Surface soil water content plays an important role in driving the exchange of latent and sensible heat between the atmosphere and land surface through transpiration and evaporation processes, regulating key physiological processes affecting plants growth. Given the high impact of water scarcity on yields, and of irrigated agriculture on the overall withdrawal rate of freshwater, it is important to define models that help to improve water resources management for agricultural purposes, and to optimize rainfed crop yield. Recent advances in satellite-based remote sensing have led to valuable solutions to estimate soil water content based on microwave or optical/thermal-infrared data. This study aims at improving soil water content estimation at high spatial and temporal resolution, by means of the Optical Trapezoid Model (OPTRAM) driven by Copernicus Sentinel-2 data. Two different model variations were considered, based on linear and nonlinear parameters constraints, and validated against in situ soil water content measurements made with time domain reflectometry (TDR) on irrigated maize in central Italy and on rainfed maize and pasture in northern Italy. For the first site the non-linear model shows a better correlation between measured and estimated soil water content values (r = 0.80) compared to the linear model (r = 0.73). In both cases the modeled soil moisture tends to overestimate the measured values at medium to high water content level, while both models underestimate soil moisture at low water content level. Estimated versus measured normalized surface soil water for rainfed pasture plots from nonlinear OPTRAM parametrized based on irrigated maize parameterization (SIM1), and site-specific parametrization for rainfed pasture (SIM2), indicate that both models (SIM1 and SIM2) are comparable for rotational grazing pasture (RMSEsim1 = 0.0581 vs. RMSEsim2 = 0.0485 cm3 cm-3) and the continuous grazing pasture (RMSEsim1 = 0.0485 vs. RMSEsim2 = 0.0602 cm3 cm-3), while for the rainfed maize plots SIM1 shows lower RMSE (average for all plots RMSE = 0.0542 cm3 cm-3) compared to the site-specific calibration model (SIM2 – average for all plots RMSE = 0.0645 cm3 cm-3). Finally, OPTRAM estimations are close to in situ measurement values while Surface Soil Moisture at 1 km (SSM1 km) tends to underestimate the measurements during maize crop growing season. Soil moisture retrieval from high-resolution Sentinel-2 optical images allows water stress conditions to be effectively mapped, supporting decision making in irrigation scheduling and other crop management.

表层土壤水分在通过蒸腾和蒸发过程驱动大气与土地表面之间的潜热和显热交换，调节影响植物生长的关键生理过程中起着重要作用。鉴于缺水对单产和灌溉农业对淡水总体取水率的巨大影响，重要的是确定有助于改善农业水资源管理和优化雨育作物产量的模型。基于卫星的遥感技术的最新进展导致了有价值的解决方案，可基于微波或光学/热红外数据估算土壤含水量。这项研究旨在以高时空分辨率改善土壤含水量估算，通过哥白尼Sentinel-2数据驱动的光学梯形模型（OPTRAM）。基于线性和非线性参数约束，考虑了两种不同的模型变体，并针对原位在意大利中部的灌溉玉米和意大利北部的雨养玉米和牧场上，通过时域反射计（TDR）对土壤水分含量进行了测量。对于第一个站点，与线性模型（r = 0.73）相比，非线性模型显示出测得的土壤水分含量和估计的土壤水分含量值之间的相关性更好（r = 0.80）。在这两种情况下，建模的土壤水分都倾向于高估中高水分含量的测量值，而两种模型都低估了低水分含量的土壤水分。通过基于灌溉玉米参数化（SIM1）和现场特定雨水牧场参数化（SIM2）参数化的非线性OPTRAM参数化估计的雨水牧场图的标准化地面土壤水量 表示两种模型（SIM1和SIM2）在旋转放牧牧场（RMSEsim1 = 0.0581 vs. RMSEsim2 = 0.0485 cm3 cm-3）和连续放牧牧场（RMSEsim1 = 0.0485 vs. RMSEsim2 = 0.0602 cm3 cm-3）方面均相当与定点校准模型（SIM2 –所有地块RMSE的平均值= 0.0645 cm3 cm-3）相比，雨育玉米地块SIM1的RMSE较低（所有地块的平均值RMSE = 0.0542 cm3 cm-3）。最后，OPTRAM估计接近原位测量值而表层土壤水分为1公里（SSM1公里）倾向于在玉米作物生长期低估了测量。通过高分辨率的Sentinel-2光学图像检索土壤水分，可以有效地绘制水分胁迫条件，支持灌溉计划和其他作物管理中的决策。