Assessment of the Relationship between NDVI and EVI and Leaf Area Index in Commercial Wheat Fields Using Sentinel-2

Resumen

Objective: To evaluate the relationship between the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI), derived from satellite imagery, and the Leaf Area Index (LAI) in wheat crops, in order to develop an LAI estimation model and validate its performance. Design/methodology/approach: The study was conducted in 16 commercial wheat fields in the Yaqui Valley, Sonora, using an observational, non-experimental, multilevel design. Eight fields were sown with the durum wheat variety CIRNO C2008, and eight with the bread wheat variety Borlaug 100. LAI was measured during the booting and grain-filling stages using an AccuPAR LP-80 ceptometer. NDVI and EVI were obtained from Sentinel-2 imagery using the VICAL tool. Measurements were matched using a ±5-day time window. Data were analyzed through linear regression, Random Forest models, and linear mixed models, with performance assessed by cross-validation, R², and RMSE (Root Mean Square Error).

Results: NDVI and EVI exhibited positive relationships with LAI, with moderate predictive capacity (R² = 0.33-0.41; RMSE = 1.18-1.35), and comparable accuracy between linear and machine learning models. The linear mixed models revealed that EVI explained LAI variability more consistently at the population level. Phenological stage had a significant effect, whereas variety did not exert a relevant influence.

Limitations on study/implications: The study was constrained by the limited number of LAI observations per field and by heterogeneity in planting dates, agronomic management, and environmental conditions. Findings/conclusions: According to the linear mixed modeling approach, EVI showed greater consistency in explaining LAI variability at the average population level.