Advancing Satellite Precipitation Retrievals With Data Driven Approaches: Is Black Box Model Explainable?

Abstract

Satellite‐based precipitation retrieval is an essential and long‐standing scientific problem. With an increase of observational satellite data, the advances of data‐driven approaches such as machine learning (ML)/deep learning (DL) are favored to deal with large data sets and potentially improve the accuracy of precipitation estimates. In this study, we took advantage of new technologies by wrapping up a ML/DL‐based model pipeline (LinkNet segmentation + tree ensemble). This approach is applied to the Advanced Microwave Sounding Unit (AMSU) on National Oceanic and Atmospheric Administration 18 and 19 flight, and compared with the MultiRadar MultiSensor. Four simulations were configured to examine the performance gain by incorporating three components: (1) precipitation identification, (2) nonlocal features, and (3) precipitation classification. More importantly, we examined the interpretability of the “black box” model to get a better understanding of the underlying physical connections. First, the results by this model pipeline suggest the advantages of the ML model by reducing the systematic error and instantaneous error to a factor of two. Second, identifying precipitation pixels helps to reduce the systematic error by 130%, and predicting precipitation classification benefits improved correlations by 32%. Last, channels at higher frequencies (beyond 150 GHz) are favored to identify precipitation regions, and also channels at 89 and 150 GHz are ranked as the two most important features to precipitation retrieval. This study explores the potentials of AMSU precipitation estimates with ML algorithms and provides means of interpreting the models to facilitate the better prediction of precipitation.

Ground Validation and Error Sources Identification for GPM IMERG Product over the Southeast Coastal Regions of China

“…However, SSMIS demonstrates large random errors, while GMI is generally reliable in the IMERG study of Tan et al in the mid-Atlantic region of the United States [10]. In this research, SSMIS performs well on precipitation detectability, which might be due to sensitivity to precipitation in high frequency channels [27], and GMI stands out from other PMW sensors with high detectability over For the second level of assessment, coastal areas (<100 km from the sea) had significant random errors, as the RMSEs for four IMERG estimates are above the regression line, as shown in Figure 11b. As for the CC score in Figure 12b, the precipitation estimations in inland places suffered from a smaller CC score than coastal areas.…”

“…However, SSMIS demonstrates large random errors, while GMI is generally reliable in the IMERG study of Tan et al in the mid-Atlantic region of the United States [10]. In this research, SSMIS performs well on precipitation detectability, which might be due to sensitivity to precipitation in high frequency channels [27], and GMI stands out from other PMW sensors with high detectability over water bodies. Based on the distinctive characteristics of each PMW sensors, except for the algorithm update of precipitation retrieval from PMW satellites, the improvement of the merged PMW estimate can exploit a conditional merging method with different trust weights for each sensor, according to their performance in each landscape, to assimilate a high quality merged PMW product.…”

Evaluation of Soil Moisture-Based Satellite Precipitation Products over Semi-Arid Climatic Region

“…Although the performance of the selected soil moisture-based precipitation is better on the monthly scale and can be employed for hydro-climatic applications, the daily estimations were very poor against the gauge-daily estimations. Therefore, the algorithm retrievals of SPPs should be enhanced by applying advanced techniques and models [9,43,44] (data-driven approaches such as machine learning/deep learning, downscaling of precipitation products, bias correction of SPPs) for more efficient utilization of satellite data. Moreover, sub-daily data were not available in the research, and the lowest temporal scale used in this study to examine the performances of four SPPs was daily resolution.…”

Mapping the representativeness of precipitation measurements in Mainland China

“…Since the GPM-era IMERG product started in June 2014, we choose the full years of 2015-2020 as our study period. Here, 0.1 mm h −1 is considered the threshold of precipitation to avoid large uncertainties in light precipitation (Li et al 2020, Tapiador et al 2020.…”