Development of Neural Networks and Performance Appraisal of Supervised Learning Models for Predicting Organic Carbon in Soils Under Different Cropping Systems

Journal of soil science and plant nutrition, Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 19, 2024

Language: Английский

---

Comparison of principal component analysis algorithms for imputation in agrometeorological data in high dimension and reduced sample size

PLoS ONE, Journal Year: 2024, Volume and Issue: 19(12), P. e0315574 - e0315574

Published: Dec. 31, 2024

Meteorological data acquired with precision, quality, and reliability are crucial in various agronomy fields, especially studies related to reference evapotranspiration (ETo). ETo plays a fundamental role the hydrological cycle, irrigation system planning management, water demand modeling, stress monitoring, balance estimation, as well environmental studies. However, temporal records often encounter issues such missing measurements. The aim of this study was evaluate performance alternative multivariate procedures for principal component analysis (PCA), using Nonlinear Iterative Partial Least Squares (NIPALS) Expectation-Maximization (EM) algorithms, imputing time series meteorological variables. This carried out on high-dimensional reduced-sample databases, covering different percentages data. collected between 2011 2021, originated from 45 automatic weather stations São Paulo region, Brazil. They were used create daily ETo. Five scenarios (10%, 20%, 30%, 40%, 50%) simulated, which datasets randomly withdrawn base. Subsequently, imputation performed NIPALS-PCA, EM-PCA, simple mean (IM) procedures. cycle repeated 100 times, average indicators calculated. Statistical evaluation utilized following indicators: correlation coefficient (r), Mean Absolute Error (MAE), Percentage (MAPE), Square (MSE), Normalized Root (nRMSE), Willmott Index (d), index (c). In scenario 10% data, NIPALS-PCA achieved lowest MAPE (15.4%), followed by EM-PCA (17.0%), while IM recorded 24.7%. 50% there reversal, showing (19.1%), (19.9%). approaches demonstrated good results (10% ≤ nRMSE < 20%), excelling 10%, 30% scenarios, 40% scenarios. Based statistical evaluation, models proved suitable estimating PCA NIPALS EM algorithms most promise. Future research should explore effectiveness methods diverse climatic geographical contexts, develop new techniques considering spatial structure advance understanding climate prediction.

Language: Английский

---

Artificial Neural Networks Estimate Evapotranspiration for Miscanthus × Giganteus as Effectively as Empirical Model But with Fewer Inputs

Published: Jan. 1, 2023

Estimating actual evapotranspiration (ET) is particularly crucial for addressing how vegetation affects the water balance of ecosystems. ET can be estimated with empirical models, but their need many parameters and dependence on aridity make them complex less adaptable across different regions climates. In contrast, artificial neural networks (ANNs) could potentially estimate fewer more common meteorological parameters. this study, we trained two ANNs, one using a feed-forward approach (FFN) other nonlinear auto-regressive network (NARX), to predict compared commonly used model Granger Gray (GG). We our models nine-year eddy covariance (EC) dataset Miscanthus × giganteus (M. giganteus) from Illinois (UIEF), then tested out-of-sample data both UIEF location in Iowa (SABR) compare accuracy FFN, NARX, GG estimating daily ET.A combination air temperature (Ta) solar radiation (Rs) was chosen as inputs due highest R2 FFN (R2= 0.79, 0.81, 0.79 training, testing, validation, respectively) only Ta NARX 0.70 validation). examined optimum historic range prediction power found that three years best (R2 = 0.70). The predictive superior at site 0.84, 0.70, 0.83 respectively). However, generalization capability does not appear good when applied SABR because performance decreased 070, 0.60 GG). Our analysis showed ANN approaches are accurate use inputs.

Language: Английский

---

Internet of Things and Artificial Intelligence for Sustainable Agriculture: A Use Case in Citrus Orchards

Published: Oct. 12, 2023

With climate changes, the agricultural sector will soon face significant challenges due to increasing water scarcity, extreme weather conditions, and shrinking arable land. Accurate estimations of crop requirements are thus essential improve usage in agriculture. This paper provides a successful application Internet Things (IoT) Artificial Intelligence (AI) technologies for developing Smart Sustainable Agriculture. In particular, presents an example IoT system monitor predict soil contents, actual evapotranspiration other environmental variables, with objective use AI precise irrigation scheduling Mediterranean tree crops. The data collected during monitoring period is used training Machine Learning (ML) models daily $(\text{ET}_{a})$ citrus orchard regulated deficit (RDI) strategy, using different feature combinations. Results show that accuracy proposed ML remains acceptable even when number input features reduced from 10 4, making cost such systems more affordable sustainable

Language: Английский

---

Assessing the Performance of the Gaussian Process Regression Algorithm to Fill Gaps in the Time-Series of Daily Actual Evapotranspiration of Different Crops in Temperate E Continental Zones Using Ground and Remotely Sensed Data

Published: Jan. 1, 2023

The knowledge of crop evapotranspiration is crucial for several hydrological processes, including those related to the management agricultural water sources. In particular, estimations actual fluxes within fields are essential managing irrigation strategies save and preserve resources. Among indirect methods estimate evapotranspiration, ETa, eddy covariance (EC) method allows acquire continuous measurement latent heat flux (LE). However, time series EC measurements sometimes characterized by a lack data due sensors' malfunctions. At this aim, Machine Learning (ML) techniques could represent powerful tool fill possible gaps in series. paper, ML technique was applied using Gaussian Process Regression (GPR) algorithm daily evapotranspiration.The tested six different plots, two Italy, three United States America, one Canada, with crops climatic conditions order consider suitability model various contexts. For each site, climate variables were not same, therefore, performance investigated on basis available information. Initially, comparison ground reanalysis data, where both databases available, between satellite products, when have been conducted. Then, GPR tested. mean functions set considering database variables, soil status measurements, remotely sensed vegetation indices. five combinations analyzed verify approach limited input or weather replaced data. Cross-validation used assess procedure. performances assessed based statistical indicators: Root Mean Square Error (RMSE), coefficient determination (R2), Absolute (MAE), regression (b), Nash-Sutcliffe efficiency (NSE). quite high Nash Sutcliffe Efficiency (NSE) coefficient, root square error (RMSE) low values confirm proposed algorithm.

Language: Английский

---