Precision irrigation in tropical durian cultivation requires accurate classification of meteorological variables. Existing approaches rely on complex predictions that are difficult for farmers to implement. This study develops practical classification frameworks for reference evapotranspiration (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">$\hbox {ET}_o$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mtext>ET</mtext> <mi>o</mi> </msub> </math></EquationSource> </InlineEquation>) and vapor pressure deficit (VPD) monitoring using IoT sensor technology. Four machine learning algorithms including support vector machines (SVM), K-nearest neighbors (KNN), Bayesian networks (BNG) and artificial neural networks (ANN) were evaluated using eight-month IoT data from commercial durian orchards in Chanthaburi, Thailand (October 2023–May 2024). The study implemented quantile-based <InlineEquation ID="IEq2"> <EquationSource Format="TEX">$\hbox {ET}_o$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mtext>ET</mtext> <mi>o</mi> </msub> </math></EquationSource> </InlineEquation> classification (low, medium and high) and agricultural threshold-based VPD classification (green, yellow and orange zones) across daily and hourly temporal resolutions. SVM achieved superior <InlineEquation ID="IEq3"> <EquationSource Format="TEX">$\hbox {ET}_o$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mtext>ET</mtext> <mi>o</mi> </msub> </math></EquationSource> </InlineEquation> classification with 99.25% hourly accuracy (F1-score = 0.99) while ANN excelled in VPD classification with 99.62% hourly accuracy (F1-score = 0.99). Hourly data consistently outperformed daily aggregation for real-time applications. The validated hybrid approach (SVM for <InlineEquation ID="IEq4"> <EquationSource Format="TEX">$\hbox {ET}_o$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mtext>ET</mtext> <mi>o</mi> </msub> </math></EquationSource> </InlineEquation> and ANN for VPD) successfully translates complex meteorological calculations into intuitive management categories, enabling automated irrigation systems that optimize water use efficiency while ensuring appropriate growing conditions for tropical fruit cultivation. The hybrid classification framework achieved exceptional performance with accuracies up to <InlineEquation ID="IEq5"> <EquationSource Format="TEX">$\sim $</EquationSource> <EquationSource Format="MATHML"><math> <mo>∼</mo> </math></EquationSource> </InlineEquation> 99% at hourly resolution (SVM for <InlineEquation ID="IEq6"> <EquationSource Format="TEX">$\hbox {ET}_o$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mtext>ET</mtext> <mi>o</mi> </msub> </math></EquationSource> </InlineEquation> and ANN for VPD), while daily performance remained high (up to <InlineEquation ID="IEq7"> <EquationSource Format="TEX">$\sim $</EquationSource> <EquationSource Format="MATHML"><math> <mo>∼</mo> </math></EquationSource> </InlineEquation> 98% for <InlineEquation ID="IEq8"> <EquationSource Format="TEX">$\hbox {ET}_o$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mtext>ET</mtext> <mi>o</mi> </msub> </math></EquationSource> </InlineEquation> and <InlineEquation ID="IEq9"> <EquationSource Format="TEX">$\sim $</EquationSource> <EquationSource Format="MATHML"><math> <mo>∼</mo> </math></EquationSource> </InlineEquation> 93% for VPD). These findings establish intelligent irrigation management foundations for sustainable tropical agriculture.

Statistical classifications for evapotranspiration and vapor pressure deficit monitoring using machine learning in tropical agriculture

期刊论文
发表日期：

2025年10月8日

查看全文

Pattharaporn Thongnim,
Sueppong Mueanchamnong

摘要

Precision irrigation in tropical durian cultivation requires accurate classification of meteorological variables. Existing approaches rely on complex predictions that are difficult for farmers to implement. This study develops practical classification frameworks for reference evapotranspiration ( $\hbox {ET}_o$ ${ET}_{o}$ ) and vapor pressure deficit (VPD) monitoring using IoT sensor technology. Four machine learning algorithms including support vector machines (SVM), K-nearest neighbors (KNN), Bayesian networks (BNG) and artificial neural networks (ANN) were evaluated using eight-month IoT data from commercial durian orchards in Chanthaburi, Thailand (October 2023–May 2024). The study implemented quantile-based $\hbox {ET}_o$ ${ET}_{o}$ classification (low, medium and high) and agricultural threshold-based VPD classification (green, yellow and orange zones) across daily and hourly temporal resolutions. SVM achieved superior $\hbox {ET}_o$ ${ET}_{o}$ classification with 99.25% hourly accuracy (F1-score = 0.99) while ANN excelled in VPD classification with 99.62% hourly accuracy (F1-score = 0.99). Hourly data consistently outperformed daily aggregation for real-time applications. The validated hybrid approach (SVM for $\hbox {ET}_o$ ${ET}_{o}$ and ANN for VPD) successfully translates complex meteorological calculations into intuitive management categories, enabling automated irrigation systems that optimize water use efficiency while ensuring appropriate growing conditions for tropical fruit cultivation. The hybrid classification framework achieved exceptional performance with accuracies up to $\sim $ $\sim$ 99% at hourly resolution (SVM for $\hbox {ET}_o$ ${ET}_{o}$ and ANN for VPD), while daily performance remained high (up to $\sim $ $\sim$ 98% for $\hbox {ET}_o$ ${ET}_{o}$ and $\sim $ $\sim$ 93% for VPD). These findings establish intelligent irrigation management foundations for sustainable tropical agriculture.