Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Doctorat → États-Unis → 2022 → Advancing Urban Landscape Irrigation Management using Smart Controllers and Machine Learning-Based Models

University of California Riverside (2022)

Advancing Urban Landscape Irrigation Management using Smart Controllers and Machine Learning-Based Models

Singh, Amninder

Titre : Advancing Urban Landscape Irrigation Management using Smart Controllers and Machine Learning-Based Models

Auteur : Singh, Amninder

Université de soutenance : University of California Riverside

Grade : Doctor Philosophy (PhD) in Environmental Sciences 2022

Résumé
Efficient urban landscape irrigation management is critical in California and depends on the reliable estimation of soil hydraulic properties and reference evapotranspiration (ETo). Direct measurements of these properties are time-consuming, challenging, and often expensive. Therefore, it is desirable to estimate indirectly using readily available data. The ET-based smart irrigation controllers used for landscape irrigation often rely on temperature-based ETo models ; thus, a comprehensive evaluation of these models across climate regions is required in California. Furthermore, studies are needed to evaluate the response of turfgrass to soil moisture sensor (SMS) based deficit irrigation treatments and assess the efficacy of smart controller for autonomous irrigation scheduling in semi-arid conditions of California.This dissertation addresses the first two challenges by developing artificial neural network (ANN) based models for accurate estimations of soil hydraulic properties and ETo. For the first time, we utilized an international high resolution dataset measure by evaporation methods using HYPROPTM (Hydraulic Property Analyzer, Meter Group Inc., USA) to develop the pseudo-continuous neural network PTF (PCNN-PTF) models. We assessed the accuracy and reliability of the PCNN-PTF approach for estimating the soil water retention curve (SWRC) and soil hydraulic conductivity curve (SHCC) The best performing PCNN-PTF shoed root mean square error (RMSE) of 0.043 cm3 cm−3 for SWRC, and RMSE of 0.520 for SHCC estimation. The subsequent study evaluated eight temperature-based empirical ETo models and four ANN models for ETo estimation in California. A total of 101 active California Irrigation Management Information System (CIMIS) weather stations were selected for this study, with more than 725,000 data points expanding from 1985 to 2019. The ANN model outperformed the widely used Hargreaves and Samani (HSa) model using the same input variables (i.e., air temperature and extraterrestrial solar radiation) with 11% lesser RMSE. Lastly, a three-year (2019–20121) irrigation research trial was conducted to evaluate the response of bermudagrass to soil moisture sensor (SMS) based deficit irrigation treatments and assess the smart controller for autonomous irrigation scheduling using recycled water. By the end of the research period, turfgrass quality was below the acceptable NDVI of 0.5, suggesting that bermudagrass generally does not perform well when deficit-irrigated with recycled water in a long-term basis in semi-arid climate. Further investigation is needed to substantiate SMS-based autonomous deficit irrigation scheduling when recycled water is used.

Présentation et version intégrale

Page publiée le 24 décembre 2022