Purpose Spatial estimates of crop water use during the growing season are crucial for precision irrigation management, especially under conditions of water scarcity and climate change. The on-farm trial detailed in this paper focuses on a processing tomato field in the Sacramento Valley of California. Methodology Different meteorological parameters, including temperature, precipitation, relative humidity, shortwave radiation, and wind speed, were measured hourly at three specific locations in the field using on-the-ground sensors. Additionally, these sensors captured canopy surface reflectance properties in multiple bands, covering the wavelength range of 420–956 nm. The collected hourly meteorological and crop biophysical data were used to predict reference evapotranspiration (ETo) and calculate crop coefficients (Kc), which were then used to derive point-based estimates of canopy evapotranspiration (ETc). High-resolution multispectral and thermal aerial imagery were collected eight times during the growing season of 2021 (June–August). Several canopy properties, including leaf area index and crop height, were computed from the aerial imagery. The spatial data was combined with meteorological information from nearby weather stations to map actual evapotranspiration (ETa) with the High-Resolution Mapping of EvapoTranspiration (HRMET) model at 5 m resolution. Analysis The eight ETa maps and the daily sensor-based ETc estimates for the entire growing season were combined to predict high-resolution ETa maps for days when aerial imagery was not collected. Three different algorithms for prediction were implemented: Linear Regression, Weighted Piecewise Linear Interpolation, and Forward Interpolation. Results The performance of these algorithms was evaluated using metrics such as Nash–Sutcliffe Efficiency (NSE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Kling-Gupta Efficiency (KGE). Results indicate that Forward Interpolation is the most suitable algorithm (NSE = 0.694; MAE = 0.06; RMSE = 0.07; KGE = 0.894) for gap-filling and approximating daily high-resolution ETa maps (average hourly estimates for each day) for the entire growing season for our study site. Conclusion This innovative procedure aims to establish a decision support system for efficient irrigation scheduling, fusing different sources of spatiotemporal information to optimize irrigation in drought-prone areas.

Data fusion approach for predicting high resolution estimates of crop evapotranspiration

期刊论文
发表日期：

2025年8月22日

查看全文

Farshina Nazrul Shimim,
Mathieu Pagé Fortin,
Mallika Nocco,
Bradley Whitaker,
Andrew Gal,
Dawson Diaz,
Radomir Schmidt,
Gaurav Jha

摘要

Purpose

Spatial estimates of crop water use during the growing season are crucial for precision irrigation management, especially under conditions of water scarcity and climate change. The on-farm trial detailed in this paper focuses on a processing tomato field in the Sacramento Valley of California.

Methodology

Different meteorological parameters, including temperature, precipitation, relative humidity, shortwave radiation, and wind speed, were measured hourly at three specific locations in the field using on-the-ground sensors. Additionally, these sensors captured canopy surface reflectance properties in multiple bands, covering the wavelength range of 420–956 nm. The collected hourly meteorological and crop biophysical data were used to predict reference evapotranspiration (ET_o) and calculate crop coefficients (K_c), which were then used to derive point-based estimates of canopy evapotranspiration (ET_c). High-resolution multispectral and thermal aerial imagery were collected eight times during the growing season of 2021 (June–August). Several canopy properties, including leaf area index and crop height, were computed from the aerial imagery. The spatial data was combined with meteorological information from nearby weather stations to map actual evapotranspiration (ET_a) with the High-Resolution Mapping of EvapoTranspiration (HRMET) model at 5 m resolution.

Analysis

The eight ET_a maps and the daily sensor-based ET_c estimates for the entire growing season were combined to predict high-resolution ET_a maps for days when aerial imagery was not collected. Three different algorithms for prediction were implemented: Linear Regression, Weighted Piecewise Linear Interpolation, and Forward Interpolation.

Results

The performance of these algorithms was evaluated using metrics such as Nash–Sutcliffe Efficiency (NSE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Kling-Gupta Efficiency (KGE). Results indicate that Forward Interpolation is the most suitable algorithm (NSE = 0.694; MAE = 0.06; RMSE = 0.07; KGE = 0.894) for gap-filling and approximating daily high-resolution ET_a maps (average hourly estimates for each day) for the entire growing season for our study site.

Conclusion

This innovative procedure aims to establish a decision support system for efficient irrigation scheduling, fusing different sources of spatiotemporal information to optimize irrigation in drought-prone areas.