Modeling and Evaluating Systematic and Random Errors in Multiscale GPM IMERG Summer Precipitation Estimates Over the Sichuan Basin

Shunxian Tang,Rui Li,Jianxin He

DOI: https://doi.org/10.1109/jstars.2021.3076197

IF: 4.715

2021-01-01

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

Abstract:The Integrated Multi-satellitE Retrievals for the Global Precipitation Measurement (IMERG) is widely used in hydrological and meteorological studies, owing to its high spatial resolution and accuracy. The quantification of systematic and random errors for the IMERG estimates at different temporal resolutions is beneficial to the calibration of observation instruments and correction of estimates, especially for regions such as the Sichuan Basin where frequent geological disasters occur in the summer. At present, there are two most commonly used models, namely the additive and multiplicative models, for modeling the systematic and random errors of precipitation. However, it is unknown which model is more suitable for the IMERG summer hourly, daily, and monthly precipitation estimates. Therefore, in this study, two models' separative capability of the systematic and random errors and predictive capability in the IMERG estimates are investigated, upon the evaluation of the models' applicability. Results show that for the hourly and daily precipitation estimates, the multiplicative error model has better separative and predictive capabilities than the additive error model and is recommended to quantify the systematic and random errors. Conversely, as for the monthly precipitation estimates, the additive error model is a relatively better choice by comparing the overall performance of both models. However, it still has some weaknesses for heavy monthly precipitation, such as nonconstant fluctuations and reduced predictive capability. So, the additive model should be used with caution in analyzing the systematic and random errors of the heavy monthly precipitation.

imaging science & photographic technology,remote sensing,engineering, electrical & electronic,geography, physical

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the quantification and modeling of systematic and random errors in GPM IMERG precipitation estimates at different time resolutions (hourly, daily, and monthly) in the Sichuan Basin during summer. Specifically, the study aims to evaluate the separation and prediction capabilities of the additive error model and the multiplicative error model at these different time resolutions to determine which model is more suitable for error analysis of IMERGE summer precipitation estimates. This research is of great significance for calibrating observational instruments and correcting estimated values, especially in areas such as the Sichuan Basin, where geological disasters occur frequently in summer and higher accuracy of precipitation data is required.

### Main research content:
1. **Research background**:
- GPM IMERG precipitation estimates are widely used in hydrological and meteorological research due to their high spatial resolution and accuracy.
- The quantification of systematic and random errors is particularly important for calibrating observational instruments and correcting estimated values, especially in areas such as the Sichuan Basin where geological disasters occur frequently in summer.

2. **Research objectives**:
- Evaluate the applicability of the additive error model and the multiplicative error model in IMERGE summer precipitation estimates at different time resolutions (hourly, daily, and monthly).
- Compare the separation and prediction capabilities of the two models and determine which model is more suitable for error analysis of IMERGE summer precipitation estimates.

3. **Research methods**:
- Fit the IMERGE summer precipitation estimates using the additive error model and the multiplicative error model.
- Evaluate the separation capabilities of the models through standardized residuals (SR) and standard deviation of residuals (SD SRs).
- Use the observational data of the first four years to evaluate the separation capabilities of the models and use the data of the last year to verify the prediction capabilities of the models.

4. **Research results**:
- For hourly and daily precipitation estimates, the separation and prediction capabilities of the multiplicative error model are better than those of the additive error model.
- For monthly precipitation estimates, the overall performance of the additive error model is relatively good, but there are still some problems when dealing with large monthly precipitation amounts, such as non - constant fluctuations and decreased prediction capabilities.
- Therefore, it is recommended to use the additive error model with caution when analyzing the systematic and random errors of large monthly precipitation amounts.

### Formula explanation:
- **Additive error model**:
\[
Y_{i,j}=a + b\times X_{i,j}+\epsilon_{i,j}
\]
where \(Y_{i,j}\) is the satellite precipitation estimate, \(X_{i,j}\) is the ground reference value, and \(\epsilon_{i,j}\) is a random error with a mean of 0 and a standard deviation of \(\sigma\).

- **Multiplicative error model**:
\[
Y_{i,j}=a\times X_{i,j}^b\times e^{\epsilon_{i,j}}
\]
where \(\epsilon_{i,j}\) is a random error with a mean of 0 and a variance of \(\sigma^{2}\).

- **Standardized residuals**:
\[
SR = \sqrt{\frac{1}{n}\sum_{i,j}\left(\epsilon_{i,j}-\frac{1}{n}\sum_{i,j}\epsilon_{i,j}\right)^{2}}
\]

### Conclusion:
By comparing the performance of the additive error model and the multiplicative error model at different time resolutions, the study found that the multiplicative error model performs better in hourly and daily precipitation estimates, while the additive error model is more applicable in monthly precipitation estimates. This conclusion helps data producers and users better understand the uncertainty of precipitation estimates, thereby promoting the improvement and application of products.