Non Technical Summary
The accurate estimation of high-resolution geospatial crop evapotranspiration (ETc) is crucial for precise irrigation management and development of ETc prescription maps for variable-rate irrigation systems. One favored approach involves utilizing unmanned aerial systems (UAS)-based imagery within the METRIC energy balance model to calculate daily ETc at a resolution less than 10 cm/pixel. However, previous research has revealed inaccuracies in ETc estimation, primarily due to difficulties in identifying hot and cold pixels representing dry, bare agricultural soil and fully irrigated canopies within UAS imagery. This project aims to improve high-resolution ETc estimation using UAS imagery with exploring alternative solutions for the hot and cold anchor pixels used in the model's internal calibration and develop machine learning (ML) models based on UAS-imagery for ETc estimation. The specific objectives include: 1) Establishing empirical equations for temperature estimation of fully irrigated canopies and dry, bare agricultural soil based on meteorological parameters. 2) Designing and constructing artificial hot and cold reference surfaces with controlled automated temperature for UAS data acquisition. 3) Evaluating the application of reference surfaces with automated controlled temperature for ETc estimation. 4) Developing UAS-imagery ML models for ETc estimation and investigating their performance. Field experiments will be conducted with winter canola, cover crops, grass, alfalfa, and corn in the hot and humid climate of Tennessee, as well as alfalfa and soybean in hot and dry climate of Texas to collect essential data for this study. This collaborative project strengthens research, education, and extension capabilities by fostering partnerships between minority-serving and other institutions.

Classification

Knowledge Area (KA)	Subject of Investigation (SOI)	Field of Science (FOS)	Percent
102	7210	2020	60%
111	0199	2070	40%

Knowledge Area
102 - Soil, Plant, Water, Nutrient Relationships; 111 - Conservation and Efficient Use of Water;

Subject Of Investigation
0199 - Soil and land, general; 7210 - Remote sensing equipment and technology;

Field Of Science
2070 - Meteorology and climatology; 2020 - Engineering;

Goals / Objectives
Goal 1: Determining the empirical equation for fully irrigated canopy temperature and dry bare agricultural soil that can be used for automated artificial hot and cold reference surfacesObjective 1: Developing a model to determine the relationship between the radiometric temperature of the canopy and aerodynamic temperature of the canopyObjective 2: Developing a model that relates the radiometric temperature of the fully irrigated crop to the meteorological parameters from the study site (Crop under the study are: alfalfa, soybean crops (TN, and TX) and winter canola only in TNObjective 3: Developing a model that relates the radiometric temperature of dry bare agricultural soil to the site's meteorological parameters.Goal 2: Designing/constructing controlled automated temperature artificial hot and cold reference surfaces for UAS data acquisitionObjective 1: Investigating a material with emissivity close to the bare agricultural soil with the capability of heating systems that can be used for construction of artificial hot referenceObjective 2: Investigating a material with emissivity close to the healthy vegetation with the capability of cooling systems that can be used for construction of artificial cold referenceObjective 3: Design and constructing controlled automated temperature artificial hot and cold reference surfaces with implementing the developed models for the objectives of the goal 1Goal 3: Evaluating the suitability of artificial hot and cold reference surfaces with automated controlled temperature for estimating ETc using the UAS-METRIC modelObjective 1: Evaluating the performance of constructed artificial hot and cold reference surfaces in real field conditionsObjective 2: Estimating the accuracy of artificial hot and cold reference surfaces with automated controlled temperature for estimating ETc using the UAS-METRIC modelGoal 4: Development of UAS-based imagery machine learning models for ETc estimation and investigation of their performance compared to the UAS-METRIC modelObjective 1: Developing a machine learning model for estimating ETc using the vegetation indices and hourly reference weatherObjective 2: Comparing the accuracy of the new developed ML model with UAS-METRIC model for estimating ETc

Project Methods
The following goals are set to conduct research study, provide experiential learning for grad and undergraduate students, develop tools and models that can be used for site specific irrigation management. We will use the developed materials in this project for educational purposes to the different audience such as students, extension agents during in-service training, farmers and stakeholders during the field days.Goal 1. Determining the empirical equation for fully irrigated canopy temperature and dry bare agricultural soil:Weinvestigate this range of radiometric temperature in the different research plots in Ashland City, and Nashville, TN, and in Bushland, Texas and compare it with the aerodynamic temperature of these surfaces to find out whether we can use this equation for the aerodynamic equation for the artificial cold reference surfaces or not.Research plots for data collection and demonstration will be developed with crops such as winter canola, alfalfa, grass, and soybean at the TSU research farm in Nashville, TN, in addition to a commercial irrigated soybean and corn field in Franklin County, TN. Another research plot with soybean and alfalfa will be established at the USDA-ARS in Bushland, TX. We need to install some sensors in the field to close a soil water balance and energy balance for ground truth measurement of ET at the research sites.Soil-based: We will close a soil water balance for each research experimental plot under the study in TSU, and USDA-ARS suingvarious wireless soil sensors such as TEROS 12, TEROS 21.Weather-based: One research grade standardized grass reference weather station will be installed in the field. Fescue grass will be planted under grass will station and the crop height will be maintained at 0.12 m to represent the standard reference surface.Eddy Covariance-Based: Eddy covariance stations will be placed adjacent to the weather station, directly measuring ETc of the reference grass.USDA-ARS at Bushland, Texas: Two fields (5 ha each) are considered for this experiment which will be irrigated with a Lindsay lateral move sprinkler.Lysimeter Soil-based: Each field, named NorthEast (NE) and SouthEast (SE), is equipped with a weighing lysimeter measuring 1.5 meters by 1.5 meters with a depth of 2.3 meters. The mass of the lysimeters will be continuously measured at 6-second intervals and reported as 15-minute averages.UAS-based Data Acquisition:A small UASwill be used with different onboard sensors. The multispectral camera considered for this study is a Micasense Dual camera with 10 bands and the thermal imagery sensor.Evaluation: Different methods will be followed to statistically measure the error of the generated ML models such as Nash-Sutcliffe Efficiency (NSE) Root Mean Square Error (RMSE), and Mean bias error (MBE).Goal 2. Designing/constructing controlled automated temperature artificial hot and cold reference surfaces.Artificial hot reference surface: This surface will be constructed of terra cotta ceramic plate, a material recommended in previous study that has an emissivity close to the brick and dry dark clay. To allow for adequate air circulation, this surface will be elevated above the ground by four legs. Below the surface, a heated plate and five surface-mounted thermistors will be affixed, all of which will be connected to a temperature controller. It was observed that dry bare agricultural soil exhibited temperatures ranging from 40 to 50 °C. The automated control system for the hot surface allows us to set it to the desired temperature and maintain this temperature consistently during field data acquisition. When the surface temperature falls below the desired threshold, the heated plate will elevate the surface temperature and maintain it at the specified level while UAS imagery is being collected over the agricultural field. This ensures optimal "hot" temperature conditions (40-50°C) during data collection, particularly when the reference panel requires heating.Artificial cold reference surface: The primary design of this surface comprises terra cotta ceramic plates with a shallow layer of water and supported by four legs, allowing for elevation above the ground to ensure proper air circulation around the surface. Surface-mounted thermistors are affixed beneath the surface and connected to a temperature controller and a water sprayer. Previous studies indicated that fully irrigated canopies typically maintain temperatures within the range of 27°C to 29°C. These temperatures will serve as the starting point for the algorithm to set the temperature of the surface. If the surface temperature exceeds the set temperature, the system activates a water pump that sprays water and a fan onto the surface to effectively cool the surface until the surface temperature reaches the predetermined threshold (e.g., the temperature of the fully irrigated canopy at 29°C) during UAS imagery collection over the agricultural field. In situations where the surface is cooler than the threshold, the heated plate will be activated to raise the surface temperature.Evaluation: A thermal infrared imaging sensor will be mounted on a ground vehicle at a height of 10 meters to capture images of artificial hot and cold reference surfaces and a blackbody instrument. The temperatures of the cold surfaces and blackbody will be adjusted from 15°C to 35°C in 5°C increments, with three images taken at each interval. Similarly, the temperatures of the hot surfaces and blackbody will be adjusted from 35°C to 65°C in 5°C increments, also with three images captured at each interval. The collected thermal images will be analyzed to develop an empirical calibration equation by comparing the reference surface temperatures with those of the blackbody.Goal 3. Evaluating the suitability of artificial hot and cold reference surfaces with automated controlled temperatureWe will continue to collect data during the growing season on a weekly basis from both the TSU research plots (winter canola, alfalfa/grass, soybean) and USDA-ARS research plots (soybean/alfalfa) to evaluate the functionality of these reference surfaces. During each day of UAS data acquisition, the automated controlled artificial hot and cold reference surfaces will be positioned within the field of view of the UAS imagery, as illustrated in Figure 1.C from a previous study. The steps for UAS data acquisition, image preparation, and image processing will be consistent with those described in Objective 1.Evaluation:We will evaluate the performance of ETc estimation by UAS-METRIC model when it uses the automated controlled temperature of artificial hot and cold reference surfaces for the internal calibration of the model by using the different methods described in Figure 8.Goal 4. Development of UAS-based imagery machine learning models for ETc estimationPredicting crop evapotranspiration (ET) is key to improving irrigation strategies. Traditional methods estimate ET using weather data and validate it through techniques like soil water balance, eddy covariance, and lysimeters. However, machine learning (ML) models using satellite imagery often have lower accuracy due to data quality issues like sensor noise, cloud cover, and spatial resolution limitations. To address this, UAS-based imagery from research sites in Tennessee and Texas will be used to develop an ML model based on vegetation indices and weather data collected during UAS flights. The model's accuracy will be validated against direct ET measurements from lysimeters and eddy covariance systems. Various ML techniques, including SVM, ANN, XGBoost, and CNN, will be tested to determine the most accurate model.Evaluation:The model's performance will be evaluated using statistical measures like NSE, RMSE, and MBE to assess prediction errors and biases.