Vegetation indices (VIs) derived from optical sensors have been used as proxies for monitoring gross primary productivity (GPP). In contrast to satellite-based VIs, whose temporal resolution is typically limited, especially in cloudy areas, in situ derived VIs may have higher temporal resolution. This fine temporal frequency implies much larger sample sizes to test the performance of VIs by comparing with eddy covariance-based in situ GPP estimates. Here, we tested the potential of in situ measured VIs to estimate GPP in a temperate heathland ecosystem. We compared half-hourly GPP values derived from an eddy covariance CO₂ flux measurements with several greenness-, structure- and chlorophyll-sensitive VIs (e.g., the Terrestrial Chlorophyll Index [TCI] and the Inverted Red-Edge Chlorophyll Index [IRECI], among others) obtained by multiband sensors mounted at the tower. Results showed that tested VIs differed in their ability to capture the temporal variability of GPP during non-drought condition, and that all VIs failed in describing GPP during an extreme drought event. After integrating a drought indicator (e.g., soil water content, precipitation/potential evapotranspiration ratio or actual/potential evapotranspiration ratio) in the regression model, the performances of VIs drastically improved. Among all tested indices, IRECI and TCI were the most promising VIs, capturing best the temporal variation in GPP (R² = 0.73 and RMSE = 1.85, and R² = 0.77, RMSE = 1.69, respectively), provided that drought stress was properly accounted for. Our findings have implications for the development and improvement of global ecological models for drought monitoring based on proximal and remote sensing data. Our results have also a strong impact on our ability to upscale CO₂ fluxes using satellite sensors (e.g. Sentinel-2 and Sentinel-3) of heathland ecosystem characterized by heather vegetation which is relatively resistant to the structural changes in the canopy and strongly affects the interpretation of the remote sensing signals.

源自光学传感器的植被指数 (VI) 已被用作监测总初级生产力 (GPP) 的代理。与基于卫星的 VI 相比，其时间分辨率通常是有限的，尤其是在多云地区，原位衍生的 VI 可能具有更高的时间分辨率。这种精细的时间频率意味着通过与基于涡流协方差的原位GPP 估计进行比较来测试 VI 性能的样本量要大得多。在这里，我们测试了原位测量 VI 在温带荒地生态系统中估计 GPP 的潜力。我们比较了源自涡流协方差 CO _{2的半小时 GPP 值}通过安装在塔上的多波段传感器获得的几种绿色、结构和叶绿素敏感 VI（例如，陆地叶绿素指数 [TCI] 和倒置红边叶绿素指数 [IRECI] 等）的通量测量。结果表明，测试的 VI 在非干旱条件下捕捉 GPP 时间变化的能力不同，并且所有 VI 都未能描述极端干旱事件期间的 GPP。在回归模型中加入干旱指标（如土壤含水量、降水/潜在蒸散比或实际/潜在蒸散比）后，VI 的性能显着提高。在所有测试的指数中，IRECI 和 TCI 是最有希望的 VI，最好地捕捉 GPP 的时间变化（R ² = 0.73 和 RMSE = 1.85，R ²  = 0.77，RMSE = 1.69），前提是适当考虑干旱胁迫。我们的研究结果对开发和改进基于近端和遥感数据的干旱监测全球生态模型具有重要意义。我们的研究结果也对我们使用石南生态系统的卫星传感器（例如 Sentinel-2 和 Sentinel-3）提升 CO ₂通量的能力产生了强烈影响，石南植被的特征是相对抵抗冠层结构变化并强烈影响遥感信号的解释。