Evaluation of Satellite Precipitation Estimates over Australia

remote sensing
Article
Evaluation of Satellite Precipitation Estimates
over Australia
Zhi-Weng Chua 1 , Yuriy Kuleshov 1,2, * and Andrew Watkins 1
 1 Long-Range Forecasts, Bureau of Meteorology, Docklands 3008, Australia;
 zhi-weng.chua@bom.gov.au (Z.-W.C.); andrew.watkins@bom.gov.au (A.W.)
 2 SPACE Research Centre, School of Science, Royal Melbourne Institute of Technology (RMIT) University,
 Melbourne 3000, Australia
 * Correspondence: yuriy.kuleshov@bom.gov.au or yuriy.kuleshov@rmit.edu.au; Tel.: +61-3-9669-4896
 



 Received: 16 January 2020; Accepted: 14 February 2020; Published: 19 February 2020 


 Abstract: This study evaluates the U.S. National Oceanographic and Atmospheric Administration’s
 (NOAA) Climate Prediction Center morphing technique (CMORPH) and the Japan Aerospace
 Exploration Agency’s (JAXA) Global Satellite Mapping of Precipitation (GSMaP) satellite precipitation
 estimates over Australia across an 18 year period from 2001 to 2018. The evaluation was performed
 on a monthly time scale and used both point and gridded rain gauge data as the reference dataset.
 Overall statistics demonstrated that satellite precipitation estimates did exhibit skill over Australia
 and that gauge-blending yielded a notable increase in performance. Dependencies of performance on
 geography, season, and rainfall intensity were also investigated. The skill of satellite precipitation
 detection was reduced in areas of elevated topography and where cold frontal rainfall was the main
 precipitation source. Areas where rain gauge coverage was sparse also exhibited reduced skill.
 In terms of seasons, the performance was relatively similar across the year, with austral summer
 (DJF) exhibiting slightly better performance. The skill of the satellite precipitation estimates was
 highly dependent on rainfall intensity. The highest skill was obtained for moderate rainfall amounts
 (2–4 mm/day). There was an overestimation of low-end rainfall amounts and an underestimation in
 both the frequency and amount for high-end rainfall. Overall, CMORPH and GSMaP datasets were
 evaluated as useful sources of satellite precipitation estimates over Australia.

 Keywords: satellite precipitation estimates; Australia; rain gauge precipitation measurements;
 satellite precipitation validation

1. Introduction
 Precipitation is an essential climate variable and is one of the most important climate variables
affecting human activities [1]. Variations in the intensity, duration, and frequency of precipitation
directly impact water availability for many millions of people and industries. Measuring rainfall over
broad areas enables efficient water management and disaster response and recovery.
 The conventional method of using rain gauges to estimate spatial patterns of rainfall provides
a direct measurement of surface rainfall but spatial density can be an issue across many parts of
the world, including over the oceans, where the installation of an adequate rain gauge network is
economically or physically unfeasible [2]. This greatly affects the ability to accurately assess rainfall
across a region as it is a variable that exhibits a high degree of spatial variation and a point-based
measurement may not provide an ideal representation of an area. Rain gauge estimates are subject
to instrumental errors with many relying on manual sampling methods. Clock synchronization and
mechanical faults are examples of potential issues [3]. Furthermore, they are also affected by localised
effects including wind (precipitation can be prevented from entering the gauge), evaporation (some

Remote Sens. 2020, 12, 678; doi:10.3390/rs12040678 www.mdpi.com/journal/remotesensing

Remote Sens. 2020, 12, 678 2 of 17

of the precipitation is evaporated before it can be recorded), wetting (some of the precipitation can
be left behind in the gauge) and splashing effects (precipitation can incorrectly splash in and out of
the gauge) [4,5]. For example, Groisman and Legates (1994) found biases due to wind-induced effects
could be quite significant, especially around mountainous areas where the bias was as large as 40% [6].
 Alternative physical-based precipitation datasets include those derived from ground-based radars
and satellites. Radar estimates are derived from the detected reflectivity of hydrometeors but suffer from
problems such as topography blockage, beam ducting, range-related and bright band effects [7,8]. Van
De Beek et al. (2016) found that for a 3-day rainfall event in the Netherlands, the radar underestimated
rainfall amount by over 50%, though after correction the difference was only 5-8% [9]. Correction
to rain gauges is critical but this means radars are likely to perform poorly in areas that lack gauge
coverage, hindering their ability to replace gauges. As a ground-based source, they are also affected by
the same physical and economical limitations that are applicable to installation of rain gauge networks.
 The use of meteorological satellites to monitor rainfall was introduced in the 1970s, providing a
means to estimate rainfall across most of the globe. The first methods inferred precipitation intensity
based on visible or infrared (IR) data by linking cloud-top temperature or reflectivity to rain rates
through empirical relationships. Later methods used passive microwave (PMW) sensors that detect
the radiation from hydrometeors and link this to rainfall rates, thereby providing a more direct
interpretation of precipitation. However, the coverage of PMW satellites is much less than that of their
IR counterparts.
 Consequently, techniques have been developed to combine the increased accuracy of PMW
estimates with the coverage provided by IR satellites. One method which can be referred to as the
cloud-motion advection method involves using IR images to derive cloud-motion vectors and then
using these vectors to advect PMW-based precipitation estimates to cover areas lacking in PMW
coverage. The Climate Prediction Center morphing technique (CMORPH) developed by the National
Oceanic and Atmospheric Administration (NOAA) was the first product of this kind and was followed
by the Japan Aerospace Exploration Agency’s (JAXA) Global Satellite Mapping of Precipitation
(GSMaP) dataset [10,11]. CMORPH and GSMaP have undergone multiple advancements since their
inception. The key improvements have been the introduction of the Kalman filter to modify the
shape and intensity of the advected rainfall and the implementation of bias-correction using gauge
data [10,12,13].
 Many past verification studies have been performed with modern-day satellite technology,
showing that, at least on a monthly basis, the technology can possess good potential [14]. However,
few studies have been performed over Australia, with even fewer, if any, using the latest CMORPH
and GSMaP datasets.
 Continental studies in the past have indicated that performance varies greatly with rainfall type,
amount, and season, with performance tending to be better for heavier rainfall regimes including
those during summer and in the tropics [15,16]. Ebert et al. (2007) evaluated the performance of
multiple satellite datasets, including CMORPH, over Australia for a two-year period and found the
estimates were relatively unbiased over summer but the accuracy greatly deteriorated in winter [15].
Pipunic et al. (2015) examined Tropical Rainfall Measuring Mission (TRMM) 3B42RT satellite data over
mainland Australia across a nine-year period and found a similar conclusion with detection of light
rainfall (

Remote Sens. 2020, 12, 678 3 of 17

April 2014 to March 2016 and showed that GSMaP overestimated light precipitation (32 mm/day) [20]. Hit bias rather than false or missed
event bias was noted as the major error with false event bias also being more significant than missed
event bias. The introduction of a bias-correction scheme is largely able to correct a positive bias by
scaling down the magnitudes, but the inability to correct missed events means there has been much
less success in correcting the negative bias [13].
 Previous studies have also indicated that a significant degradation of performance occurred over
orography, with satellites underestimating rainfall over higher elevations [18,21]. The bias can be
worse during winter where the poor detection of snowfall, as well as rainfall, over cold surfaces leads
to both missed events and an underestimation of intensity [22]. Derin et al. (2016) performed an
evaluation over the western Black Sea region of Turkey, an area featuring complex topography in the
form of a mountain range, from 2007 to 2011, and found that CMORPH exhibited a bias of −54% for the
windward side of the region during the warm season, increasing to −82% during the cold season [21].
 Kubota et al. (2009) found that the greatest biases in GSMaP were over coastal areas with frequent
orographic rainfall and that estimates were generally better over the ocean than over the land [23].
Coastal regions are likely to present difficulties as the retrieval algorithm struggles to account for both
ocean and land surfaces in a single grid point.
 This study aims to contribute to the validation of satellite rainfall data. It differs from earlier
studies by evaluating satellite precipitation estimates over a relatively long period of record (18 years)
with a focus on Australia, which has a relatively dense rain gauge network over a large area when
compared to other world regions [15]. The use of a percentile-based verification statistic is an innovative
feature of this study, while the use of both gridded and point gauge data as a reference adds additional
insight compared to using just one. The CMORPH and GSMaP datasets were chosen due to their
provision as part of the World Meteorological Organization (WMO) Space-based Weather and Climate
Extremes Monitoring Demonstration Project (SEMDP) [24]. This project aims to introduce operational
satellite rainfall monitoring products based on these two datasets, to East Asia and Western Pacific
countries, of which many lack adequate rainfall monitoring capabilities due to the absence of an
extensive and accurate rain gauge network. The verification of these datasets is thus an important step
for the creation of these products. Moreover, the cloud-motion advection method used to blend PMW
and IR data ranks amongst the best in terms of performance across various satellite methods used
to estimate precipitation [25]. The variance of the errors in the satellite precipitation estimates with
location, season, and rainfall intensity was investigated.
 The paper is organised as follows. Section 2 describes the study area, datasets, and methods used
in the study. Section 3 presents the results while Section 4 discuss the findings. Section 5 summarises
the major findings and provides directions for future work.

2. Materials and Methods

2.1. Study Area
 Australia has a land area of around 8.6 million km2 , making it the sixth-largest country in the
world by land size. Its large geographical size means that it experiences a variety of climates, including
temperate zones to the south east and south west, tropical zones to the north, and deserts or semi-arid
areas across much of the interior [26]. The main orographic feature occurs in the form of the Great
Dividing Range (GDR), a mountain range along the eastern side of the country that extends more than
3500 km from the north-eastern tip of Queensland, towards and along the coast of New South Wales,
and into the eastern and central parts of Victoria. The width of the GDR ranges from about 160 to 300
km with a maximum elevation of 2228 m, though the typical elevation range for the highlands is from
300 to 1600 m [27]. In Figure 1, the domain of analysis is shown, with the stations used in the study
also marked.

Remote Sens. 2020, 12, 678 4 of 17
Remote Sens. 2020, 12, x FOR PEER REVIEW 4 of 18

 (a) (b)
 Figure
 Figure1.1.Domain
 Domainofofanalysis.
 analysis.(a)(a)
 Stations are
 Stations aremarked
 markedasasblue
 bluedots;
 dots;(b)
 (b)scale
 scaleofoftopography.
 topography.

 2.2.Datasets
2.2. Datasets

 AsAspart
 partofofthetheWMO
 WMOSEMDP,SEMDP,access accesstotoGSMaP
 GSMaPand andCMORPH
 CMORPHdata datawere
 wereprovided
 providedby byJAXA
 JAXAand and
 NOAA,respectively.
NOAA, respectively.Both Bothdatasets
 datasetsofofsatellite
 satelliteprecipitation
 precipitationestimates
 estimatesemploy
 employthe thecloud-advection
 cloud-advection
 techniqueintroduced
technique introduced in Section
 in Section 1. The1.GSMaP
 The GSMaPversion used version wasused
 GSMaP was GSMaP Gauge-adjusted
 Gauge-adjusted Near-Real-
 Near-Real-Time (GNRT) Version 6. To allow for a faster data
Time (GNRT) Version 6. To allow for a faster data latency, gauge adjustment over land latency, gauge adjustment over landwas
 was
 performedagainst
performed againstthethegauge-calibrated
 gauge-calibratedversion version(GSMaP
 (GSMaPgauge) gauge)from fromthethepast
 pastperiod,
 period,which,
 which,itself,
 itself,
 is calibrated by matching daily satellite rainfall estimates to a global
is calibrated by matching daily satellite rainfall estimates to a global gauge analysis, CPC Unified gauge analysis, CPC Unified
 Gauge-BasedAnalysis
Gauge-Based AnalysisofofGlobal
 GlobalDaily DailyPrecipitation
 Precipitation(CPC (CPCUnified)
 Unified)[28].[28].Further
 Furtherdetails
 detailscancanbebefound
 found
ininthe
 theGSMaP
 GSMaPtechnical
 technicaldocumentation
 documentation[28]. [28].
 Two versions of CMORPH
 Two versions of CMORPH were used. These were used. These
 werewere the bias-corrected
 the bias-corrected CMORPH CMORPH(CMORPH (CMORPH
 CRT)
 CRT)
and theand the gauge-blended
 gauge-blended CMORPH CMORPH (CMORPH(CMORPH BLD)BLD) datasets.
 datasets. BiasBias correction
 correction overover land
 land waswasalso
 also
 performedusing
performed usingthetheCPC
 CPCUnified
 Unifiedanalysis
 analysisbut butusing
 usinga adifferent
 differentalgorithm
 algorithmthat thatinvolves
 involvesmatching
 matchingtoto
 probabilitydistribution
probability distributionfunction
 function(PDF) (PDF)tables
 tablesfromfromthe thepast
 past30 30days.
 days.The Thegauge-blended
 gauge-blendedversion versionuses
 uses
 thebias-corrected
the bias-correctedversionversionas as aa first
 first guess
 guess and then incorporates
 incorporates the the gauge
 gaugedatadatabased
 basedon onthethedensity
 densityof
 the observations; further details can be found
of the observations; further details can be found in [10,13]. in [10,13].
 Consequently,this
 Consequently, thisstudy
 studyused usedtwo twogauge-corrected
 gauge-correctedsets sets(GSMaP
 (GSMaPand andCMORPH
 CMORPHCRT) CRT)and andone
 one
 thathad
that hadbeen
 beenfurther
 furtherprocessed
 processedby bycombining
 combiningCMORPH CMORPHCRT CRTwithwithgauge
 gaugedatadata(CMORPH
 (CMORPHBLD). BLD).
 The reference datasets used were both based on the Bureau
 The reference datasets used were both based on the Bureau of Meteorology (BoM) rain gauges of Meteorology (BoM) rain gauges
 withthe
with theAustralian
 AustralianWater WaterAvailability
 AvailabilityProject
 Project(AWAP)(AWAP)analysis
 analysisbeingbeingused
 usedasasthe thereference
 referencedataset
 dataset
 forthe
for thegridded
 griddedcomparison
 comparisonand andthe thevalues
 valuesfrom fromthe thestations
 stationsthemselves
 themselvesbeing beingused
 usedfor forthe
 thepoint
 point
 comparison. The AWAP rainfall analysis is generated by decomposing
comparison. The AWAP rainfall analysis is generated by decomposing the field into a climatology the field into a climatology
 componentand
component andan ananomaly
 anomalycomponent
 componentbased basedon onthe
 theratio
 ratioofofthe theobserved
 observedrainfall
 rainfallvalue
 valuetotothethe
 climatology[29].
climatology [29].The
 The Barnes
 Barnes successive-correction
 successive-correction technique techniqueisisapplied
 appliedtoto thetheanomaly
 anomaly component
 component and
 added
and addedto the
 to monthly
 the monthly climatological
 climatological averages, whichwhich
 averages, were were
 derived using using
 derived a three-dimensional
 a three-dimensional smooth
 splice approach
smooth [29]. The
 splice approach climatological
 [29]. averagesaverages
 The climatological were generated from 30 years
 were generated from of 30 monthly
 years of totals
 monthly[29].
 For the point comparison, only ‘Series 0’ stations were chosen as these stations
totals [29]. For the point comparison, only ‘Series 0’ stations were chosen as these stations are Bureau- are Bureau-maintained
 and conform
maintained andto International
 conform to Civil Aviation Civil
 International Organization
 Aviation(ICAO) standards.
 Organization The minimum
 (ICAO) standards. number
 The
 of stations
minimum used across
 number the period
 of stations was 4764.
 used across As discussed
 the period was 4764. earlier, even though
 As discussed earlier,rain
 evengauge
 though network
 rain
 measurements
gauge can be taken ascan
 network measurements ‘truth’, they as
 be taken still‘truth’,
 contain errors,
 they still which
 containwill artificially
 errors, which inflate the errors
 will artificially
 attributed
inflate to satellite
 the errors measurements.
 attributed to satellite measurements.
 Detailson
 Details onthe
 thespatial
 spatialandandtemporal
 temporalresolutions
 resolutionsofofthe thegridded
 griddeddatasets
 datasetsalongalongwith
 withtheir
 theirdomains
 domains
 are shown in
are shown in Table 1.Table 1.

Remote Sens. 2020, 12, 678 5 of 17

 Table 1. Details about dataset used.

 Dataset Resolution (◦ ) Start of Temporal Domain Latitude Range (◦ ) Longitude Range (◦ )
 CMORPH BLD 0.25 Jan 1998 (−45, 40) (50, 200)
 CMORPH CRT 0.25 Jan 1998 (−45, 40) (50, 200)
 GSMaP 0.25 Apr 2000 (−45, 40) (50, 200)
 AWAP 0.05 Jan 1900 (−44.525, −9.975) (111.975, 156.275)

 The longest common period across the datasets using full years was chosen for the analysis (i.e.,
January 2001 to December 2018). A spatial domain of latitude from −44.625◦ N to −10.125◦ N and
longitude from 112.125◦ E to 156.125◦ E was chosen as this domain centers on Australia. Ocean data
were masked.

2.3. Method
 The satellite datasets were compared against the gauge-based datasets. Both a gridded comparison
and point comparison were performed. When performing the comparisons, all the datasets were
linearly interpolated to the same spatial resolution. An interpolation to the coarsest resolution was
chosen (i.e., 0.25◦ ). Values at each grid box from these interpolated grids could then be compared
against each other for the gridded comparison.
 For the point comparison, values corresponding to the location of a station were linearly
interpolated from each grid. These values could then be compared to the actual station value. Inclusion
of the AWAP dataset was done to provide an additional reference. A complication arose from the fact
that the gauge-based data values were 24 h accumulated values to 0900 local standard time (LST), while
the satellite data values were values to 00 UTC. As this study is focused on monthly comparisons, the
longer period greatly reduces the impact of this timing inconsistency. An elementary remedy would
be to have shifted the gauge and AWAP values one day ahead of their satellite counterparts, reducing
the inconsistency to two hours or less. Doing this adjustment resulted in improvements of less than 2%
and so the unadjusted datasets were used for simplicity.
 Both continuous and percentile-based statistics were calculated. The continuous statistics
calculated were the mean bias (MB), root-mean-square error (RMSE), mean average error (MAE),
and the Pearson correlation coefficient (R). The MB is the average difference between the estimated
and observed values, which gives an indicator of the overall bias. The MAE measures the average
magnitude of the error. To remove the effect of higher rainfall averages leading to larger errors, the
MAE was also normalised through division by the average rainfall producing the normalised mean
average error. The RMSE also measures the average error magnitude but is weighted towards larger
errors. R is commonly known as the linear correlation coefficient as it measures the linear association
between the estimated and observed datasets.
 In addition to continuous verification statistics, a percentile-based verification can also be
performed to measure how well the datasets reproduce the occurrence of low- and high-end values.
This is a novel verification metric that the authors have deemed useful to assess because, even if the
satellites performs poorly in terms of absolute values, they may still produce accurate values relative
to their own climatology, meaning there is the potential to produce percentile-based products. Such
products have already been produced (e.g., both NOAA and JAXA have generated satellite-derived
versions of the Standardized Precipitation Index, as well as rainfall values expressed as high-end
percentiles). The quintile for an observed month at a location could be derived by ranking that value
against the same month but for different years across the verification period. The ranking can then be
converted to a percentile through linear interpolation. If a bottom or top quintile was observed, the
value from the satellite dataset was then investigated. If it was also registered in the same quintile, this
was recorded as a success; otherwise, it was recorded as a failure. The number of successes was then
converted to a hit rate. This hit rate was only calculated for the gridded comparison as the varying
number of stations across the verification period made a point-based comparison more difficult. The

Remote Sens. 2020, 12, x FOR PEER REVIEW 6 of 18
Remote Sens. 2020, 12, 678 6 of 17

 The equations for the metrics are summarised in Table 2 with Ei representing the estimated value
at a of
use point or grid
 quintiles box i, Ogreater
 provided i being the observed value,
 differentiation and Nvalues
 of extreme being than
 the number of samples
 terciles or quartiles,(across the
 while the
whole length
record domainwas andconsidered
 period) fortoo
 theshort
 continuous metrics.
 for the use of deciles.
 The equations for the metrics are summarised in Table 2 with Ei representing the estimated value
at a point or grid box i, Oi being the Table 2. Summary
 observed value,ofand
 metrics used.the number of samples (across the
 N being
whole domain and period) for the continuous Equation
 Metric metrics. Range Perfect Value Unit
 1
 Mean bias (MB) of 
 Table 2. Summary metrics used. [0, ∞) 0 mm/day
 
 Metric Equation
 1 Range Perfect Value Unit
 Mean average error (MAE) | | [0, ∞) 0 mm/day
 N 
 Mean bias (MB) 1 P
 (Ei − Oi ) (−∞, ∞) 0 mm/day
 N 1
 i=1 ∑ | |
 Normalised mean average error N
 [0, ∞) 0
 1 P 1i| [0, ∞)
 Mean average error (MAE) N |Ei − O ∑ 0 mm/day
 i=1 
 1 PN
 Normalised mean average error i=1 Ei −Oi [0, ∞) 0
 1
 N
 1 PN
 i = 1 Ei
 Root-mean-square error (RMSE) N [0, ∞) 0 mm/day
 s 
 N
 Root-mean-square error (RMSE) 1 ( E i − Oi ) 2 [0, ∞) 0 mm/day
 P
 i∑
 N
 =1 
 Pearson correlation coefficient (R) [0, 1] 1
 ∑i=1 [(Ei 
 −E)(O i −O)] ∑ 
 PN

 Pearson correlation coefficient (R) [−1, 1] 1
 q q
 PN 2 PN 2
 i=1 ( i
 E −E ) i=1 (Oi −O)

3. Results
3. Results
 The results of the gridded continuous comparison against AWAP data are presented in Figure
2. The
 Thelinear correlation
 results of thecontinuous
 of the gridded satellite rainfall estimates
 comparison ranges
 against AWAPfromdata0.77are
 to presented
 0.88, whileinthe MAE
 Figure 2.
ranges from 0.61 to 0.43 mm/day. The trend amongst all the metrics is the same with performance
The linear correlation of the satellite rainfall estimates ranges from 0.77 to 0.88, while the MAE ranges
being0.61
from thetobest
 0.43for CMORPH
 mm/day. BLD, amongst
 The trend then CMORPH CRT, and
 all the metrics lastly
 is the sameGSMaP. CMORPH being
 with performance CRT and
 the
GSMaP
best display similar
 for CMORPH BLD, performances,
 then CMORPHwhile therelastly
 CRT, and is a clear
 GSMaP. increase
 CMORPH in performance for CMORPH
 CRT and GSMaP display
BLD. performances, while there is a clear increase in performance for CMORPH BLD.
similar

 Figure 2. Gridded continuous comparison of satellite datasets against Australian Water Availability
 Figure 2. Gridded continuous comparison of satellite datasets against Australian Water Availability
 Project (AWAP) from January 2001 to December 2018. Mean bias (MB), root-mean-square error (RMSE),
 Project (AWAP) from January 2001 to December 2018. Mean bias (MB), root-mean-square error
 mean average error (MAE), Pearson correlation coefficient (R) and normalised mean average error
 (RMSE), mean average error (MAE), Pearson’s correlation coefficient (R) and normalised mean
 (Norm. MAE) are displayed.
 average error (Norm. MAE) are displayed.
 The gridded percentile-based comparison against AWAP data is shown in Figure 3. The satellite
 The gridded percentile-based comparison against AWAP data is shown in Figure 3. The satellite
datasets obtain around a 70%–80% hit rate for the bottom quintile whilst scoring around 10% less for
datasets obtain around a 70%–80% hit rate for the bottom quintile whilst scoring around 10% less for

Remote Sens. 2020, 12, x FOR PEER REVIEW 7 of 18

Remote Sens. 2020, 12, 678 7 of 17
the top
Remote quintile.
 Sens. 2020, 12,This suggests
 x FOR the rainfall
 PEER REVIEW values produced by the satellites are relatively accurate 7 of in
 18
terms of climatological occurrence, with better performance exhibited for low-end extremes. There
the top quintile.
appears This suggests
 to be potential the rainfall
 in generating values produced
 percentile-based by the
 products satellites
 from satellitearedata.
 relatively accurate in
 climatological occurrence, with
terms of climatological with better
 better performance
 performance exhibited
 exhibited forfor low-end
 low-end extremes.
 extremes. There
appears to be potential
 potential in
 in generating
 generating percentile-based
 percentile-based products
 products from
 fromsatellite
 satellitedata.
 data.

 Figure 3. Gridded percentile-based comparison of satellite datasets against AWAP from January 2001
 to December 2018. Bottom- and top-quintile hit rates are displayed.
 Figure
 Figure 3. Gridded percentile-based
 3. Gridded percentile-based comparison
 comparison ofof satellite
 satellite datasets
 datasets against
 against AWAP
 AWAP from
 from January
 January 2001
 2001
 to December 2018. Bottom- and top-quintile hit rates are displayed.
 to December
 Figure 2018. Bottom-
 4 displays and top-quintile
 the continuous hit rates
 statistics are point
 using displayed.
 gauge data as truth. The comparison
against point4 gauge
 Figure displaysdata supports
 the the gridded
 continuous comparison
 statistics with
 using point the CMORPH
 gauge BLD error
 data as truth. being about
 The comparison
 Figure
50% larger 4
 than displays
 the datathe
 AWAP continuous
 error and statistics
 CMORPH using point
 CRT and with gauge
 GSMaP data as truth. The comparison
against point gauge supports the gridded comparison the being
 CMORPH aboutBLD
 150% larger.
 error The
 being MB
 about
against
is point
 negative forgauge
 all data
 the supports
 satellite the gridded
 datasets, comparison
 indicating a slightwith the CMORPH
 tendency for BLD error being
 underestimation of about
 overall
50% larger than the AWAP error and CMORPH CRT and GSMaP being about 150% larger. The MB is
50% larger than the AWAP error and CMORPH CRT and GSMaP being about 150% larger. The MB
rainfall.
negative for all the satellite datasets, indicating a slight tendency for underestimation of overall rainfall.
is negative for all the satellite datasets, indicating a slight tendency for underestimation of overall
rainfall.

 Figure 4. Comparison
 Figure 4. Comparisonofofsatellite
 satellitedatasets
 datasetsagainst point
 against gauge
 point datadata
 gauge fromfrom
 January 2001 2001
 January to December 2018.
 to December
 2018.
 The ranking
 Figure of performance
 4. Comparison between
 of satellite theagainst
 datasets satellite datasets
 point gaugeremains theJanuary
 data from same for both
 2001 tothe continuous
 December
and the
 Thepercentile-based
 2018. statistics. between
 ranking of performance The benefittheofsatellite
 blending in gauge
 datasets data isthe
 remains again displayed,
 same for bothwith
 the
CMORPH
continuousBLD andshowing significant improvement
 the percentile-based statistics. over
 The the unblended
 benefit datasetsinand
 of blending skill comparable
 gauge data is again to
AWAP. As the
 The ranking
displayed, trend
 with CMORPH between
 of performance MB,
 BLD showing MAE,
 betweenRMSE, and R
 the satellite
 significant is the same, future
 datasets over
 improvement remains references
 the same for
 the unblended to continuous
 both and
 datasets the
statistics
continuous will refer
 and to
 the just MAE, normalised
 percentile-based MAE
 statistics. and
 The R for brevity.
 benefit
skill comparable to AWAP. As the trend between MB, MAE, RMSE, and R is the same, future of blending in gauge data is again
 The values
displayed,
references with
 to and
 CMORPH
 continuous residualsBLD of showing
 statisticsthe
 willdatasets toagainst
 significant
 refer just MAE, point gauge data
 improvement
 normalised over are
 MAEtheshown in Figure
 unblended
 and R for 5. There
 datasets
 brevity. and
appears
skill The tovalues
 be a tendency
 comparable to AWAP. towards
 As an
 the overestimation
 trend between for
 MB,low rainfall
 MAE, months
 RMSE, and and
 and residuals of the datasets against point gauge data are shown in Figure 5. There R an
 is underestimation
 the same, future
for high to
references
appears rainfall
 to months.statistics
 becontinuous
 a tendency AWAP an
 towards and,
 will to a to
 refer lesser
 overestimation extent,
 just MAE, CMORPH
 normalised
 for low BLD and
 MAE
 rainfall months were able
 andRan to capture the
 forunderestimation
 brevity.
high-end
for high rainfall
 Therainfall months
 values months.
 and residuals more
 AWAP accurately,
 ofand,
 the datasets with
 to a lesser observation
 against
 extent,point
 CMORPH of months
 gauge BLD where
 data were
 are shownmore than
 able toincapture
 Figure40 the
 mm/day
 5. There
 high-
was recorded,
appears to be being
 a distinctly
 tendency better.
 towards an All datasets
 overestimation appear forto struggle
 low with
 rainfall
end rainfall months more accurately, with observation of months where more than 40 mm/day was very
 months high-end
 and an rainfall months
 underestimation
(>60
for mm/day).
 high
recorded, being These
 rainfall months.
 distinctlygauge
 AWAP totals
 better. Allsitdatasets
 and, toalong
 a lesserthe lower
 extent,
 appear boundary,
 toCMORPH which
 BLD
 struggle with indicates
 were
 very that
 able to capture
 high-end the the
 rainfall datasets
 high-
 months
observed
end
(>60 rainfalllittle
 mm/day). rainfall
 months
 Thesemore while
 gauge the gauges
 accurately,
 totals sitwith observed
 along a significant
 observation
 the lower of amount.
 months where
 boundary, The fact
 which more that even
 thanthat
 indicates AWAP
 40 mm/day does
 was
 the datasets
not depict
recorded, these
 being totals
 distinctly well suggests
 better. All that
 datasetsgridded
 appear datasets
 to systematically
 struggle with
observed little rainfall while the gauges observed a significant amount. The fact that even AWAP very struggle
 high-end with these
 rainfall very
 months
high-end
(>60 not values.
does mm/day).depictTheseA likely
 these gaugereason
 totals is suggests
 totals
 well that the gridded
 sit along datasets
 thegridded
 that lower smoothsystematically
 boundary,
 datasets down point
 which values
 indicates that
 struggleas the
 part of these
 their
 datasets
 with
objective
observed
very high-endanalysis process
 little values.
 rainfall andthe
 Awhile
 likely so it is expected
 gauges
 reason thethat
 observed
 is that ahigh-end
 gridded totals
 significant
 datasets will beThe
 amount.
 smooth underrepresented
 down fact thatvalues
 point even AWAP bypart
 as the
grids.
does The
 not impact
 depict from
 these this
 totals effect
 well would
 suggests be worse
 that if there
 gridded were nearby
 datasets gauges
 systematically
of their objective analysis process and so it is expected that high-end totals will be underrepresented with low
 struggle totals.
 with these
very
by thehigh-end values.
 grids. The A from
 impact likelythis
 reason is would
 effect that thebegridded
 worse ifdatasets smooth
 there were down
 nearby pointwith
 gauges values
 lowas part
 totals.
of their objective analysis process and so it is expected that high-end totals will be underrepresented
by the grids. The impact from this effect would be worse if there were nearby gauges with low totals.

Remote Sens. 2020, 12, 678 8 of 17
Remote
Remote Sens.
 Sens.2020,
 2020,12,
 12,xxFOR
 FORPEER
 PEERREVIEW
 REVIEW 8 of 188 of 18

 Figure 5. Scatterplot comparisons of satellite datasets against point gauge data from January 2001 to
 Figure5.5. Scatterplot
 Figure
 December 2018.
 Scatterplot comparisons
 comparisons of
 of satellite
 satellite datasets
 datasets against
 against point
 point gauge
 gauge data
 data from
 fromJanuary
 January2001
 2001to
 to
 December 2018.
 December 2018.
 3.1. Variation with Geography
3.1. Variation with Geography
 3.1. Variation
 A griddedwith Geography was performed over the Australian domain with the geographical
 comparison
 A gridded
 representations comparison
 the MB andwas performed
 shown in over the Australian domain withCMORPH
 the geographical
 A griddedofcomparison MAE
 was performed Figure
 over the6. The CMORPH
 Australian CRT and
 domain BLD
 with the geographical
representations
 datasets were of the MB
 chosen to and MAE
 allow an shown in Figure
 investigation into 6. The
 the CMORPH
 effects of CRT
 gauge and CMORPH
 correction. BLD datasets
 Generally, the BLD
 representations of the MB and MAE shown in Figure 6. The CMORPH CRT and CMORPH
were chosen to allow
 satellite-derived dataan investigation
 overestimate into the
 rainfall, effects
 except of western
 over gauge correction.
 Tasmania Generally,
 where therethe
 is asatellite-derived
 significant
 datasets were chosen to allow an investigation into the effects of gauge correction. Generally, the
data overestimate rainfall, except over western Tasmania where there is a significant underestimation.
 underestimation.
 satellite-derived data overestimate rainfall, except over western Tasmania where there is a significant
 underestimation.

 Figure 6. Cont.

Remote Sens. 2020, 12, 678 9 of 17
 Remote Sens. 2020, 12, x FOR PEER REVIEW 9 of 18
 Remote Sens. 2020, 12, x FOR PEER REVIEW 9 of 18

 Figure
 Figure 6. Mean
 Mean
 6. 6. bias,
 bias, meanmean average
 average error,
 error, and normalised
 and normalised mean
 meanmean average
 average error
 error for for bias-corrected
 bias-corrected Climate
 Figure Mean bias, mean average error, and normalised average error for bias-corrected
 Climate
 Prediction Prediction
 Center Center
 morphing morphing
 technique technique
 (CMORPH (CMORPH
 CRT) and CRT) and gauge-blended
 gauge-blended CMORPH CMORPH
 (CMORPH
 Climate Prediction Center morphing technique (CMORPH CRT) and gauge-blended CMORPH
 (CMORPH
 BLD) BLD)
 datasetsBLD) datasets
 fromdatasets from
 Januaryfrom January
 2001January 2001 to December
 to December 2018 using AWAP as truth.
 (CMORPH 2001 to2018 using AWAP
 December as truth.
 2018 using AWAP as truth.

 The The effects of normalisationare
 effects are indicated alongalong the
 the northern coast
 coastofofAustralia
 Australiaand andin western
 The effectsofofnormalisation
 normalisation areindicated
 indicated along the northerncoast
 northern of Australia and ininwestern
 western
 Tasmania
Tasmania where
 where thethe unnormalised
 unnormalised errors
 errors were
 were previously
 previously the
 the greatest
 greatest but
 but improve
 improve to about
 to about average
 average
 Tasmania where the unnormalised errors were previously the greatest but improve to about average
 after
after thethe adjustment,
 adjustment, atat leastfor
 least forthe
 theCMORPH
 CMORPHBLD BLD dataset.
 dataset.
 after the adjustment, at least for the CMORPH BLD dataset.
 The The effect of gaugecorrection
 effect correction is especially evident
 evident around western
 westernTasmania,
 Tasmania,as well asas
 around
 The effectofofgauge
 gauge correctionisisespecially aroundwestern
 especially evident around Tasmania, asaswell
 well around
 as around
 western parts of Western Australia, the southern Australian coastline, the northern coastline of New
western parts of Western Australia, the southern Australian coastline, the northern
 western parts of Western Australia, the southern Australian coastline, the northern coastline of New coastline of New
 South Wales, the Australian Alps, and the southwestern coast of Western Australia. In these areas,
South
 SouthWales,
 Wales,thethe
 Australian
 AustralianAlps,
 Alps,andand
 thethe
 southwestern
 southwestern coast of Western
 coast of WesternAustralia. In these
 Australia. areas,
 In these there
 areas,
 there are significant improvements in the normalised errors from the uncorrected dataset to the
arethere
 significant improvements
 are significant in the normalised
 improvements errors from
 in the normalised thefrom
 errors uncorrected dataset to
 the uncorrected the corrected
 dataset to the
 corrected one, indicating that there is a problem with satellite rainfall detection that cannot be
one, indicating
 corrected one,that there is athat
 indicating problem
 there with satellite rainfall
 is a problem detection
 with satellite that cannot
 rainfall be accounted
 detection that cannotfor beby
 accounted for by higher rainfall averages. Possible reasons will be discussed in the next section.
 accounted
higher foraverages.
 rainfall by higherPossible
 rainfall averages.will Possible reasons will be next
 discussed in the next section.
 A point-based comparison reasons
 using rain gaugesbe discussed in the
 categorised section.
 by states supported the gridded
 Apoint-based
 point-basedcomparison
 comparison using using rain
 rain gauges
 gauges categorised by states supported the
 thegridded
 comparison with the results shown in Figure 7. The unnormalised MAEsupported
 A categorised by states values suggest gridded
 that
 comparison
comparison with
 withisthe the results
 results shown shown in Figure 7. The unnormalised MAE values suggest that
 performance decreased in theintropical
 Figure 7.regions
 The unnormalised MAE values
 and in Tasmania, suggest
 but after that performance
 normalisation, the
 performance
is decreased in is
 the decreased
 tropical in
 regionsthe tropical
 and in regions
 Tasmania, and
 but in
 after Tasmania, but
 normalisation, after
 the normalisation,
 performance is the
 much
 performance is much higher even across the states. The performance is slightly worse in Queensland
 performance
moreandeven
 South is much
 across
 Australia, higher
 the states. evenperformance
 The
 while gauge across the states.
 correction The
 to performance
 is slightly
 appears worse
 have the in is slightly
 effect worse
 Queensland
 greatest inand in Queensland
 South
 Tasmania. Australia,
 and South Australia, while gauge correction appears
while gauge correction appears to have the greatest effect in Tasmania. to have the greatest effect in Tasmania.

 Figure 7. Point-based comparison categorised by states from January 2001 to December 2018 using
 Figure
 Figure Point-based
 7. 7.gauges
 station Point-based comparison
 comparison categorised
 as truth. categorised by states
 states from
 from January
 January2001
 2001totoDecember
 December2018
 2018using
 using
 station gauges as truth.
 station gauges as truth.

Remote Sens. 2020, 12, x FOR PEER REVIEW 10 of 18
Remote Sens. 2020, 12, 678 10 of 17
 3.2. Variation with Seasons
 A seasonal
 Remote 2020,analysis
 Sens. with 12, wasREVIEW
 x FOR PEER completed by categorising the data into four seasons with December, 10 of 18
3.2. Variation Seasons
 January, and February (DJF); March, April, and May (MAM); June, July, and August (JJA); and
 A seasonal
 3.2. Variation
 September, analysis
 with
 October, andwas
 Seasons completed
 November (SON)byrepresenting
 categorisingaustral
 the data into four
 summer, seasons
 autumn, with December,
 winter, and spring
January, and
 respectively. February (DJF); March, April, and May (MAM); June, July, and August (JJA);
 A seasonal analysis was completed by categorising the data into four seasons with December,
 and September,
October, and and
 A gridded
 January, November
 February (SON)
 comparison (DJF);representing
 showing the austral
 March, April, summer,
 normalised
 and May MAE
 (MAM); autumn,
 June, winter,
 from the and
 July,CMORPH
 and spring
 August BLDrespectively.
 (JJA);dataset
 and is
 A gridded
 displayed in comparison
 Figure 8. Theshowing
 greatest the normalised
 seasonal MAE
 variation offrom
 the the CMORPH
 error is BLD
 observed
 September, October, and November (SON) representing austral summer, autumn, winter, and spring dataset
 towards is displayed
 the interior
in Figure 8. The
 andrespectively. greatest seasonal variation of the error is observed towards the
 around the northern coastline with winter possessing the worst performance and summer havinginterior and around
the
 thenorthern coastline
 best.A gridded with winter
 comparison possessing
 showing the worst performance
 the normalised MAE from the and summer BLD
 CMORPH having the best.
 dataset is
 displayed in Figure 8. The greatest seasonal variation of the error is observed towards the interior
 and around the northern coastline with winter possessing the worst performance and summer having
 the best.

 Figure8.8. Gridded
 Figure Gridded comparison
 comparison categorised
 categorised by
 by seasons
 seasons from
 from January
 January 2001
 2001totoDecember
 December2018
 2018using
 using
 Figure 8. Gridded comparison categorised by seasons from January 2001 to December 2018 using
 AWAP
 AWAP as truth.
 as truth. Normalised
 Normalised mean
 mean average
 average error from CMORPH BLD is displayed.
 AWAP as truth. Normalised mean averageerror
 errorfrom
 from CMORPH
 CMORPH BLD BLDisisdisplayed.
 displayed.

 Ananalysis
 An analysis using point gauge data
 datawas also performed with thethe results shown in Figure 9. The
 An analysis usingpoint pointgauge
 gauge data wasalso
 was also performed
 performed withwiththe resultsresults
 shownshown in Figurein 9.Figure
 The 9.
 MAE
The MAE is the
 MAE isis thesmallest
 the smallest in SON
 smallestininSON and
 SONand largest
 andlargest in DJF
 largestininDJF
 DJF where
 where
 where the
 thethe error
 error
 error is approximately
 is approximately
 is approximately 50%
 50% greater. greater.
 greater.
 Normalisation
Normalisation of the errors results in the smallest relative error occurring in DJF
 Normalisation of the errors results in the smallest relative error occurring in DJF and the largest in in
 of the errors results in the smallest relative error occurring in DJF and
 and the
 the largest
 largest in
 MAM
MAM MAM and
 andand JJA,
 JJA,JJA,supporting
 supporting
 supporting the gridded
 thethegridded comparison.
 griddedcomparison. The linear
 Thelinear
 comparison. The correlation
 linearcorrelation
 correlation coefficients
 coefficients
 coefficients across
 across
 across the
 the the
 seasons
seasons also
 seasons
 also suggest
 also suggest
 suggest that
 that
 that DJF
 DJFDJF has
 hashas the
 thethe best
 best
 best performance across
 performanceacross
 performance across the
 the
 the seasons.
 seasons.
 seasons. The
 TheThe performance
 performance
 performance increase
 increase
 increase is
 is more
more prominent
 isprominent
 more prominent inin
 in the the
 thenon-gauge
 non-gauge
 non-gauge blended
 blended
 blended datasets,
 datasets,
 datasets, where
 where
 where the the
 the improvement
 improvement
 improvement atisleast
 is at
 is at least
 least 10%.
 10%. 10%.

 Figure 9. Cont.

Remote Sens. 2020, 12, x FOR PEER REVIEW 11 of 18

Remote Sens. 2020, 12, 678 11 of 17
 Remote Sens. 2020, 12, x FOR PEER REVIEW 11 of 18

 Figure 9. Point-based comparison categorised by seasons from January 2001 to December 2018 using
 Figure 9. Point-based
 Figure 9. Point-basedcomparison
 comparisoncategorised
 categorised by
 by seasons
 seasons fromJanuary
 from January2001
 2001 to December 2018 using
 station gauges as truth. Mean average error (MAE), normalised MAE, and to December
 R are 2018
 displayed. using
 station gauges
 station as truth.
 gauges Mean
 as truth. average
 Mean averageerror
 error(MAE),
 (MAE), normalised MAE,and
 normalised MAE, andR R are
 are displayed.
 displayed.

 Overall, the
 Overall, the performance
 performanceappears
 appearstotobe
 berelatively
 relativelysimilar
 similar across the seasons withthethe exception
 Overall, the performance appears to be relatively similaracross
 across the
 the seasons with
 seasons with the exception
 exception of
of DJF,
DJF, of which
 which shows
 showsshows
 DJF, which
 a somewhat
 a somewhat superior
 superior
 a somewhat
 performance
 performance
 superior performance
 to
 to the the rest.
 rest.
 to the rest.

3.3. Variations with
 3.3. Variations Rainfall
 with Intensity
 Rainfall Intensity
 TheThe
 The effects
 effects ofofthe
 effects the intensity
 of intensity
 the ofof
 intensity ofthe
 the rainfall
 rainfall
 the onon
 rainfall thethe
 on accuracy
 accuracy
 accuracy ofthe
 of of
 the the
 data data
 datawerewere
 werealso also analysed.
 analysed.
 also analysed. The The
 Thedata
were categorised into these bins: 0–0.2, 0.2–1, 1–2, 2–3, 3–4, 4–6, 6–9, and >9 mm/day. These rainfall
 data were
 data categorised
 were categorised into these
 into thesebins:
 bins:0–0.2,
 0–0.2, 0.2–1,
 0.2–1, 1–2,
 1–2, 2–3,
 2–3, 3–4,
 3–4, 4–6,
 4–6, 6–9,
 6–9, and
 and >9 >9 mm/day.
 mm/day. TheseThese
rangesrainfall
 rainfall were ranges
 ranges were
 chosen were
 to chosen
 chosen
 ensure to ensure
 tothere
 ensure
 were there
 there were a reasonable
 were
 a reasonable reasonable
 amount amount
 amount
 of values ofofvalues
 in eachin
 values each
 in
 bin binthe
 each
 with with
 bin the the
 with
 values of
0.2 values
 values
 and of
 1 mmof 0.2
 0.2 and and
 being1 mm1 mm beingspecifically
 being
 specifically specifically chosen
 chosen as chosen
 they as they
 as they correspond
 correspond correspond totothe
 to the rainy-day rainy-day
 the rainy-day
 thresholdthreshold forand
 threshold
 for BoM for
aBoM BoM
 andand
 commonly a commonly
 a commonly
 used inused
 value used valueinincontingency
 value
 contingency contingency
 statistics statistics
 statistics
 studies, studies,
 studies,respectively
 respectively respectively [15].
 [15]. Continuous Continuous
 [15]. Continuous
 statistics for
 statistics
 statistics for these bins were calculated along with a comparison of occurrence frequencies and and
these binsfor
 were these bins were
 calculated alongcalculated along with
 with a comparison a comparison
 of occurrence of occurrence
 frequencies frequencies
 and cumulative volumes.
 cumulative volumes. These are shown in Figure 10.
 cumulative
These volumes.
 are shown These10.
 in Figure are shown in Figure 10.

 Figure 10. Point-based comparison categorised by rainfall intensities from January 2001 to December
 2018 using station gauges as truth. Occurrence frequencies, cumulative volumes, mean average error
 (MAE), and normalised MAE are displayed.

Remote Sens. 2020, 12, x FOR PEER REVIEW 12 of 18

Remote Sens. 2020, 12, 678
 Figure 10. Point-based comparison categorised by rainfall intensities from January 2001 to December 12 of 17
 2018 using station gauges as truth. Occurrence frequencies, cumulative volumes, mean average error
 (MAE), and normalised MAE are displayed.
 The datasets appear to capture the correct frequency best for rainfall amounts between 3 and
6 mm/day. For higher
 The datasets amounts,
 appear the the
 to capture satellite-derived
 correct frequency data underestimate
 best the frequency,
 for rainfall amounts between while
 3 and 6for
lower amounts,
 mm/day. the frequency
 For higher amounts, is the
 underestimated
 satellite-derived for data
 veryunderestimate
 low values (

Remote Sens. 2020, 12, x FOR PEER REVIEW 13 of 18

 Figure 11. Gridded comparison of satellite datasets and AWAP against ERA5 reanalysis from January
 2001 to December 2018. Mean bias (MB), root-mean-squared error (RMSE), mean average error
Remote Sens. 2020, 12, 678 13 of 17
 (MAE), R and normalised MAE are displayed.

 Figure
 Figure 12. Gridded comparison
 12. Gridded comparison ofof satellite
 satellite datasets
 datasets and
 and AWAP
 AWAP against
 against ERA5
 ERA5 reanalysis
 reanalysis from
 from January
 January
 2001
 2001 to
 to December
 December 2018. Quintile comparison
 2018. Quintile comparison is is displayed.
 displayed.

 The
 The results
 results of of the
 the error
 error analysis
 analysis of of the
 the gridded
 gridded comparison
 comparison are are supported
 supported by by the
 the point-based
 point-based
comparison
comparison where both satellite datasets and AWAP were compared to station gauges with
 where both satellite datasets and AWAP were compared to station gauges with the errors
 the errors
in AWAP being smaller but still within the same order of magnitude
in AWAP being smaller but still within the same order of magnitude as those from the satellite as those from the satellite dataset.
This highlights
dataset. the caution
 This highlights theneeded
 cautionin understanding
 needed that the gridded
 in understanding comparison
 that the gridded results are
 comparison unlikely
 results are
to be a proper depiction of the true error of the satellite
unlikely to be a proper depiction of the true error of the satellite datasets. datasets.
 There
 There areare certain
 certain regions
 regions where
 where the the performance
 performance of of satellite
 satellite rainfall
 rainfall detection
 detection is is decreased.
 decreased. Past Past
studies
studies have
 have indicated
 indicated that that the
 the detection
 detection of of cold
 cold frontal-based
 frontal-based rainfall
 rainfall is is poor
 poor [15,17].
 [15,17]. TheThe absence
 absence of of
ice
ice crystals
 crystals inin the relatively low
 the relatively low precipitating
 precipitating cloudsclouds typically
 typically associated
 associated with with frontal
 frontal rainfall
 rainfall hinders
 hinders
the
the ability
 ability ofof satellites
 satellites to to detect
 detect rainfall
 rainfall via via scattering
 scattering [15]. This is
 [15]. This is aa likely
 likely factor
 factor behind
 behind the the large
 large errors
 errors
over Tasmania, Western Australia, South Australia, and central Australia,
over Tasmania, Western Australia, South Australia, and central Australia, areas where the prevalent areas where the prevalent
rainfall
rainfall generation
 generation mechanism
 mechanism is is cold
 cold frontal
 frontal systems. Errors are
 systems. Errors are pronounced
 pronounced over over thethe western
 western halfhalf
of
of Tasmania
 Tasmania and and the
 the southwestern
 southwestern coast coast of of Western
 Western Australia,
 Australia, areas
 areas of of relatively
 relatively high
 high rainfall
 rainfall duedue to
 to
increased
increased exposure
 exposure to to westerly
 westerly flowflow and and associated
 associated coldcold fronts.
 fronts.
 Performance
 Performance is is also
 also known
 known to to be
 be decreased
 decreased over over topography
 topography [18,21].
 [18,21]. Decreased
 Decreased performance
 performance is is
observed
observed along
 along the the eastern
 eastern coastline
 coastline near near the
 the Great
 Great Dividing
 Dividing Range.
 Range. The The errors
 errors are
 are greatest
 greatest alongalong thethe
northern
northern NSWNSW coastline
 coastline and and thethe Australian
 Australian Alps Alps where
 where thethe Great
 Great Dividing
 Dividing RangeRange is is atat its
 its highest
 highest
elevations, leading to a strong orographic influence
elevations, leading to a strong orographic influence on rainfall. on rainfall.
 A high-qualityrain
 A high-quality raingauge
 gauge network
 network is extremely
 is extremely valuable
 valuable for improving
 for improving the accuracy
 the accuracy of
 of satellite-
satellite-derived rainfall estimates as satellite estimates rely on gauges to
derived rainfall estimates as satellite estimates rely on gauges to calibrate or correct their raw values.calibrate or correct their raw
values. The significantly
The significantly greatergreater
 number number of gauges
 of gauges towards
 towards thethe coastline
 coastline wheremost
 where most of of Australia’s
 Australia’s
population
population resides allows for a much greater improvement from gauge correction in contrast to
 resides allows for a much greater improvement from gauge correction in contrast to the
 the
interior of the continent. Consequently, areas towards the coastlines
interior of the continent. Consequently, areas towards the coastlines that experience problematic that experience problematic
regimes
regimes such
 such as cold-frontal rainfall
 as cold-frontal rainfall and and orographically
 orographically influenced
 influenced rainfall
 rainfall greatly
 greatly benefit
 benefit fromfrom gauge
 gauge
correction,
correction, resulting in a performance similar to unproblematic regimes. However, the lack of
 resulting in a performance similar to unproblematic regimes. However, the lack of rain
 rain
gauges
gauges towards the interior means there are still large normalised errors in this region, even in the
 towards the interior means there are still large normalised errors in this region, even in the
gauge-corrected
gauge-corrected dataset.dataset. ThisThis is is compounded
 compounded by by the
 the tendency
 tendency of of rainfall
 rainfall toto be
 be lighter
 lighter towards
 towards the the
interior
interior compared
 comparedtotothe thecoast
 coast as as
 light rainfall
 light has has
 rainfall beenbeen
 shown to beto
 shown a problematic
 be a problematicregimeregimeas well as[17,18].
 well
 Low mean rainfall is another factor that would contribute to a large normalised MAE. Some
[17,18].
areasLow
 of large
 mean normalised
 rainfall is MAE another around
 factorthe interior
 that wouldofcontribute
 the continent to a can
 large benormalised
 seen to generallyMAE. Some align
with
areasareas of low
 of large mean rainfall
 normalised MAEvalues,aroundasthe seen in Figure
 interior 13, continent
 of the which depicts can betheseen
 seasonal mean rainfall
 to generally align
across Australia. This is especially true during the austral winter ‘dry’ season
with areas of low mean rainfall values, as seen in Figure 13, which depicts the seasonal mean rainfall for central Australia and
northwards towards the Northern Territory coast.
across Australia. This is especially true during the austral winter ‘dry’ season for central Australia
and northwards towards the Northern Territory coast.

Remote Sens. 2020, 12, 678 14 of 17
Remote Sens. 2020, 12, x FOR PEER REVIEW 14 of 18

 Figure
 Figure13.
 13.Mean
 MeanCMORPH
 CMORPHBLD
 BLDrainfall
 rainfallby
 byseason
 seasonfrom
 fromJanuary
 January2001
 2001totoDecember
 December2018.
 2018.

 The
 Theimportance
 importanceofofgauge gaugecorrection
 correctionisisreduced
 reducedfor for unproblematic
 unproblematic regimes.
 regimes. ForForexample,
 example,the the
normalised
 normalised errors for forthethecorrected
 corrected andand uncorrected
 uncorrected datasets
 datasets around around the northern
 the northern coastlinecoastline of
 of Australia
Australia are relatively similar, highlighting how the raw satellite algorithms exhibit decent
 are relatively similar, highlighting how the raw satellite algorithms exhibit decent performance in these
performance
 areas, leading intothese
 gauge areas, leadingbeing
 correction to gauge correction
 less crucial. being less crucial.
 Tropical-based rainfallTropical-based
 has been notedrainfall
 to be onehasof
been noted to be one ofregimes
 the better-performing the better-performing regimes
 for satellite rainfall for satellite
 detection [15]. rainfall detection [15].
 Satellite-derived
 Satellite-derivedprecipitation
 precipitationestimates
 estimates for for austral
 austral winter demonstrate the the worst
 worst performance,
 performance,a
aresult
 resultthat
 thatagrees
 agreeswith with pastpast studies
 studies [15].
 [15]. TheThe difficulty
 difficulty of detecting
 of detecting cold-frontal
 cold-frontal rainfall,
 rainfall, which which is
 is more
more
 frequentfrequent
 during during
 winter,winter,
 is mostislikely
 mosta keylikely a key
 factor. The factor. The introduction
 introduction of snowchallenge
 of snow is another is another for
challenge for satellite
 satellite detection detection of
 of precipitation andprecipitation and is likely
 is likely a contributing a contributing
 factor factor to the
 to the poor performance poor
 observed
performance
 in western Tasmaniaobserved and inthe
 western Tasmania
 Australian Alps. By and the Australian
 contrast, the greater Alps. By contrast,
 prevalence the greater
 of convective-based
prevalence of convective-based
 rainfall in summer is a reason for rainfall in summer
 this season is a reason
 performing the best for[15,17].
 this season performing the best
[15,17].An overestimation (underestimation) of low (high) rainfall rates was observed and is consistent
 withAn pastoverestimation
 literature [19,20]. (underestimation) of low (high) rainfall rates was observed and is consistent
with past
 It is literature
 natural to[19,20].
 expect that CMORPH BLD (a gauge-blended dataset) should have at least equal
 It is natural
 performance to to expect(athat
 AWAP CMORPHanalysis)
 gauge-based BLD (a gauge-blended
 as it relies on dataset) shouldwhere
 using gauges have at
 theleast
 dataequal
 exist
performance
 whilst depending to AWAP more(a gauge-based
 heavily analysis)
 on satellites where asthere
 it relies on using
 is little gaugesdata.
 to no gauge where the datathe
 However, exist
 key
whilst depending
 assumption here is more
 that heavily
 satelliteon satellites
 depiction ofwhere
 rainfallthere is littleto
 is superior tointerpolation
 no gauge data. However,
 methods the key
 in areas with
assumption here This
 little to no data. is that satellite
 is not depiction
 necessarily true,of as,rainfall is superior
 even though to interpolation
 satellites methods
 are sourcing their in areasa
 data through
with littlesensor,
 physical to no data. This is still
 this process not necessarily
 relies heavily true, as, even though
 on calibration to rainsatellites are sourcing
 gauge data. theirwhere
 For locations data
through a physical
 there is little sensor,data,
 to no gauge thiscalibration
 process still and,relies heavily onperformance
 subsequently, calibration will
 to rain gauge data.
 be severely For
 hindered.
locations
 Furthermore, whereAWAP there usedis little to no gauge
 a minimum data, of
 number calibration and, subsequently,
 stations exceeding 3000 whileperformance will be
 the satellite datasets
severely hindered.
 are calibrated Furthermore,
 to the CPC Unified AWAP
 gaugeused a minimum
 analysis, which has number of stations
 a minimum exceeding
 number 3000 while
 of stations across
the satelliteatdatasets
 Australia least anare calibrated
 order to the CPC
 of magnitude less Unified
 than that gauge
 of AWAPanalysis, which
 [29,30]. Thehas a minimum
 ingestion number
 of less data is
oflikely
 stations across Australia
 to contribute at least an observed.
 to the discrepancies order of magnitude less than that of AWAP [29,30]. The
ingestion of less data is likely to contribute to the discrepancies observed.