High-resolutioninputsofrainfallareimportantinhydrologicalsciences, especiallyforur- ban hydrology. This is mainly because heavy rainfall-induced events such as flash floods can have a tremendous impact on society given their destructive nature and the short time scales in which they develop. The spatiotemporal resolutions at which rainfall can be retrieved are becoming higher and higher, with the development of technologies like radars, satellites and (commercial) microwave links (CML). Such high resolutions provide not only more accurate rainfall estimates but also allow a better understanding of its spa- tiotemporal distribution
For the land surface of the Netherlands, we evaluate here four rainfall products, i.e., link- derived rainfall maps, IMERG Final Run (Integrated Multi-satellitE Retrievals for GPM - Global Precipitation Measurement mission), NIPE (Nighttime Infrared Precipitation Esti- mation), and MSG CPP (Meteosat Second Generation Cloud Physical Properties). All rain- fall products are compared against gauge-adjusted radar data, considered as the ground truthgivenitshighquality,resolutionandavailability. Theevaluationisdonefor7months
at30min and24h. Overall, we found that link-derived rainfall maps outperform the sa-
tellite products, and that GPM outperforms CPP and NIPE.
WealsoexplorethepotentialofaCMLnetworktovalidatesatelliterainfallproducts. Usu- ally, satellite derived products are validated against radar or rain gauge networks. If data from CML would be available, this would be highly relevant for ground validation in areas with scarce rainfall observations, since link-derived rainfall is truly independent from sa- tellite-derived rainfall. The large worldwide coverage of CML potentially offers a more extensive platform for the ground validation of satellite estimates over the land surface of the Earth.
This chapter was submitted after revision toIEEE Trans. Geosci. Remote Sens.as:
M.F. Rios Gaona, A. Overeem, N. Brasjen, J.F. Meirink, H. Leijnse, and R. Uijlenhoet, “Evaluation of Rainfall Pro- ducts Derived from Satellites and Microwave Links for the Netherlands,” Mar. 2017.
3.1
Introduction
Rainfall is the key input in environmental applications such as hydrological modeling, flash-flood and crop forecasting, landslide triggering, waterborne disease propagation, and fresh water availability. Rainfall is the product of hydrometeorological processes that extensivelyvaryintimeandspaceHou et al.(2008);Sene(2013a). Itisextremelydifficultto
capture its inherent variability at high spatiotemporal resolutions, i.e., in the order of mi- nutes and kilometers. The requirements of spatiotemporal resolutions are related to the applicationsforwhichrainfallinputsaredeemed,forinstance,urbanvs. ruralcatchments
Schilling(1991). In urban hydrology, rainfall-related events such as flash floods drastically
impact society at short time scales. Thus, temporal resolutions of30min or less and spa-
tial resolutions of10km or less are necessaryBerne et al.(2004). Over history, devices and
techniques have been developed not only to measure or estimate rainfall quantities but also to fill the spatiotemporal gaps in the data collected.
Rain gauges are the only devices that directly measure rainfall. They are a relatively low- cost technology due to their compact size and simplicity; consequently, large networks of rain gauges are found in many places over the land surface of the EarthBecker et al.(2013).
Ifwellmaintained,measurementsfromraingaugesareaccurate. However,theyarepoint measurements, only representative of the area in their direct vicinity. The spatial variabi- lity of rainfall is not well captured by gauges given their sparsity, especially over the oce- ans, where practically no gauges are placed. Often, depending on the type of gauge, it is onlypossibletocollectdailyrainfallvalues, whichisnotdesirableifhourly(orsub-hourly) analyses are needed. Yet still, rain gauges are the main and most reliable devices to vali- date more advanced platforms like weather radars and meteorological satellites. Weather radar has greatly improved the spatiotemporal resolution of rainfall retrievals. After World War II radar (RAdio Detection And Ranging) technology became popular for meteorological purposesRogers and Smith(1996);Atlas and Batan(1990). Weather radar
measures the power of the backscatter (reflection) produced by hydrometeors encounte- red by an electromagnetic pulse along its propagation path. The backscatter produced by raindrops can later be processed and transformed into precipitation intensity. Hence, rainfall is estimated for any given volume that radar scans up in the atmosphere. In ge- neral, weather radars scan azimuths (0◦−360◦) with an angular resolution of∼1◦, and
reach distances of∼100−300km with horizontal resolutions of∼1km. Such scans are
also done for several elevations (0◦−90◦) with regard to the horizon. Radar retrieves in-
formation from all directions with a typical time resolution of5min.
Satellite constellations provide global estimates of rainfall on a quasi real-time basis. Sa- tellite rainfall estimates are obtained from remotely sensing the radiation emitted or re- flected by atmospheric hydrometeors. This can be done by several techniques which de- pendonthetypeofsensorandtherangeofthespectrumtheysenseKidd et al.(2010): 1)Vi-
sible (VIS) and infrared (IR) methods−in which measurements of reflectance and brig- htness temperature, or retrieved cloud properties such as water path, particle effective radius, and height are related to occurrence and intensity of precipitation; 2) Passive Mi-
3.1. Introduction
crowave methods (PMW)−in which liquid or solid precipitation is identified through measurements of microwave radiation transmitted by both hydrometeors and the Earth, which is partly scattered and absorbed by hydrometeors; 3) Active Microwave methods (AMW)−inwhichrainfallinformationisretrievedfrompowermeasurementsoftheback- scatter produced by the interaction of hydrometeors and electromagnetic pulses emitted by satellite-bornesensors, namely radars; 4) Multisensor techniques−in whichthe com- bination of data collected within a satellite constellation overcomes some of the deficien- cies of the above techniques, for instance, temporal sampling, spatial resolution, and in- direct measurements. An example of the latter is the Global Precipitation Measurement mission(GPM),whichistherecentupgradeoftheTropicalRainfallMeasurementMission (TRMM). The GPM constellation consists of9satellites which offer global precipitation
estimates between60◦N−60◦S at a spatial resolution of0.1◦×0.1◦every30min, with
a minimum latency of6h, i.e., in near-real time.Hou et al.(2014);Skofronick-Jackson et al.
(2013) give a detailed account of all the technicalities of the GPM mission and its core ob- servatory launched on27February2014.
Meteorological satellites can be divided into two categories: Geostationary Earth Orbit (GEO), and Low Earth Orbit (LEO or polar). GEO satellites orbit the Earth at∼36,000km,
and maintain a fixed position to match Earth's rotation; thus, always the same area is scanned. Their instrumentation typically includes VIS and IR sensors, which provide ob- servations at resolutions of∼10−30min (or hourly), and1−4kmSene(2013b);Wang
(2013). ExamplesofGEOsatellitesarethosepartofNOAAGOES(NationalOceanicandAt- mospheric Administration - Geostationary Operational Environmental Satellite system); and EUMETSAT (European Organisation for the Exploitation of Meteorological Satelli- tes) with the MSG, which carries the Spinning Enhanced Visible and InfraRed Imager (SE- VIRI) aboard. LEO satellites orbit the Earth mostly in polar (sun-synchronous) orbits at ∼800km(407kmforGPM'scoreobservatory). Theresolutionoftheirobservationsstron-
gly depends on the type of sensor. For instance, LEO VIS-NIR (near IR) imagers typically have∼1kmorbetter; whereasPMWimagersarecoarser, e.g., GMI(GPMMicrowaveIma-
ger) is∼6×13km2(depends on the channel). Since TRMM, AMW sensors such as preci-
pitation radar (PR) or DPR (dual PR) are also placed on meteorological satellites. Exam- ples of LEO satellites are those part ofEUMETSAT MetOp and NOAA-N NASA (National Aeronautics and Space Administration) Aqua and Terra, which carries the MODerate re- solution Imaging Spectroradiometer (MODIS) aboard. For more in-depth details on the state-of-the-art of rainfall estimation from satellites seeKidd et al.(2010) and references
therein.
CML are a promising alternative for rainfall estimation, especially for places in which rain gauges are scarce or poorly maintained, or where ground-based weather radars are not yet deployed or cannot be affordedDoumounia et al.(2014). Rainfall rates can be retrie-
ved from power measurements because rain drops attenuate the electromagnetic signal along the link path, i.e., between transmitter and receiver. Rainfall estimation via ML da- tes back nearly four decadesAtlas and Ulbrich(1977);Crane(1977), but with the worldwide
spread of wireless communication in the last two decades, this technique has become in- creasingly popular.Giuli et al.(1991) had previously reconstructed rainfall fields from si-
mulated microwave attenuation measurements. Nonetheless,Messer et al.(2006);Leijnse
et al.(2007a) were the first to actually use CML to estimate rainfall rates. Schleiss et al.
(2013);Chwala et al.(2014) developed methods to improve rainfall estimates from link
measurements.Overeem et al.(2013);Zinevich et al.(2008) advanced in link rainfall maps
for large datasets. When compared to satellite, weather radar or even rain gauges, links may outperform them in one or more of the following areas: they are just tens of meters above the ground, their spatial resolution is in the order of hundreds of meters to kilo- meters, and their temporal sampling is in the order of seconds to minutes. The massive availability of CML can potentially provide complementary rainfall information. Never- theless, the success of this technology depends on the availability of received signal level (RSL) measurements from the providers of cellular communication.
We divide our study in two parts: 1) evaluation of satellite- and link-derived rainfall pro- ducts, and2)thepotentialofaCMLnetworktovalidatesatelliterainfallestimates. Forthe first part, we evaluate the accuracy of four gridded rainfall products: Cloud Physical Pro- perties (CPP) and the experimental Nighttime Infrared Precipitation Estimation (NIPE), both from the GEO platform Meteosat Second Generation (MSG); Integrated Multi-satel- litE Retrievals for GPM (IMERG), of which the core satellite is a LEO platform; and rainfall maps from a Dutch CML network. The evaluation is done through pixel-by-pixel compa- risons against radar rainfall maps at half-hourly (30min) and daily (24h) durations from 1April2014to1November2014, i.e.,7months. Radar rainfall maps are considered here
to be the ground truth. For the second part, we focus our analysis on the spatiotempo- ral availability of a Dutch CML network. Once we have established the accuracy of link rainfall estimates (first part), we are able to evaluate their performance against satellite estimates. We then assess the potential of a CML network as ground validation for satel- lite products for different link densities. This is of relevance due to the very high coverage of CML over the Earth's land surface, also in areas where no other infrastructure to mea- sure rainfall is availableOvereem et al.(2013).
This chapter is organized as follows: Section 3.2 describes the rainfall datasets, jointly with the evaluation metrics, and the methodology to assess the spatiotemporal perfor- mance of a Dutch CML network. Section 3.3 presents the evaluation of satellite and link- derived rainfall products. Results from the spatiotemporal analysis of a Dutch CML net- work are also presented in section 3.3. Section 3.4 highlights our major findings. Sum- mary, conclusions and recommendations are provided in Section 3.5.
3.2
Measurements and Methods
The evaluation period is from0000UTC1April2014to0000UTC1November2014. This
period is limited by the availability of GPM data for the start date, and CML data for the end date.
3.2. Measurements and Methods