Prediction Of Evapotranspiration Research Articles

Improved forest dynamics leads to better hydrological predictions in watershed modeling

This study explored how the characterization of forest processes in hydrologic models affects watershed hydrological responses. To that end, we applied the widely used Soil and Water Assessment Tool (SWAT) model to two forested watersheds in the southeastern United States. Although forests can cover a large portion of watersheds, tree attributes such as leaf area index (LAI), biomass accumulation, and processes such as evapotranspiration (ET) are rarely calibrated in hydrological modeling studies. The advent of freely and readily available remote-sensing data, combined with field observations from forestry studies and published literature, allowed us to develop an improved forest parameterization for SWAT. We tested our proposed parameterization at the watershed scale in Florida and Georgia and compared simulated LAI, biomass, and ET with the default model settings. Our results showed major improvements in predicted monthly LAI and ET based on MODIS reference data (NSE > 0.6). Simulated forest biomass also showed better agreement with the USDA forest biomass gridded data. Through a series of modeling experiments, we isolated the benefits of LAI, biomass, and ET in predicting streamflow and baseflow at the watershed level. The combined benefits of improved LAI, biomass, and ET predictions yielded the most optimal model configuration where terrestrial and in-stream processes were simulated reasonably well. We performed automated model calibration using two calibration strategies. In the first calibration scheme (M0), SWAT was calibrated for daily streamflow without adjusting LAI, biomass, and ET. In the second calibration scheme (MLAI+BM+ET), previously calibrated parameters constraining LAI, biomass, and ET were incorporated into the model and daily streamflow was recalibrated. The MLAI+BM+ET model showed superior performance and reduced uncertainties in predicting daily streamflow, with NSE values ranging from 0.52 to 0.8. Our findings highlight the importance of accurately representing forest dynamics in hydrological models.

Context Aware Evapotranspiration (ETs) for Saline Soils Reclamation

Accurate Evapotranspiration for saline soils (ETs) is important as well as challenging for the reclamation of saline soils through an effective leaching process. Evapotranspiration (ET) by FAO-56 Penman-Monteith standard method is complex, especially for saline soils. Moreover, existing studies focus on the use of the Internet of Things (IoT) and machine learning-enabled smart and precision irrigation water recommendation systems along with the ET estimation by limited parameters. The ETs for saline soils are also equally important for the reclamation of saline soils, which is ignored by the existing literature. The study proposed IoT and machine leaching-based architecture of context-aware monthly ETs estimations for saline soil reclamation with the effective leaching process. The IoT-enabled crop field contexts in terms of crop field temperature, soil salinity, and irrigation water salinity are used as input features to the Long Short-Term Memory (LSTM) and ensembled LSTM models for monthly ETs predictions. The performance of the proposed solution is observed in terms of the accuracy of the machine learning models along with the comparison against the FAO-56 PM-based standard method. The implementation of the proposed solution reveals that the ensembled LSTM-based approach for ETs is more accurate as compared to the LSTM model with accuracies of 92 and 90% for the training and validation datasets, respectively. The predictions made by the ensembled LSTM are more in line with the FAO-56 PM-based method with a Pearson correlation of 0.916 as compared to LSTM models. The implementation of the proposed solution in real-time environments reveals that the proposed solution is more effective in reducing the soil salinity as compared to the traditional method.

Pan evaporation based geometrical similarity model for prediction of evapotranspiration

Pan evaporation based geometrical similarity model for prediction of evapotranspiration

An effective evapotranspiration estimation scheme based on statistical indicators for sustainable environments in humid and semi-arid area of Khyber Pakhtunkhwa, Pakistan

Abstract Reference evapotranspiration (ETo) is critical for irrigation design and water management in rainfed and irrigated agriculture. The Penman-Monteith (FAO-56(PM)) equation was demonstrated to be the most reliable and adaptive to a wide range of humid to semi-arid climates. However, it requires several environmental parameters (e.g., wind speed, solar radiation), rarely available in developing countries. Therefore, numerous temperature-based formulas have been designed to address this issue for various environments. Their calibration and validation against the local climate frequently lead to increases in performance. We revised the Hargreaves exponent (EH) and substituted a value of (0.16) for the original value (0.5). The modified Hargreaves formula enhances the ETo predictions with a mean absolute error ranging from (0.791) mm per day for Balakot to (2.36) mm per day in Risalpur, averaging (3.797) mm per day, as compared to the Hargreaves-Samani (16.827) mm per day. In general, all the selected models showed high accuracy. However, the modified Hargreaves equation appeared to give the most promising results. It ranked first in (50%) of the whole area based on the standard error of estimate for estimating ETo in Khyber Pakhtunkhwa. Additional research must be conducted to determine the study's relevance to other regions.

Methods to estimate evapotranspiration in humid and subtropical climate conditions

Selecting appropriate reference evapotranspiration (ETo) methods is crucial for managing water resources efficiently. Statistical criteria commonly used to assess the performance of empirical ETo models on a station level may produce inconsistent results, making ranking approaches a complex process. This study was conducted in India's Humid and Subtropical region, considering 11 years of mean daily data from 2009 to 2019. We evaluated thirty empirical ETo models, which were categorized into four groups based on the input parameters, namely, temperature-based (10), radiation-based (10), mass transfer-based (9), and combination model (1). The results show that the observed ETo reached maximum magnitude during Monsoon followed by Spring, Summer, Autumn, Winter, and pre-Winter season; the average observed ETo from 2009 to 2019 was ≈ 1163 mm. Among temperature-based and radiation-based models, the Hargreaves model with RMSE of 1.45 mm/day and the Turc model with RMSE of 1.01 mm/day yielded the best ETo predictions under the humid, sub-tropical climate conditions. The radiation-based models demonstrate higher accuracy in the prediction of ETo than the temperature-based and mass transfer-based models. The FAO56-PM technique, Turc model, Hargreaves model, Makkink model, and Papadakis model were ranked as the five best models among all 30 tested models. Overall, the FAO56-PM method outperformed among all 30 selected models. Thus, the exact calculation of ETo is essential for many agricultural water engineering applications, particularly in developing countries with a lack of meteorological data records and limited resources to conduct long-term in-situ observation of evapotranspiration. The methodological approach proposed in this work applies to any other location for a simple yet rigorous evaluation of evapotranspiration empirically.

Statistical Modeling to Predict Climate Change Effects on Watershed Scale Evapotranspiration

Estimation of satellite-based remotely sensed evapotranspiration (ET) as consumptive use has been an integral part of agricultural water management. However, less attention has been given to future predictions of ET at watershed-scales especially since with a changing climate, there are additional challenges to planning and management of water resources. In this paper, we used nine years of total seasonal ET derived using a satellite-based remote sensing model, Mapping Evapotranspiration at Internalized Calibration (METRIC), to develop a Random Forest machine learning model to predict watershed-scale ET into the future. This statistical model used topographic and climate variables in agricultural areas of Lower Yakima, Washington and had a prediction accuracy of 88% for the region. This model was then used to predict ET into the future with changed climatic conditions under RCP4.5 and RCP8.5 emission scenarios expected by 2050s. The model result shows increases in seasonal ET across some areas of the watershed while decreases in other areas. On average, growing seasonal ET across the watershed was estimated to increase by +5.69% under the low emission scenario (RCP4.5) and +6.95% under the high emission scenario (RCP8.5).

Evapotranspiration and Runoff Patterns Across California's Sierra Nevada

Spatially resolved annual evapotranspiration was calculated across the 14 main river basins draining into California's Central Valley, USA, using a statistical model that combined satellite greenness, gridded precipitation, and flux-tower measurements. Annual evapotranspiration across the study area averaged 529 mm. Average basin-scale annual precipitation minus evapotranspiration was in good agreement with annual runoff, with deviations in wet and dry years suggesting withdrawal or recharge of subsurface water storage. Evapotranspiration peaked at lower elevations in the colder, northern basins, and at higher elevations in the southern high-Sierra basins, closely tracking the 12.3°C mean temperature isocline. Precipitation and evapotranspiration are closely balanced across much of the study region, and small shifts in either will cause disproportionate changes in water storage and runoff. The majority of runoff was generated below the rain-snow transition in northern basins, and originated in snow-dominated elevations in the southern basins. Climate warming that increases growing season length will increase evapotranspiration and reduce runoff across all elevations in the north, but only at higher elevations in the south. Feedback mechanisms in these steep mountain basins, plus over-year subsurface storage, with their steep precipitation and temperature gradients, provide important buffering of the water balance to change. Leave-one-out cross validation revealed that the statistical model for annual evapotranspiration is sensitive to the number and distribution of measurement sites, implying that additional strategically located flux towers would improve evapotranspiration predictions. Leave-one-out with individual years was less sensitive, implying that longer records are less important. This statistical top-down modeling of evapotranspiration provides an important complement to constraining water-balance measurements with gridded precipitation and unimpaired runoff, with applications such as quantifying water balance following forest die-off, management or wildfire.

Estimating reference evapotranspiration using hybrid adaptive fuzzy inferencing coupled with heuristic algorithms

Hybrid heuristic algorithm (HA), an innovative technique in the machine learning field, enhances the accuracy of reference evapotranspiration (ETo) prediction, which is of paramount significance for regional water management, agricultural planning, and irrigation designing. However, the new hybrid HA techniques, namely Moth-Flame Optimization Algorithm (MFO) and Water Cycle Optimization Algorithm (WCA) are rarely applied to estimate ETo in the earlier literature. Therefore, this study assessed prediction and the estimation abilities of a novel hybrid adaptive neuro-fuzzy inference system (ANFIS-WCAMFO) for monthly ETo of Dhaka and Mymensing stations with data-limited humid regions of south-central Bangladesh. Prediction precision of the ANFIS-WCAMFO model is compared with other state-of-art models, i.e., ANFIS-WCA and ANFIS-MFO using a 4-fold cross-validation method including root-mean-square-error (RMSE), mean absolute error (MAE), and Nash-Sutcliffe efficiency (NSE), and coefficient of determination (R2). Nine input combinations of meteorological datasets, including extraterrestrial radiation (Ra), solar radiation (Rs), maximum and minimum temperatures (Tmax and Tmin), relative humidity (RH), and wind speed (U2), were employed for model training and testing purposes. The ANFIS-WCAMFO performed superior to the other state-of-arts methods in estimating monthly ETo in all input combinations. The ANFIS-WCA, ANFIS-MFO, and ANFIS-WCAMFO hybrid models for estimating ETo improved RMSE as 2.7%, 6.9%, and 15.1% for Dhaka station and 0.6%, 7.3%, 12.4% for Mymensingh station, respectively. The use of the Ra variable with the temperature inputs considerably improved the models’ accuracy in ETo; improvements in RMSE, MAE, NSE, and R2 of the hybrid ANFIS-WCMFO models were 26.6%, 30.9%, 17.6%, and 8.2% for Dhaka station and by 28.8%, 34.2%, 18.8% and 22.1% for Mymensingh station, respectively. The proposed hybrid neuro-fuzzy model has been suggested as a promising technique due to high predictive accuracy and less error for monthly ETo prediction in a data-limited tropical humid region.

Geo-CropSim: A Geo-spatial crop simulation modeling framework for regional scale crop yield and water use assessment

Remote sensing derived datasets (e.g. Leaf Area Index (LAI)) are increasingly being used in process based cropping system models to improve the prediction skill of the simulations when implementing operationally at regional scale. However, challenges such as inadequate quality of the available remote sensing data products and high reliance of models on climate variables and their uncertainties still exist. To address these challenges, we developed Geo-CropSim, a spatial modeling framework to use high quality remote sensing products in the Environmental Policy Integrated Climate (EPIC) agroecosystem model to regulate simulated processes and improve predictions of crop yield and evapotranspiration. Geo-CropSim comprises three main features 1) pixel level model initialization using crop emergence dates; 2) ability of the EPIC model to read in the PROSAIL (i.e. combined PROSPECT leaf optical properties model and SAIL canopy bidirectional reflectance model) inversion-based crop type LAI; and 3) a stress adjustment function to regulate simulated stress using LAI anomalies. To understand its performance, we implemented it over the State of Nebraska to estimate corn (Zea mays L.) and soybean (Glycine max [Merr.]) yields and evapotranspiration (ET) for 2012 (drought year) and 2015 (normal year) at 500-m resolution. Results showed that emergence dates and seasonal LAI captured spatial and temporal differences in crop progression (e.g. delayed planting in 2015) and growth (e.g. declined LAI in 2012) driven by regional differences in crop management and weather conditions very well. These differences were reflected in Geo-CropSim yield estimates, and showed improved spatial and temporal details compared over those from EPIC simulations obtained without using remote sensing derived emergence and LAI. Results revealed that Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE) of Geo-CropSim yield estimates, computed based on USDA-NASS reported yields, were 18.85% and 1.22 Mg ha−1 for corn, and 17.90% and 0.46 Mg ha−1 for soybeans, respectively, which are substantially lower than those of original EPIC estimates (MAPE = 33.74% and RMSE = 2.18 Mg ha−1 for corn; and MAPE = 40.71% and RMSE = 0.98 Mg ha−1 for soybeans). Further, Geo-CropSim was able to capture ET and transpiration dynamics reasonably well (e.g. 10–12 % lower values for soybeans compared to corn values), and showed good agreement with flux measurements (i.e. R2 values of 0.63 and 0.72, RMSE values of 29.88 and 33.41 mm, and MAPE values of 5.0% and 6.8% for corn and soybean, respectively). Overall, this study demonstrated that Geo-CropSim has considerable potential to serve as a reliable operational tool to assess crop yields and water use under various cropping systems and to help in regional yield monitoring and water resource management.

Hierarchical Fuzzy Systems Integrated with Particle Swarm Optimization for Daily Reference Evapotranspiration Prediction: a Novel Approach

Reference evapotranspiration (ET0) is a crucial element for deriving irrigation scheduling of major crops. Thus, precise projection of ET0 is essential for better management of scarce water resources in many parts of the globe. This study evaluates the potential of a Hierarchical Fuzzy System (HFS) optimized by Particle Swarm Optimization (PSO) algorithm (PSO-HFS) to predict daily ET0. The meteorological variables and estimated ET0 (using FAO-56 Penman–Monteith equation) were employed as inputs and outputs, respectively, for the PSO-HFS model. The prediction accuracy of PSO-HFS was compared with that of a Fuzzy Inference System (FIS), M5 Model Tree (M5Tree), and a Regression Tree (RT) model. Ranking of the models was performed using the concept of Shannon’s Entropy that accounts for a set of performance evaluation indices. Results revealed that the PSO-HFS model performed better (with Entropy weight = 0.93) than the benchmark models (Entropy weights of 0.77, 0.74, and 0.90 for the FIS, RT, and M5Tree, respectively). Furthermore, the generalization capabilities of the proposed models were evaluated using the dataset from a test station. Generalization performances revealed that the models performed equally well with the unseen test dataset and that the PSO-HFS model provided superior performance (with R = 0.93, RMSE = 0.59 mm d−1 and IOA = 0.94) while the RT model (with R = 0.82, RMSE = 0.90 mm d−1, and IOA = 0.83) exhibited the worst performance for the test dataset. The overall results imply that the PSO-HFS model could effectively be utilized to model ET0 quite efficiently and accurately.

A Deep Neural Network Architecture to Model Reference Evapotranspiration Using a Single Input Meteorological Parameter

Hydro-agrological research considers the reference evapotranspiration (ETo), driven by meteorological variables, crucial for achieving precise irrigation in precision agriculture. ETo modelling based on a single meteorological parameter would be beneficial in places where the collection of climatic parameters is challenging. The aim of this research is to develop a deep neural network (DNN) architecture that predicts daily ETo with a single input parameter selected based on the feature importance (FI) score generated by the machine learning techniques, random forest (RF), and extreme gradient boosting (XGBoost). This study also investigated the potential of SHapley Additive exPlanations to interpret and validate the outcomes of the feature selection methods by assessing the contributions of each feature to the ETo prediction. These methods recommended solar radiation as a significant parameter in the datasets of three California Irrigation Management System (CIMIS) weather stations located in distinct ETo zones. Three ETo models (DNN-Ret, XGB-Ret, and RF-Ret) were built using solar radiation as the sole input, and CIMIS ETo as the output. The performance evaluation of the developed models proved that DNN-Ret outperformed XGB-Ret and RF-Ret regardless of the dataset, with coefficients of determination (R2) ranging from 0.914 to 0.954 in the local scenario, with an average decrease of 8–9.5% in mean absolute error and root mean squared error, and an improvement of 2.6–2.9% in Nash–Sutcliffe efficiency and 1.7–2% increase in R2. The overall result analysis highlighted the efficiency of DNN-Ret in the single input parameter based ETo modelling in diverse climatic zones.

Reference Evapotranspiration Prediction from Limited Climatic Variables Using Support Vector Machines and Gaussian Processes

Climatic variables collected from weather stations evenly distributed in all regions of Turkey were used to study the potential of Gaussian Process Regression (GPR) and Support Vector Regression (SVR) in predicting reference evapotranspiration (ET0). The variables used as input features for the GPR and SVR models were solar radiation, mean temperature, wind speed, relative humidity, and month of the year. The corresponding ET0 values were calculated using the Food and Agriculture Organization recommended equation FAO 56 PM using climatic measurements collected from the same stations. Results show that regression models with high accuracies are possible using GPR and SVR models. The most effective input variable for ET0 prediction was found to be solar radiation. Relative humidity had the lowest impact on model accuracies.

A Framework for Actual Evapotranspiration Assessment and Projection Based on Meteorological, Vegetation and Hydrological Remote Sensing Products

As the most direct indicator of drought, the dynamic assessment and prediction of actual evapotranspiration (AET) is crucial to regional water resources management. This research aims to develop a framework for the regional AET evaluation and prediction based on multiple machine learning methods and multi-source remote sensing data, which combines Boruta algorithm, Random Forest (RF), and Support Vector Regression (SVR) models, employing datasets from CRU, GLDAS, MODIS, GRACE (-FO), and CMIP6, covering meteorological, vegetation, and hydrological variables. To verify the framework, it is applied to grids of South America (SA) as a case. The results meticulously demonstrate the tendency of AET and identify the decisive role of T, P, and NDVI on AET in SA. Regarding the projection, RF has better performance in different input strategies in SA. According to the accuracy of RF and SVR on the pixel scale, the AET prediction dataset is generated by integrating the optimal results of the two models. By using multiple parameter inputs and two models to jointly obtain the optimal output, the results become more reasonable and accurate. The framework can systematically and comprehensively evaluate and forecast AET; although prediction products generated in SA cannot calibrate relevant parameters, it provides a quite valuable reference for regional drought warning and water allocating.

Prediction model for daily reference crop evapotranspiration based on hybrid algorithm and principal components analysis in Southwest China

Accurately estimating reference crop evapotranspiration (ETo) is crucial for water resource management and precise irrigation. The FAO-56 Penman–Monteith model has the highest accuracy in terms of ETo prediction, but its application is restricted by a lack of complete meteorological data. To develop a high accuracy prediction model for ETo with incomplete meteorological data, this study used four principal component analysis methods (i.e. path analysis, stepwise linear regression analysis, logistic regression analysis and factor analysis) to determine few factors that strongly affect ETo as the input to established the ETo model with machine learning (ML). The prediction model for ETo was established using support vector machine (SVM), gradient boosting decision tree (GBDT), particle swarm optimisation (PSO)- SVM, and PSO-GBDT algorithms. This study demonstrated that SVM, GBDT, PSO-SVM and PSO-GBDT models can maintain the highest accuracy by using the five most important affect factors extracted on the basis of path analysis (PA) as input, for various models, the precision ranges of NSE, R2, RMSE, MAE, and GPI were approximately 0.53–0.70, 0.63–0.84, 0.27–0.47, 0.27–0.69, and 0.72–1.01 respectively. The accuracy of SVM and GBDT significantly improved through optimisation using PSO. When inputting the top five factors of importance extracted by PA algorithm in each model, the SVM and GBDT model optimized by PSO algorithm increased the average value of GPI by 17.74% and 46.71% respectively. When inputting the top five factors extracted by the PA algorithm in PSO-GBDT, the average value of NSE, R2, RMSE and MAE is 22.37%, 2.83%, 9.05% and 6.84% higher than that of PSO-SVM, and the GPI is 8.30% higher than that of PSO-SVM. This study shows that the ETo prediction model constructed by the PSO-GBDT algorithm can obtain the highest accuracy when the top five factors extracted by the PA algorithm are used as input. This provides a reference for ETo estimation in Southwest China.

DNN-MET: A deep neural networks method to integrate satellite-derived evapotranspiration products, eddy covariance observations and ancillary information

Accurate estimates of the spatiotemporal distribution of evapotranspiration (ET) are essential for understanding terrestrial energy, carbon and water cycles. Station-based observations are limited for their spatial coverage whereas satellite-derived ET products exhibit large discrepancies and uncertainties. Here we presented a Deep Neural Networks based Merging ET (DNN-MET) framework that combines information from satellite-derived ET products, eddy covariance (EC) observations and ancillary surface properties to improve the representation of the spatiotemporal distribution of ET, especially in data-sparse regions. DNN-MET was implemented over the Heihe River Basin (HRB) from 2008 to 2015, and the performance of DNN-MET and eight input state-of-the-art satellite-derived ET products (i.e., MOD16, ET-SEMI, ET-JPL, ET-MS, ET-HF, GLEAM, ETMonitor and EB-ET) was evaluated against observations from 19 EC flux tower sites. The results showed that DNN-MET improved ET estimates over HRB, and decreased the RMSE by 0.13 to 1.02 mm/day (14%-56%) when compared with eight products. DNN-MET also yielded superior performance compared to the products derived by other merging methods (i.e., Random Forest, Bayesian model averaging and a simple averaging method). When DNN-MET was validated for data-scarce regions, its performance remained better even when the training samples were decreased to 20% of the available EC sites. An innovation of our approach is by building a multivariate merging model with ancillary surface properties, DNN-MET incorporated geographical proximity effects and spatial autocorrelations into merging procedure, which can be used as a “spatial knowledge engine” to improve ET predictions. The approach can be readily and effectively applied elsewhere to improve the spatiotemporal representation of various hydrometeorological variables.

Developing a method for integrating canopy measurements into evapotranspiration predictions

The availability of affordable instruments that measure surface radiative temperature and canopy cover have made it possible, in principle, to extend and improve existing methods of estimating potential evapotranspiration (ET) to predictions where assumptions of ‘full cover’ and ‘well-watered’ are violated. The aim of this study was to develop a model for ET whose input data requirements were not large, and would make use of surface temperature and cover measurements to predict ET in both potential and sub-potential conditions. A randomised experiment was conducted with two cultivars of barley (‘Dash’ and ‘Omaka’, erect and prostrate leaved, respectively) in combination with three irrigation treatments (High, Mid and Low) in a rain-out shelter. The treatments produced a wide range of surface temperature, cover and ET. Water fluxes in the soil were measured by methods sensitive enough to detect changes on the timescale of hours. Meteorological variables, short wave radiation albedo, ground heat flux density and surface temperature were logged on a sub-hour timescale. The Priestley and Taylor potential ET equation is modified and additional equations are derived to allow the inclusion of a crop coefficient to account for incomplete cover and to vary the alpha coefficient based on surface temperature to account for soil drying. Equations were also derived for estimating short-wave radiation, albedo and soil heat flux from canopy cover and meteorological variables. The resulting equations were able to quantify a wide range of ET values measured by detailed water balance in barley crops that often had incomplete canopy cover and encountered differing degrees of water stress. We conclude the approach presented has promise but requires broader testing. The successful integration of canopy measurements with an ET model will enable the production of ET maps which will have broad applications in research and irrigation management applications.

Daily scale evapotranspiration prediction over the coastal region of southwest Bangladesh: new development of artificial intelligence model

Daily scale evapotranspiration prediction over the coastal region of southwest Bangladesh: new development of artificial intelligence model

Predictions of soybean harvest index evolution and evapotranspiration using STICS crop model

AbstractIn the last century, soybean [Glycine max (L.) Merr.] genetic improvements have resulted in increased yield partly due to an increase in harvest index (HI). To account for these genetic improvements, an update of the soybean calibration of the STICS soil‐crop model was carried out. The model was calibrated and evaluated for two sets of soybean plant parameters using datasets from the Ottawa region (ON, Canada); a low HI cultivar calibrated using datasets (1993–2008) of cultivars of maturity groups (MGs) 00 and 0 and a high HI cultivar calibrated with more recent datasets (2016–2017) of cultivars of MGs 0 and I. The model succeeded in reproducing the HI increase. Leaf area index (LAI), shoot biomass, and yield were also well predicted for the high HI cultivars with a normalized root mean square error (NRMSE) of 34, 10, and 14%, respectively, which was a great improvement compared to the default parametrization proposed in STICS for soybean. Under rainfed conditions, accurate simulation of evapotranspiration is a critical point to achieve good model performance. A comparison of the two crop evapotranspiration approaches available in STICS was also carried out. It showed that the resistive approach (NRMSE of 36%) was more efficient than the crop coefficient approach (NRMSE of 67%). This good performance of the model in predicting evapotranspiration allowed the model to perform equally well under water stress and non‐water stressed conditions. The model could therefore be used in future studies to simulate the impact of water stress on soybean growth in eastern Canada.

Forecasting evapotranspiration in different climates using ensembles of recurrent neural networks

Accurate ahead evapotranspiration forecasting is crucial for irrigation planning, for wetlands, agricultural and forest habitats preservation, and for water resource management. Deep learning algorithms can be used to develop effective forecasting models of ahead evapotranspiration. In this study, three Recurrent Neural Network-based models were built for the prediction of short term ahead actual evapotranspiration. Two variants of each model were developed changing the employed algorithm, selecting between long short-term memory (LSTM) and nonlinear autoregressive network with exogenous inputs (NARX), while the modeling was performed in the context of an ensemble approach. The prediction models were trained and tested using data from two sites with different climates: Cypress Swamp, southern Florida, and Kobeh Valley, central Nevada. With reference to the subtropical climatic conditions of South Florida, LSTM models proved to be more accurate than NARX models, while some exogenous variables such as sensible heat flux and relative humidity did not affect the results significantly. An increase of the forecast horizon from 1 to 7 days resulted in a slight reduction in the accuracy of both the LSTM- and NARX based models. Considering instead the semi-arid climate of Central Nevada, NARX models generally provided more accurate results, which were only slightly affected by relative humidity, sensible heat flux, and forecast horizon. On the other hand, LSTM models performance decayed if sensible heat flux and relative humidity were neglected, and if the forecast horizon was increased from 1 to 7 days. Deep learning-based models can provide very accurate predictions of actual evapotranspiration, but the performance of the models can be significantly affected by local climatic conditions.

Evaluation of prediction and forecasting models for evapotranspiration of agricultural lands in the Midwest U.S

Evapotranspiration (ET) prediction and forecasting play a vital role in improving water use in agriculturally intensive areas. Metrological and biophysical predictors that drive ET in managed landscapes have complex nonlinear relationships. Deep learning and data-driven methods have shown promising performance for identifying the dependencies among variables. Here, we evaluated the potentials of random forest (RF) and long short-term memory (LSTM) neural networks to estimate and forecast daily ET for corn, soybeans, and potatoes in diverse agricultural farms during 2003–2019. The modeling framework was applied for nineteen fields where eddy covariance ET and meteorological observations in the Midwest USA for growing season (April-October) is available. In this study, we applied data-driven models (RF and LSTM) with 3 sets of predictors (5, 11, and 16 predictors). Results show that a 16 predictor RF model (RF_16 R2 = 0.7, Willmott’s skill score = 0.90) outperformed a process-based land surface model (LSM R2 = 0.57, Willmott’s skill score = 0.86) for predicting daily ET, while LSTM performance was lower (LSTM_16 R2 = 0.65, Willmott’s skill score = 0.89 and LSTM_11 R2 = 0.62, Willmott’s skill score = 0.86) than RF using the same sets of predictors. Vapor pressure and crop coefficients were identified as the most important predictors for irrigated crops, while short wave radiation and enhanced vegetation index were key predictors for non-irrigated crops. For certain crop types, such as corn and soybeans on fine-grained soils (silt loam), a simpler version RF, using only 11 drivers, can provide comparable results (R2 = 0.70 vs 0.69 and Willmott’s skill score = 0.90 vs 0.88). For short-term 3-day ET forecasting, LSTM is more sensitive to uncertainty in ensemble forecast meteorology than RF. ET forecasts were strongly sensitive to forecast uncertainty of vapor pressure. The proposed modeling architecture provides a field-scale, locally calibrated tool for accurate prediction and short-term forecasting of daily ET in areas where in situ ET, metrological, and biophysical data are lacking.