Mapping Walnut Water Stress with High Resolution Multispectral UAV Imagery and Machine Learning

TL;DR

This study demonstrates that combining high-resolution UAV multispectral imagery with weather data and Random Forest learning enables accurate, scalable mapping of walnut water stress. By extracting vegetation indices such as NDVI, NDRE, and PSRI from UAV data and integrating wind, temperature, and VPD, the authors built regression and classification models that predict stem water potential (SWP) and stress levels across an orchard. Across four flight dates, the full RF regression using all dates and weather achieved $R^2$ around $0.65$ with the NoRedEdge variants performing worse, while the full RF classification reached about $85\%$ accuracy and $AUC\approx 0.86$, indicating strong practical utility for precision irrigation. The work highlights the value of red-edge indices, the importance of weather data for cross-date alignment, and the potential to replace labor-intensive SWP measurements with data-driven, plant-level irrigation guidance.

Abstract

Effective monitoring of walnut water status and stress level across the whole orchard is an essential step towards precision irrigation management of walnuts, a significant crop in California. This study presents a machine learning approach using Random Forest (RF) models to map stem water potential (SWP) by integrating high-resolution multispectral remote sensing imagery from Unmanned Aerial Vehicle (UAV) flights with weather data. From 2017 to 2018, five flights of an UAV equipped with a seven-band multispectral camera were conducted over a commercial walnut orchard, paired with concurrent ground measurements of sampled walnut plants. The RF regression model, utilizing vegetation indices derived from orthomosaiced UAV imagery and weather data, effectively estimated ground-measured SWPs, achieving an $R^2$ of 0.63 and a mean absolute error (MAE) of 0.80 bars. The integration of weather data was particularly crucial for consolidating data across various flight dates. Significant variables for SWP estimation included wind speed and vegetation indices such as NDVI, NDRE, and PSRI.A reduced RF model excluding red-edge indices of NDRE and PSRI, demonstrated slightly reduced accuracy ($R^2$ = 0.54). Additionally, the RF classification model predicted water stress levels in walnut trees with 85% accuracy, surpassing the 80% accuracy of the reduced classification model. The results affirm the efficacy of UAV-based multispectral imaging combined with machine learning, incorporating thermal data, NDVI, red-edge indices, and weather data, in walnut water stress estimation and assessment. This methodology offers a scalable, cost-effective tool for data-driven precision irrigation management at an individual plant level in walnut orchards.

Figures (22)

• Figure S1: The study area (inset) and experimental layout of a mature commercial walnut orchard near UC Davis, California, are illustrated in the false color UAV imagery taken on July 09, 2018 (the upper plot). The orchard is segmented into twenty-five blocks, with their boundaries marked as orange rectangles on the UAV imagery. The locations of sampled walnut plants, where ground measurements were conducted, are indicated by red dots. The lower plot provides a detailed illustration of the irrigation treatment assigned to each block.
• Figure S2: Daily measured SWP statistics across the five irrigation treatments. Each treatment is characterized by a reduction in irrigation intensity, set at $X$ bar below the standard level, with $X$ belonging to the set $\{0, 1, 2, 3, 4\}$. This graph provides a comparative analysis of the soil water potential under varying irrigation conditions.
• Figure S3: Variations in SWP values across different flight dates. This figure illustrates the diverse patterns in SWP value distributions observed on the five flight dates. These variations can be attributed to fluctuating environmental conditions, including changes in wind, temperature, and humidity, which significantly impact the SWP readings.
• Figure S4: Comparison of UAV imagery before and after the application of the DSM and NExG non-canopy masks, illustrated in the upper and lower images, respectively.
• Figure S5: A walnut field segmented into grids, each approximately encompassing a single canopy. The upper image shows UAV imagery of the field with grids overlaid, while the lower image displays the same area after the application of the DSM and NExG non-canopy masks, illustrating the effect of these masks on canopy identification.