Combining Sentinel-2 data with an optical-trapezoid approach to infer within-field soil moisture variability and monitor agricultural production stages

Abstract

Soil moisture is an important precision agricultural variable that can be used to identify optimal growth conditions, infer vegetation stress, and maximize crop yield. However, obtaining soil moisture information with sufficient spatial and temporal resolution for such applications remains a challenge. The optical trapezoid model (OPTRAM), a shortwave infrared transformed reflectance and normalized difference vegetation index (STR-NDVI) method, has previously been used for retrieving soil moisture from optical remote sensing data. However, the capacity of OPTRAM for mapping the high-resolution spatial heterogeneity of soil moisture at individual agricultural field scales has yet to be explored. Here, we advance an approach for retrieving and quantifying the spatial heterogeneity of soil moisture for individual fields using high spatiotemporal resolution Sentinel-2 imagery. We also propose the concept of a dynamic STR-NDVI space for identification of irrigation events and crop growth stages, such as irrigation start and end dates, and dates of initial crop growth, maturity and harvest. Pre-existing OPTRAM parameterization schemes were evaluated for several crops (maize, carrot, alfalfa and Rhodes grass) and a new scheme was proposed. Results show that the original OPTRAM model can derive surface soil moisture (∼ 1 cm in depth) with acceptable coefficients of determination (R2 ≥ 0.43) and root mean square errors (RMSE ≤ 0.09 m3/m3) against ground measurements when uniform dry and wet edge parameters are applied for all crop types. The proposed scheme produced improved soil moisture retrievals with an R2 ≥ 0.65 and RMSE ≤ 0.05 m3/m3 when the specific dry and wet edge parameters were applied for each specific crop type. By analyzing time series of STR-NDVI, we demonstrate that the dynamic STR-NDVI spaces not only traces the coevolution of surface processes but also quantifies the spatial heterogeneity of soil moisture and vegetation status. Thus, the proposed dynamic STR-NDVI spaces based on the high spatiotemporal resolution Sentinel-2 imagery can both advance and extend the application of OPTRAM and improve the interpretation of soil moisture spatial heterogeneity at agricultural field scales, supporting efforts towards irrigation and crop growth monitoring in precision agriculture.