Evaluation of Satellite-Based Precipitation Products over Complex Topography in Mountainous Southwestern China

Abstract

Satellite-based precipitation products (SBPPs) are essential for rainfall quantification in areas where ground-based observation is scarce. However, the accuracy of SBPPs is greatly influenced by complex topography. This study evaluates the performance of Integrated Multi-satellite Retrievals for GPM (IMERG) and Global Satellite Mapping of Precipitation (GSMaP) in characterizing rainfall in a mountainous catchment of southwestern China, with an emphasis on the effect of three topographic variables (elevation, slope, aspect). The SBPPs are evaluated by comparing rain gauge observations at eight ground stations from May to October in 2014–2018. Results show that IMERG and GSMaP have good rainfall detection capability for the entire region, with POD = 0.75 and 0.93, respectively. In addition, IMERG overestimates rainfall (BIAS = −48.8%), while GSMaP is consistent with gauge rainfall (BIAS = −0.4%). Comprehensive analysis shows that IMERG and GSMaP are more impacted by elevation, and then slope, whereas aspect has little impact. The independent evaluations suggest that variability of elevation and slope negatively correlate with the accuracy of SBPPs. The accuracy of GSMaP presents weaker dependence on topography than that of IMERG in the study area. Our findings demonstrate the applicability of IMERG and GSMaP in mountainous catchments of Southwest China. We confirm that complex topography impacts the performance of SBPPs, especially for complex topography in mountainous areas. It is suggested that taking topographical factors into account is needed for hydrometeorological applications such as flood forecasting, and SBPP evaluations and retrieval technology require further improvement in the future for better applications.

1. Introduction

Precipitation data are one of the most crucial variables in climatic and hydrologic processes [1]. They are vital for flood forecasting as the primary inputs [2,3,4]. However, many small mountainous catchments worldwide, including southwestern China are ungauged, particularly above 1500 m [5,6]. Previous research has highlighted the importance of satellite-based precipitation products (SBPPs) as they can provide a spatial measurement of precipitation at large and regional scales, which are useful for hydrological simulation, water resource management and other applications [4,7,8]. Their advantages of consistency over longer-period, global coverage and uninterrupted data availability compensate, particularly, for the insufficient and uneven distribution of ground precipitation observations in poorly gauged basins [9]. Therefore, there is considerable interest in using indirect Satellite-based precipitation products.

Multiple SBPPs were freely available and widely used in previous researches, such as Climate Hazards Group InfraRed Precipitation with Station Data (CHIRPS) [10], Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis (TMPA) product [11], National Oceanic and Atmospheric Administration/Climate Prediction Centre (NOAA/CPC) morphing technique (CMORPH) [12] and Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN) [13]. The Integrated Multi-satellite Retrievals for GPM (IMERG) and Global Satellite Mapping of Precipitation (GSMaP) have been widely applied in recent years as they are at a more precise temporal resolution of 30 min and spatial resolution of 0.1° [14,15,16]. Many studies have shown that the IMERG and GSMaP product outperformed other datasets in both statistical and hydrological evaluation [17,18,19,20,21].

Several previous studies demonstrated that the performance of SBPPs is strongly associated with climatic conditions and topographic complexity [22,23,24,25,26,27,28,29]. Sakib et al. [22] evaluated GPM-IMERG over three typical climatic conditions. The western region is arid and dry, the central and eastern regions are under humid climatic conditions, and the southeastern region is predominantly wet, with subtropical weather conditions. Significant underestimation was found in the southeastern region with subtropical weather conditions. IMERG showed poor performance in Northern China [24]. TRMM 3B42V7 was validated against rain gauges in Middle Qilian Mountain and found the TMPA product was unreliable in high mountainous areas [25]. Gautam et al. [23] found that both TMPA and IMERG performed better in the Thar desert area, but poor in the mountainous regions. Due to the fact that SBPP algorithms are prone to all these errors for certain tropical/subtropical or mountainous regions, evaluating the performance of SBPPs in detail is crucial.

However, previous studies focused mainly on the influence of just one factor such as elevation [30,31,32,33,34]. Xu et al. [30] used a high-density rain gauge network to validate GPM IMERG and TRMM 3B42V7 rainfall products over the southern Tibetan Plateau. They demonstrated that the bias of TRMM 3B42V7 is positively correlated with mean elevation. Lu et al. [31] found that the amplitude and occurrence consistency of the GPM IMERG products increased with elevation. Bharti et al. [32] revealed 3100 m altitude as the breakpoint for the satellite overestimating and underestimating rainfall amount for elevation ranges below and above it, respectively. TRMM 3B42V7 performed better for 1000–2000 m but deviated from true values over higher-altitude regions. Few studies explored the co-influence of several topographic variables. Moreover, limited studies of SBPPs evaluation have been carried out in Southwest China, which has a humid subtropical climate and is one of the most complex terrain regions with plenty of small mountainous catchments. The relationship between the topography and the performance of SBPPs has not been widely understood yet. Therefore, this region’s comprehensive evaluation of SBPPs over complex topography is still lacking.

This study aims to evaluate the SBPPs over complex topography in mountainous regions of southwestern China and explore the impact of three topographic variables (elevation, slope, and aspect) on SBPPs performances. The studied area (Yingjing catchment) is a typical mountainous catchment with a subtropical climate and the lack of gauged precipitation data has been a serious issue for flood forecasting. The rest of the paper is arranged as follows. Materials are detailed in Section 2, describing the study area and datasets. Next, evaluation criteria are presented in Section 3. Then, Section 4 and Section 5 present the obtained results and discussions. Finally, the main conclusions and suggestions for future studies are drawn in Section 6.

2. Materials

2.1. Study Area

The Yingjing River (shown in Figure 1), which is a first-grade tributary of the Qingyi River in southwest China, is 105 km long and spans an area of 1958 km². It is located between the longitudes of 102°20′–103° E and the latitudes of 29°20′–30°N and has a humid subtropical climate. The annual precipitation is approximately 1253 mm, and the annual temperature ranges between 10 and 20 °C.

The catchment is located in the center of the Longmen Mountain rainstorm, where heavy rain and flooding events frequently happen during the flood season from June to October. According to historical data [35], the maximum daily rainfall has reached 229.5 mm, and the record maximum 24 h rainfall was as high as 371.6 mm. The measured maximum flood was 3450 m³/s, reaching 33 times the average flow. Rainstorms and flash floods cause severe threats to the local economy, social stability, and native lives. Furthermore, the elevation varies from 746 m to 3652 m, which is a typical mountainous catchment. Therefore, the extreme rainfall amounts, the severity of the flood response, and the complex terrain have made this catchment a case study for evaluating SBPPs.

2.2. Data Set

2.2.1. Digital Elevation Model

A 30 × 30 m digital elevation model (DEM) was obtained from the Geospatial Data Cloud at http://www.gscloud.cn (accessed on 1 August 2022), which was used to extract three key terrain factors (elevation, slope, and aspect) of all rain gauges (

Table 1. The terrain information of all rain gauges.

Gauge	Elevation (m)	Slope (°)	Aspect (°)
Kuhao	1531	37.18	296.42
Jinshan	1380	19.69	12.94
Sanhe	1343	24.81	239.56
Luchi	1169	14.43	335.56
Datongqiao	1156	20.07	41.73
Shizi	1093	14.91	277.88
Siping	1029	7.19	255.07
Yingjing	750	2.16	94.4

) by ArcGIS.

2.2.2. In Situ Data of Precipitation

The daily precipitation records of 8 in situ gauges in the Yingjing catchment, are used to evaluate the performance of SBPPs in this study area, as the gauged data are more precise which are usually used for evaluation [36,37]. The location of the gauges is shown in Figure 1. The rain gauge network is jointly maintained by the local meteorological agencies. We focus on the rainy-season period (1 May to 31 October) over 2014–2018. The eight gauges all have a complete 6-month time series of daily observed rainfall records.

2.2.3. IMERG

The IMEGR products are launched by Global Precipitation Measurement Mission (GPM) based on integrating passive microwave (PMW) satellite-based precipitation estimates, microwave-calibrated infrared (IR) satellite estimates, and precipitation gauge analyses, etc.

In addition, the GPM mission provides three types of IMERG products, including near-real-time (NRT) Early (IMERG-E) and Late (IMERG-L), and post-real-time (PRT) Final (IMERG-F) products. The NRT IMERG products (IMERG-E and IMERG-L) are calibrated by climatological gauge data, while the PRT IMERG (IMERG-F) adopts the Global Precipitation Climatology Centre (GPCC) gauge analysis. Although there is a 3.5 month latency for producing IMERG-F, the reliability and accuracy of IMERG-F are supposed to be better than others. Thus, the daily accumulated IMERG version 6 GPM-Level 3 Final Run product (abbreviated as IMERG below) is employed in the study. The data can be downloaded from the NASA Goddard Earth Sciences Data and Information Services Center (GES DISC, https://disc.gsfc.nasa.gov/ (accessed on 23 July 2022)).

2.2.4. GSMaP

The GSMaP products are generated by the Japan Science and Technology Agency (JAXA) based on the combination of the PMW-IR algorithm. Three GSMaP products (GSMaP _NRT, GSMaP_MVK and GSMaP_Gauge) can be obtained at the website of JAXA (https://sharaku.eorc.jaxa.jp/GSMaP/index.htm (accessed on 23 July 2022)). In the process of GSMaP_NRT production, the forward motion scheme of the cloud vector is adopted. For GSMaP_MVK, both forward and backward bidirectional cloud vector motion schemes are considered. GSMaP_Gauge is produced based on the calibration of GSMaP_MVK by the global surface rain gauge data. Thus, the daily accumulated GSMaP_Gauge version 6 (abbreviated as GSMaP below) is employed for evaluation.

Two SBPPs used in the study have spatial resolutions of 0.1° × 0.1°. (

Table 2. Data type, resolution, and sources of precipitation used in this study.

Data Type	Temporal Resolution	Spatial Resolution	Source
DEM	-	30 × 30 m	http://www.gscloud.cn (accessed on 1 August 2022)
Precipitation gauge data	Daily (1 May to 31 October 2014–2018)	-	Local meteorological agencies
IMERG-Final data	Daily (1 May to 31 October 2014–2018)	0.1° × 0.1°	https://disc.gsfc.nasa.gov/ (accessed on 23 July 2022)
GSMaP_Gauge data	Daily (1 May to 31 October 2014–2018)	0.1° × 0.1°	https://sharaku.eorc.jaxa.jp/GSMaP/index.htm (accessed on 23 July 2022)

). We recalculate the daily IMERG and GSMaP, which use coordinated universal time (UTM), and shift them to China Standard Time (CST, UTM + 8 h), in order to retain the consistency of temporal unit between the SBPPs and gauged observations.

3. Methodology

Correlation coefficient (CC), relative bias (BIAS) and root-mean-square error (RMSE) are used to evaluate the performance of the two-satellite data. The degree of agreement between gauge data and satellite data at the site and regional scale is represented by the CC index. BIAS is a systemic bias index that describes the difference between gauge and satellite data. Positive (negative) values of BIAS represent the underestimation (overestimation). The average error magnitude and the accuracy of SBPPs compared with gauge data are represented by RMSE. The ideal values for CC, BIAS, and RMSE are 1, 0, 0 (see

Table 3. List of the statistical indexes used to quantify the performance of the satellite rainfall products statistical index.

Statistical Index	Unit	Equation	Perfect Value
Correlation coefficient (CC)	NA	CC = $\frac{{[{\sum }_{\mathrm{i}=1}^{\mathrm{n}}({\mathrm{S}}_{\mathrm{i}}-\overline{\mathrm{S}})({\mathrm{G}}_{\mathrm{i}}-\overline{\mathrm{G}})]}^2}{{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}({\mathrm{S}}_{\mathrm{i}}-\overline{\mathrm{S}})}^2{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{({\mathrm{G}}_{\mathrm{i}}-\overline{\mathrm{G}})}^2}$	1
Relative bias (BIAS)	NA	$\mathrm{BIAS}=\frac{{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n}}{(\mathrm{G}}_{\mathrm{i}}-{\mathrm{S}}_{\mathrm{i}})}{\mathrm{n}}$	0
Root-mean-square error (RMSE)	mm	$\mathrm{RMSE}=\sqrt{\frac{{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n}}{{(\mathrm{G}}_{\mathrm{i}}-{\mathrm{S}}_{\mathrm{i}})}^2}{\mathrm{n}}}$	0
Probability of detection (POD)	NA	$\mathrm{POD}=\frac{\mathrm{h}}{\mathrm{h}+\mathrm{m}}$	1
Frequency of hit (FOH)	NA	$\mathrm{FOH}=\frac{\mathrm{h}}{\mathrm{h}+\mathrm{f}}$	1
False alarm ratio (FAR)	NA	$\mathrm{FAR}=\frac{\mathrm{f}}{\mathrm{h}+\mathrm{f}}$	0
Critical success index (CSI)	NA	$\mathrm{CSI}=\frac{\mathrm{h}}{\mathrm{h}+\mathrm{m}+\mathrm{f}}$	1
Heidke skill score (HSS)	NA	$\mathrm{HSS}=\frac{2(\mathrm{hc}-\mathrm{fm})}{\left(\mathrm{h}+\mathrm{f}\right)\left(\mathrm{f}+\mathrm{c}\right)+(\mathrm{h}+\mathrm{m})(\mathrm{m}+\mathrm{c})}$	1

where S and $\overline{\mathrm{S}}$ are satellite product data (mm/d) at ith grid; G and $\overline{\mathrm{G}}$ are gauge precipitation data at ith grid; n is the number of gauge stations. h, m, f, and c stand for hits, misses, false detections, and correct rejections by occurrence, respectively.

for equations).

In addition, the contingency statistical metrics are employed to evaluate the precipitation detective ability of SBPPs, including the probability of detection (POD), the frequency of hit (FOH), the false alarm ratio (FAR), the critical success index (CSI) and Heidke skill score (HSS). POD determines the likelihood of correctly detecting rainfall. FOH measures how frequently satellite products detect rainfall when it occurs. FAR calculates false events that gauge stations do not observe. CSI measures the accuracy when satellites detect real-world events. HSS measures the accuracy of the estimates accounting for matches due to random chance. A high value of POD, FOH, CSI and HSS indicates a more accurate detection ability of a satellite precipitation product. In contrast, a high value of FAR refers to a high ratio of falsely detected events (see Table 3 for equations).

4. Results

4.1. Temporal and Spatial Distribution of Gauge and Satellite Data sets

Thiessen polygon method is used for the calculation of the areal average daily rainfall in the rainy season from May to October in 2014–2018. An overview of the variation in rainfall from gauge and satellite products is shown in Figure 2. For gauge data sets, rainfall is mainly concentrated in July and August, caused by the humid monsoon from Pacific. The two satellite data sets capture the seasonal pattern correctly. GSMaP is basically consistent with gauge rainfall in dry and wet years. However, IMERG overestimates in all five rainy seasons, particularly in the dry year of 2015 (shown in Figure 2b), where the average daily rainfall of IMERG is almost twice as that of the gauge.

The spatial distribution pattern of Gauge (Figure 3a) is interpolated by Kriging, which is a commonly used interpolation method, indicating average values of eight stations over the study area. By obtaining the center point coordinates, resolution (0.1° × 0.1°) and precipitation amount of each grid, the spatial distribution of satellite data sets can be interpolated, as shown in Figure 3b,c. It is noted that IMERG and GSMaP rainfall data concentrate on the Yingjing station, which is at the lowest altitude. While the spatial distribution of daily average gauge rainfall data is the opposite. Different distributions of gauge and satellite rainfall over high and low elevations indicate the influence of orography on the performance of SBPPs which can be further explored as follows.

4.2. Evaluation of Satellite Precipitation

Table 4. Summary of evaluation of IMERG and GSMaP products at daily step.

Gauge	CC		BIAS (%)		RMSE (mm/d)
	IMERG	GSMaP	IMERG	GSMaP	IMERG	GSMaP
Kuhao	0.52	0.53	−43.5	−3.6	16.48	11.66
Jinshan	0.55	0.40	−13.7	26.0	16.69	14.61
Sanhe	0.58	0.57	−55.1	−5.3	15.18	10.14
Luchi	0.52	0.44	−42.8	6.3	16.19	12.49
Datongqiao	0.39	0.54	−48.3	3.1	18.95	11.72
Shizi	0.56	0.42	−48.4	1.7	16.80	14.04
Siping	0.58	0.59	−48.8	−3.5	16.54	10.58
Yingjing	0.61	0.51	−89.6	−27.9	17.02	12.28
Whole catchment	0.54	0.50	−48.8	−0.4	16.73	12.19

shows the intercomparison of daily rainfall from all eight rain gauges and two satellite rainfall products at the corresponding grids collocated with rain gauges. The two SBPPs have good agreements with daily rain gauges (CCs ≥ 0.5).In terms of bias, IMERG provide overall larger daily rainfall estimates than rain gauges with an average BIAS of −48.8%. For GSMaP, daily rainfall estimates are overestimated at Yingjing station (−27.9%) but underestimated at Jinshan station (26%). GSMaP is slightly larger than rain gauges with a relative bias of −0.4% at the regional scale. GSMaP performs better on BIAS and RMSE.

The observed rainfall detective abilities of IMERG and GSMaP are shown in Figure 4 and

Table 5. Summary of rainfall detective ability evaluation of IMERG and GSMaP products at daily step.

Gauge	POD		FAR		CSI		HSS		FOH
	IMERG	GSMaP	IMERG	GSMaP	IMERG	GSMaP	IMERG	GSMaP	IMERG	GSMaP
Kuhao	0.68	0.94	0.23	0.26	0.26	0.32	0.77	0.74	0.57	0.71
Sanhe	0.73	0.94	0.19	0.22	0.30	0.37	0.81	0.78	0.62	0.75
Siping	0.76	0.94	0.21	0.26	0.34	0.33	0.79	0.74	0.63	0.71
Datongqiao	0.72	0.94	0.24	0.21	0.17	0.41	0.76	0.79	0.59	0.75
Yingjing	0.80	0.96	0.27	0.30	0.35	0.37	0.73	0.70	0.62	0.68
Luchi	0.76	0.90	0.19	0.16	0.08	0.29	0.81	0.84	0.64	0.77
Jinshan	0.76	0.91	0.15	0.13	0.16	0.35	0.85	0.87	0.66	0.80
Shizi	0.79	0.93	0.21	0.21	0.23	0.34	0.79	0.79	0.65	0.74
Whole catchment	0.75	0.93	0.21	0.22	0.24	0.35	0.79	0.78	0.62	0.74

. GSMaP has significantly good capabilities in detecting precipitation events. The POD of both gauge stations and the whole region are larger than 0.9. The value of FOH, FAR, CSI and HSS all perform well which are close to the perfect value. The POD of IMERG is smaller than GSMaP, as well as HSS, CSI and FOH, while IMERG obtains fewer false events (FAR values perform better).

4.3. Evaluation of the Dependence of the Performance of Rainfall Products on Topography Factors

To explore the influence of topography factors on the accuracy of remote sensing precipitation data, ArcGIS is used to extract three key terrain factors (namely elevation, slope and aspect) of the study area. CC, POD and BIAS are used to represent the performance of IMERG and GSMaP.

4.3.1. Comprehensive Evaluation of the Influence of Elevation, Slope and Aspect

Preprocessing of terrain factors is carried out to comprehensively reflect the influence. Firstly, the Z-score method is used to standardize elevation, slope, and aspect. Then, principal components analysis (PCA) is used to reduce the redundancy in the topographic data set based on values of elevation, slope and aspect, so as to the new principal components a, b and c were constructed. Finally, Linear regression analysis is conducted on the newly constructed principal components as independent variables, and the accuracy evaluation metrics CC and POD are taken as the dependent variable. The larger the absolute values of the regression coefficient, the greater the influence.

As shown in Figure 5 (

Table 6. Multiple regression analysis results. The newly constructed principal components a, b, c are independent variables, and the accuracy evaluation metrics CC and POD are taken as the dependent variable.

The Dependent Variable	Satellite Rainfall Data	Regression Model
CC	IMERG	CC = 0.123 − 0.303a − 0.221b − 0.005c
	GSMaP	CC = 0.216 − 0.144a − 0.119b − 0.009c
POD	IMERG	POD = 0.337 − 0.410a − 0.243b − 0.012c
	GSMaP	POD = 0.512 − 0.251a − 0.210b − 0.011c

), the absolute values of the regression coefficient of elevation (a) are larger than these of slope (b), regardless of CC or POD of IMERG/GSMaP as the dependent variable. The coefficients of aspect (c) are almost 0 for both SBPPs, which illustrates accuracy’s dependence on aspect is not significant. The comprehensive analysis results reveal that elevation has a greater influence on SBPP performance than the slope. The performance of IMERG and GSMaP presents weak dependence on the aspect. So, the accuracy of IMERG and GSMaP in the Yingjing catchment is discussed from two aspects (elevation and slope) in the following, respectively.

4.3.2. Dependence of the Performance of Rainfall Products on Elevation

All the rain gauges are grouped in the entire region into four main categories: (i) <1000 m, (ii) 1000–1250 m, (iii) 1250–1500 m, (iv) >1500 m. Figure 6 shows the relationship between evaluation metrics and elevation for IMERG and GSMaP. The size of the circle indicates elevation. The larger the circle, the higher the altitude. Color represents CC/POD/BIAS of satellite products.

Significant negative correlations are demonstrated between elevation variation and CC/POD for IMERG. In Figure 6a,c,e and

Table 7. Summary of dependence of the performance of rainfall products on topography factors (elevation and slope).

Topography Factors		CC		POD		BIAS (%)
		IMERG	GSMaP	IMERG	GSMaP	IMERG	GSMaP
Elevation (m)	<1000	0.61	0.51	0.80	0.96	−89.62	−27.90
	1000–1250	0.57	0.50	0.76	0.93	−47.09	1.90
	1250–1500	0.56	0.49	0.74	0.93	−34.41	10.35
	>1500	0.52	0.53	0.68	0.94	−43.54	−3.60
Slope (°)	<10°	0.65	0.61	0.82	0.95	−69.20	−15.70
	10°−20°	0.54	0.42	0.77	0.91	−35.00	11.30
	20°−30°	0.49	0.56	0.73	0.94	−51.70	−1.10
	>30°	0.52	0.53	0.68	0.94	−43.50	−3.60

, CC decreases from 0.61 to 0.52 (POD decreases from 0.80 to 0.68) with elevation increasing. In

Table 8. Single regression analysis between topographic variables and CC, POD, and BIAS, respectively.

Topography Aspect	Satellite Rainfall Data	Regression Coefficient
		CC	POD	BIAS
Elevation	IMERG	−0.94	−0.95	0.14
	GSMaP	−0.62	−0.52	0.11
Slope	IMERG	−0.81	−0.98	0.23
	GSMaP	−0.58	−0.21	0.19

, the results of single regression analysis between elevation and CC/POD also illustrate the significantly negative correlation, the coefficient of CC is −0.94 (the coefficient of POD is −0.95). IMERG severely overestimates precipitation in all elevation categories for error distribution with elevation.

For GSMaP, the variation of its performance with elevation is not significant. The regression coefficient of CC is −0.62 and that of POD is −0.52. As shown in Figure 6b,d,f and Table 7, GSMaP has a good consistency (CC = 0.51) and rainfall detectivity (POD = 0.96) when the elevation is below 1000, but the performance is getting better/worse when above 1500m, with the CC of 0.53 and POD of 0.94. GSMaP trends slightly overestimate or underestimate precipitation in different elevations.

Overall, the performance of IMERG shows significantly negative correlations with elevation in this study area, while the correlation for GSMaP is insignificant.

4.3.3. Dependence of the Performance of Rainfall Products on Slope

The slope was divided into (i) <10°, (ii) 10°–20°, (iii) 20°–30°, (iv) >30°. Figure 7 indicates the effect of slope on the accuracy of IMERG and GSMaP. The daily IMERG and GSMaP data both perform best when the slope is below 10°, while CC is 0.65 and POD is 0.82 for IMERG (CC is 0.61 and POD is 0.95 for GSMaP). With the slope degree greater than 10°, the accuracy of IMERG tends to decrease. Especially, the POD of IMERG is only 0.68 when the slope is above 30°. However, the dependence of the performance of GSMaP on the slope decreases from <20°, then increases at 20°–30°, while finally decreases again at >30°. GSMaP’s performance does not show an obvious trend and the variations are not consistent with the slope. In terms of BIAS distribution, IMERG shows evident overestimation, while GSMaP slightly deviates from ground truth with slope variation.

In Table 8, the single regression analysis between the slope and CC/POD/BIAS shows the negative correlations between the performance of SBPPs and the slope. Moreover, the regression coefficient of CC (POD) for IMERG is −0.81 (−0.98), the absolute of which is more than that of GSMaP. It indicates that the accuracy of IMERG presents greater dependence on the slope than that of GSMaP.

5. Discussion

5.1. Influence of Terrain Factors

Many studies have shown that the detection ability of the microwave/infrared sensors or precipitation retrieval algorithms are impacted in mountainous regions, leading to negative relations between SBPPs’ accuracy and elevation [30,38,39,40]. In this study, CC and POD are found to get smaller with the increase of slope, but an inflection point would appear at 10°. Shi at el. [40] evaluated the dependence of the performance of CMPA remote sensing precipitation data on the slope in China and found that CMPA maintained a high accuracy below 10°, while the CMPA accuracy decreased significantly in the area with a slope greater than 10°. One reason may be that meteorological stations are less distributed in mountainous areas, resulting in random error increases. Another reason may be the complexity of dynamic lifting and hindering combined action. Some researchers [31,41,42] illustrated that the dynamic lifting effect of terrain on airflow and weather systems could promote topographic rain formation on windward slopes. However, when the slope is too large, the hindering effect of topography would be stronger than the lifting effect, so as to decrease the precipitation.

In Section 4.3.1, we find that aspect has little impact on the accuracy of SBPPs, which is in accordance with previous studies [43,44]. The influence of slope direction on precipitation is related to the geographical location, mountain structure and prevailing airflow direction of precipitation. Aspect does not directly affect the distribution of precipitation. Instead, the uplift of the windward slope is the main factor affecting regional precipitation.

5.2. Uncertainty of the Rainy Season

In China as well as Asia, the rainy season mainly concentrate on summer, when many severe underestimations were found in SBPPs due to more convective precipitation events [45,46]. For example, Sunilkumar et al. [46] showed that IMERG underestimated precipitation which is contributed by the well-known Baiu frontal system during summer over Japan. However, we found that IMERG significantly overestimated precipitation in southwestern China in this study. Moreover, Zhang et al. [47] found that the TRMM and IMERG products overestimated precipitation, whereas MSWEP underestimated rainfall during summer over the Min Jiang watershed. Several factors may explain the uncertainty of SBPPs’ accuracy in summer: (1) although convective precipitation events may reduce the detection accuracy by affecting satellite signals, the overestimation of IMERG may also appear in summer, when the scale of convective precipitation is not as large as the resolution of GPM [45,48]; (2) it may be reasonable for a higher concentration of aerosols over southwestern China, causing the increased cloud drop size [49].

5.3. Uncertainty of Rainfall Intensity

Previous studies have shown that performance of SBPPs may vary with rainfall intensity. For example, Gao et al. [50] found that the performance of the TRMM 3B42V6 product depends more on rainfall intensities rather than topography. Therefore, we further evaluated the performance of IMERG and GSMaP in detecting the frequency and total volume of rainfall under different rainfall categories.

The PDF by occurrence (PDFc) and by rain volume (PDFv) are used to provide details about the frequency of rainfall events with different rainfall intensities [51]. PDFc is equal to the ratio of the precipitation count under different intensities to the total precipitation count. It indicates the probability of occurrence of precipitation intensity. PDFv is equal to the ratio of the volume contribution of each precipitation intensity stage to the total precipitation volume and indicates the probability of occurrence of precipitation volume. According to the classification standard of daily precipitation intensity used by Tan et al. [52] and Liu et al. [53], the precipitation intensity is classified into six groups: light rain events (0–1 mm/d and 1–5 mm/d), moderate rain events (5–10 mm/d and 10–50 mm/d), and heavy rain events (50–100 mm/d and >100 mm/d).

Figure 8 (

Table 9. Summary of PDFc and PDFv of IMERG, GSMaP and rain gauges over different elevation ranges with different rainfall intensities.

Elevation Range (m)	Rainfall Intensity (mm/d)	PDFc (%)			PDFv (%)
		Gauge	IMERG	GSMaP	Gauge	IMERG	GSMaP
<1000	0–1	48.6	42.9	32.0	0.8	0.4	1.1
	1–5	23.6	17.5	29.7	9.8	4.1	10.7
	5–10	11.3	10.8	16.1	14.0	7.0	15.7
	10–50	14.9	24.2	20.9	53.5	54.6	58.7
	50–100	1.3	3.7	1.2	15.9	23.0	10.2
	>100	0.3	0.9	0.2	6.1	10.9	3.7
1000–1250	0–1	35.7	41.3	31.1	0.5	0.5	1.1
	1–5	27.0	18.7	31.1	9.1	4.4	11.5
	5–10	15.2	10.4	15.7	14.4	7.0	15.6
	10–50	20.3	25.0	20.7	58.8	55.9	59.5
	50–100	1.6	4.0	1.1	14.4	25.1	10.4
	>100	0.2	0.5	0.1	2.7	7.1	1.9
1250–1500	0–1	33.3	41.5	31.5	0.5	0.4	1.2
	1–5	24.8	19.5	31.2	7.3	4.9	12.0
	5–10	15.4	10.3	15.1	13.1	7.1	15.6
	10–50	24.3	24.7	21.3	61.0	57.4	60.8
	50–100	2.0	3.4	0.9	16.4	21.7	8.4
	>100	0.1	0.7	0.1	1.7	8.4	2.0
1500>	0–1	42.6	46.3	30.2	0.7	0.2	1.1
	1–5	24.6	17.4	30.0	8.3	4.2	11.0
	5–10	13.3	9.3	18.4	13.6	6.7	18.6
	10–50	16.8	22.3	20.3	51.8	53.6	59.3
	50–100	2.7	4.2	1.1	25.5	29.0	10.0
	>100	0.0	0.4	0.0	0.0	6.4	0.0
All gauge	0–1	37.6	42.2	31.2	0.6	0.4	1.1
	1–5	25.7	18.6	30.8	8.6	4.5	11.4
	5–10	14.5	10.3	15.9	13.9	7.0	16.0
	10–50	20.2	24.5	20.8	58.0	55.9	59.7
	50–100	1.8	3.9	1.1	16.4	24.4	9.8
	>100	0.2	0.6	0.1	2.4	7.8	1.9

) shows PDFc and PDFv for the entire study region. The frequency of light and moderate rainfall events (1–10 mm/d) is underestimated by IMERG, while the frequency of intense rainfall events (>10 mm/d) is overestimated. The overestimation in the frequency of heavy rainfall events leads to the overestimation in total accumulative rainfall of heavy intense rainfall categories (>50 mm/d). The distribution of PDFc and PDFv of GSMaP is the opposite of IMERG. The occurrence/volume of light and moderate rainfall events are overestimated but the occurrence/volume of heavy rainfall events are underestimated. PDFc and PDFv in different elevation ranges are similar to those for the whole catchment. In particular, the total rainfall volume of IMERG is much larger than rain gauges when the rain rate is larger than 50 mm/d in the <1000 m range, which is responsible for the abnormal BIAS of −89.6%.

6. Conclusions

In this study, IMERG and GSMaP are evaluated over a typical mountainous catchment of southwestern China and the effect of complex topography on the performance of two SBPPs is examined. Both IMERG and GSMaP have good accuracy and precipitation detective ability in the study area, which indicate that IMERG and GSMaP can provide reasonable precipitation estimations in mountainous southwestern China. The multiple and single regression analyses of topographical variables with CC/POD/BIAS reveal the relationship between the topographic complexity and the performance of SBPPs. The accuracy of SBPPs tends to decrease with elevation and slope. This study demonstrates a strong dependency of IMERG and GSMaP products on topographic transition. It provides valuable insights that the topographic and geographic information can be used to correct satellite rainfall to improve flash flood forecasting accuracy. Moreover, we find SBPPs could vary significantly with seasons and rainfall intensity. So, the relationship between topography and precipitation is not a simple linear relationship in mountainous regions. This non-linear relationship is suggested to be recognized and further validated in more catchments in the future studies.