880
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 54
Daily Rainfall Detection and Estimation over Land Using Microwave Surface Emissivities CAMILLE BIRMAN AND FATIMA KARBOU CNRM/GAME, UMR 3589, CNRS/Météo-France, Saint Martin d’Hères, France
JEAN-FRANÇOIS MAHFOUF CNRM/GAME, UMR 3589, CNRS/Météo-France, Toulouse, France (Manuscript received 24 June 2014, in final form 16 January 2015) ABSTRACT Surface emissivities computed at 89 GHz from AMSU-A, AMSU-B, and SSMI/S instruments are used to detect rain events and to estimate a daily precipitation rate over land surfaces. This new retrieval algorithm, called the emissivity rainfall retrieval (EMIRR) algorithm, is evaluated over France and compared with several other precipitation products. The precipitation detection is performed using temporal changes in daily surface emissivities. A statistical fit, derived from a rainfall analysis product using rain gauge and radar data, is devised to estimate a daily precipitation rate from surface emissivities. Rain retrievals are evaluated over a 1-yr period in 2010 against other precipitation products, including rain gauge measurements. The EMIRR algorithm allows a reasonable detection of rainy events from daily surface emissivities. The number of rainy days and the daily rainfall rates compare well to estimates from other precipitation products. However, the algorithm tends to overestimate low precipitation amounts and to underestimate higher ones, with reduced performances in the presence of snow. Despite such limitations, this new method is very promising and provides a demonstration of the potential use of the 89-GHz surface emissivities to infer relevant information (occurrence and amounts) related to daily precipitation over land surfaces.
1. Introduction Rainfall is a key parameter for many applications related to meteorology, hydrology, and climate. Given the fact that many areas of the globe are not covered by surface networks of rain gauges or radars, rainfall retrievals from satellite data are currently the only means of providing global coverage at various spatial and temporal scales. During the last 30 years, there has been significant progress made in the use of satellite remote sensing for the detection and quantitative estimation of surface precipitation. The scientific community is organized within the International Precipitation Working Group (http://www.isac.cnr.it/;ipwg) that allows exchanges on existing techniques, validation studies, and intercomparison projects. The retrieval of rainfall area and rainfall rate from satellite measurements can be
Corresponding author address: Camille Birman, CNRM/GAME, UMR 3589, CNRS/Météo-France, 1441 rue de la Piscine, 38400 Saint Martin d’Héres, France. E-mail:
[email protected] DOI: 10.1175/JAMC-D-14-0192.1 Ó 2015 American Meteorological Society
achieved through a variety of methods using visible/ infrared and/or microwave (both active and passive) observations, and a number of operational products are available such as the Climate Prediction Center morphing method (CMORPH; Joyce et al. 2004), the Global Precipitation Climatology Project (Adler et al. 2003), or Precipitation Estimation from Remotely Sensed Information Using Artificial Neural Networks (PERSIANN; Sorooshian et al. 2000). However, each method has limitations coming either from the observations themselves or from the assumptions of the retrieval technique. For example, it has been shown that the microwave region of the electromagnetic spectrum is the most suitable one for deriving quantitative information on instantaneous precipitation whereas combined techniques using visible, infrared, and microwave regions lead to the best daily and monthly estimates (Kidd and Levizzani 2011; Tapiador et al. 2012). In the microwave region, rainfall satellite retrievals are easier to perform over ocean surfaces that have a rather low surface emissivity (close to 0.5). Indeed, at low frequencies (below 35 GHz), the emission from liquid precipitation enhances
APRIL 2015
BIRMAN ET AL.
the low background radiances at the top of the atmosphere. Since microwave instruments are only available on board low-orbiting satellites, they suffer from rather poor temporal sampling. Over oceans, microwave observations have been used to detect precipitating events and to retrieve rain products (e.g., Hilburn and Wentz 2008; Funatsu et al. 2007). Over land surfaces, rainfall retrieval from microwave radiances can only been achieved through the scattering signal produced at high frequencies (above 35 GHz) by large liquid or frozen hydrometeors (Bennartz et al. 2002). This indirect signature of rainfall is thus identified by a reduction of the radiances at the top of the atmosphere. A number of scattering-based techniques have been proposed to detect precipitation over continents (e.g., Spencer et al. 1989; Ferraro and Marks 1995). Some empirical algorithms have been devised based on the construction of an a priori database (e.g., Petty and Li 2013), but most of the microwave retrieval methods that have been developed in order to derive quantitative precipitation over land use observations along with radiative transfer models (Defer et al. 2008; DiTomaso et al. 2009). An important source of uncertainty of such methods lies in the specification of scattering properties of frozen hydrometeors in the radiative transfer model (Bennartz and Petty 2001; Geer and Baordo 2014). An alternative is to consider land surface properties instead of top-of-the-atmosphere radiances. Such an approach has been scarcely used for rain-rate retrievals, except by Brocca et al. (2013, 2014) using soil moisture time series to estimate rainfall rates, or Crow et al. (2009, 2011) and Pellarin et al. (2008, 2013), who use satellite-derived soil moisture data for rainfall correction. The present work aims at retrieving useful information on surface precipitation over continents from the signature of surface properties in the microwave region. Even though such an approach is still indirect, it allows us to bypass the direct simulation of the radiative transfer model in scattering atmospheres. More specifically, we study the potential of Advanced Microwave Sounding Unit-A (AMSU-A), and -B [AMSU-B or Microwave Humidity Sounder (MHS)] and the Special Sensor Microwave Imager/Sounder (SSMI/S) microwave emissivities derived at 89 GHz over land surface to detect rainfall events and to retrieve daily rainfall rates over France, where independent data are available to calibrate and evaluate the performance of the method. The use of a multisensor approach in the microwave reduces sampling issues associated with low-Earth-orbiting satellites. The paper is organized as follows. In section 2 the data and models used are presented. In section 3, a new precipitation retrieval algorithm called the emissivity rainfall retrieval (EMIRR) algorithm is described and its performance assessed against several other rainfall products. Section 4 provides the main conclusions of the study.
881
2. Data and models The EMIRR algorithm proposed in this paper was developed over France and established over a 1-yr period in 2010. This particular location, where a number of precipitation products already exist, has been chosen to facilitate the calibration and the evaluation of the algorithm, which could afterward be applied over other regions of the globe. The algorithm consists of producing gridded fields of daily rainfall amounts and occurrences at 0.258 resolution.
a. Land surface emissivity Satellite brightness temperatures measured by AMSU-A, AMSU-B, and SSMI/S instruments are considered in this study. These sensors are passive microwave radiometers on board several polar-orbiting satellites. AMSU-A has 15 channels, among which 12 are located near the 50– 60-GHz oxygen absorption band and provide information on atmospheric temperature. AMSU-B (or MHS) is a humidity sounder with five channels, three of which are situated near the 183-GHz strong water vapor absorption line. The SSMI/S sensor is carried on the latest generation of the Defense Meteorological Satellite Program (DMSP) satellites. This instrument provides information on atmospheric temperature and humidity using a conical scanning technique by making measurements at 24 frequencies, 14 of them being located in the oxygen band near 50–60-GHz range. SSMI/S is also informative about atmospheric humidity with frequencies close to 183 GHz. In addition to sounding channels, AMSU and SSMI/S instruments have so-called window channels that are often used for observing the surface (temperature, humidity, sea wind, etc.) and for cloud detection. In this study, land surface emissivity estimates were chosen rather than observed brightness temperatures Tb , thus lessening the atmospheric contribution to the radiometric signal and therefore bypassing the complexity of modeling cloudy atmospheres. Under several assumptions (nonscattering plane-parallel atmosphere, specular surface), the Tb observed by a microwave sensor for a given zenith angle u and frequency n can be expressed as Tb (n, u) 5 Ts «(n, u)G 1 [1 2 «(n, u)]GTaY (n, u) 1 Ta[ (n, u) 2t(0, H) G 5 exp , cos(u) (1) where «(n, u) represents the surface emissivity and Ts , TaY (n, u), and Ta[ (n, u) are the skin temperature and the atmospheric downwelling and upwelling radiances, respectively. The net atmospheric transmissivity G is expressed as a function of the atmospheric opacity
882
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
t(0, H) and the observation zenith angle u. The height of the top of atmosphere is H. Emissivity can be retrieved using Eq. (1) by «(n, u) 5
Tb (n, u) 2 Ta[ (n, u) 2 TaY (n, u)G [Ts 2 TaY (n, u)]G
.
(2)
The contribution of the atmosphere to the measured signal is estimated using the radiative transfer model RT TOV (Eyre 1991; Saunders et al. 1999; Matricardi et al. 2004) in clear-sky conditions fed by short-range forecasts (6 h) of surface temperature, air temperature, and humidity from the French global numerical weather prediction model ARPEGE (Courtier et al. 1991). More details about the land surface retrieval and evaluation can be found in Karbou et al. (2006), Choudhury (1993), Felde and Pickle (1995), and Prigent et al. (1997). In Karbou et al. (2005), a sensitivity study was conducted to estimate the AMSU emissivity sensitivity to errors in the input parameters (brightness temperature, air temperature, air moisture, and skin temperature). The sensitivity was inferred by looking at the emissivity change due to a variation in one of the input parameters, the other parameters remaining unchanged. The standard deviations remain low for all the parameters. The sensitivity of emissivity to errors in the skin temperature was evaluated by calculating the emissivity variation due to 64 K in the skin temperature. The latter study shows that, for 89 GHz, errors in humidity profiles can have a nonnegligible impact on the calculated emissivity (up to 1%). Errors in the skin temperature influence the computed emissivity with emissivity relative error of 3.5% in dry conditions. The effect of air temperature on emissivity retrieval at 89 GHz is quite small. The main errors are induced by uncertainties on surface temperature; however, studies on sea ice showed that we can greatly improve the assimilation of microwave observations, sensitive to air temperature and moisture, over polar regions using emissivity estimates at 89 GHz despite a relatively poor knowledge of the surface temperature (Karbou et al. 2014). Emissivity at 89 GHz varies according to sea ice types, and this property was used in order to improve the assimilation of AMSU-B– MHS observations over polar regions with a positive impact on analyses and forecasts (more details in Karbou et al. 2014). As the emissivity is retrieved in all sky situations assuming model clear-sky conditions, the quantity derived from Eq. (2) can be considered as an ‘‘effective emissivity’’ that includes the atmospheric contribution in cases with cloudy/rainy conditions. When precipitation is present in the atmospheric profile, the effective emissivity will be lower than the actual value since the
VOLUME 54
scattering by frozen hydrometeors will decrease the brightness temperature at the top of the atmosphere [as shown in Eq. (2)]. This decrease with respect to clear-sky situations can be interpreted as a consequence of precipitating hydrometeors in the field of view of the satellite. The changes in surface emissivity are induced by combined effects of surface and atmospheric changes. Rainfall produces a decrease in the emissivity due to soil moisture, but also to scattering by frozen hydrometeors in the atmosphere taken into account in the effective emissivity retrieved using the radiative transfer model RT TOV in clear-sky conditions. Recently, Turk et al. (2014) developed a similar technique for detecting the presence of precipitation, using a principal component analysis instead of a radiative transfer model to retrieve microwave land surface emissivities. One can also notice that after a rainy event, the soil being wet and the vegetation covered by liquid water, these effects can also contribute to lowering the surface emissivity, as shown by Calvet et al. (2011) for a wide range of microwave frequencies. For AMSU-A and AMSU-B window channels, the polarization is assumed to be vertical at nadir and the emissivity is observed with a mixed polarization. For SSMI/S the emissivity is observed with both vertical V and horizontal H polarizations, at 458 scan angle, and the mixed emissivity can thus be obtained using «(n, u)mixedpolar 5
«(n, u)V 1 «(n, u)H . 2
(3)
b. Choice of frequency We first examined the possible links between surface emissivity from available microwave window channels (from 19 to 150 GHz) and rain occurrence to select the most relevant frequencies for our purpose. This choice is illustrated using a time series of retrieved surface emissivities against rain gauge measurements. Figure 1 shows daily time series of rain gauge data and of surface emissivity at 31 GHz (from AMSU-A), 89 GHz (from AMSUB), and 150 GHz (from AMSU-B). The period is from January to December 2010, near the town of Agen, which is located in the southwestern part of France (Fig. 2). A rather high variability of the emissivity in space and time, as well as frequency, is observed, which is mainly induced by the state of the surface (vegetation, moisture, roughness, etc.). Coastal areas have a lower emissivity with a larger day-to-day variability coming from mixed land–sea pixels. At 31 GHz, which is a frequency more sensitive to surface changes, the retrieved emissivity exhibits temporal variations, mostly coincident with decreases in the 89-GHz emissivity. However, these
APRIL 2015
BIRMAN ET AL.
883
FIG. 1. Time series of daily rain rates (blue line) and daily emissivities (black line) at (top) 31 GHz from AMSU-A, (middle) 89 GHz from AMSU-B, and (bottom) 150 GHz from AMSU-B over 1 yr near Agen. In the middle panel, the black dashed line represents a dynamic mean dry emissivity.
changes are less pronounced than those noticed at 89 GHz, and are not always associated with rain occurrence. On the contrary, the emissivity at 150 GHz, which is a frequency sensitive to scattering by hydrometeors in the atmosphere and their amount at the top of the cloud cover, shows larger day-to-day variations that do not correlate only with rainy events. The use of 89 GHz is the best compromise between a good sensitivity to the surface and the amount of data available over our domain. Indeed, 89 GHz is a common frequency to AMSU-A, AMSU-B, MHS, and SSMI/S, which make it possible to have good daily coverage of our study area. Using this frequency allows us to better sample the diurnal cycle and minimize the chance of missing a rain event. It is also relevant to examine the consistency between the emissivities at 89 GHz derived from the AMSU-A sensor and those derived from the AMSU-B and SSMI/S instruments since the various instruments have been merged in order to increase the sample size within each grid cell. Figure 3 (left) shows the distribution of the
emissivities at 89 GHz from AMSU-A and -B and at 91 GHz from SSMI/S. Across our study area and throughout the month of February 2010, the average emissivity near 89 GHz is close to 0.898 (standard deviation, denoted as std hereinafter: 0.062), 0.897 (std: 0.066) and 0.883 (std: 0.071) for AMSU-A, AMSU-B, and SSMI/S, respectively. Note that the numbers of observations are different for each sensor: with a total number of observations exceeding 450 000 for the month of February 2010, 69% of the data come from AMSU-B/ MHS (on board NOAA-16, MetOp-A, NOAA-19, NOAA-18, and NOAA-17), 25% are from AMSU-A (on board NOAA-15, NOAA-16, NOAA-18, MetOp-A, Aqua, and NOAA-19), and only 6% are from SSMI/S (on board DMSP-16). The difference between AMSU-A and AMSU-B/MHS sensors comes from the difference of the satellite footprint (48 and 16 km, respectively, at nadir). The advantage of merging all these observations is to ensure a homogeneous coverage of the study area over a period of 24 h. For example, when examining the
884
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
FIG. 2. Domain of study and locations of specific areas.
average distribution of AMSU/SSMIS observation times during 24 h for the first 4 days of February 2010 (the distribution for 3 February is shown on the righthand side of Fig. 3), one can see that the observations from SSMI/S are acquired early in the morning or late evening in contrast with AMSU-A and AMSU-B, for which observations are available at all times (with few exceptions). Using frequencies near 89 GHz allows us to consider observations at 89 GHz from AMSU-A and AMSU-B and at 91 GHz from SSMI/S. This process substantially increases the amount of data available at each location for a given day, with observations spanning the entire diurnal cycle, thus minimizing the chance of missing a rain event. For other frequencies since fewer instruments are available, sampling issues could potentially degrade the quality of the rainfall retrievals.
VOLUME 54
At Agen (shown in Fig. 1, middle panel), a dynamical mean ‘‘dry soil’’ emissivity is also plotted (dashed line together with standard deviations). It is calculated using a sliding window of 10 clear-sky days identified from shortrange numerical weather forecasts (by convenience). The dry soil emissivity is generally close to 0.95 but the instantaneous effective emissivity can fall down by 0.1 during rainy days. Figure 4 (left) shows the time series of the emissivity near 89 GHz and of Analyse par Spatialisation Horaire des Précipitations (ANTILOPE) precipitation during the month of February 2010 near Agen (see Fig. 2). The emissivity of the first days is close to 0.95 and drops to values close to 0.92 on the third day when rain occurs. This effective emissivity includes effects of scattering from ice particles, as the radiative transfer model assumes a clear, nonscattering atmosphere, and the decrease in brightness temperature due to precipitation is interpreted as a signal from the surface. Effects from the surface moistening are also included in this quantity. Indeed, after a rainy event, the effective emissivity gradually increases as the surface becomes drier, and it reaches back to its clear-sky value, as shown in Fig. 4 (left) from the 5 to 10 February.
c. Precipitation data Several datasets of precipitation available over France have been used for calibration and validation of the methodology: rain gauges over France and analyses of 24-h accumulated rainfall rates from ANTILOPE and Spatialisation des Précipitations en Zone de Montagne (SPAZM). ANTILOPE rain rates are an operational rainfall product based on the combined use of radar and rain gauge measurements. This analysis was developed at Météo-France to provide rain rates over France with
FIG. 3. (left) Normalized histogram of emissivities at 89 GHz from AMSU-A and -B, and at 91 GHz from SSMI/S, over the whole domain during February 2010. (right) Daily distribution of observations at 89 GHz from AMSU-A and -B, as well as at 91 GHz from SSMI/S, for 3 Feb 2010.
APRIL 2015
BIRMAN ET AL.
885
FIG. 4. (left) Time series of daily emissivities at 89 GHz (red) and precipitation (blue) over February 2010 near Agen (see Fig. 2). (right) Scatterplot of instantaneous emissivities at 89 GHz from AMSU-A and -B, as well as at 91 GHz from SSMI/S, versus ANTILOPE precipitation analysis near Orléans from January to December 2010, for rainy days. The statistical fit illustrates the negative correlation between emissivities and precipitation.
a spatial resolution of 1 km and a time step of 1 h (Laurantin 2008). A kriging method is used to interpolate rain gauges on the 1-km analysis grid, providing large-scale precipitation. Radar measurements are related to convective cells over smaller areas. The radar data are corrected to account for ground echoes or masked areas by mountains, before converting the measured reflectivity into an effective rain rate. These two inputs are merged together, after suppressing the convective part from the rain gauge interpolation. The SPAZM analyses were originally developed to retrieve precipitation rates in mountainous areas to assess the hydrological potential of watersheds in French mountains (Gottardi et al. 2012). They result from rain gauge measurements over the French Alps, the Pyrénées, and the Massif Central available from Météo-France and the French Electricity Company (EDF), along with data from the Spanish, Italian, and Swiss meteorological services. The rain gauge data are interpolated with a method that accounts for altitude and the local climate at a resolution of 1 km, considering a statistical relationship between orography and precipitation. The SPAZM analyses are only available over an area covering the southern part of France for
daily accumulations over mountains and large neighboring areas. In addition to ground-based rainfall products, two satellite-based rainfall products are used. First are TRMM3B42-RT products, which combine daily Tropical Rainfall Measuring Mission (TRMM) and other rainfall estimates, including gauge data. TRMM is a joint U.S.–Japanese satellite mission designed to monitor tropical and subtropical precipitation. The second satellite product is the CMORPH, version 1 (V1.0), ‘‘raw’’ that combines microwave and infrared satellite data. CMORPH consists of combining microwave and infrared observations to produce global precipitation analyses. Precipitation estimates are derived from low-orbiter microwave observations and from geostationary satellite infrared observations. The geostationary data come from GOES-9, -10, and -12, as well as from Meteosat-5 and -7 satellites. Microwave rainfall estimates are derived from the TRMM Microwave Imager on board TRMM and SSM/I on board DMSP-13, -14, and -15 platforms with the Goddard profiling algorithm and from AMSR-E on board the Aqua platform and AMSU-B on board NOAA-15, -16, and -17 with the NESDIS algorithm. Table 1 summarizes the rainfall products and their uses in this study.
TABLE 1. List of rainfall products used in this study.
Product list
Used for rainfall detection?
ANTILOPE
No
SPAZM Rain gauges TRMM 3B42-RT CMORPH V1.0
No No No No
Used to quantify rain rates? Yes, a single function is derived to link emissivity changes and rain rates No No No No
Used for evaluation? Yes Yes Yes Yes Yes
886
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
3. From land emissivity to daily rainfall rate
TABLE 2. Best-fit function and its parameters for rainfall-rate estimation.
a. Estimation of precipitation rates The EMIRR algorithm developed in this study follows a two-step approach. First, a rain detection process is performed using the differences between land surface emissivity and its dry soil value, and then an accumulated rain rate is computed at 0.258 resolution using an ‘‘emissivity to rain’’ statistical fit derived from the ANTILOPE analyses and daily surface emissivities. The 0.258 resolution is a compromise chosen to reduce sample noise by obtaining a sufficient number of points inside each pixel without degrading the resolution too much. This choice was also supported by the similar resolution of several satellite precipitation products, including TRMM and CMORPH. The rain detection is performed on the basis of the strong negative correlations observed between the daily emissivity values and the occurrence of rainfall events (see Fig. 4, left). A reliable rain detection method requires the knowledge of a dry soil emissivity reference from which daily variations of the emissivity can be attributed to the occurrence of rain. A dynamical surface emissivity atlas (18 3 18) at 89 GHz obtained by screening rainy/cloudy events using short-range forecasts from a numerical weather prediction model has been produced. A sliding window of 10 days is used to calculate the mean dry soil emissivity in each grid cell and the corresponding standard deviation (see the dashed lines in the middle panel of Fig. 1 for an example near Agen) The choice of a 10-day window corresponds to the best compromise to account for temporal variations of the soil and to diminish the noise. Preliminary tests involved a monthly atlas, but the results of rain detection were improved by using a dynamical atlas. Rain detection is then performed by comparing the emissivity of a given day D with the dry soil emissivity and with the daily emissivity of days D 2 1 and D 2 2, leading to different scenarios: 1) when the emissivity decreases during three consecutive days, then a precipitation event is detected if the decrease exceeds a prescribed threshold (1.5 3 std gives the best results, where std is the standard deviation of the dry soil emissivity); 2) when the emissivity increases over three consecutive days, then there is no rain event; and 3) for other situations, the dry soil emissivity atlas is used as a reference to make the detection decision following several prescribed thresholds, which are compared to the difference between the emissivities of the two last successive days. The thresholds used are based on the standard deviation of the dry soil emissivity, which varies spatially. The choice of two or three successive days to detect the occurrence of rain is based on the fact that rainfall induces a decrease in the emissivity
VOLUME 54
RR 5 a0 1 a1 cos(a2 ) exp(a3 ) 1 a4 sin(a5 ) exp(a6 ) a0
a1
a2
a3
a4
a5
a6
770.42
2296.44
1.02
1.59
23.31
42.88
233.76
almost instantaneously, whereas on a dry day after a rain event the emissivity increases weakly and remains lower that its mean dry value. In that case only the knowledge of the fact that rain was present on the previous days allows us to correctly discriminate between rainy and dry days. The emissivity-to-rain function is a best-fit function derived to estimate the rain rates from retrieved emissivity, once rain is detected. Table 2 shows the best-fit function and its parameters. Figure 4 (right) shows the scatterplot of daily rainfall rates from an ANTILOPE analysis as a function of the emissivities at 89 GHz near the town of Orléans, located south of Paris during the year 2010. The global fit function that has been derived for the whole domain is also shown. As stated earlier, a strong negative relationship is observed between emissivity and precipitation: precipitation is associated with lower emissivities, and the heavier the precipitation, the lower the emissivity. A global fit has been chosen between emissivity and rainfall rate to simplify the design of the algorithm since individual scatterplots had very similar characteristics over France. However, the best-fit function has not been tested outside of France, where it reproduces rather well the weather regimes in temperate regions, but it may be necessary to adapt it for areas with other precipitation regimes. This statistical relationship is used to estimate daily rainfall rates from daily surface emissivities when a rain event is detected.
b. Evaluation of results The evaluation of results is done in two stages: 1) evaluation of the detection using time variations of emissivities at 89 GHz and 2) evaluation of the retrieved daily rainfall rates. The results of EMIRR in terms of precipitation occurrence and amounts are evaluated against rainfall rates from TRMM-3B42-RT, CMORPH-V1.0, rain gauges, and SPAZM analysis (which has not been used for the calibration of the algorithm) over the whole domain and we focus on particular locations near the towns of Agen, Grenoble, Tours, and Orléans (see Fig. 2b for their locations). Note that rain gauge data are indirectly used in the ANTILOPE analyses (with radar
APRIL 2015
887
BIRMAN ET AL.
TABLE 3. Scores for daily rainfall detection (1-mm threshold) with respect to ANTILOPE over the whole domain.
TABLE 4. Scores for 5-day rainfall accumulation detection (1-mm threshold) with respect to ANTILOPE over the whole domain.
Daily rates
EMIRR
SPAZM
TRMM
CMORPH
5-day accumulations
EMIRR
SPAZM
TRMM
CMORPH
POD max POD median FAR min FAR median HSS max HSS median
0.75 0.50 0.31 0.48 0.46 0.27
1.0 0.96 0.03 0.11 0.98 0.89
0.79 0.59 0.07 0.21 0.68 0.52
0.71 0.48 0.05 0.15 0.64 0.48
POD max POD median FAR min FAR median HSS max HSS median
0.97 0.84 0.04 0.15 0.51 0.33
1.0 0.99 0.009 0.04 0.95 0.98
0.97 0.85 0.005 0.06 0.77 0.71
0.92 0.71 0.004 0.06 0.81 0.40
measurements), and ANTILOPE analyses were only used to derive a statistical fit over the domain to quantify the relationship between daily rainfall rates and emissivity. Rainfall detection is evaluated on a daily basis and for 5-day accumulated rainfall. Contingency tables and objective scores [probability of detection (POD), falsealarm ratio (FAR), and Heidke skill score (HSS)] for rainfall detection (with a threshold of 1 mm between an event and a nonevent) have been computed over the whole domain for the different precipitation products available, with respect to ANTILOPE. Tables 3 and 4 summarize the results for daily rainfall and 5-day accumulations, respectively. Higher scores are systematically observed for SPAZM, which is a product based on rain
gauges, which to be is expected since ANTILOPE rainfall estimations are based on both rain gauges and radars. The scores of EMIRR, TRMM, and CMORPH are similar, except for the false alarm rate, which is higher for EMIRR. Heidke skill scores are also slightly lower for EMIRR than for TRMM and CMORPH. Results are systematically improved for 5-day accumulations. Later, we will display scores at the four evaluation sites (Agen, Grenoble, Tours, and Orléans) for daily rates and for 5-day rainfall accumulations, respectively (see Tables 7 and 8). On these particular sites the scores are computed with respect to rain gauges, which are not used to derive the best-fit function. Note, however, that rain gauges provide point measurements
FIG. 5. The 24-h precipitation amounts on 9 Jun 2010 from (a) the ANTILOPE analysis (b) EMIRR, (c) CMORPH, and (d) TRMM.
888
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 54
FIG. 6. As in Fig. 5, but for 7 Sep 2010.
whereas the simulations and other products are spatialized. The evaluation assumes that a single rain gauge is representative of the whole area of a cell (0.258 3 0.258). The scores at the four particular sites of study are rather good, especially for the POD, which exceeds or is close to 90% for 5-day accumulations (73%, 55%, 62%, and 61%, respectively, for Agen, Grenoble, Tours, and Orléans for daily detection, and 97%, 86%, 92%, and 89%, respectively, for 5-day accumulations), showing that few rainfall events are missed. However, the FAR is relatively high but the scores remain only slightly lower than those for TRMM and CMORPH (22%, 11%, 24%, and 21% for daily detection, and 54%, 46%, 53%, and 51% for 5-day accumulations). The FAR depends on location and the larger results are likely due to differences in the amplitude of the emissivity changes for a given rain event. In addition, precipitation remaining on the ground after a rainfall event may induce false detections. The HSS results remain relatively good for daily rainfall and 5-day accumulations, close to 40% (36%, 39%, 43%, and 30% for daily rainfall and 34%, 32%, 35%, and 34% for 5-day accumulations). Rainfall rates are first evaluated by comparing the retrieved precipitation from EMIRR over France with CMORPH, TRMM, and ANTILOPE products. As an
illustration of the behavior of EMIRR, Figs. 5 and 6 present two cases study: on 9 June and 7 September 2010. In both figures, the panels show the 24-h accumulated rainfall rate from ANTILOPE, EMIRR, CMORPH, and TRMM, respectively. Rather good agreement can be noticed between all rainfall products in representing the spatial distribution of the rain and its amounts. The rainfall amounts derived from surface emissivities are closer to those from the ANTILOPE simulation than to the other satellite products with more small-scale patterns. A number of false detections occur in coastal regions associated with mixed land–sea pixels in the microwave emissivities. They could be filtered out by an improved data preprocessing routine. The quality of the EMIRR retrievals seems better on 7 September than on 9 June, because of heavier rain, which is better delineated on 7 September than light rain on 9 June. To provide a quantitative comparison of these products, monthly mean rainfall histograms over a large domain in France (448–488N, 18–58E) are displayed in Fig. 7 for 2010. Despite some differences, rather good agreement can be noticed between the EMIRR rain retrievals, in terms of the monthly average (Fig. 7) and monthly accumulated precipitation (not shown), and results derived from the ANTILOPE, TRMM, and CMORPH products. These
APRIL 2015
BIRMAN ET AL.
889
FIG. 7. Monthly mean rainfall histograms for 2010 computed over a large region in France (448–488N, 18W–58E) using 24-h rain rates from TRMM (black), CMORPH (green), EMIRR (red), and ANTILOPE (gray filled).
results are very encouraging given the relative simplicity of the EMIRR algorithm. However, during winter periods the larger disparity of EMIRR with other products for monthly accumulated rain rates highlights a tendency of this algorithm to produce more intense rainy events during the cold winter period. For a quantitative assessment of EMIRR rainfall retrieval, correlation coefficients R and root-mean-square
errors have been computed over the whole domain with respect to ANTILOPE. Tables 5 and 6 summarize the results over the whole domain, for daily rain rates and 5-day accumulations, respectively. These scores confirm the conclusions on rainfall detection, showing rather good results of EMIRR with respect to other products. The EMIRR correlation coefficients are lower than those of TRMM and CMORPH, but they reach 0.6 in
890
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
TABLE 5. Correlation coefficients and root-mean-square errors for daily rainfall rates with respect to ANTILOPE over the whole domain.
VOLUME 54
TABLE 6. Correlation coefficients and root-mean-square errors for 5-day rainfall accumulations with respect to ANTILOPE over the whole domain.
Daily rates
EMIRR
SPAZM
TRMM
CMORPH
5-day accumulations
EMIRR
SPAZM
TRMM
CMORPH
R max R median RMSE min RMSE median
0.46 0.26 2.74 4.18
0.99 0.96 0.81 1.79
0.78 0.62 2.18 3.47
0.83 0.64 2.14 3.27
R max R median RMSE min RMSE median
0.65 0.33 7.17 12.41
1.0 0.98 2.07 4.73
0.84 0.71 6.02 8.86
0.91 0.66 6.17 9.53
many areas. Tables 7 and 8 display the same scores for the four evaluation sites (Agen, Grenoble, Tours, and Orléans) for daily rain rates and 5-day accumulations, respectively. The results show that the root-mean-square errors of EMIRR are only slightly lower than those of TRMM and CMORPH, despite the lower correlation coefficients. The POD, FAR, and HSS results have also been computed for different thresholds, for 5-day accumulations. Tables 9 and 10 display these scores for thresholds of 5 and 10 mm, respectively, computed over the whole domain with respect to ANTILOPE; Tables 11 and 12 show the scores for thresholds of 5 and 10 mm, respectively, for the four sites of study, with respect to rain gauges. The scores of EMIRR deteriorate with increasing rainfall thresholds, which is to be expected since the rainfall detection algorithm is rather robust but the rainfall quantification is still subject to improvements. We focus on the four areas of study (Agen, Grenoble, Tours, and Orléans) to further evaluate the results of EMIRR and their variations over the whole year. Figures 8 and 9
TABLE 7. Scores for daily rainfall detection (1-mm threshold) with respect to rain gauges for four locations near the towns of Agen, Grenoble, Tours, and Orléans. Daily rates POD
FAR
HSS
R
RMSE
Agen Grenoble Tours Orléans Agen Grenoble Tours Orléans Agen Grenoble Tours Orléans Agen Grenoble Tours Orléans Agen Grenoble Tours Orléans
EMIRR
SPAZM
TRMM
CMORPH
0.73 0.55 0.62 0.61 0.54 0.46 0.53 0.51 0.34 0.32 0.35 0.34 0.36 0.30 0.29 0.30 3.85 6.42 3.97 4.08
0.97 0.89 0.82 00.93 0.22 0.27 0.45 0.29 0.80 0.70 0.51 0.72 0.87 0.85 0.49 0.67 2.06 3.45 3.85 3.11
0.60 0.68 0.52 0.55 0.34 0.36 0.34 0.29 0.50 0.49 0.46 0.50 0.63 0.70 0.48 0.47 3.26 4.73 3.64 3.80
0.59 0.46 0.48 0.57 0.26 0.30 0.19 0.21 0.54 0.41 0.45 0.54 0.60 0.70 0.35 0.36 3.35 4.97 4.44 4.14
display the time series of 5-day accumulations simulated by EMIRR, and from rain gauges, SPAZM, TRMM, and CMORPH, over the months from January to June, as well as from July to December, respectively. The location of Grenoble is characterized by a relatively high elevation (Alpine range) whereas Tours and Orléans are situated in flat regions in the northwestern and western parts of the domain, respectively, and Agen is situated in the southwestern part and is characterized by more intense precipitation events. The EMIRR algorithm has a tendency to slightly overestimate light rain events and to underestimate large amounts. The algorithm is more successful in the plains than in the Alpine regions, because of the reduced orographic variability. In mountainous regions, the microwave pixel encompasses regions of high elevation due to the small-scale variability of the surface conditions, which are not well captured by the lower boundary condition of the radiative transfer model. Moreover, this first version of the EMIRR algorithm does not discriminate between liquid and frozen hydrometeors, TABLE 8. Scores for 5-day rainfall detection (1-mm threshold) with respect to rain gauges for four locations near the towns of Agen, Grenoble, Tours, and Orléans. 5-day accumulations POD
FAR
HSS
R
RMSE
Agen Grenoble Tours Orléans Agen Grenoble Tours Orléans Agen Grenoble Tours Orléans Agen Grenoble Tours Orléans Agen Grenoble Tours Orléans
EMIRR
SPAZM
TRMM
CMORPH
0.97 0.86 0.92 0.89 0.22 0.11 0.24 0.21 0.36 0.39 0.43 0.30 0.45 0.32 0.44 0.48 9.68 17.07 10.13 9.82
0.99 0.97 0.98 0.97 0.11 0.09 0.21 0.14 0.75 0.64 0.56 0.62 0.93 0.86 0.63 0.78 4.41 7.86 9.21 6.95
0.88 0.92 0.73 0.76 0.12 0.09 0.17 0.10 0.59 0.54 0.43 0.48 0.72 0.76 0.56 0.65 7.69 10.46 9.69 8.67
0.82 0.71 0.67 0.84 0.09 0.14 0.09 0.10 0.59 0.17 0.21 0.50 0.69 0.67 0.55 0.36 8.33 13.84 10.59 11.81
APRIL 2015
891
BIRMAN ET AL.
TABLE 9. Scores for 5-day rainfall accumulations with a 5-mm threshold with respect to ANTILOPE over the whole domain.
TABLE 11. Scores for 5-day rainfall accumulations with 10-mm threshold with respect to ANTILOPE over the whole domain.
5-day accumulations
EMIRR
SPAZM
TRMM
CMORPH
5-day accumulations
EMIRR
SPAZM
TRMM
CMORPH
POD max POD median FAR min FAR median HSS max HSS median
0.83 0.57 0.09 0.29 0.56 0.29
1.0 0.99 0.02 0.08 0.96 0.86
0.89 0.70 0.01 0.13 0.75 0.53
0.80 0.54 0.01 0.09 0.73 0.43
POD max POD median FAR min FAR median HSS max HSS median
0.63 0.31 0.13 0.51 0.51 0.22
1.0 0.99 0.05 0.15 0.93 0.85
0.86 0.60 0.02 0.20 0.71 0.50
0.78 0.42 0.02 0.14 0.71 0.41
which is crucial in mountainous areas, especially during the cold season. Figures 8 and 9 shows that the rain rates are far too large in January and December in the northern part of France, as well as in mountainous areas; this is because of snow remaining on the ground after snowfall, leading to false detections and to significantly overestimated rainfall rates. This aspect will require a specific study in order to propose a more robust version of EMIRR during the winter season.
An algorithm called the emissivity rainfall retrieval algorithm (EMIRR) devised to retrieve daily precipitation rates from microwave emissivities computed at 89 GHz has been developed using observations from the AMSU-A, AMSU-B, and SSMI/S instruments. This algorithm has been calibrated and evaluated over France over a 1-yr period spanning 2010. First, EMIRR provides a precipitation detection algorithm using daily surface emissivity time series, with a ‘‘dry soil’’ reference varying in space and time. Then, a best-fit function has been derived from ANTILOPE precipitation analyses (combining rain gauges and radar data) to estimate daily precipitation amounts from daily retrieved surface emissivities. An
evaluation of EMIRR has been undertaken using independent products: SPAZM rainfall analyses, rain gauge measurements, and CMORPH and TRMM satellite-based rainfall products during the year 2010. Rather good results for the detection of rain events using daily surface emissivities have been obtained. The probability of detection is successful in about 70% of cases, and very few rainy events are missed. The use of a dynamical dry soil emissivity atlas that varies in time has allowed us to better describe the seasonal changes in vegetation and soil conditions throughout the year. Significant false alarm ratios have been noticed, most of them resulting from snow events. The algorithm is more successful in cases of heavy rain than in cases of light precipitation, which is problematic for rainfall detection. The daily retrieved rain rates compare well to a number of independent rainfall products, but with a tendency to overestimate low amounts and to underestimate larger ones. This highlights the need to develop an appropriate bias correction procedure. Other comparisons with satellite-based products revealed the good performance of EMIRR despites its relative simplicity. This study should be considered as a demonstration of the potential use of surface emissivities at 89 GHz to provide indirect information on precipitation over land
TABLE 10. Scores for 5-day rainfall accumulations with 5-mm threshold with respect to rain gauges near the towns of Agen, Grenoble, Tours, and Orléans.
TABLE 12. Scores for 5-day rainfall accumulations with a 10-mm threshold with respect to rain gauges for the towns of Agen, Grenoble, Tours, and Orléans.
4. Conclusions
5-day accumulations POD
FAR
HSS
Agen Grenoble Tours Orléans Agen Grenoble Tours Orléans Agen Grenoble Tours Orléans
EMIRR
SPAZM
TRMM
CMORPH
0.78 0.58 0.76 0.68 0.30 0.13 0.40 0.22 0.42 0.40 0.39 0.42
0.96 0.97 0.86 0.89 0.10 0.12 0.37 0.15 0.85 0.78 0.48 0.69
0.66 0.84 0.69 0.64 0.20 0.11 0.27 0.11 0.48 0.66 0.51 0.52
0.59 0.59 0.59 0.54 0.29 0.35 0.43 0.36 0.49 0.46 0.36 0.44
5-day accumulations POD
FAR
HSS
Agen Grenoble Tours Orléans Agen Grenoble Tours Orléans Agen Grenoble Tours Orléans
EMIRR
SPAZM
TRMM
CMORPH
0.52 0.27 0.45 0.43 0.50 0.27 0.56 0.40 0.27 0.19 0.23 0.31
0.93 0.91 0.71 0.75 0.21 0.18 0.43 0.30 0.78 0.72 0.47 0.59
0.68 0.74 0.30 0.48 0.23 0.18 0.41 0.29 0.59 0.59 0.26 0.42
0.54 0.41 0.52 0.39 0.25 0.10 0.27 0.23 0.49 0.37 0.42 0.29
892
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 54
FIG. 8. Time series of 5-day precipitation accumulations from EMIRR (red), SPAZM (blue line), TRMM (black), and CMORPH (green) in areas near (top) Agen, (top middle) Grenoble, (bottom middle) Tours, and (bottom) Orléans, from January to June 2010. The rain gauge accumulations are gray filled.
surfaces. With a rather simple method, quite satisfactory results have been obtained for the detection and estimation of precipitation with regard to the high degree of complexity of such an inversion from microwave satellite radiances over continental surfaces. Nevertheless, the proposed method still has a number of weaknesses that need be corrected before the algorithm can be applied effectively over France and also over other parts of the world. A continuation of this work is already planned with several areas of possible improvements, including 1) the use of an advanced learning method (such as a neural network system or a variational method), which could lead to a better description of the nonlinear relationship between daily surface emissivities and precipitation amounts; 2) a priori knowledge of precipitation rates provided by a numerical weather prediction model,
which could benefit such a system and could also combine the 89-GHz emissivities with other lower-frequency emissivities (close to 31 GHz) from new instruments now available (such as the Advanced Technology Microwave Sounder, ATMS), to better separate atmospheric effects from those related to the surface; and 3) an extension of the learning database from ANTILOPE analysis to a suitable combination of all available precipitation products. A surface classification should be developed to associate a given surface type with a climatological value of ‘‘dry emissivity’’ varying in time over the year. Each surface type would also be associated with a regression function to derive quantitative rain rates from emissivities, which would allow the algorithm to be applied over various regions of the earth, especially where in situ data are scarce. Specific studies will also be
APRIL 2015
BIRMAN ET AL.
893
FIG. 9. As in Fig. 8, but from July to December 2010.
conducted to decorrelate the fraction of the soil moisture signal that does not come from precipitation (e.g., the diurnal cycle) by exploiting long time records of soil moisture measurements. For a reader who wishes to reproduce our methodology and who does not work within a numerical weather prediction (NWP) framework that ensures privileged access to NWP models and their functionality, we recommend the following procedure: 1) use the latest version of RTTOV made available by the Satellite Application Facility for Numerical Weather Prediction (SAF-NWP; http://nwpsaf.eu/deliverables/rtm/), 2) feed the radiative transfer model with short-range forecasts or analyses coming from ERA-Interim, for instance (surface temperature, air temperature, and moisture profiles), and 3) use the main RTTOV module to store some outputs of radiative transfer modeling, the transmission, the
downwelling, and the upwelling radiation, to calculate the emissivity. The skin temperature was obtained in our study from short-range forecasts of the surface temperature (6-h forecasts), but it is possible to use other surface temperature products such as those from the International Satellite Cloud Climatology Project (based on the use of IR data). Rainfall detection can then be performed using a monthly emissivity atlas as a dry soil reference (available online at http://nwpsaf.eu/ deliverables/rtm/emissivity/). The use of a monthly atlas slightly deteriorates the results of rainfall detection, but it avoids the use of the short-range forecasts of an NWP model. The best-fit function and its parameters are provided in Table 2, but for an application of the algorithm over regions other than France, it may be necessary to derive a new function with corresponding parameters.
894
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
Acknowledgments. The authors thank Jean Maziejewski for his help in revising the manuscript and Frederic Gottardi for providing the SPAZM data. They also acknowledge useful discussions with Yves Durand. REFERENCES Adler, R. F., and Coauthors, 2003: The version 2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–present). J. Hydrometeor., 4, 1147–1167, doi:10.1175/ 1525-7541(2003)004,1147:TVGPCP.2.0.CO;2. Bennartz, R., and G. W. Petty, 2001: The sensitivity of microwave remote sensing observations of precipitation to ice particle size distributions. J. Appl. Meteor., 40, 345–364, doi:10.1175/ 1520-0450(2001)040,0345:TSOMRS.2.0.CO;2. ——, A. Thoss, A. Dybbroe, and D. B. Michelson, 2002: Precipitation analysis using the Advanced Microwave Sounding Unit in support of nowcasting applications. Meteor. Appl., 9, 177–189, doi:10.1017/S1350482702002037. Brocca, L., T. Moramarco, F. Melone, and W. Wagner, 2013: A new method for rainfall estimation through soil moisture observations. Geophys. Res. Lett., 40, 853–858, doi:10.1002/grl.50173. ——, and Coauthors, 2014: Soil as a natural rain gauge: Estimating global rainfall from satellite soil moisture data. J. Geophys. Res., 119, 5128–5141, doi:10.1002/2014JD021489. Calvet, J.-C., J.-P. Wigneron, J. P. Walker, F. Karbou, A. Chanzy, and C. Albergel, 2011: Sensitivity of passive microwave observations to soil moisture and vegetation content: L-band to W-band. IEEE Trans. Geosci. Remote Sens., 49, 1190–1199, doi:10.1109/TGRS.2010.2050488. Choudhury, B. J., 1993: Reflectivities of selected land surface types at 19 and 37 GHz from SSM/I observations. Remote Sens. Environ., 46, 1–17, doi:10.1016/0034-4257(93)90028-V. Courtier, P., C. Freydier, J. F. Geleyn, F. Rabier, and M. Rochas, 1991: The ARPEGE project at Météo-France. Proc. ECMWF Workshop in Numerical Methods in Atmospheric Modelling, Vol. II, Reading, United Kingdom, ECMWF, 193–231. Crow, W. T., G. F. Huffman, R. Bindlish, and T. J. Jackson, 2009: Improving satellite rainfall accumulation estimates using spaceborne soil moisture retrievals. J. Hydrometeor., 10, 199– 212, doi:10.1175/2008JHM986.1. ——, M. J. van Den Berg, G. F. Huffman, and T. Pellarin, 2011: Correcting rainfall using satellite-based surface soil moisture retrievals: The Soil Moisture Analysis Rainfall Tool (SMART). Water Resour. Res., 47, W08521, doi:10.1029/ 2011WR010576. Defer, E., C. Prigent, F. Aires, J. R. Prado, C. J. Walden, O.-Z. Zanif, J.-P. Chaboureau, and J.-P. Pinty, 2008: Development of precipitation retrievals at millimeter and sub-millimeter wavelength for geostationary satellites. J. Geophys. Res., 113, D0811, doi:10.1029/2007JD008673. DiTomaso, E., F. Romano, and V. Cuomo, 2009: Rainfall estimation from satellite passive microwave observations in the range 89 GHz to 190 GHz. J. Geophys. Res., 114, D18203, doi:10.1029/ 2009JD011746. Eyre, J. R., 1991: A fast radiative transfer model for satellite sounding systems. ECMWF Tech. Memo. 176, 28 pp. [Available online at http://old.ecmwf.int/publications/library/ ecpublications/_pdf/tm/001-300/tm176.pdf.] Felde, G. W., and J. D. Pickle, 1995: Retrieval of 91 and 150 GHz Earth surface emissivities. J. Geophys. Res., 100 (D10), 20 855– 20 866, doi:10.1029/95JD02221.
VOLUME 54
Ferraro, R. R., and G. F. Marks, 1995: The development of SSM/I rain-rate retrieval algorithms using ground-based radar measurements. J. Atmos. Oceanic Technol., 12, 755–770, doi:10.1175/ 1520-0426(1995)012,0755:TDOSRR.2.0.CO;2. Funatsu, B. M., C. Claud, and J.-P. Chaboureau, 2007: Potential of Advanced Microwave Sounding Unit to identify precipitating systems and associated upper-level features in the Mediterranean region: Case studies. J. Geophys. Res., 112, D17113, doi:10.1029/ 2006JD008297. Geer, A., and F. Baordo, 2014: Improved scattering radiative transfer for frozen hydrometeors at microwave frequencies. Atmos. Meas. Tech. Discuss., 7, 1749–1805, doi:10.5194/amtd-7-1749-2014. Gottardi, F., C. Obled, J. Gailhard, and E. Paquet, 2012: Statistical reanalysis of precipitation fields based on ground network data and weather patterns: Application over French mountains. J. Hydrol., 432–433, 154–167, doi:10.1016/j.jhydrol.2012.02.014. Hilburn, K. A., and F. J. Wentz, 2008: Intercalibrated passive microwave rain products from the Unified Microwave Ocean Retrieval Algorithm (UMORA). J. Appl. Meteor. Climatol., 47, 778–794, doi:10.1175/2007JAMC1635.1. Joyce, R. J., J. E. Janowiak, P. A. Arkin, and P. Xie, 2004: CMORPH: A method that produces global precipitation estimates from passive microwave and infrared data at high spatial and temporal resolution. J. Hydrometeor., 5, 487–503, doi:10.1175/1525-7541(2004)005,0487:CAMTPG.2.0.CO;2. Karbou, F., C. Prigent, L. Eymard, and J. Pardo, 2005: Microwave land emissivity calculations using AMSU-A and AMSU-B measurements. IEEE Trans. Geosci. Remote Sens., 43, 948– 959, doi:10.1109/TGRS.2004.837503. ——, E. Gérard, and F. Rabier, 2006: Microwave land emissivity and skin temperature for AMSU-A and -B assimilation over land. Quart. J. Roy. Meteor. Soc., 132, 2333–2355, doi:10.1256/ qj.05.216. ——, F. Rabier, and C. Prigent, 2014: The assimilation of observations from the Advanced Microwave Sounding Unit over sea ice in the French global numerical weather prediction system. Mon. Wea. Rev., 142, 125–140, doi:10.1175/MWR-D-13-00025.1. Kidd, C., and V. Levizzani, 2011: Status of satellite precipitation retrievals. Hydrol. Earth Syst. Sci., 15, 1109–1116, doi:10.5194/ hess-15-1109-2011. Laurantin, O., 2008: ANTILOPE: Hourly rainfall analysis merging radar and rain gauge data. Proc. Int. Symp. on Weather Radar and Hydrology, Grenoble, France, P2-008. Matricardi, M., F. Chevallier, G. Kelly, and J.-N. Thépaut, 2004: An improved general fast radiative transfer model for the assimilation of radiance observations. Quart. J. Roy. Meteor. Soc., 130, 153–173, doi:10.1256/qj.02.181. Pellarin, T., A. Ali, F. Chopin, I. Jobard, and J.-C. Bergès, 2008: Using spaceborne surface soil moisture to constrain satellite precipitation estimates over West Africa. Geophys. Res. Lett., 35, L02813, doi:10.1029/2007GL032243. ——, S. Louvet, C. Gruhier, G. Quantin, and C. Legout, 2013: A simple and effective method for correcting soil moisture and precipitation estimates using AMSR-E measurements. Remote Sens. Environ., 136, 28–36, doi:10.1016/ j.rse.2013.04.011. Petty, G. W., and K. Li, 2013: Improved passive microwave retrievals of rain rate over land and ocean. Part II: Validation and intercomparison. J. Atmos. Oceanic Technol., 30, 2509– 2526, doi:10.1175/JTECH-D-12-00184.1. Prigent, C., W. B. Rossow, and E. Matthews, 1997: Microwave land surface emissivities estimated from SSM/I observations. J. Geophys. Res., 102, 21 867–21 890, doi:10.1029/97JD01360.
APRIL 2015
BIRMAN ET AL.
Saunders, R. W., M. Matricardi, and P. Brunel, 1999: An improved fast radiative transfer model for assimilation of satellite radiance observations. Quart. J. Roy. Meteor. Soc., 125, 1407–1425, doi:10.1256/smsqj.55614. Sorooshian, S., K. L. Hsu, X. Gao, H. V. Gupta, B. Imam, and D. Braithwaite, 2000: Evaluation of PERSIANN system satellite-based estimates of tropical rainfall. Bull. Amer. Meteor. Soc., 81, 2035–2046, doi:10.1175/1520-0477(2000)081,2035: EOPSSE.2.3.CO;2. Spencer, R. W., H. M. Goodman, and R. E. Hood, 1989: Precipitation retrieval over land and ocean with the SSM/I:
895
Identification and characteristics of the scattering signal. J. Atmos. Oceanic Technol., 6, 254–273, doi:10.1175/1520-0426(1989) 006,0254:PROLAO.2.0.CO;2. Tapiador, F. J., and Coauthors, 2012: Global precipitation measurement: Methods, datasets and applications. Atmos. Res., 104–105, 70–97, doi:10.1016/j.atmosres.2011.10.021. Turk, F. J., Z. S. Haddad, and Y. You, 2014: Principal components of multifrequency microwave land surface emissivities. Part I: Estimation under clear and precipitating conditions. J. Hydrometeor., 15, 3–19, doi:10.1175/ JHM-D-13-08.1.