Identifying Attainable-ET Benchmarks to Guide Irrigation Water Optimization in Arizona Agriculture

Laxman Bokati; Saurav Kumar; Nisarg Shah

In Arizona, where irrigation accounts for ~72% of water demand and water supplies are stressed by prolonged drought, optimizing irrigation to reduce overall water demand is essential. Yet substantial yield variability across farms makes it difficult to pinpoint the optimal water levels that reliably sustain crop production. We introduce a spatial framework that leverages integrated remote-sensing and statistical records to map field-scale Attainable Evapotranspiration (ETAT) benchmarks across Arizona, defined as the 5th percentile of current-ET among high-yielding farms under similar agro-environmental conditions.

County-level USDA Quick Stats yields and USDA Crop Sequence Boundaries (CSB) are paired with monthly Sentinel‑2 NDVI mosaics aligned to each crop’s growth-to-harvest cycle to calibrate crop-specific yield models for corn, wheat, cotton, barley, and alfalfa. Predicted yields are aggregated to field polygons, and the distribution of predicted yields within each crop–county–year group is used to classify each farm-year into Good, Average, or Poor yield cohorts. OpenET ensemble ET data (30 m) are summed to annual depth and aggregated over the same polygons to produce a multi-year, spatially explicit dataset of current-ET, which serves as a proxy for current water use. For each unique combination of precipitation regime, soil taxonomy, crop type, and annual temperature, we compute the 5th percentile of current-ET among high-performing farms and assign that value as the Attainable-ET benchmark.
The resulting benchmark maps highlight where crops already achieve sustainable yields with efficient water use, and reveal fields where current-ET substantially exceeds attainable levels, indicating opportunities for improved efficiency. The difference between current and attainable-ET provides a spatially explicit estimate of potential water savings at the field scale. It also enables regional analysis of whether Arizona’s Active Management Areas (AMAs) are approaching optimal water-use efficiency. We present results across crops, years, and AMAs using an open-source dashboard that supports policy insights. This approach offers a scalable, data-driven tool for guiding irrigation efficiency and basin-scale planning in arid agricultural regions.

Detection of Cadmium Stress in Spinach Using Fine-Scale Aerial Hyperspectral Imagery and Deep Learning

Mahdis Khorram; Saurav Kumar; Debankur Sanyal

Cadmium (Cd) contamination in irrigated vegetable systems is both a soil- and water-quality concern. Cd can arrive via contaminated irrigation water, fertilizers, and soil amendments. It may also be elevated where agricultural soils form on Cd-rich parent materials. Leafy vegetables such as spinach readily accumulate Cd in edible tissues, creating food-safety risks and complicating management in data-poor fields. This study evaluates whether fine-scale aerial hyperspectral imagery combined with deep learning can detect Cd stress in spinach canopies under experimental field conditions. Spinach was grown in 36 Cd-treated and 6 control plots, and an unmanned aerial vehicle carrying a HySpex Mjolnir sensor (400–2500 nm) was flown at four growth stages to acquire high-resolution hyperspectral data. Image data are processed through a workflow that includes radiometric and geometric corrections, fractional-order spectral derivatives, multi-dimensional vegetation indices, and feature selection. Deep learning and other machine-learning models are trained to predict plot-level Cd concentrations from destructive measurements at harvest and to rank plots by relative Cd stress. Preliminary results show that fractional-order derivatives enhance spectral separability between Cd treatments across key wavelength regions, improving model sensitivity to subtle canopy-level stress. The resulting models are intended to enable spatial mapping of Cd stress in irrigated spinach and provide a transferable framework for monitoring heavy-metal contamination in other leafy vegetables. This work demonstrates the potential of fine-scale aerial hyperspectral sensing and deep learning to support early detection of contaminant stress in crops and to inform precision management in irrigated agricultural systems.

A Multi-Temporal Remote Sensing Analysis of Land Cover and Geomorphological Change in Ecuador’s Dulcepamba Basin

Jorge Espinosa; Rachel Conrad; John Ramirez-Avila

Hydroelectric projects are often promoted as clean energy solutions, but their long-term impacts on rivers and nearby communities are not well understood. This study examines land cover and river changes in Ecuador’s Dulcepamba River basin after the construction of the San José del Tambo small-scale hydroelectric plant. Using satellite images from 2001 to 2020 and supervised classification, we tracked changes in land use and channel form, with special attention to the 2015 flood event. Results show the river shifted from a stable condition to persistent instability, marked by channel widening, sediment buildup, and loss of natural and populated areas. After the flood, river surface area grew by 113%, while populated areas declined by 9.6%. The classification method showed high accuracy (Kappa = 0.89), confirming reliable results. This is the first long-term, multi-temporal assessment of a small hydropower project in an Andean basin, linking physical river changes to social vulnerability. Findings challenge the idea that small hydropower has minimal impact and highlight the need for environmental reviews that include geomorphological monitoring and community involvement to ensure sustainable energy development.