Uncovering the importance of spatiotemporal resolution in satellite-based rice yield estimation using a simple but effective proxy

• High resolution NIRvP accum distinguishes rice yields of neighboring fields. • Low resolution can underestimate rice yields beyond inter-annual differences. • Yield estimation needs high resolution to capture VI magnitude and peak timing. • Enhanced spatial detail reduces mixed pixel effects in small farms. Accurate crop yield mapping is essential for assessing climate change impacts on agriculture and identifying yield gaps. While high spatiotemporal resolution satellite products such as Planet Fusion (PF) with daily 3 m resolution imagery, offer new opportunities for detailed crop monitoring, the impact of the spatiotemporal resolution of satellite data on crop yield estimation remains underexplored. In this study, we create a benchmark dataset consisting of a 3 m resolution rice yield map for a heterogeneous paddy landscape with different cultivars, using PF-based accumulated near-infrared radiation from vegetation (NIRvP accum ) between heading and harvest stages. Comparisons against plot-level rice yield measurements yield an R² of 0.76. We cross-compare yield estimates from other satellite products—MODIS, Sentinel-2, Landsat 8, and a spatial-temporal Savitzky-Golay product—against the PF-based benchmark yield data resampled to relevant coarser spatiotemporal scales. We find that, compared to PF-based yield estimation, lower spatiotemporal resolution leads to higher yield underestimation. Additionally, the downsampled PF data exhibit patterns similar to those observed in the coarser-resolution products. High-spatiotemporal resolution PF data captures peak growth stages more accurately, alleviating the mixed-pixel problem and mitigating nonlinear effects where reflectance-yield relationships deviate from linear scaling. In contrast, coarser-spatiotemporal resolution products, such as Landsat 8 has longer revisit intervals, often miss critical phenological phase transitions (e.g., peak growing season), resulting in substantial yield underestimations compared to PF. Notably, we find that yield underestimations caused by lower spatiotemporal resolutions can surpass inter-annual yield variations. These findings underscore the importance of using satellite imagery with both high spatial resolution and frequent revisits to achieve sufficiently accurate yield estimates in smallholder-dominated, heterogeneous landscapes. By highlighting the trade-offs associated with different satellite-based spatiotemporal resolutions, the study underscores the importance of considering resolution impacts on yield estimation, offering insights for optimizing Earth observation-based agricultural management, particularly in smallholder farming settings.