ET0 Values Research Articles

Prediction of reference crop evapotranspiration based on improved convolutional neural network (CNN) and long short-term memory network (LSTM) models in Northeast China

The accurate prediction of reference crop evapotranspiration (ET0) is essential to better manage crop irrigation water consumption and improve crop water use efficiency. To effectively improve the accuracy of ET0 simulated by machine learning models, five meteorological stations in Hailaer, Harbin, Hohhot, Changchun, and Dalian were taken as representative stations, daily and monthly ET0 data from 1952 to 2020 were used, and empirical mode decomposition (EMD) and wavelet threshold denoising (WD) were considered. The convolutional neural network (CNN) and long short-term memory network (LSTM) models were improved, and two new hybrid neural network models (EMD–WD–CNN and EMD–WD–LSTM) were constructed. Using the ET0 calculated using the FAO-56 Penman–Monteith (P–M) formula as the standard value, the applicability of the improved machine learning model was evaluated. Results showed the following: i) the daily ET0-PM minimum values of five stations were close to 0, the average values were not significantly increased, and the maximum values significantly fluctuated (the fluctuations in Hailaer and Hohhot showed an upward trend, and the fluctuations in Harbin, Changchun, and Dalian showed a downward trend). The annual average monthly ET0-PM varied seasonally, with the peak in June in the Hailaer station and May in all other stations (the peak in Hohhot was the largest, and the peak in Dalian was the smallest). ii) The daily and monthly ET0 values predicted by the proposed EMD–WD–CNN and EMD–WD–LSTM models were highly consistent with the calculated results of the P–M model, showing high accuracy on the daily and monthly ET0 of the simulated five stations (daily: mean absolute error (MAE), 0.30–0.41 mm/day; root mean square error (RMSE), 0.46–0.60 mm/day; R2, 0.86–0.95; monthly: MAE, 5.66–13.71 mm/month; RMSE, 8.97–18.04 mm/month; R2, 0.91–0.95). iii) The EMD–WD–CNN model was suitable for daily scale ET0 simulation and prediction in Northeast China and monthly scale in Harbin, Changchun, and Hohhot. The EMD–WD–LSTM model was suitable for monthly ET0 simulation and prediction in Hailaer and Dalian in Northeast China. The mixed models of EMD–WD–CNN and EMD–WD–CNN can effectively improve the prediction accuracy of ET0 and can provide a new method for agricultural development and irrigation regulation in Northeast China.

Future Reference Evapotranspiration Trends in Shandong Province, China: Based on SAO-CNN-BiGRU-Attention and CMIP6

One of the primary factors in the hydrological cycle is reference evapotranspiration (ET0). The prediction of ET0 is crucial to manage irrigation water in agriculture under climate change; however, little research has been conducted on the trends of ET0 changes in Shandong Province. In this study, to estimate ET0 in the entire Shandong Province, 245 sites were chosen, and the monthly ET0 values during 1901–2020 were computed using the Hargreaves–Samani formula. A deep learning model, termed SAO-CNN-BiGRU-Attention, was utilized to forecast the monthly ET0 during 2021–2100, and the predictions were compared to two CMIP6 climate scenarios, SSP2-4.5 and SSP5-8.5. The hierarchical clustering results revealed that Shandong Province encompassed three homogeneous regions. The ET0 values of Clusters H1 and H2, which were situated in inland regions and major agricultural areas, were the highest. The SAO-CNN-BiGRU-Attention and SSP5-8.5 forecasting results generally displayed a monotonically growing trend during the forecast period in the three regions; however, the SAO-CNN-BiGRU-Attention model displayed a declining tendency at a few points. According to the SAO-CNN-BiGRU-Attention and SSP5-8.5 results, during 2091–2100, H1, H2, and H3 will reach their peaks; the SSP2-4.5 results show that H1, H2, and H3 will peak in 2031–2040. At the end of the forecast period, for Clusters H1, H2, and H3, the prediction rate of SAO-CNN-BiGRU-Attention increased by 1.31, 1.56%, and 1.80%, respectively, whereas SSP2-4.5’s prediction rate increased by 0.31%, 0.95%, and 1.57%, respectively, and SSP5-8.5’s prediction rate increased by 10.88%, 10.76%, and 10.69%, respectively. The prediction results of SAO-CNN-BiGRU-Attention were similar to those of SSP2-4.5 (R2 > 0.96). The SAO-CNN-BiGRU-Attention deep learning model can be used to forecast future ET0.

ANALYSIS OF THE CALCULATION OF REFERENCE EVAPOTRANSPIRATION ACCORDING TO THE DATA OF THE STATE METEOROLOGICAL STATION

Since direct measurement of reference evapotranspiration (ET0) is a complex, time-consuming and expensive process, the most common procedure is to estimate ET0 from climate data. The purpose of this study was to perform reference evapotranspiration calculations based on the data of the state meteorological station Askania-Nova and compare them with the actual ET0 data obtained using an automatic Internet meteorological station. The data for the study were taken from the state meteorological station Askania-Nova (township Askania-Nova, Kakhovsky district, Kherson region, 46.45°N 33.88°E) and the automatic Internet meteorological station iMetos IMT 300 from the company "Pessl Instruments", which is located at the meteorological site of the Askaniysk DSDS (Tavrychanka village, Kakhovsky district, Kherson region, 46.55°N, 33.83°E). Standard evapotranspiration was calculated using the Penman-Monteith method (FAO56-RM). To assess the accuracy of ET0 calculations, mean absolute percent error (MAPE), root mean square error (RMSE) and Standard Error of Estimate (SEE ) were determined. According to the results of the comparison of indicators from two meteorological stations, it was found that the smallest errors are inherent in the daily average and maximum temperature and relative air humidity (MAPE<10%), for the minimum temperature and relative air humidity, the MAPE errors are 18,1 and 13,7%, respectively. The MAPE error for water vapor pressure deficit and solar radiation is 20,2 and 26,3%, respectively. The largest MAPE error of 40,3% was established for wind speed measurements. The average MAPE error between the calculated ET0, based on the meteorological data of the Askania-Nova station, and the actual ET0 data obtained from the automatic Internet meteorological station iMetos is 16,8%, RMSE – 0,65 mm, SEE – 0,56 mm. Applying a coefficient of 0,92 when calculating ET0 reduces the errors of MAPE, RMSE, and SEE by 3,2%, 0,15 mm, and 0,05 mm, respectively, for all calculation periods. For the May-August period, the MAPE error was 10,7%, which brings the calculations close to high accuracy (MAPE <10%). Based on the results of the calculations, it was established that on average over the years of research, the actual ET0 was 68 mm less than the calculated one. The absolute errors of determination of ETc depended on the crop and the average over the years of research ranged from 33 mm (winter wheat) to 68 mm (early tomatoes). The application of the refined value of ET0 in calculations reduces the absolute errors in the determination of c over the years of research, this error did not exceed 6 mm (early tomato). Research results confirm the possibility of using meteorological indicators obtained from state meteorological stations to calculate ET0. To increase the accuracy of calculations, it is necessary to use a refinement coefficient.

Information Entropy-Based Hybrid Models Improve the Accuracy of Reference Evapotranspiration Forecast

Accurate forecasting of reference crop evapotranspiration (ET0) is vital for sustainable water resource management. In this study, four popularly used single models were selected to forecast ET0 values, including support vector regression, Bayesian linear regression, ridge regression, and lasso regression models, respectively. They all had advantages of low requirement of data input and good capability of data fitting. However, forecast errors inevitably existed in those forecasting models due to data noise or overfitting. In order to improve the forecast accuracy of models, hybrid models were proposed to integrate the advantages of the single models. Before the construction of hybrid models, each single model’s weight was determined based on two weight determination methods, namely, the variance reciprocal and information entropy weighting methods. To validate the accuracy of the proposed hybrid models, 1–30 d forecast data from January 2 to February 1, 2022, were used as a test set in Xinxiang, North China Plain. The results confirmed the feasibility of the information entropy-based hybrid model. In detail, the information entropy model generated the mean absolute percentage errors of 11.9% or a decrease by 48.9% compared to the single and variance reciprocal hybrid models. Moreover, the model generated a correlation coefficient of 0.90 for 1–30 d ET0 forecasting or an increase by 13.6% compared to other models. The standard deviation and the root mean square error of the information entropy model were 1.65 mm·d−1 and 0.61 mm·d−1 or had a decrease by 16.4% and 23.7%. The maximum precision and the F1 score were 0.9618 and 0.9742 for the information entropy model. It was concluded that the information entropy-based hybrid model had the best midterm (1–30 d) ET0 forecasting performance in the North China Plain.

Comparative Analysis of Climate Change Impacts on Climatic Variables and Reference Evapotranspiration in Tunisian Semi-Arid Region

Systematic biases in general circulation models (GCM) and regional climate models (RCM) impede their direct use in climate change impact research. Hence, the bias correction of GCM-RCMs outputs is a primary step in such studies. This study compares the potential of two bias correction methods (the method from the third phase of the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP3) and Detrended Quantile Matching (DQM)) applied to the raw outputs of daily data of minimum and maximum air temperatures and precipitation, in the Cap-Bon region, from eight GCM-RCM combinations. The outputs of GCM/RCM combinations were acquired from the European branch of the coordinated regional climate downscaling experiment (EURO-CORDEX) dataset for historical periods and under two representative concentration pathway (RCP4.5 and RCP8.5) scenarios. Furthermore, the best combination of bias correction/GCM-RCM was used to assess the impact of climate change on reference evapotranspiration (ET0). Numerous statistical indicators were considered to evaluate the performance of the bias correction/historical GCM-RCMs compared to the observed data. Trends of the Hargreaves–Samani_ET0 model during the historical and projected periods were determined using the TFPMK method. A comparison of the bias correction methods revealed that, for all the studied model combinations, ISIMIP3 performs better in reducing biases in monthly precipitation. However, for Tmax and Tmin, the biases are greatly removed when the DQM bias correction method is applied. In general, better results were obtained when the HadCCLM model was used. Before applying bias correction, the set of used GCM-RCMs projected reductions in precipitation for most of the months compared to the reference period (1982–2006). However, Tmin and Tmax are expected to increase in all months and for the three studied periods. Hargreaves–Samani ET0 values obtained from the best combination (DQM/ HadCCLM) show that RCP8.5 (2075–2098) will exhibit the highest annual ET0 increase compared to the RCP4.5 scenario and the other periods, with a change rate equal to 11.85% compared to the historical period. Regarding spring and summer seasons, the change rates of ET0 are expected to reach 10.44 and 18.07%, respectively, under RCP8.5 (2075–2098). This study shows that the model can be used to determine long-term trends in ET0 patterns for diverse purposes, such as water resources planning, agricultural crop management and irrigation scheduling in the Cap-Bon region.

Development of daily gridded Penman-Monteith reference crop evapotranspiration data for Karnataka State, India

ABSTRACT Reference crop evapotranspiration (ET0) estimates are often unavailable at fine spatial resolutions due to the lack of dense climatic networks. The present study was taken up to develop a methodology for obtaining a gridded Penman-Monteith (PM) ET0 dataset for Karnataka State, India, through spatial interpolation of estimates obtained at climate stations. Three interpolation techniques, Inverse Distance Weighted (IDW), Kriging, and Point-Based Sentinel Hospitals Area Distance (P-BSHADE), were employed to derive PM ET0 estimates. Considering the historical climate records of 67 stations located in Karnataka State for the period of 2006–2016, the interpolation methodologies were evaluated using a cross-validation procedure. Results showed that the performance of IDW was better with significant error reduction at validation sites. Subsequently, the IDW algorithm was used to derive daily PM ET0 values at 0.25° × 0.25° grids over the Karnataka State, India. Finally, the comparative analysis of developed PM ET0 gridded data product against three global ET0 products indicated the developed gridded ET0 to be most accurate across different agroclimatic zones. Web links of the developed product have been created in an effort to share the data on ET0, which is a critical input in a variety of studies in earth sciences.

Scaling solar-induced chlorophyll fluorescence by using VPD0.5 improves the simulation of reference crop evapotranspiration in the arid and semiarid regions of northern China

Accurately estimating reference crop evapotranspiration (ET0) is crucial for decision-making regarding irrigation, improving farmland water use efficiency, planning water resources, and simulating the global water cycle. However, because traditional ET0 estimation models often require a large number of meteorological parameters as input variables, their application potential is limited, especially when simulating arid and semiarid regions where meteorological data are lacking. Therefore, it is still very challenging to estimate ET0 at large scales in arid and semiarid areas. In this study, we first proposed a semi-mechanistic model for estimating ET0 by deriving the relationship between canopy conductance to water vapor (gw) and solar-induced chlorophyll fluorescence (SIF) based on Fick's law and the light use efficiency (LUE) model, combining theories on optimal stomatal behaviour. Second, the relationship between SIF × VPD0.5 (VPD0.5: 0.5-power of the vapor pressure deficit) and ET0 in the model was verified at the site scale. On this basis, the semi-mechanistic model was applied to three machine learning models to simulate ET0. Four statistical indicators, including the coefficient of determination (R2), root mean square error (RMSE), normalized root mean square error (NRMSE) and mean absolute error (MAE), were used to verify the model performance and screen out the optimal model. Finally, the optimal model was compared with five traditional empirical models. The results showed that (1) the relationship between ET0 and the product of SIF and the 0.5-power vapor pressure deficit (SIF × VPD0.5) was closer than the relationship between ET0 and SIF at the site scale. (2) Compared to the case in which SIF was used as the only input variable, three machine learning models performed better when SIF was scaled using VPD0.5 as the input variable, and the random forest regression (RFR) model in which SIF × VPD0.5 was the input variable was the optimal model (ET0opt). (3) The total mean value of ET0 at 22 sites simulated by ET0opt is 15.46 mm/4 days. The total mean ET0 value for the 22 sites calculated by the five traditional empirical models vary from 10.70 mm/4 days to 20.96 mm/4 days, while our results were in this range and close to their mean value. Our study provides a valuable reference basis for large-scale ET0 estimations using remote sensing in arid and semiarid areas.

Performance evaluation of simpler reference crop evapotranspiration estimation equations with and without local calibration

The present study was taken up to demonstrate an overall methodology for local calibration of simpler reference crop evapotranspiration (ET0) estimation equations relative to the physically-based, albeit more data-intensive Penman-Monteith (PM) method. Climate records for the historical period 2006-2016 of 67 stations located across 10 agro-climatic zones of Karnataka State, India, were used for this purpose. The numerical coefficients in the original forms of the simpler equations namely, Blaney-Criddle (BC), Priestley-Taylor (PT), Radiation (RAD), and Turc (TC) were treated as unknown parameters and optimized for each station using PM ET0 estimates as reference. Results of local calibration indicated significant improvement in prediction accuracies and bias reduction for all the equations with the calibrated PT equation exhibiting the best performance. Spatial maps depicting the variations in the optimal values of parameters were developed to aid the accurate estimation of monthly ET0 values in the study area using only minimal climate inputs.

ANN-based PCA to predict evapotranspiration: a case study in India

Abstract The Penman–Monteith evapotranspiration (ET) model has superior predictive ability to other methods, but it is challenging to apply in several Indian stations, owing to the need for a large number of climatic variables. The study investigated an artificial neural network (ANN) model for calculating ET for various agro-climatic regions of India. Sensitivity analysis showed that the overall average changes in ET0 values for 25% change in the climatic variables were 18, 16, 14, 7, 5, and 4%, respectively, for Tmax, RHmean, Rn, wind speed, Tmin, and sunshine hours. The dominant climatic variables were identified from the principal component analysis (PCA) and ET0 was computed using an ANN with dominant climatic variables. The ANN architecture with backpropagation technique had one hidden layer and neurons ranging from 10 to 30 for all climatic variables and from 5 to 10 for PCA variables. The new ET models were statistically compared with Penman–Monteith ET estimate, and found reliable. PCA variables guaranteed an estimate of ET0 accounting for 98% of the variability. The average values of coefficient of determination, standard error of estimate, and percentage efficiency were observed as 0.96, 0.24, and 94%, respectively.

Development of machine learning-based reference evapotranspiration model for the semi-arid region of Punjab, India

Evapotranspiration (ET) is a critical element of the hydrological cycle, and its proper assessment is essential for irrigation scheduling, agricultural and hydro-meteorological studies, and water budget estimation. It is computed for most applications as a product of reference crop evapotranspiration (ET0) and crop coefficient, notably using the well-known two-step method. Accurate predictions of reference evapotranspiration (ET0) using limited meteorological inputs are critical in data-constrained circumstances. Due to the unavailability and heterogeneity of broad parameters of the FAO PM method, it becomes a major constraint for accurately estimating ET0. To overcome the complexity of calculation, the present study was focused on developing a Random Forest-based (RF) ET0 model to estimate the crop ET for the semi-arid region of northwest India. The RF-based model was developed by focusing on the easily available data at the farm level. For comparative study existing models like Hargreaves–Samani, Modified Penman and modified Hargreaves–Samani were used to estimate the ET0. The models' calibration and validation were done using meteorological data collected from the weather station of Punjab Agricultural University for 21 years (1990–2010) and nine years (2011–2019), respectively, and the FAO PM model was taken as a standard. The mean absolute error (MAE) and root-mean-square error (RMSE) were found to be least as 0.95 mm/d and 1.32 mm/d, respectively for the developed RF model, with an r2 value of 0.92. The seasonal ET0 estimated by modified Hargreaves–Samani (MHS) and RF were found as 498.3, 482.1 mm in rabi season and 755, 744.8 mm in kharif season respectively, whereas the annual ET0 was 1380.2 and 1355.7 mm respectively. The predicted ET0 values by RF-based model were used for irrigation scheduling of two growing seasons (2020–2021) of maize and wheat crops. The outcome of the field trial also demonstrates that there was no appreciable yield drop in the crop when compared to irrigation scheduling by the FAO PM model, demonstrating the applicability of the developed model for irrigation in the semiarid region of the Punjab in India.

Modeling Daily Reference Evapotranspiration from Climate Variables: Assessment of Bagging and Boosting Regression Approaches

The increasing frequency of droughts and floods due to climate change has severely affected water resources across the globe in recent years. An optimal design for the scheduling and management of irrigation is thus urgently needed to adapt agricultural activities to the changing climate. The accurate estimation of reference crop evapotranspiration (ET0), a vital hydrological component of the water balance and crop water need, is a tiresome task if all the relevant climatic variables are unavailable. This study investigates the potential of four ensemble techniques for estimating precise values of the daily ET0 at representative stations in 10 agro-climatic zones in the state of Karnataka, India, from 1979 to 2014. The performance of these models was evaluated by using several combinations of climatic variables as inputs by using tenfold cross-validation. The outcomes indicated that predictions of ET0 by all four ensemble models based on all climatic variables were the most accurate in comparison with other input combinations. The random forest regressor was found to deliver the best performance among the four models on all measures considered (Nash–Sutcliffe efficiency, 1.0, root-mean-squared error, 0.016 mm/day, and mean absolute error, 0.011 mm/day). However, it incurred the highest computational cost, whereas the computational cost of the bagging model for linear regression was the lowest. The extreme gradient-boosting model delivered the most stable performance with a modified training dataset. The work here shows that these models can be recommended for daily ET0 estimation based on the users’ interests.

Investigating the roles of different extracted parameters from satellite images in improving the accuracy of daily reference evapotranspiration estimation

Agricultural water management, crop modeling, and irrigation scheduling are all dependent on the accurate estimation of reference evapotranspiration (ET0). A satellite image can also compensate for the lack of reliable weather information. So, in this study, stochastic gradient descent (SGD) has been implemented for optimizing multilayer perceptron (MLP) and developing SGD-MLP to estimate daily ET0 in Tabriz (semi-arid climate) and Babolsar (humid climate) stations, Iran, using extracted data from satellite images. The estimated ET0 values were compared to the determined ET0 based on the FAO-Penman–Monteith equation. Based on satellite image data collected from 2003 to 2021, the database was constructed. During the development of the abovementioned models, data from 2003 to 2016 (70%) were used for training purposes, and residual data (30%) were used for testing purposes. Additionally, the input variables, including land surface temperature (LST) day and night, normalized difference vegetation index (NDVI), leaf area index (LAI), and a fraction of photosynthetically active radiation (FPAR) from MODIS sensor, were utilized to estimate the daily ET0. Thus, there are three studied models; first is based on the LST, second on the vegetation indices, and third on the combination of the LST and the vegetation indices. Additionally, four performance indexes, including the coefficient of determination (R2), the root-mean-square error (RMSE), Willmott’s index of agreement (WI), and Nash–Sutcliffe efficiency, were utilized in order to measure the implemented model’s accuracy. According to the obtained results, the SGD-MLP-3 with input parameters of LSTday&night, LSTmean, LAI, NDVI, and FPAR gave the most accurate results with RMSE and WI values of as 0.417 mm/day, 0.973, for Tabriz and 0.754 mm/day, 0.922 for Babolsar stations, respectively. Conclusively, LST of daytime, nighttime, and average may be suggested as the most influential parameter for ET0 estimation.

Parametric calibration of Hargreaves–Samani (HS) reference evapotranspiration equation with different coefficient combinations under the humid environment

Evapotranspiration (ET) is a major hydrological component in irrigation system design, irrigation scheduling, and hydrologic and drainage design. Estimating crop ET requires accurate reference evapotranspiration (ET0) measurement. In this work, numerous modified Hargreaves models for ET0 were tested and calibrated against the standard FAO-56 Penman-Monteith eq. (F-56 PM) for further improvement in performance in the humid subtropical environment of Imphal West District, Manipur.The Martinez-Cob and Tejero-Juste (2004) equation resulted in best region-wide ET0 values of 3.536 mm d−1, mean absolute error (MAE) = 0.367 mm d−1, standard error of estimate (SEE) = 0.348 mm d−1, correlation coefficient (r) = 0.855, coefficient of determination (R2) =0.731, and index of agreement (d) = 0.891. Trajkovic (2007) equation has the second-best ET0 values of 3.383, MAE = 0.413 mm d−1, SEE = 0.3.77 mm d−1, r = 0.827, R2 = 0.685, and d = 0.852. The third-best equation, Ravazzani et al. (2012), with average ET0 values of 4.015 mm d−1, MAE = 0.506 mm d−1, SEE = 0.395 mm d−1, r = 0.855, R2 = 0.731, and d = 0.79.The observed wind speed data were sorted into five groups based on the average wind speed in different months in the region (1 m/s,2.5 m/s, 3.5 m/s, 4.5 m/s, and 5.5 m/s), and the estimated evapotranspiration was used to calibrate 14 selected combinations of the Hargreaves models. The calibration process significantly improved the performance of the different identified Hargreaves models.The analysis indicated that the improvement in estimation decreases with increasing wind speed. The revised coefficient predicted evapotranspiration is closer to F-56PM, with MAE (mm d−1) ranging from 0.125 at 1 m/s to 0.248 at 4.5 m/s and SEE(mm d−1) from 0.16 at 1 m/s to 0.245 at 5.5 m/s, as compared to the original Hargreaves equation, whose MAE is 0.543, and SEE is 0.400.More research must be conducted to extend the applicability of this equation to other regions of the State.

Modelling of Reference Evapotranspiration using Neural Network and Regression Approaches for Semi-humid Region of Sikkim

This study compared the applicability of multiple linear regression (MLR) and artificial neural network (ANN) approaches for estimating the weekly reference evapotranspiration (ET0) in a semi-humid area of Tadong in Gangtok district of Sikkim state in India. Daily meteorological data (1991 - 2020) were used for developing MLR and ANN model using various combinations of meteorological data. Predicted ET0 values were compared with the FAO-56 Penman Monteith (PM) equation estimated ET0. The coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE), and Nash Sutcliffe efficiency (NSE) were used to assess the performance of MLR and ANN models. For ANN model, one hidden layer and four neurons was found as the best ANN architecture. ANN-4 model with inputs parameters of maximum air temperature (Tmax), minimum air temperature (Tmin), maximum relative humidity (RHmax), minimum relative humidity (RHmin), and sunshine hours (SSH) had the smallest RMSE (0.609 mm.day-1 and 0.880 mm.day-1), MAE (0.473 mm.day-1 and 0.551mm.day-1) and the highest R2 (0.752 and 0.657) and NSE (0.782 and 0.696) during training and testing phases, respectively. The MLR-4 model with same combination of input parameters (Tmax, Tmin, RHmax, RHmin, SSH) resulted in smallest RMSE (0.704 mm.day-1 and 0.714 mm.day-1), MAE (0.562 mm.day-1and 0.583 mm.day-1); and the highest R2 (0.679 and 0.602) and NSE (0.708 and 0.597) during training and testing phase, respectively. Study also showed increase in model performance of ANN and MLR models with increase in number of meteorological parameters. All the ANN models gave comparable results, but ANN 4 model resulted in higher prediction accuracy for the sub-humid region of Tadong in Gangtok district of North-east India. The developed ANN-4 model could be helpful for preparing irrigation schedules and planning, and management of water resources in the data scarce northeast region of India and other identical climatic regions.

Generalized Daily Reference Evapotranspiration Models Based on a Hybrid Optimization Algorithm Tuned Fuzzy Tree Approach

Reference evapotranspiration (ET0) is an important driver in managing scarce water resources and making decisions on real-time and future irrigation scheduling. Therefore, accurate prediction of ET0 is crucial in the water resources management discipline. In this study, the prediction of ET0 was performed by employing several optimization algorithms tuned Fuzzy Inference System (FIS) and Fuzzy Tree (FT) models, for the first time, whose generalization capability was tested using data from other stations. The FISs and FTs were developed through parameters tuning using Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Pattern Search (PS), and their combinations. The FT was developed by combining several FIS objects that received ranked meteorological variables. A total of 50 FIS and FT models were developed and the model ranking was performed utilizing Shannon’s Entropy (SE). Evaluation outcomes revealed the superiority of the hybrid PSO-GA tuned Sugeno type 1 FT model (with R = 0.929, NRMSE = 0.169, accuracy = 0.999, NS = 0.856, and IOA = 0.985) over others. For evaluating the generalization capability of the best model, three different parts of datasets (all-inclusive, 1st half, and 2nd half) of the five test stations were evaluated. The proposed hybrid PSO-GA tuned Sugeno type 1 FT model performed similarly well, according to the findings, on the datasets of the test stations. The study concluded that the hybrid PSO-GA tuned Sugeno type 1 FT approach, which was composed of several standalone FIS objects, was suitable for predicting daily ET0 values.

Evaluation of artificial intelligence models for daily prediction of reference evapotranspiration using temperature, rainfall and relative humidity in a warm sub-humid environment

Accurate estimation of reference evapotranspiration is essential for agricultural management and water resources engineering applications. In the present study, the ability and precision of three artificial intelligence (AI) models (i.e., Support Vector Machines (SVMs), Adaptive Neuro-Fuzzy Inference System (ANFIS) and Categorical Boosting (CatBoost)) were assessed for estimating daily reference evapotranspiration (ET0) using limited weather data from five locations in a warm sub-humid climate in Mexico. The Penman–Monteith FAO-56 equation was used as a reference target for ET0 values. Three different input combinations were investigated, namely: temperature-based (minimum and maximum air temperature), rainfall-based (minimum air temperature, maximum air temperature and rainfall), and relative humidity-based (minimum air temperature, maximum air temperature and relative humidity). Extraterrestrial radiation values were used in all combinations. The temperature-based AI models were compared with the conventional Hargreaves–Samani (HS) model commonly used to estimate ET0 when only temperature records are available. The goodness of fit for all models was assessed in terms of the coefficient of determination (R2), Nash–Sutcliffe model efficiency coefficient (NSE), root mean square error (RMSE) and mean absolute error (MAE). The results showed that among the AI models evaluated, the SVM models outperformed ANFIS and CatBoost for modeling ET0. Further, the influence of relative humidity and rainfall on the performance of the models was investigated. The analysis indicated that relative humidity significantly improved the accuracy of the models. Finally, the results showed a better response of the temperature-based AI models over the HS method. AI models can be an adequate alternative to conventional models for ET0 modeling.

ClimaSG: A weather generator for crop modelling and water requirements studies

A new stochastic weather generator, ClimaSG, is presented and evaluated under a wide range of climates. The model requires monthly means data as inputs to generate daily weather data. Validation tests revealed that ClimaSG performed well in capturing monthly rainfall variability but showed some limitations for generating extreme temperatures. Solar radiation, air humidity, and wind speed were reproduced acceptably. The weather generator was particularly accurate in predicting mean and extreme values of ET0, which indicates that ClimaSG may be used as the basis for calculating crop water requirements and for irrigation design and planning purposes. The suitability of synthetic weather series generated with ClimaSG for crop modelling applications was further assessed by running the model OliveCan for low density rainfed (LD) and super high density irrigated (SHD) olive orchards for several locations in Spain. Yield distributions simulated with synthetic weather rarely differed significantly from those obtained with actual weather data, although the coefficient of variation (CV) was usually underestimated. Considering its low input requirements and satisfactory performance, ClimaSG represents a promising source of weather data for different practical applications.

Evaluation of Data-driven Hybrid Machine Learning Algorithms for Modelling Daily Reference Evapotranspiration

ABSTRACT Reference evapotranspiration (ET0) is one of the crucial variables used for irrigation scheduling, agricultural production, and water balance studies. This study compares six different models with sequential inclusion of six meteorological input variables such as minimum temperature (Tmin), maximum temperature (Tmax), mean relative humidity (RH), wind speed (SW), sunshine hours (HSS), and solar radiation (RS), which are necessarily used in physical or empirical-based models to estimate ET0. Each model utilized three variants of machine learning algorithms, i.e. Additive Regression (AdR), Random Subspace (RSS), M5 Pruning tree (M5P) independently and four novel permutated hybrid combinations of these algorithms. To evaluate the efficacy of these hybridizations and the stability of machine learning models, a comprehensive evaluation of independent and hybrid models was performed. With more input variables, the model performances were found to be superior in terms of prediction accuracies. The model AdR6 that included all the 6 selected meteorological variables outperformed other models during the testing period, exhibiting statistical performance of MAPE (1.30), RMSE (0.07), RAE (2.41), RRSE (3.10), and R 2 (0.998). However, the AdR algorithm, alone, was found to capture about 86% of variance in the observed data conforming to the 95% confidence band across all models irrespective of the number of input variables used to predict ET0. The RSS algorithm, in comparison to other algorithms, failed to capture the observed trends even with all the input variables. The hybrid combinations of algorithms with AdR as a constituent were better performers in terms of their prediction accuracies but remained inferior to AdR as an individual performer. All the algorithms are better predictors of the higher values of ET0 that included values beyond the 75% quartile.

DeepEvap: Deep reinforcement learning based ensemble approach for estimating reference evapotranspiration

Precision agriculture aims to increase crop yield by employing an efficient resource management scheme, such as estimating irrigation requirements. Reference evapotranspiration (ET0), defined as the process of water loss from the soil and reference plant, is one of the indispensable components on which crop irrigation requirement depends. It is mainly calculated by using empirical models. However, these models require a large climate dataset that is sometimes unavailable in data-scarce regions. The present study focuses on the estimation of ET0 values by using three climate parameters as input variables i.e., minimum temperature (Tmin), maximum temperature (Tmax), and solar radiation (Rs). Moreover, to consider the effect of time-varying characteristics of the ET0 process, deep reinforcement learning (DRL) based ensemble approach, DeepEvap, is introduced to estimate ET0 values. The whole modeling procedure of the proposed ensemble model incorporates three phases. In phase I, the data preprocessing technique is performed on the meteorological data to clean the existing impurities as it affects the performance of any machine learning (ML) based approach. In phase II, four different deep neural network-based models are used to build the estimation model of ET0 and calculate the prediction results. In phase III, the DRL approach is used to ensemble the prediction results of these four models. The meteorological dataset of two stations of India: Ludhiana and Patiala, is selected to validate the proposed approach. The results of the conducted study depict that: (a) The proposed DeepEvap approach is competitive for ET0 prediction by achieving a coefficient of determination (R2) = 0.96. It significantly outperforms four baseline models; (b) The proposed technique also integrates four deep neural network models and works better than existing ensemble approaches.

Data intelligence and hybrid metaheuristic algorithms-based estimation of reference evapotranspiration

For developing countries, scarcity of climatic data is the biggest challenge, and model development with limited meteorological input is of critical importance. In this study, five data intelligent and hybrid metaheuristic machine learning algorithms, namely additive regression (AR), AR-bagging, AR-random subspace (AR-RSS), AR-M5P, and AR-REPTree, were applied to predict monthly mean daily reference evapotranspiration (ET0). For this purpose, climatic data of two meteorological stations located in the semi-arid region of Pakistan were used from the period 1987 to 2016. The climatic dataset includes maximum and minimum temperature (Tmax, Tmin), average relative humidity (RHavg), average wind speed (Ux), and sunshine hours (n). Sensitivity analysis through regression methods was applied to determine effective input climatic parameters for ET0 modeling. The results of performed regression analysis on all input parameters proved that Tmin, RHAvg, Ux, and n were identified as the most influential input parameters at the studied station. From the results, it was revealed that all the selected models predicted ET0 at both stations with greater precision. The AR-REPTree model was located furthest and the AR-M5P model was located nearest to the observed point based on the performing indices at both the selected meteorological stations. The study concluded that under the aforementioned methodological framework, the AR-M5P model can yield higher accuracy in predicting ET0 values, as compared to other selected algorithms.