Accounting for spatiotemporal errors of gauges: A critical step to evaluate gridded satellite precipitation products

Rain gauge observations are commonly used to evaluate the quality of satellite precipitation products. However, point-scale gauge measurements and areal satellite precipitation have inherent differences. Gauge observations are at a point of space in time accumulation, while satellite estimates are for a snapshot of time in space aggregation. Such difference has an important effect on the accuracy and precision of qualitative and quantitative evaluation results. Therefore, the objective of this study is to quantify the uncertainty of using rain gauges to evaluate the state-of-the-art satellite precipitation, the Integrated Multi-satellite Retrievals for Global Precipitation Measurement (IMERG). Gauge networks at various spatiotemporal scales (0.1°-0.8° and 1-24h) with different gauge densities are designed in the Ganjiang River basin, South China, based on observations from a highly dense gauge network that have undergone strict quality control. For comparison with the dense gauge network serving as "ground truth", 500 sparse gauge networks are generated through random combinations of gauge numbers at each set of spatiotemporal scales. Results show that all sparse gauge networks persistently underestimate the performance of IMERG according to most metrics. However, the probability of detection is overestimated because hit and miss events are more likely fewer than the reference numbers derived from dense gauge networks. A nonlinear error function of spatiotemporal scales and the number of gauges in each grid pixel is developed to estimate the errors of using gauges to evaluate satellite precipitation. Coefficients of determination of the fitting are above 0.9 for most metrics. The error function can also be used to estimate the required minimum number of gauges in each grid pixel to meet a predefined error level. This study suggests that the actual quality of satellite precipitation products could be better than conventionally evaluated or expected, and hopefully enables non-subject-matter-expert researchers to have better understanding of the explicit uncertainties when using point-scale gauge observations to evaluate areal products.