget closure. We thoroughly assessed the advantages of the multitask framework and the proposed constrained schemes. Results showed that multitask models with different loss-weighted strategies produced comparable or better performance compared to the single-task model. The multitask model with a scaling factor of 5 achieved the best among all multitask models and performed better than the single-task model over 70.5% of grids. In addition, the hybrid constrained scheme took advantage of both soft and hard constrained models, providing physically consistent predictions with better model performance. The hybrid constrained models performed the best among different constrained models in terms of both general and extreme performance. Moreover, the hybrid model was affected the least as the training data were artificially reduced, and provided better spatiotemporal extrapolation ability under different artificial prediction challenges. These findings suggest that the hybrid model provides better performance compared to previously reported constrained models when facing limited training data and extrapolation challenges.